Parameter optimization method and device of variational quantum feature solver, equipment and product

Abstract

The application discloses a parameter optimization method and device of a variational quantum feature solver, equipment and products, relates to the technical field of quantum computing, and comprises the following steps: acquiring molecular interaction information of a target molecular system, and constructing a parameterized quantum circuit according to the molecular interaction information; constructing a noise model according to noise characteristic parameters; acquiring current circuit parameters and initial circuit parameters of the parameterized quantum circuit; correcting the energy expectation value of the parameterized quantum circuit through the noise model to obtain a corrected energy expectation value; updating the current circuit parameters according to the corrected energy expectation value, and obtaining optimized preset circuit parameters when the updated current circuit parameters meet a preset convergence condition. The application corrects the energy expectation value by introducing the noise model, and improves the parameter optimization convergence stability of the variational quantum feature solver in a neutral atomic quantum computing platform noise environment.

Description

Technical Field

[0001] This application relates to the field of quantum computing technology, and in particular to a parameter optimization method, apparatus, device and product for a variable quantum characteristic solver. Background Technology

[0002] In variational quantum characteristic solvers, parameter optimization of parameterized quantum circuits is a key factor determining the algorithm's convergence speed and solution accuracy.

[0003] Existing parameter optimization methods typically employ classical optimizers such as Adam and L-BFGS-B. These optimizers adjust the learning rate and update direction based on gradient statistics, without considering the noise characteristics unique to quantum hardware platforms. Neutral atom quantum computing platforms experience various noise sources during operation, including laser phase noise, atomic leakage, and Rydberg state decay. This noise directly affects the fidelity of the circuit's execution results, causing energy measurements to deviate from the true values, which in turn affects the accuracy of gradient calculations, making the optimization process prone to getting trapped in local optima or diverging.

[0004] Therefore, how to improve the convergence stability of parameter optimization of the variable quantum feature solver in the noisy environment of the neutral atom quantum computing platform is an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a parameter optimization method, apparatus, device, and product for a variable quantum characteristic solver, aiming to solve the technical problem of how to improve the convergence stability of parameter optimization of the variable quantum characteristic solver in the noisy environment of a neutral atom quantum computing platform.

[0006] To achieve the above objectives, this application proposes a parameter optimization method for a variable quantum characteristic solver. The method is applied to a parameter optimization device for a variable quantum characteristic solver in a neutral atom quantum computing platform. The method comprises: Obtain molecular interaction information of the target molecular system, and construct a parameterized quantum circuit based on the molecular interaction information; Obtain noise characteristic parameters and construct a noise model based on the noise characteristic parameters; Obtain the current circuit parameters and initial circuit parameters of the parameterized quantum circuit; Based on the initial circuit parameters and the current circuit parameters, the energy expectation value of the parameterized quantum circuit is corrected using the noise model to obtain the corrected energy expectation value; The current circuit parameters are updated based on the corrected energy expectation value, and the optimized preset circuit parameters are obtained when the updated current circuit parameters meet the preset convergence condition.

[0007] In one embodiment, the step of obtaining noise feature parameters and constructing a noise model based on the noise feature parameters includes: Noise data is acquired through quantum process tomography, and noise characteristic parameters are obtained based on the noise data; A noise model is constructed based on the noise characteristic parameters. The noise model is used to describe the deviation distribution of noise on the execution result of the parameterized quantum circuit.

[0008] In one embodiment, the step of correcting the energy expectation value of the parameterized quantum circuit based on the initial circuit parameters and the current circuit parameters using the noise model to obtain the corrected energy expectation value includes: The initial energy expectation value is obtained by measuring the energy expectation value of the target molecular system through the parameterized quantum circuit. Based on the initial energy expectation value, the noise offset under the current noise conditions is calculated using the noise model; The weighting coefficient of the noise offset is determined based on the initial circuit parameters and the current circuit parameters; The initial energy expectation value is compensated and corrected based on the weighting coefficient and the noise offset to obtain the corrected energy expectation value.

