Quantum circuit cutting and fusing distributed quantum computing method and system of classical computing

Abstract

The application provides a distributed quantum computing method and system for cutting and fusing classical computing of a quantum circuit, and relates to the technical field of quantum computing. The method first completes standardization preprocessing on an original quantum circuit, then constructs a double-target optimization function for fusing classical simulating to complete optimization of a circuit cutting point, and splits to obtain a sub-circuit set; the sub-circuit is intelligently classified through a multi-dimensional quantum advantage degree evaluation system, and is divided into a classical simulation sub-circuit and a quantum advantage sub-circuit; classical high-precision simulation and quantum hardware execution are respectively performed on the two types of sub-circuits, then unbiased fusion and noise correction of classical-quantum results are completed, and finally the calculation result of the original circuit is output. The application realizes low-overhead and high-fidelity distributed execution of a large-scale quantum circuit on a NISQ device, greatly reduces quantum sampling overhead, improves result fidelity, is compatible with a main flow sub-compilation framework and a quantum computing platform, and has strong engineering landing performance.

Description

Technical Field

[0001] This invention relates to the field of quantum computing technology, specifically to quantum circuit slicing technology in distributed quantum computing, particularly applicable to the splitting and hybrid execution of large-scale quantum circuits in the scenario of medium-scale noisy intermediate-scale quantum (NISQ) devices, and especially to a distributed quantum computing method and system that integrates quantum circuit slicing with classical computing. Background Technology

[0002] Quantum computing, with its properties such as quantum superposition and entanglement, offers exponential speedup advantages unmatched by classical computing in scenarios such as large number factorization, quantum simulation, and optimization problem solving. However, current quantum computing is in the NISQ (Noisy Medium-Scale Quantum) era, characterized by a limited number of physical qubits, short decoherence times, and inherent noise in gate operations, making it impossible to directly execute large-scale quantum circuits. Distributed quantum computing and quantum circuit segmentation technology have become the core means to solve this problem: large-scale quantum circuits are segmented into several smaller sub-circuits, distributed to multiple distributed quantum processing units (QPUs) for parallel execution, and then the results of the sub-circuits are merged through classical post-processing to restore the computational output of the original circuit.

[0003] Current NISQ devices are limited by chip size, noise suppression, and control complexity, resulting in a limited number of stably usable qubits per quantum processor (QPU), making it impossible to directly execute large-scale quantum circuits exceeding the capacity of a single QPU. Distributed quantum computing is the core solution to this problem. Based on the quasi-probabilistic decomposition theory of quantum channels, it divides the large-scale original circuit into multiple sub-circuits that can be executed on a single QPU. After each sub-circuit is executed on the QPU and repeatedly sampled, the results are merged through classical post-processing to recover the expected value of the target observable of the original circuit. Existing mainstream quantum circuit segmentation schemes all adopt the technical route of "full sub-circuit QPU execution + classical sampling and merging," which has the following core drawbacks: 1. Complete decoupling of cutting optimization and execution strategy: Existing cutting point selection only takes "minimizing the number of cutting points" or "minimizing the number of quantum teleportation states" as a single objective, without incorporating the classical simulability of the sub-circuit into the optimization function. This fails to maximize the benefits of reducing sampling overhead brought by classical simulation, resulting in a disconnect between the cutting scheme and the execution strategy.

[0004] 2. Lack of a systematic intelligent classification system for sub-circuits: In existing technologies, the selection of classical / quantum execution of sub-circuits relies on human experience. There is no quantifiable and automated quantum dominance evaluation system, which cannot be adapted to the automated processing of large-scale circuits, nor can the optimality of classification be guaranteed.

[0005] 3. Unresolved technical contradiction between sampling overhead and classical post-processing: The more cutting points there are, the lower the sampling overhead, but the number of channel combinations in quasi-probabilistic decomposition grows exponentially, and the computational load of classical post-processing increases explosively. Existing technologies have not established a balance constraint between the two, resulting in a decrease in overall execution efficiency in multi-cutting point scenarios.

[0006] 4. Non-targeted classical-quantum result fusion and noise correction methods: Existing technologies only support the merging of pure QPU sampling results, without utilizing the noise-free and accurate results of classical simulation to correct quantum sampling results, thus failing to improve result fidelity while reducing sampling overhead. Summary of the Invention

[0007] In view of this, in order to overcome the core defects of existing sub-circuit cutting technologies, such as the disconnect between cutting optimization and execution strategies, the lack of quantized intelligent classification standards for sub-circuit execution, the prominent contradiction between sampling overhead and classical post-processing costs, and the lack of targeted classical-quantum result fusion and noise correction mechanisms, the purpose of this invention is to provide a distributed quantum computing method and system that integrates quantum circuit cutting with classical computing. Through the full-process design of dual-objective joint cutting optimization, multi-dimensional quantum advantage intelligent classification, unbiased result fusion, and noise correction, it achieves low-overhead, high-fidelity distributed execution of large-scale quantum circuits on NISQ devices.

[0008] This invention is achieved using the following technical solution: In a first aspect, the present invention provides a distributed quantum computing method that integrates quantum circuit cutting and classical computing, comprising the following steps: S1. Preprocessing of the original quantum circuit: Input the original quantum circuit to be executed, the classical computing resource configuration, the quantum processor (QPU) hardware parameters and the confidence level of the target result, define the target observable and complete the Pauli decomposition, decompose the original quantum circuit into the basic gate set supported by the target QPU, eliminate redundant gate operations, extract the circuit gate dependency graph, and output the standardized quantum circuit, the gate dependency graph and the Pauli decomposition results of the target observable; S2. Dual-objective line cutting point optimization that integrates classical simulability: Generate compliant candidate cutting points based on gate dependency graph, define optimization variables and dual-objective optimization function, set hardware constraints, post-processing cost constraints and line dependency constraints, solve for the optimal cutting scheme, and output the optimal cutting point information, the set of sub-lines after splitting, the quasi-probability decomposition parameters of each sub-line and the channel combination weight coefficient table. S3. Intelligent classification of sub-circuits based on quantum dominance: Construct a multi-dimensional quantum dominance quantification evaluation system, calculate the comprehensive quantum dominance score of each sub-circuit, classify the sub-circuit into classical simulation sub-circuit and quantum dominance sub-circuit based on a preset classification threshold, and output the sub-circuit classification results and corresponding execution configuration parameters; S4. Classification and execution of sub-circuits: Deploy classical simulation sub-circuits to classical computing clusters to complete high-precision simulations and output the precise expected values ​​of the target observables under the corresponding quantum channels; compile quantum advantage sub-circuits to the target QPU for execution, use the precise results of classical simulation sub-circuits as prior information to optimize the sampling strategy, and output the sampling statistics of the observables under the corresponding quantum channels. S5. Unbiased fusion and noise correction of classical-quantum results: Based on the linearity of quantum channel quasi-probabilistic decomposition, the linear weighted calculation of the full channel combination is completed through the unbiased fusion formula. At the same time, the noise-free results of classical simulation are used to correct the noise of the sampling results of the quantum circuit, and the final expected value of the original line target observable is output. S6. Result Output and Verification: Perform error analysis and verification on the results, and output the final calculation results, error analysis report and sub-line execution log.