[0009] In one embodiment, the step of updating the current circuit parameters according to the corrected energy expectation value, and obtaining the optimized preset circuit parameters when the updated current circuit parameters satisfy a preset convergence condition, includes: The gradient value of the parameterized quantum circuit is calculated based on the corrected energy expectation value; The current circuit parameters are updated based on the gradient value to obtain the updated circuit parameters; When the updated circuit parameters meet the preset convergence conditions, the optimized preset circuit parameters are obtained.

[0010] In one embodiment, the step of calculating the gradient value of the parameterized quantum circuit based on the corrected energy expectation value includes: Based on the corrected energy expectation value, the initial gradient value of the current circuit parameters is calculated using a random perturbation approximation algorithm; Based on the noise model, calculate the gradient deviation of the current circuit parameters under the current noise environment; The initial gradient value is corrected based on the gradient deviation to obtain the corrected gradient value.

[0011] In one embodiment, the step of updating the current circuit parameters based on the gradient value to obtain the updated circuit parameters includes: Based on the noise model, the learning rate corresponding to the current circuit parameters is determined, and the learning rate is used to control the update step size of the current circuit parameters; The first-order momentum term is calculated based on the gradient value, and the gradient value is smoothed based on the first-order momentum term to obtain the smoothed gradient value. The learning rate is dynamically adjusted based on the noise model and the number of iterations of the current circuit parameters to obtain the adjusted learning rate. The current circuit parameters are updated based on the smoothed gradient value and the adjusted learning rate to obtain the updated circuit parameters.

[0012] In one embodiment, the step of correcting the energy expectation value of the parameterized quantum circuit using the noise model includes: The initial circuit parameters are input into the noise model to obtain the initial noise offset; The current circuit parameters are input into the noise model to obtain the current noise offset; The energy expectation value is corrected based on the difference between the current noise offset and the initial noise offset to obtain the corrected energy expectation value.

[0013] Furthermore, to achieve the above objectives, this application also proposes a parameter optimization device for a variational quantum characteristic solver, the device comprising: A circuit construction module is used to acquire molecular interaction information of the target molecular system and construct a parameterized quantum circuit based on the molecular interaction information. The noise modeling module is used to acquire noise characteristic parameters and construct a noise model based on the noise characteristic parameters. The parameter acquisition module is used to acquire the current circuit parameters and initial circuit parameters of the parameterized quantum circuit. An energy correction module is used to correct the energy expectation value of the parameterized quantum circuit based on the initial circuit parameters and the current circuit parameters, using the noise model, to obtain the corrected energy expectation value. The parameter optimization module is used to update the current circuit parameters according to the corrected energy expectation value, and to obtain the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition.

[0014] In addition, to achieve the above objectives, this application also proposes a parameter optimization device for a variable quantum characteristic solver, which is applied to a neutral atom quantum computing platform; The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the parameter optimization method for the variational quantum characteristic solver as described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the parameter optimization method for the variable quantum feature solver as described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This application acquires molecular interaction information of the target molecular system and constructs a parameterized quantum circuit. It then obtains noise characteristic parameters and constructs a noise model, acquiring the current and initial circuit parameters. Based on these parameters, the energy expectation value of the circuit is corrected using the noise model to obtain a corrected energy expectation value. The current circuit parameters are then updated based on this corrected energy expectation value. When the updated parameters meet a preset convergence condition, the optimized circuit parameters are obtained. Furthermore, during the actual optimization process, the system utilizes the noise characteristic parameters to construct a noise model, performing reverse correction of the energy expectation value on the noisy hardware output during parameter optimization. Compared to existing parameter optimization methods that ignore hardware noise and directly use noisy measurement results for parameter updates, leading to convergence to parameter points deviating from the noise-free true values, this application, by introducing a noise model to correct the energy expectation value, compensates for the offset effect of hardware noise on the measurement results, improving the convergence stability of the variable quantum characteristic solver in the noisy environment of a neutral atom quantum computing platform. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the parameter optimization method for the variable quantum feature solver of this application. Figure 2 This is a flowchart illustrating the second embodiment of the parameter optimization method for the variable quantum feature solver of this application. Figure 3 This is a flowchart illustrating the third embodiment of the parameter optimization method for the variable quantum feature solver of this application. Figure 4 This is a schematic diagram of the module structure of the parameter optimization device for the variable quantum feature solver in an embodiment of this application; Figure 5 This is a schematic diagram of the hardware operating environment involved in the parameter optimization method of the variable quantum feature solver in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] In variational quantum characteristic solvers, parameter optimization of parameterized quantum circuits is a key factor determining the convergence speed and solution accuracy of the algorithm.

[0024] Existing parameter optimization methods typically employ classical optimizers such as Adam and L-BFGS-B. These optimizers adjust the learning rate and update direction based on gradient statistics, without considering the noise characteristics unique to quantum hardware platforms. Neutral atom quantum computing platforms experience various noise sources during operation, including laser phase noise, atomic leakage, and Rydberg state decay. This noise directly affects the fidelity of the circuit's execution results, causing energy measurements to deviate from the true values, which in turn affects the accuracy of gradient calculations, making the optimization process prone to getting trapped in local optima or diverging.

[0025] Therefore, how to improve the convergence stability of parameter optimization of the variable quantum feature solver in the noisy environment of the neutral atom quantum computing platform is an urgent problem to be solved.

[0026] The main solution of this embodiment is as follows: By acquiring molecular interaction information of the target molecular system and constructing a parameterized quantum circuit, noise characteristic parameters are obtained and a noise model is built. The current and initial circuit parameters are acquired. Based on the initial and current circuit parameters, the energy expectation value of the circuit is corrected using the noise model to obtain the corrected energy expectation value. Then, the current circuit parameters are updated based on the corrected energy expectation value. When the updated parameters meet a preset convergence condition, the optimized circuit parameters are obtained. Furthermore, in the actual optimization process, the system uses the noise characteristic parameters to construct a noise model and performs reverse correction of the energy expectation value of the noisy hardware output during parameter optimization. Compared to existing parameter optimization methods that ignore the influence of hardware noise and directly use noisy measurement results for parameter updates, leading to convergence to parameter points deviating from the noise-free true values, this embodiment, by introducing a noise model to correct the energy expectation value, can compensate for the offset effect of hardware noise on the measurement results, improving the convergence stability of the variable quantum feature solver in the noisy environment of a neutral atom quantum computing platform.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a parameter optimization device for a variable quantum characteristic solver, etc. The following description uses a parameter optimization device for a variable quantum characteristic solver (hereinafter referred to as the device) as an example to illustrate this embodiment and the subsequent embodiments.

[0028] Based on this, embodiments of this application provide a parameter optimization method based on a variable quantum feature solver, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the parameter optimization method for the variable quantum feature solver of this application.

[0029] In this embodiment, the method is applied to the parameter optimization device of the variable quantum feature solver in the neutral atom quantum computing platform.

[0030] It is understandable that the aforementioned neutral atom quantum computing platform can be a quantum computing hardware system based on optical tweezers arrays that capture and manipulate individual neutral atoms as qubits, wherein the two ground state energy levels of each atom qubit are respectively encoded as the |0| of the quantum state. and |1 Atoms can be transitioned from their ground state to a Rydberg state through laser excitation, and quantum gate operations can be achieved using the Rydberg interactions between atoms. The aforementioned parameter optimization device can be a classical computing device that works in conjunction with the neutral atom quantum computing platform. It is set up within the neutral atom quantum computing platform to perform classical computing tasks in the variational quantum characteristic solver algorithm, excluding the operation of quantum circuits. These tasks include, but are not limited to, initializing circuit parameters, receiving and processing energy expectations, calculating gradient vectors, controlling the optimization iteration logic, and determining convergence conditions.

[0031] The method includes steps S10 to S50: Step S10: Obtain molecular interaction information of the target molecular system, and construct a parameterized quantum circuit based on the molecular interaction information.