[0009] As a further aspect of the present invention, in step S2, the binary variables of the optimization variables include binary cut-point selection variables. And binary sub-circuit classical analogability determination variables ; in, Indicates the first Whether a candidate cut point is selected, the binary cut point selection variable is defined as: ; in, Indicates the first Whether a sub-circuit is a classically simulable sub-circuit, the variable for determining the classical simulability of a binary sub-circuit is defined as follows: ; The candidate cutting point is either a gate cutting candidate point at a two-bit gate or a bit cutting candidate point at a quantum bit line.

[0010] As a further aspect of the present invention, in step S2, the bi-objective optimization function is: ;in, , representing the total number of cutting points, , representing the total number of sub-circuits after cutting; , representing the number of classically simulable sub-circuits; , These are weighting coefficients, which can be adjusted according to hardware resource configuration to meet [the requirements]. .

[0011] As a further aspect of the present invention, the hardware constraints, post-processing cost constraints, and line dependency constraints set in step S2 are as follows: Hardware constraints: The number of qubits in each sub-circuit must be less than or equal to the maximum number of usable qubits in the target QPU; Post-processing cost constraint: Total computational cost of classical post-processing ≤ Maximum computational cost of classical post-processing; where the formula is: In the formula, The set of channel combinations corresponding to all cut points. Channel Combination Quasi-probability weights, This represents the classical computational cost required to process this channel combination. This represents the maximum post-processing computational load that classical computing can handle. Dependency constraints: The gate dependencies of the sub-circuits after the cut are consistent with those of the original circuits, and there is no causal reversal.

[0012] As a further aspect of the present invention, in step S3, the quantum advantage quantification evaluation system calculates the comprehensive score S of the sub-circuit from four dimensions: bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score. The value range is [0,1]. The higher the score, the more significant the advantage of classical simulation. The formula for calculating the overall score is as follows: In the formula, Score based on bit size. The score represents the percentage of non-Clifford gates. To score the degree of entanglement, To score for time advantage, These are the weight coefficients for the four dimensions, with a default value of [value]. ,satisfy .

[0013] As a further aspect of the present invention, the calculation rules for the four-dimensional scores are as follows: Bit size score ,in For the number of qubits in the sub-circuit, This represents the maximum number of bits that can be efficiently simulated using classical computation. Non-Clifford gate percentage score ,in The percentage of non-Clifford gates in a sub-circuit is given. The Gottesman-Knill theorem is satisfied only when the sub-circuit contains only Clifford gates, calculates ground state preparation, calculates basis measurements, and has no intermediate measurement feedback. To achieve classical, exact polynomial-time simulation; Entanglement score ,in The average two-part entanglement entropy of the sub-circuit. for The maximum entanglement entropy of a bit system; Time advantage score ,in This represents the high-precision simulation prediction time of the sub-circuit in classical computation. The pure Clifford circuit uses a stable sub-simulation, with a complexity of [missing information]. ,in, This represents the total number of gates; non-Clifford circuits are simulated using full amplitude, resulting in complexity... ; This indicates the total estimated execution time of the sub-line on the QPU.

[0014] As a further aspect of the present invention, in step S4, when the classical simulation sub-circuit is executed, the optimal simulation engine is selected according to the gate type characteristics of the sub-circuit: the pure Clifford circuit adopts a stable sub-simulation engine, the low-bit non-Clifford circuit adopts a full-amplitude simulation engine, and the high-bit sparse entangled circuit adopts a tensor network simulation engine; when the quantum advantage sub-circuit is executed, a dynamic sampling strategy is adopted to minimize the number of samplings while satisfying the target confidence level.

[0015] As a further aspect of the present invention, in step S5, the unbiased fusion formula is:

[0016] in: For sub-circuit Corresponding channel Quasi-probability weighting coefficients; This is a collection of classic analog sub-circuits. This is the exact expected value obtained from classical simulation; This is a collection of classic analog sub-circuits. Sub-circuit obtained from classical simulation In channel combination The precise expected value; For a set of quantum advantage sub-circuits, For sub-circuit In channel combination The expected value after noise correction; the premise of the unbiased fusion formula is that the sub-circuits are cascaded through quantum channels at the cutting points, and the output of the sub-circuit under each channel combination satisfies the tensor product structure. The correctness of the unbiased fusion formula is strictly guaranteed by the linearity of quantum operations.

[0017] As a further aspect of the present invention, in step S5, the noise correction formula is: ; in, This represents the original statistical expectation value obtained from QPU sampling. For sub-circuit The noise correction factor.

[0018] Secondly, the present invention provides a distributed quantum computing system that integrates quantum circuit cutting and classical computing, comprising an input module, a circuit cutting module, a computing task scheduling module, a distributed hybrid computing network, a result fusion and correction module, and an output module that are sequentially connected in communication. The input module is used to receive, parse, verify and distribute user input of the original quantum circuit to be executed, classical computing resource configuration, QPU hardware parameters and target result confidence level; The line segmentation module includes a line preprocessing submodule, a cutting optimization submodule, and a sub-line classification submodule. The line preprocessing submodule is used to perform standardized preprocessing of the original quantum lines, outputting standardized quantum lines, gate dependency graphs, and Pauli decomposition results of the objective observables. The cutting optimization submodule is used to construct a bi-objective optimization function that integrates classical simulability, completes line cutting point optimization and sub-line splitting, and outputs the optimal cutting scheme, sub-line set, quasi-probabilistic decomposition parameters, and channel combination weight coefficient table. The sub-line classification submodule is used to perform intelligent sub-line classification based on a multi-dimensional quantum dominance evaluation system, and outputs sub-line classification results and corresponding execution configuration parameters. The computational task scheduling module is used to receive the output of the line segmentation module, generate classical simulation task scheduling plans and quantum execution physics plans, and complete resource matching, task scheduling and data caching of sub-lines; The distributed hybrid computing network includes a classical simulation submodule and a quantum computing submodule. The classical simulation submodule is used to perform high-precision simulation of classical simulation sub-circuits and output the accurate expected value under the corresponding channel combination. The quantum computing submodule is used to perform hardware adaptation compilation, noise-aware optimization and QPU execution of quantum advantage sub-circuits and output the sampling statistics results under the corresponding channel combination. The result fusion and correction module is used to complete the unbiased fusion of classical simulation results and quantum execution results, to complete the noise correction of quantum sampling results based on classical noise-free results, and to output the final expected value of the original line target observable, the result confidence interval and the error analysis report. The output module is used to convert the final calculation results and the entire process execution data into standardized output content and distribute it.

[0019] As a further aspect of the present invention, the circuit preprocessing submodule incorporates QASM (Quantum Assembly Language), Quil, and Cirq quantum circuit format parsing plugins, which are used to decompose the non-native gates in the original circuit into the basic gate set supported by the target QPU, complete the circuit redundancy optimization, extract the circuit qubit interaction relationship and gate dependency relationship graph, and complete the Pauli term decomposition of the target observable.

[0020] As a further aspect of the present invention, the cutting optimization submodule is used to generate compliant candidate cutting points based on the gate dependency graph, construct a dual-objective optimization function with the core of minimizing the number of cutting points and maximizing the proportion of classically simulable sub-lines, embedding hardware constraints, post-processing cost constraints and line dependency constraints, and using a genetic algorithm to solve the multi-objective optimization problem to complete the sub-line splitting and quantum channel quasi-probabilistic decomposition.

[0021] As a further aspect of the present invention, the sub-circuit classification sub-module has a built-in multi-scenario weight template, which is used to calculate the bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score for each sub-circuit, calculate the comprehensive quantum dominance score based on the preset weight, and complete the determination of classical simulation sub-circuit and quantum dominance sub-circuit according to the classification threshold.

[0022] As a further embodiment of the present invention, the classical simulation submodule is deployed on a classical computer cluster and has a built-in stable sub-simulation engine, a full-amplitude simulation engine and a tensor network simulation engine. It interfaces with Qiskit Aer, Intel QuantumSimulator and NVIDIA cuQuantum quantum simulation tools to automatically select the optimal simulation engine based on the characteristics of the sub-circuit gate type and complete the parallel simulation calculation of the full channel combination.

[0023] As a further aspect of the present invention, the quantum computing submodule incorporates a mainstream quantum computing cloud platform adaptation plugin, which is used to compile the quantum advantage sub-circuit into the underlying instruction set supported by the target QPU, complete the circuit mapping and noise perception optimization, and use the accurate results of the classical simulation sub-circuit as prior information to complete the QPU execution and sampling data acquisition using a dynamic sampling strategy.

[0024] As a further aspect of the present invention, the result fusion and correction module incorporates multiple noise correction models to perform linear weighted calculations of the full channel combination based on the linearity of the quantum channel quasi-probabilistic decomposition, and uses the noise-free results of classical simulation to perform noise correction on the sampling results of the quantum circuit, thereby completing the error estimation, confidence interval calculation and result verification of the final result.

[0025] The distributed quantum computing system of this invention, which integrates quantum circuit cutting with classical computing, is compatible with mainstream quantum compilation frameworks such as Qiskit, TKET, and Cirq, and supports task submission and data retrieval from IBM Quantum, Origin Quantum Cloud Platform, and Alibaba Cloud Quantum Computing Platform.

[0026] Compared with existing technologies, the distributed quantum computing method and system that integrates quantum circuit cutting and classical computing provided by this invention has the following beneficial effects: 1. This invention incorporates the classical simulability of sub-circuits into the optimization objective of circuit cutting, constructs a dual-objective joint optimization function, realizes deep coupling between cutting point selection and sub-circuit execution strategy, solves the defect of complete decoupling between cutting optimization and execution strategy in the prior art, maximizes the benefits of reducing sampling overhead brought by classical simulation, and balances the core contradiction between sampling overhead and classical post-processing cost through multiple constraints, significantly improving the overall execution efficiency of distributed quantum computing.

[0027] 2. This invention establishes a multi-dimensional quantifiable quantum dominance intelligent classification system, realizing fully automated classification of classical analog sub-circuits and quantum dominance sub-circuits. It eliminates the dependence of existing technologies on human experience, can be adapted to the automated processing of large-scale quantum circuits, and ensures the optimality of classification through multi-dimensional evaluation, maximizing the offloading capacity of classical computing and reducing the ineffective occupation of scarce quantum computing power.

[0028] 3. This invention proposes an unbiased fusion and noise correction method for classical-quantum results. Based on the linearity of quantum operations, the unbiasedness of the results is guaranteed. At the same time, the noise-free and accurate results of classical simulation are used to correct the quantum sampling results. This not only eliminates the sampling overhead of classical simulable sub-circuits, but also significantly improves the fidelity of the final results, achieving the dual core benefits of reduced sampling overhead and improved result accuracy.

[0029] 4. The technical solution of this invention is compatible with mainstream quantum compilation frameworks and commercial quantum computing platforms. It has a mature engineering implementation path throughout the entire process, can be directly adapted to the hardware constraints of current NISQ devices, has strong engineering feasibility and scenario adaptability, and can be widely used in multiple core quantum computing application scenarios such as quantum chemical simulation, combinatorial optimization, and quantum machine learning.

[0030] These or other aspects of the invention will become more apparent from the following description of embodiments. It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the accompanying drawings used in the description of the exemplary embodiments or related technologies will be briefly introduced below. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the distributed quantum computing method that integrates quantum circuit cutting with classical computing in an embodiment of the present invention.

[0032] Figure 2This is a schematic diagram of the 4-qubit quantum circuit diagram segmentation in the distributed quantum computing method that integrates quantum circuit segmentation with classical computing in an embodiment of the present invention.

[0033] Figure 3 This is a flowchart of the distributed quantum computing method in the distributed quantum computing method that integrates quantum circuit cutting and classical computing in an embodiment of the present invention.

[0034] Figure 4 This is a structural block diagram of a distributed quantum computing system that integrates quantum circuit cutting and classical computing in an embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0036] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0037] To address the core shortcomings of existing sub-circuit splitting technologies, such as the disconnect between splitting optimization and execution strategies, the lack of quantized intelligent classification standards for sub-circuit execution, the significant contradiction between sampling overhead and classical post-processing costs, and the absence of targeted classical-quantum result fusion and noise correction mechanisms, this invention provides a distributed quantum computing method and system that integrates quantum circuit splitting with classical computation. Through a complete process design involving dual-objective joint splitting optimization, multi-dimensional quantum advantage intelligent classification, unbiased result fusion, and noise correction, it achieves low-overhead, high-fidelity distributed execution of large-scale quantum circuits on NISQ devices. It is particularly suitable for the splitting and hybrid execution of large-scale quantum circuits in medium-scale noisy quantum (NISQ) device scenarios.

[0038] See Figures 1 to 4 As shown, the present invention provides a distributed quantum computing method that integrates quantum circuit cutting and classical computing, comprising the following steps: S1. Preprocessing of the original quantum circuit: Input the original quantum circuit to be executed, the classical computing resource configuration, the quantum processor (QPU) hardware parameters and the confidence level of the target result, define the target observable and complete the Pauli decomposition, decompose the original quantum circuit into the basic gate set supported by the target QPU, eliminate redundant gate operations, extract the circuit gate dependency graph, and output the standardized quantum circuit, the gate dependency graph and the Pauli decomposition results of the target observable.

[0039] In this step, the QPU hardware parameters include maximum bit capacity, gate fidelity, readout fidelity, and gate execution time. Defining the target observable involves defining it as a linear combination of Pauli operator tensor products, determining the measurement basis vectors and measurement bit positions, generating a measurement configuration adapted to the QPU, and ensuring that the observable can be measured on the target hardware. During circuit normalization preprocessing, the original quantum circuit is decomposed into the basic gate set supported by the target QPU (single-bit gates: H / X / Y / Z / S / T gates; two-bit gates: CNOT / CZ gates). If there are other multi-bit gates, they are converted into combinations of the basic gate set to eliminate redundant gate operations. Circuit information is extracted, such as directed acyclic graphs of gate dependencies, qubit entanglement relationships, and the positions and proportions of non-Clifford gates; the normalized quantum circuit, gate dependency graph, and Pauli decomposition results of the target observable are output.

[0040] S2. Dual-objective line cutting point optimization integrating classical simulability: Based on the gate dependency graph, compliant candidate cutting points are generated, optimization variables and dual-objective optimization functions are defined, hardware constraints, post-processing cost constraints and line dependency constraints are set, the optimal cutting scheme is obtained by solving, and the optimal cutting point information, the set of sub-lines after splitting, the quasi-probabilistic decomposition parameters of each sub-line and the channel combination weight coefficient table are output.