[0032] It should be noted that the target molecular system mentioned above can be a molecular system whose ground state energy needs to be solved using quantum computing, such as hydrogen molecules, lithium hydride molecules, or tetrahydrogen linear molecular chains. The molecular interaction information mentioned above can be physical quantity information describing the types of interactions between particles within the target molecular system, including at least single-unit interactions and two-unit interactions. The single-unit interactions can be rotation operations involving only a single atom qubit, while the two-unit interactions can be pairwise interactions involving two atom qubits. The parameterized quantum circuit mentioned above can be a type of quantum circuit containing adjustable parameters, whose output quantum state can be changed by adjusting these parameters. It is used in the variable quantum characteristic solver algorithm to prepare a trial state, the expression of which is |ψ(θ)|. =U(θ)|0 n Where U(θ) is the unitary transform corresponding to the above parameterized quantum circuit, θ is the set of the above adjustable parameters, n is the number of atomic qubits used, |0 n This assumes all atomic qubits are in their ground state in the initial state. The parameterized quantum circuit described above exists on a classical computing device as a digital circuit description, which includes the sequence of quantum operations to be performed, the target of each operation, and the timing order.

[0033] In its implementation, the aforementioned device first acquires the molecular interaction information of the target molecular system. This acquisition can be achieved by receiving the structural parameters of the target molecular system from a host computer, generating a second-order quantized Hamiltonian for the system using classical quantum chemical calculations, and then decomposing this Hamiltonian to obtain monomeric and two-body interactions. These monomeric and two-body interactions constitute the molecular interaction information. Next, the device constructs a parameterized quantum circuit based on the acquired molecular interaction information. When constructing the parameterized quantum circuit, the device determines the total number of atomic qubits and the circuit's structural framework based on the type and quantity of each term in the molecular interaction information. The construction process involves assigning a corresponding quantum operation to each interaction and arranging these operations in a preset timing sequence to form a complete parameterized circuit. Based on the number and type of monomeric interactions contained in the molecular interaction information, the device configures the same number of monomeric operation positions in the circuit and assigns these monomeric operations to the corresponding atomic qubits for execution based on their atomic index numbers.

[0034] Simultaneously, based on the pairing relationship of the two atomic qubits involved in the two-body interaction terms contained in the molecular interaction information and the sign of the coupling coefficient of the term, the device configures the corresponding two-body operations in the two-body operation layer. The device stacks the configured operations in a preset order and assigns an adjustable parameter to each operation, thereby forming a complete parameterized quantum circuit. During the construction process, the specific interaction strength value of each term in the molecular interaction information is recorded and associated with the corresponding circuit parameters. These values ​​serve as initial references or constraints for parameter updates in subsequent optimization iterations, but are not directly used as the initial values ​​of the circuit parameters. After construction, the device stores the generated parameterized quantum circuit in local memory as a description file. This description file contains the number of circuit parameters, the position index of each parameter in the circuit, the target of each quantum operation, and the temporal relationship between the operations.

[0035] Step S20: Obtain noise feature parameters and construct a noise model based on the noise feature parameters.

[0036] It should be noted that the aforementioned noise characteristic parameters can be a set of physical quantities used to quantify the various non-ideal interferences experienced by the neutral atom quantum computing platform during quantum operations. The neutral atom quantum computing platform is subject to interference from various noise sources during quantum gate operations and measurements. Common noise sources include laser phase noise, atomic leakage noise, and Rydberg state decay noise. The noise characteristic parameters are the characteristic indicators of these noise sources. The aforementioned noise model can be a mathematical model or computational function constructed based on the aforementioned noise characteristic parameters, used to estimate or predict the degree of deviation or fluctuation range of measurement results when performing parameterized quantum circuits under specific noise conditions. The aforementioned noise model can assist in correcting measurement data or guiding the selection of initial parameter values ​​during subsequent parameter optimization.

[0037] In its implementation, after constructing the parameterized quantum circuit, the aforementioned device acquires the noise characteristic parameters of the neutral atom quantum computing platform. There are several ways to acquire these noise characteristic parameters. The device can obtain them by reading various noise indicators recorded in the platform's factory calibration records. These calibration records are typically provided by the platform maintenance provider upon device delivery and contain the nominal performance parameters of each core component of the platform. Alternatively, the device can establish a communication connection with the neutral atom quantum computing platform and request various noise monitoring data from the platform's operation in real-time or at regular intervals. Furthermore, the device can actively send standard test pulse sequences to the neutral atom quantum computing platform, collect the platform's response signals, and deduce the actual noise parameter values ​​from them.