[0041] In this step, the binary variables of the optimization variables include binary cut-point selection variables. And binary sub-circuit classical analogability determination variables ; in, Indicates the first Whether a candidate cut point is selected, the binary cut point selection variable is defined as: ; in, Indicates the first Whether a sub-circuit is a classically simulable sub-circuit, the variable for determining the classical simulability of a binary sub-circuit is defined as follows: ; The candidate cutting point is either a gate cutting candidate point at a two-bit gate or a bit cutting candidate point at a quantum bit line.

[0042] In this embodiment, the bi-objective optimization function is: ;in, , representing the total number of cutting points, , representing the total number of sub-circuits after cutting; , representing the number of classically simulable sub-circuits; , These are weighting coefficients, which can be adjusted according to hardware resource configuration to meet [the requirements]. .

[0043] In this embodiment, the hardware constraints, post-processing cost constraints, and line dependency constraints are set as follows: Hardware constraints: The number of qubits in each sub-circuit must be less than or equal to the maximum number of usable qubits in the target QPU; Post-processing cost constraint: Total computational cost of classical post-processing ≤ Maximum computational cost of classical post-processing; where, the formula is: In the formula, The set of channel combinations corresponding to all cut points. Channel Combination Quasi-probability weights, This represents the classical computational cost required to process this channel combination. This represents the maximum post-processing computational load that classical computing can handle. Dependency constraints: The gate dependencies of the sub-circuits after the cut are consistent with those of the original circuits, and there is no causal reversal.

[0044] When selecting the strategy, a genetic algorithm is used to solve the above multi-objective optimization problem. Based on the input hardware resource configuration, the optimal segmentation scheme is selected that balances the number of segmentation points, the proportion of classically simulable sub-lines, and post-processing costs. The output includes the optimal segmentation point location and type, the set of split sub-lines, the quasi-probabilistic decomposition parameters corresponding to each sub-line, and a channel combination weight coefficient table.

[0045] S3. Intelligent classification of sub-circuits based on quantum dominance: Construct a multi-dimensional quantum dominance quantification evaluation system, calculate the comprehensive quantum dominance score of each sub-circuit, classify the sub-circuit into classical analog sub-circuits and quantum dominance sub-circuits based on a preset classification threshold, and output the sub-circuit classification results and corresponding execution configuration parameters.

[0046] In this step, when establishing the quantum advantage quantification evaluation system, the quantum advantage quantification evaluation system calculates the comprehensive score S of the sub-circuit from four dimensions: bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score. The value range is [0,1]. The higher the score, the more significant the advantage of classical simulation.

[0047] The formula for calculating the overall score is as follows: In the formula, Score based on bit size. The score represents the percentage of non-Clifford gates. To score the degree of entanglement, To score for time advantage, These are the weight coefficients for the four dimensions, with a default value of [value]. ,satisfy .

[0048] In this embodiment, the calculation rules for the four dimensions of scores are as follows: Bit size score ,in For the number of qubits in the sub-circuit, The maximum number of bits that can be efficiently simulated using classical computation; the bit size score is used to quantify the impact of the number of bits in a sub-circuit on the difficulty of classical simulation.

[0049] Non-Clifford gate percentage score ,in The percentage of non-Clifford gates in a sub-circuit is given. The Gottesman-Knill theorem is satisfied only when the sub-circuit contains only Clifford gates, calculates ground state preparation, calculates basis measurements, and has no intermediate measurement feedback. It achieves polynomial-time classical exact simulation; the non-Clifford gate percentage score is used to quantify the theoretical feasibility of classical simulation.

[0050] Entanglement score ,in The average two-part entanglement entropy of the sub-circuit. for The maximum entanglement entropy of the bit system; the entanglement score is used to quantify the impact of the entanglement complexity of sub-circuits on classical simulations.

[0051] Time advantage score ,in This represents the high-precision simulation prediction time of the sub-circuit in classical computation. The pure Clifford circuit uses a stable sub-simulation, with a complexity of [missing information]. ,in, This represents the total number of gates; non-Clifford circuits are simulated using full amplitude, resulting in complexity... ; This represents the total estimated execution time of the sub-line on the QPU. The time advantage score is used to quantify the time cost comparison between classical simulation and QPU execution.

[0052] The classification rule is as follows: a configurable classification threshold can be set, which can be set to 0.6 by default. Sub-circuits that meet the threshold are classified as classical analog sub-circuits, and the rest are classified as quantum advantage sub-circuits. The output includes the sub-circuit classification results, the simulation configuration parameters of the classical analog sub-circuits, and the QPU compilation configuration parameters of the quantum advantage sub-circuits.

[0053] S4. Classification and execution of sub-circuits: Deploy classical simulation sub-circuits to classical computing clusters to complete high-precision simulations and output the precise expected values ​​of the target observables under the corresponding quantum channels; compile quantum advantage sub-circuits to the target QPU for execution, use the precise results of classical simulation sub-circuits as prior information to optimize the sampling strategy, and output the sampling statistics of the observables under the corresponding quantum channels.

[0054] In this step, when the classical simulation sub-circuit is executed, the optimal simulation engine is selected according to the gate type characteristics of the sub-circuit: the pure Clifford circuit uses a stable sub-simulation engine, the low-bit non-Clifford circuit uses a full-amplitude simulation engine, and the high-bit sparse entangled circuit uses a tensor network simulation engine; when the quantum advantage sub-circuit is executed, a dynamic sampling strategy is adopted to minimize the number of samplings while meeting the target confidence level.

[0055] When executing classical simulation sub-circuits, these sub-circuits are deployed to classical computing clusters. Using mature quantum simulation tools (Qiskit Aer, Intel Quantum Simulator, NVIDIA cuQuantum, etc.), high-precision simulations of full-amplitude / tensor networks are performed, outputting the precise expected values ​​of the target observables for each sub-circuit across all corresponding quantum channels. This process requires no resampling and eliminates statistical errors. When executing quantum advantage sub-circuits, these sub-circuits are compiled into an instruction set supported by the target QPU and deployed to the QPU for execution. Simultaneously, using the precise results of the classical simulation sub-circuit as prior information, sampling strategies are optimized, such as dynamically defining queries and sampling only in high-probability regions to reduce the number of resampling attempts. The resulting observable sampling statistics for each sub-circuit under the corresponding quantum channel are then output.

[0056] S5. Unbiased Fusion and Noise Correction of Classical-Quantum Results: Based on the linearity of the quasi-probabilistic decomposition of quantum channels, the linear weighted calculation of the full channel combination is completed through the unbiased fusion formula. At the same time, the noise-free results of classical simulation are used to correct the noise of the sampling results of the quantum circuit, and the final expected value of the original line target observable is output.

[0057] In this step, the unbiased fusion formula is:

[0058] in: For sub-circuit Corresponding channel Quasi-probability weighting coefficients; This is a collection of classic analog sub-circuits. This is the exact expected value obtained from classical simulation; This is a collection of classic analog sub-circuits. Sub-circuit obtained from classical simulation In channel combination The precise expected value; For a set of quantum advantage sub-circuits, For sub-circuit In channel combination The expected value after noise correction; the premise of the unbiased fusion formula is that the sub-circuits are cascaded through quantum channels at the cutting points, and the output of the sub-circuit under each channel combination satisfies the tensor product structure. The correctness of the unbiased fusion formula is strictly guaranteed by the linearity of quantum operations.

[0059] In step S5, the sampling results of the quantum circuit are corrected using the noise-free results of classical simulation. The noise correction formula is as follows: ; in, This represents the original statistical expectation value obtained from QPU sampling. For sub-circuit The noise correction factor. Output the final expected value of the original line target observables, the confidence interval of the result, and the error estimate.

[0060] S6. Result Output and Verification: Perform error analysis and verification on the results, and output the final calculation results, error analysis report and sub-line execution log.

[0061] In this step, during result verification, for classic simulable line segments, the results of the classic simulation of the entire line are compared with the output results of this scheme to verify whether the error is within the preset allowable range; the final output is the final expected value of the original line target observable quantity, error analysis report, and sub-line execution log.

[0062] See Figure 4 As shown, the present invention also provides a distributed quantum computing system that integrates quantum circuit cutting and classical computing, used to implement the above-mentioned distributed quantum computing method that integrates quantum circuit cutting and classical computing. The distributed quantum computing system is compatible with mainstream quantum compiler frameworks such as Qiskit, TKET, and Cirq, and includes an input module, a circuit cutting module, a computing task scheduling module, a distributed hybrid computing network, a result fusion and correction module, and an output module that are connected in sequence.

[0063] In this embodiment, the input module serves as the only standardized external input entry point for the system and is the core front-end unit for user-system interaction. It is responsible for receiving, parsing, validating, and normalizing all user inputs, and for receiving, parsing, validating, and distributing user inputs including the original quantum circuit to be executed, classical computing resource configuration, QPU hardware parameters, and target result confidence level.

[0064] The circuit segmentation module includes a circuit preprocessing submodule, a cutting optimization submodule, and a sub-circuit classification submodule. The circuit preprocessing submodule is used to perform standardized preprocessing of the original quantum circuits, outputting standardized quantum circuits, gate dependency graphs, and Pauli decomposition results of the target observables. Specifically, the execution logic of the circuit preprocessing submodule is as follows: First, the advanced quantum gates and non-native gates in the original circuits are decomposed into the basic gate set supported by the target QPU (single-qubit H / X / Y / Z / S / T gates, two-qubit CNOT / CZ gates). Circuit redundancy optimization is performed through operations such as unitary gate elimination and gate merging. Then, the qubit interaction relationships and gate dependency graphs of the circuits are extracted. Simultaneously, Pauli decomposition is performed on the target observables, outputting standardized quantum circuits, gate dependency graphs, and observable Pauli decomposition results. The module has built-in parsing plugins for mainstream quantum circuit formats such as QASM, Quil, and Cirq, and is compatible with the circuit input formats of most quantum computing platforms on the market. The processed standardized data is synchronously output to the dual-target cutting optimization module.

[0065] The input to the segmentation optimization submodule is the standardized quantum circuit and gate dependency graph output by the line preprocessing submodule, as well as the user-configured hardware constraint parameters and post-processing cost upper limit. It solves the core defect of complete decoupling between line segmentation and sub-line execution strategy in the prior art, realizes joint optimization of segmentation point selection and classical simulability, and is used to construct a bi-objective optimization function that integrates classical simulability. It completes the line segmentation point optimization and sub-line splitting, and outputs the optimal segmentation scheme, sub-line set, quasi-probabilistic decomposition parameters and channel combination weight coefficient table. The specific execution logic of the cutting optimization submodule is as follows: First, based on the gate dependency graph, compliant candidate cutting points are generated at the two-bit gates and quantum bit lines to ensure that the causal dependencies of the original lines are not destroyed after cutting. Then, a dual-objective optimization function is constructed with the core objective of "minimizing the number of cutting points and maximizing the proportion of classically simulated sub-lines". At the same time, three hard constraints are embedded: single QPU bit capacity constraint, classical post-processing computational constraint, and line dependency constraint. A genetic algorithm is used to solve the multi-objective optimization problem. Finally, based on the user's hardware resource configuration, the optimal cutting scheme that balances cutting efficiency, classical simulation proportion, and post-processing cost is selected. The sub-line splitting and the quasi-probability decomposition of the quantum channel corresponding to each cutting point are completed. The optimal cutting point position and type, the set of split sub-lines, the quasi-probability decomposition parameters corresponding to each sub-line, and the full channel combination weight coefficient table are output. The module has a built-in compliance verification unit for cutting schemes, which can automatically filter invalid schemes that violate hardware constraints. The processed results will be output to the sub-line classification and evaluation module simultaneously.

[0066] The sub-circuit classification sub-module is used to complete intelligent classification of sub-circuit based on a multi-dimensional quantum dominance evaluation system, and outputs the sub-circuit classification results and corresponding execution configuration parameters. As the core decision-making unit for realizing classical-quantum intelligent classification, the core input of the sub-circuit set and quasi-probability decomposition parameters output by the dual-objective cutting optimization module, as well as the user-configured classification threshold, weight coefficients, and hardware resource configuration, is the sub-circuit set. The core function of the module is to establish a systematic and quantifiable sub-circuit quantum dominance evaluation system, realize fully automated classification of classical analog sub-circuit and quantum dominance sub-circuit, and avoid the subjectivity and inefficiency of manual experience classification. The specific execution logic of the sub-circuit classification sub-module is as follows: First, for each sub-circuit, quantitative indicators are calculated in four dimensions: bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score. Then, the comprehensive quantum dominance score of the sub-circuit is calculated based on the weight coefficients configured by the user. Finally, the sub-circuit is judged based on the preset classification threshold—sub-circuits with a comprehensive score reaching the threshold are marked as classical simulation sub-circuits, and the rest are marked as quantum dominance sub-circuits. The sub-circuit classification results, simulation configuration parameters of classical simulation sub-circuits, and QPU compilation configuration parameters of quantum dominance sub-circuits are finally output. The module has built-in multi-scenario weight templates, which preset appropriate weight configurations for different application scenarios such as quantum chemistry, combinatorial optimization, and quantum machine learning. It also supports user-defined weights and thresholds. The classification results will be synchronized to the classical simulation execution module and the quantum execution module respectively.

[0067] In this embodiment, the computing task scheduling module is used to receive the output of the line segmentation module, provide task scheduling, data caching and resource matching capabilities for the distributed hybrid computing network, and output classical simulation task scheduling plan, quantum execution physics plan, corresponding sub-line and rule data packets, and send them to the distributed hybrid computing network.