[0038] After obtaining the aforementioned noise characteristic parameters, the device constructs a noise model based on these parameters. The device extracts the intensity coefficients of each noise source from the acquired noise characteristic parameters and combines these noise sources into a comprehensive noise function according to a preset functional relationship. The input to the noise model is circuit parameters or quantum circuit execution conditions, and the output is the predicted noise impact level under those conditions. When constructing the noise model, the device uses the various noise characteristic parameters as the basic input parameters of the model and employs pre-defined mathematical expressions to superimpose or couple the effects of each noise source, thereby obtaining a computable model that can output different noise impact estimates as the input conditions change. In subsequent parameter optimization, the device can utilize the aforementioned noise model to assess the degree to which measurement results are affected by noise under the current parameter settings.

[0039] To facilitate understanding, the following examples are provided for illustration, but they do not constitute a specific limitation on this embodiment. For instance, the above device obtains the following noise characteristic parameters by reading the platform calibration record: laser phase noise intensity is 0.01 rad², atomic leakage rate is 0.5%, and Rydberg state decay rate parameter is 0.001 μs. - ¹. The aforementioned device substitutes the three parameters into a preset noise function expression to construct a comprehensive noise model. This model can accept a set of circuit parameters as input and output a predicted noise offset estimate. In subsequent optimization iterations, when the device obtains the expected energy value in a measurement, it inputs the measured value into the aforementioned noise model. The noise model calculates the predicted noise offset under the current parameters, and then subtracts the offset from the measured value to obtain the corrected energy estimate.

[0040] Step S30: Obtain the current circuit parameters and initial circuit parameters of the parameterized quantum circuit; Step S40: Based on the initial circuit parameters and the current circuit parameters, the energy expectation value of the parameterized quantum circuit is corrected using the noise model to obtain the corrected energy expectation value.

[0041] It should be explained that the aforementioned current circuit parameters can be the specific values ​​taken by each adjustable parameter in the parameterized quantum circuit in the current iteration round. At the beginning of each iteration, the device sends the current circuit parameters to the neutral atom quantum computing platform for execution of the quantum circuit. The aforementioned initial circuit parameters can be the initial values ​​assigned to each adjustable parameter in the parameterized quantum circuit by the device before performing optimization iterations. The selection of these values ​​affects the convergence speed and final result of subsequent iterations. The aforementioned energy expectation value can be the average energy output value of the Hamiltonian of the target molecular system in the quantum trial state prepared by the parameterized quantum circuit. This value is obtained by running the parameterized quantum circuit multiple times on the neutral atom quantum computing platform and statistically averaging the measurement results. The aforementioned correction can be an operation based on the aforementioned noise model to process the measured energy expectation value to eliminate or reduce the influence of hardware noise on the measurement results.

[0042] In its implementation, after constructing the noise model, the device acquires the current and initial circuit parameters of the parameterized quantum circuit. The current circuit parameters are read from memory and stored in a preset register after each iteration update, while the initial circuit parameters are obtained from the parameter records generated during the initialization phase.

[0043] Subsequently, the device first inputs the current circuit parameters into the noise model, which then calculates the expected noise offset under the current noise environment based on the values ​​of the current circuit parameters. Simultaneously, the device also inputs the initial circuit parameters into the noise model to calculate the noise offset under the initial conditions. The device compares the noise offset under the current conditions with the noise offset under the initial conditions, and the difference is taken as the incremental offset of the current circuit parameters relative to the initial circuit parameters in terms of noise influence. The device obtains the energy expectation value measurement result fed back by the neutral atom quantum computing platform after running the parameterized quantum circuit under the current circuit parameters, and then subtracts the incremental offset from this energy expectation value measurement result to obtain the corrected energy expectation value. The essence of the above correction operation is to eliminate the bias component introduced by hardware noise in the current measurement result, making the corrected energy expectation value closer to the true value under noise-free conditions.