[0068] The distributed hybrid computing network, serving as the classical-quantum hybrid execution carrier of the system, consists of a classical simulation submodule and a quantum computing submodule, deployed on classical computers and quantum computer clusters, respectively. The classical simulation submodule performs high-precision simulation of classical simulation sub-circuits, outputting accurate expected values ​​under corresponding channel combinations. As the execution unit of the classical simulation sub-circuits, it is deployed on the classical computer cluster, with its core inputs being the set of classical simulation sub-circuits, corresponding channel combination parameters, and classical computer resource configuration output by the computing task scheduling module. The core function of this module is to perform high-precision, sampling-free simulation of the classical simulation sub-circuits, outputting noise-free accurate expected value results, completely eliminating the quantum sampling overhead of the corresponding sub-circuits. The specific execution logic of the classical simulation submodule is as follows: First, the optimal simulation engine is automatically selected based on the gate type characteristics of the sub-circuit—a stable sub-simulation engine is used for pure Clifford circuits, a full-amplitude simulation engine is used for low-bit non-Clifford circuits, and a tensor network simulation engine is used for high-bit sparse entangled circuits. Simultaneously, it interfaces with mainstream quantum simulation tools such as Qiskit Aer, Intel Quantum Simulator, and NVIDIA cuQuantum. Through the multi-threaded / multi-GPU parallel capabilities of the computing cluster, it completes parallel simulation calculations for all channel combinations, traversing the expected value of the target observable of the sub-circuit under each channel combination. It also incorporates a built-in simulation result self-verification unit, ensuring the numerical accuracy of the results through repeated simulations and numerical verification. Finally, it outputs the precise expected value results of the classical simulation sub-circuit under all corresponding channel combinations. The module supports cloud-native elastic scaling, automatically scheduling computing resources according to the sub-circuit size and computational load. Simulation results are synchronously output to the result fusion and correction module.

[0069] The quantum computing submodule is used to complete the hardware adaptation and compilation, noise-aware optimization, and QPU execution of the quantum advantage sub-circuit, and outputs the sampling statistics results under the corresponding channel combination. As the hardware adaptation and execution unit for the quantum advantage sub-circuit, the core input of the quantum computing submodule is the set of quantum advantage sub-circuit output by the sub-circuit classification and evaluation module, the corresponding channel combination parameters, the target QPU hardware parameters, and the target result confidence requirement. The core function of the module is to complete the hardware adaptation and compilation, noise-aware optimization, and QPU execution of the quantum advantage sub-circuit, while simultaneously optimizing the sampling strategy based on classical prior information to reduce the required number of samples. The specific execution logic of this quantum computing submodule is as follows: First, the quantum advantage sub-circuit is compiled into the QASM underlying instruction set supported by the target QPU, completing the quantum circuit mapping and SWAP gate insertion within the node, adapting to the physical bit connectivity constraints of the QPU. At the same time, based on the gate fidelity and read fidelity parameters of the QPU, noise-aware circuit optimization is completed to reduce the impact of quantum noise on the results. Then, using the accurate results of the classical simulation sub-circuit as prior information, a dynamic sampling strategy is adopted to minimize the number of samplings while meeting the target confidence level. Finally, QPU execution, sampling data acquisition, statistical analysis, and confidence interval calculation are completed, outputting the original statistical expectation value and sampling confidence interval of the quantum advantage sub-circuit under the corresponding channel combination. The module has built-in adaptation plugins for mainstream quantum computing cloud platforms on the market, supporting task submission and data retrieval from multiple platforms such as IBM Quantum, Origin Quantum Cloud Platform, and Alibaba Cloud Quantum Computing Platform. The sampling statistical results will be synchronously output to the result fusion and correction module.

[0070] The result fusion and correction module is used to perform unbiased fusion of classical simulation results and quantum execution results, perform noise correction on quantum sampling results based on classical noise-free results, and output the final expected value, confidence interval, and error analysis report of the original line target observables. As the core data processing and result output unit of the system, the core inputs of the result fusion and correction module are the precise expected value results output by the classical simulation submodule, the sampling statistics results output by the quantum computing submodule, the channel combination weight coefficient table output by the dual-objective segmentation optimization module, and the QPU noise parameters. The core function of the module is to perform unbiased fusion of classical simulation results and quantum execution results, and simultaneously perform noise correction based on classical prior information, outputting the final result and error analysis of the original line target observables. The specific execution logic is as follows: First, based on the linearity of the quantum channel quasi-probabilistic decomposition, the linear weighting of the full channel combination is performed through the unbiased fusion core formula. The module performs calculations and simultaneously uses the noise-free results of classical simulations to perform first-order noise correction on the sampling results of the quantum sub-circuit, eliminating the result bias caused by quantum gate noise and readout noise. Then, it completes the error estimation and confidence interval calculation of the final result. At the same time, it performs result verification through the full-circuit simulation results of classically simulable circuit segments. Finally, it outputs the final expected value of the original circuit target observable, the result confidence interval, and the error analysis report. The module has built-in multiple noise correction models, which can automatically select the optimal correction strategy according to the scale and simulability of the sub-circuit. It also supports user-defined error tolerance and verification rules. The final result will be synchronously output to the storage and scheduling module.

[0071] The output module is used to convert the final calculation results and the entire process execution data into standardized output content and distribute it. As the only external standardized output outlet of this system, the output module is the unit for encapsulating, displaying and distributing the system's calculation results, and is responsible for converting the system's final calculation results and the entire process execution data into standardized output content that users can read and reuse.

[0072] This embodiment of the distributed quantum computing system, which integrates quantum circuit segmentation with classical computation, starts with an input general quantum circuit. The circuit segmentation module preprocesses the circuit, splitting it into a pure Clifford classical simulable sub-circuit and a non-Clifford quantum execution core circuit without cross-node gates. Subsequently, the computational task scheduling module performs resource matching, task scheduling, and full data caching for the sub-circuits, sending both types of sub-circuits into a distributed hybrid computing network. The classical simulation unit performs classical offloading and Pauli propagation rule pre-computation for the pure Clifford circuit, while the quantum computing unit performs distributed quantum execution of the core circuit and collects and measures the original results. Finally, the result fusion and correction module performs global Pauli Frame unified correction, classical-quantum result fusion, and fidelity verification, outputting a final computational result completely equivalent to the original circuit.

[0073] In this embodiment, the circuit preprocessing submodule has built-in QASM, Quil, and Cirq quantum circuit format parsing plugins, which are used to decompose the non-native gates in the original circuit into the basic gate set supported by the target QPU, complete the circuit redundancy optimization, extract the circuit qubit interaction relationship and gate dependency relationship graph, and complete the Pauli term decomposition of the target observable.

[0074] The cutting optimization submodule is used to generate compliant candidate cutting points based on the gate dependency graph, construct a dual-objective optimization function with the core of minimizing the number of cutting points and maximizing the proportion of classically simulable sub-lines, embedding hardware constraints, post-processing cost constraints and line dependency constraints, and using a genetic algorithm to solve the multi-objective optimization problem to complete the sub-line splitting and quantum channel quasi-probabilistic decomposition.

[0075] The sub-circuit classification submodule has a built-in multi-scenario weight template, which is used to calculate the bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score for each sub-circuit. Based on the preset weights, it calculates the comprehensive quantum dominance score and completes the determination of classical simulation sub-circuits and quantum dominance sub-circuits according to the classification threshold.

[0076] For example, a 4-qubit fully-participating hybrid quantum circuit is used to completely reproduce the entire execution logic of this invention; it includes both pure Clifford gates (which can be efficiently simulated classically) and non-Clifford gates (for scenarios where quantum execution has advantages), clearly and rigorously demonstrating the core technical effects of this invention. All processes, parameters, and data can be reproduced 1:1. The specific implementation steps are as follows: Step 1: Preprocessing of raw quantum circuits.

[0077] like Figure 2 As shown, the input is the original quantum circuit. The circuit execution timing is from left to right, and the initial state of the qubits is... Observable definition: The observable measure of achieving the goal. Pauli decomposition confirmed that it contains only one independent Pauli term with no overlapping terms. The original circuit is decomposed into the basic gate set (H, CNOT, T) supported by the target QPU. Redundancy is checked for monotonically redundant operations, and a standardized quantum circuit is output. The execution timing of the circuit gates is extracted, and a gate dependency graph is constructed to clarify the interaction relationship of each qubit.

[0078] Step 2: Optimize line cutting points.

[0079] Candidate cut point generation: Based on the gate dependency graph, two compliant candidate cut points are generated in time sequence. Both are time sequence cut points that vertically penetrate all four qubits, ensuring that the causal dependencies of the circuit are not destroyed after the cut. Candidate point 1: located after CNOT(q1,q2) gate and before T(q2) gate; Candidate point 2: located after T(q2) gate and before the second CNOT(q1,q2) gate.

[0080] Dual-objective optimization solution: The dual objective functions are to minimize the number of cutting points and maximize the proportion of classical simulable sub-circuits. The single QPU bit capacity constraint and the classical post-processing cost constraint are embedded, and the genetic algorithm is used for optimization. The optimal segmentation scheme was determined: Candidate point 1 was ultimately selected as the sole segmentation point, splitting the original 4-bit line into two independent sub-lines (front-end line A and back-end line B), resulting in a total of 3 quantum channel combinations. 1 =3, which meets the upper limit constraint of classical post-processing computation.

[0081] Step 3: Intelligent classification of sub-circuits based on quantum dominance.

[0082] For the two sub-circuits after the segmentation, the four-dimensional equal-weighted quantum dominance evaluation system of this invention is adopted. (weight) The evaluation and classification results show that circuit A is a classical analog sub-circuit; circuit B is a quantum advantage sub-circuit.

[0083] Step 4: Sub-line classification and execution.

[0084] Sub-line A was deployed to a classical computing environment and, based on the Qiskit Aer stable sub-simulation engine, parallel high-precision simulations of three channel combinations were completed, outputting the precise expected value of the target observables under each channel combination; there was no duplicate sampling and no statistical error throughout the process.

[0085] Sub-line B is deployed to a quantum computing cloud platform to complete the noise-aware line optimization. Using the classical simulation results of sub-line A as prior information, a dynamic sampling strategy is adopted to sample only the output interval with the top 95% contribution of the target observable. Each channel combination is sampled 1000 times, and the total number of sampling times is 3000 times. The statistical expectation value and confidence interval of the observable under each channel combination are output.

[0086] Step 5: Unbiased fusion of classical and quantum results and noise correction.

[0087] Based on the linearity of the quasi-probabilistic decomposition of quantum channels, a linear weighted calculation of the entire channel combination is completed through a core fusion formula. Using the classical full-amplitude simulation ideal value of sub-line B as a benchmark, sampling results are corrected to eliminate the influence of quantum gate noise and readout noise. The final output is the expected value of the target observable.

[0088] Step 6: Result verification and output.

[0089] The final output is compared with the theoretical precision value of 0.5, and it meets the preset error tolerance range; the final expected value, error analysis report, and full process execution log are output.

[0090] This embodiment fully reproduces the entire process technical solution of the present invention. By intelligently classifying sub-circuits, the sampling overhead of classic analog sub-circuits is completely eliminated. While reducing the number of QPU samplings by 2 / 3, noise correction is completed using classic noise-free and accurate results, and the fidelity of the results is greatly improved. It perfectly achieves the dual core benefits of reducing sampling overhead and improving result accuracy, and verifies the feasibility and advancement of the technical solution of the present invention.

[0091] In this embodiment, the classical simulation submodule is deployed on a classical computer cluster and has a built-in stable sub-simulation engine, a full-amplitude simulation engine, and a tensor network simulation engine. It interfaces with Qiskit Aer, Intel Quantum Simulator, and NVIDIA cuQuantum quantum simulation tools to automatically select the optimal simulation engine based on the characteristics of the sub-circuit gate type and complete the parallel simulation calculation of the full channel combination.

[0092] In this embodiment, the quantum computing submodule incorporates a mainstream quantum computing cloud platform adaptation plugin. This plugin compiles the quantum advantage sub-circuit into a low-level instruction set supported by the target QPU, completing circuit mapping and noise-aware optimization. Using the precise results of the classical simulation sub-circuit as prior information, a dynamic sampling strategy is employed to complete QPU execution and sampling data acquisition. The result fusion and correction module incorporates multiple noise correction models. Based on the linearity of the quasi-probabilistic decomposition of quantum channels, it performs linear weighted calculations of the full channel combination. Noise correction is applied to the sampling results of the quantum sub-circuit using the noise-free results of classical simulation, completing error estimation, confidence interval calculation, and result verification for the final result.

[0093] The distributed quantum computing system of this invention, which integrates quantum circuit cutting with classical computing, is compatible with mainstream quantum compilation frameworks such as Qiskit, TKET, and Cirq, and supports task submission and data retrieval from IBM Quantum, Origin Quantum Cloud Platform, and Alibaba Cloud Quantum Computing Platform.

[0094] The distributed quantum computing method and system of quantum circuit cutting and classical computing of the present invention significantly reduces quantum sampling overhead and the ineffective occupation of scarce quantum computing power by incorporating the classical simulability of sub-circuits into the cutting optimization objective and cooperating with an automated intelligent classification mechanism. At the same time, it uses the noise-free results of classical simulation to complete the deviation correction of quantum sampling data, achieving the dual benefits of reduced sampling overhead and improved result fidelity. It solves the core contradiction between sampling overhead and classical post-processing cost in the prior art. The solution is compatible with mainstream quantum compilation frameworks and commercial quantum computing platforms and has strong engineering applicability.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distributed quantum computing method that integrates quantum circuitry with classical computing, characterized in that, Includes the following steps: S1. Preprocessing of the original quantum circuit: Input the original quantum circuit to be executed, the classical computing resource configuration, the quantum processor (QPU) hardware parameters and the confidence level of the target result, define the target observable and complete the Pauli decomposition, decompose the original quantum circuit into the basic gate set supported by the target QPU, eliminate redundant gate operations, extract the circuit gate dependency graph, and output the standardized quantum circuit, the gate dependency graph and the Pauli decomposition results of the target observable; S2. Dual-objective line cutting point optimization that integrates classical simulability: Generate compliant candidate cutting points based on gate dependency graph, define optimization variables and dual-objective optimization function, set hardware constraints, post-processing cost constraints and line dependency constraints, solve for the optimal cutting scheme, and output the optimal cutting point information, the set of sub-lines after splitting, the quasi-probability decomposition parameters of each sub-line and the channel combination weight coefficient table. S3. Intelligent classification of sub-circuits based on quantum dominance: Construct a multi-dimensional quantum dominance quantification evaluation system, calculate the comprehensive quantum dominance score of each sub-circuit, classify the sub-circuit into classical simulation sub-circuit and quantum dominance sub-circuit based on a preset classification threshold, and output the sub-circuit classification results and corresponding execution configuration parameters; S4. Classification and execution of sub-circuits: Deploy classical simulation sub-circuits to classical computing clusters to complete high-precision simulations and output the precise expected values ​​of the target observables under the corresponding quantum channels; compile quantum advantage sub-circuits to the target QPU for execution, use the precise results of classical simulation sub-circuits as prior information to optimize the sampling strategy, and output the sampling statistics of the observables under the corresponding quantum channels. S5. Unbiased fusion and noise correction of classical-quantum results: Based on the linearity of quantum channel quasi-probabilistic decomposition, the linear weighted calculation of the full channel combination is completed through the unbiased fusion formula. At the same time, the noise-free results of classical simulation are used to correct the noise of the sampling results of the quantum circuit, and the final expected value of the original line target observable is output. S6. Result Output and Verification: Perform error analysis and verification on the results, and output the final calculation results, error analysis report and sub-line execution log.

2. The distributed quantum computing method that integrates quantum circuit segmentation and classical computing as described in claim 1, characterized in that, In step S2, the binary variables of the optimization variables include binary cut-point selection variables. And binary sub-circuit classical analogability determination variables ; in, Indicates the first Whether a candidate cut point is selected, the binary cut point selection variable is defined as: ； in, Indicates the first Whether a sub-circuit is a classically simulable sub-circuit, the variable for determining the classical simulability of a binary sub-circuit is defined as follows: ； The candidate cutting point is either a gate cutting candidate point at a two-bit gate or a bit cutting candidate point at a quantum bit line.

3. The distributed quantum computing method that integrates quantum circuit cutting and classical computing as described in claim 2, characterized in that, In step S2, the bi-objective optimization function is: ;in, , representing the total number of cutting points, , representing the total number of sub-circuits after cutting; , representing the number of classically simulable sub-circuits; , These are weighting coefficients, which can be adjusted according to hardware resource configuration to meet [the requirements]. .

4. The distributed quantum computing method that integrates quantum circuit segmentation and classical computing as described in claim 1, characterized in that, In step S2, the hardware constraints, post-processing cost constraints, and circuit dependency constraints are set as follows: Hardware constraints: The number of qubits in each sub-circuit must be less than or equal to the maximum number of usable qubits in the target QPU; Post-processing cost constraint: Total computational cost of classical post-processing ≤ Maximum computational cost of classical post-processing; Dependency constraints: The gate dependencies of the sub-circuits after the cut are consistent with those of the original circuits, and there is no causal reversal.

5. The distributed quantum computing method that integrates quantum circuit segmentation and classical computing as described in claim 1, characterized in that, In step S3, the quantum advantage quantification evaluation system calculates the comprehensive score S of the sub-circuit from four dimensions: bit size score, non-Clifford gate ratio score, entanglement score, and time advantage score. The value range is [0,1]. The higher the score, the more significant the advantage of classical simulation. The formula for calculating the overall score is as follows: In the formula, Score based on bit size. The score represents the percentage of non-Clifford gates. To score the degree of entanglement, To score for time advantage, These are the weight coefficients for the four dimensions, with a default value of [value]. ,satisfy .

6. The distributed quantum computing method that integrates quantum circuit cutting and classical computing as described in claim 5, characterized in that, The specific rules for calculating the scores in the four dimensions are as follows: Bit size score ,in For the number of qubits in the sub-circuit, This represents the maximum number of bits that can be efficiently simulated using classical computation. Non-Clifford gate percentage score ,in The percentage of non-Clifford gates in a sub-circuit is given. The Gottesman-Knill theorem is satisfied only when the sub-circuit contains only Clifford gates, calculates ground state preparation, calculates basis measurements, and has no intermediate measurement feedback. To achieve classical, exact polynomial-time simulation; Entanglement score ,in The average two-part entanglement entropy of the sub-circuit. for The maximum entanglement entropy of a bit system; Time advantage score ,in This represents the high-precision simulation prediction time of the sub-circuit in classical computation. The pure Clifford circuit uses a stable sub-simulation, with a complexity of [missing information]. ,in, This represents the total number of gates; non-Clifford circuits are simulated using full amplitude, resulting in complexity... ; This indicates the total estimated execution time of the sub-line on the QPU.

7. The distributed quantum computing method for merging quantum circuitry with classical computing as described in claim 6, characterized in that, In step S5, the unbiased fusion formula is: in: For sub-circuit Corresponding channel Quasi-probability weighting coefficients; This is a collection of classic analog sub-circuits. This is the exact expected value obtained from classical simulation; This is a collection of classic analog sub-circuits. Sub-circuit obtained from classical simulation In channel combination The precise expected value; For a set of quantum advantage sub-circuits, For sub-circuit In channel combination The expected value after noise correction; the premise of the unbiased fusion formula is that the sub-circuits are cascaded through quantum channels at the cutting points, and the output of the sub-circuit under each channel combination satisfies the tensor product structure.

8. The distributed quantum computing method for merging quantum circuit cutting and classical computing as described in claim 7, characterized in that, In step S5, the noise correction formula is: ； in, This represents the original statistical expectation value obtained from QPU sampling. For sub-circuit The noise correction factor.

9. A distributed quantum computing system that integrates quantum circuitry with classical computing, characterized in that, A distributed quantum computing method for performing quantum circuit cutting and classical computing as described in any one of claims 1-8, the distributed quantum computing system comprising an input module, a circuit cutting module, a computing task scheduling module, a distributed hybrid computing network, a result fusion and correction module, and an output module that are sequentially connected in communication. The input module is used to receive, parse, verify and distribute user input of the original quantum circuit to be executed, classical computing resource configuration, QPU hardware parameters and target result confidence level; The line segmentation module includes a line preprocessing submodule, a segmentation optimization submodule, and a sub-line classification submodule; The circuit preprocessing submodule is used to complete the standardization preprocessing of the original quantum circuit, and outputs the standardized quantum circuit, gate dependency graph, and Pauli decomposition results of the target observable; the cutting optimization submodule is used to construct a bi-objective optimization function that integrates classical simulability, completes the optimization of circuit cutting points and sub-circuit splitting, and outputs the optimal cutting scheme, sub-circuit set, quasi-probabilistic decomposition parameters, and channel combination weight coefficient table; the sub-circuit classification submodule is used to complete the intelligent classification of sub-circuit based on the multi-dimensional quantum dominance evaluation system, and outputs the sub-circuit classification results and corresponding execution configuration parameters. The computational task scheduling module is used to receive the output of the line segmentation module, generate classical simulation task scheduling plans and quantum execution physics plans, and complete resource matching, task scheduling and data caching of sub-lines; The distributed hybrid computing network includes a classical simulation submodule and a quantum computing submodule. The classical simulation submodule is used to perform high-precision simulation of classical simulation sub-circuits and output the accurate expected value under the corresponding channel combination. The quantum computing submodule is used to perform hardware adaptation compilation, noise-aware optimization and QPU execution of quantum advantage sub-circuits and output the sampling statistics results under the corresponding channel combination. The result fusion and correction module is used to complete the unbiased fusion of classical simulation results and quantum execution results, to complete the noise correction of quantum sampling results based on classical noise-free results, and to output the final expected value of the original line target observable, the result confidence interval and the error analysis report. The output module is used to convert the final calculation results and the entire process execution data into standardized output content and distribute it.

10. The distributed quantum computing system integrating quantum circuit slicing and classical computing as described in claim 9, characterized in that, The circuit preprocessing submodule has built-in QASM, Quil, and Cirq quantum circuit format parsing plugins, which are used to decompose the non-native gates in the original circuit into the basic gate set supported by the target QPU, complete the circuit redundancy optimization, extract the circuit qubit interaction relationship and gate dependency relationship graph, and complete the Pauli term decomposition of the target observable.

Images