[0044] Furthermore, it should be noted that during the above correction process, the device does not directly use the absolute value of the noise model output to correct the measured value. Instead, it uses the initial circuit parameters as a reference and performs relative correction by comparing the noise difference between the current state and the initial state. This can eliminate possible systematic biases in the noise model and make the correction results more stable and reliable.

[0045] To facilitate understanding, the following explanation uses examples, but does not impose specific limitations on this embodiment. For instance, in a hydrogen molecule simulation scenario, the parameterized quantum circuit contains one adjustable parameter, with an initial circuit parameter of 0.5 radians. Assume that after several iterations, the current circuit parameter is 1.2 radians. The device inputs 0.5 radians into the noise model, obtaining an initial noise offset of 0.02 Hartree. Then, it inputs 1.2 radians into the noise model, obtaining a current noise offset of 0.05 Hartree. The difference of 0.03 Hartree is the incremental offset relative to the initial state. Subsequently, the device obtains the energy expectation value measured at the current parameter of 1.2 radians from the neutral atom quantum computing platform, which is -1.10 Hartree. Subtracting the incremental offset of 0.03 Hartree from this measured value, the corrected energy expectation value is -1.13 Hartree.

[0046] Step S50: Update the current circuit parameters according to the corrected energy expectation value, and obtain the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition.

[0047] It should be noted that the aforementioned preset convergence conditions can be a set of pre-defined criteria for determining whether the parameter optimization process should be terminated. During the iterative optimization process of the variable quantum characteristic solver, the aforementioned device continuously adjusts the circuit parameters to gradually bring the expected energy value closer to the true ground state energy of the target molecular system. When the iteration progresses to a certain extent, the improvement in energy value from continued iteration becomes extremely limited. At this point, a set of criteria is needed to determine whether the iteration should be stopped.

[0048] The aforementioned preset convergence conditions can include various types of termination criteria, such as the expected energy value decreasing by less than a preset threshold in multiple consecutive iterations, the magnitude of the gradient vector of the current circuit parameters being less than a preset threshold, the iteration rounds reaching a preset maximum number of iterations, or a combination of the above criteria. The optimized preset circuit parameters can be the circuit parameter values ​​currently held by the device when the iteration process meets the aforementioned preset convergence conditions. The expected energy value corresponding to this set of parameter values ​​is the best estimate of the ground state energy of the target molecular system output by the variable quantum characteristic solver.

[0049] In a specific implementation, after the device obtains the corrected energy expectation value, it uses a preset optimization algorithm to calculate the adjustment amount of the circuit parameters according to the direction and magnitude of the change of the corrected energy expectation value relative to the energy expectation value after the previous round of correction, and then adds the current circuit parameters to the above adjustment amount to generate the updated circuit parameters.

[0050] After each update operation, the device saves the updated circuit parameters to memory and replaces the previous current circuit parameters as the current circuit parameters for the next iteration. Next, the device determines whether the updated circuit parameters meet preset convergence conditions. These preset convergence conditions are configured as a combination of multiple conditions, and the device checks each condition individually after each update. The device checks at least the following three conditions: the first condition is whether the rate of decrease of the corrected energy expectation value in a preset number of consecutive iterations is lower than a preset threshold; the second condition is whether the magnitude of the gradient vector of the current circuit parameters is less than a preset threshold; and the third condition is whether the current iteration round has reached a preset maximum number of iterations.

[0051] The aforementioned device checks each condition sequentially according to a preset priority order. When any one of the conditions is met, the device determines that the updated circuit parameters meet the preset convergence condition. When the updated circuit parameters meet the preset convergence condition, the device stops iterative updates and outputs the currently held updated circuit parameters as the optimized preset circuit parameters. The device can associate and store the optimized preset circuit parameters with the corresponding corrected energy expectation value, outputting it as the ground state energy calculation result of the target molecular system. When the updated circuit parameters do not meet the preset convergence condition, the device uses the updated circuit parameters as the current circuit parameters for the next iteration, re-executing the cyclic process of circuit operation, energy measurement, noise correction, and parameter updates until the preset convergence condition is met.

[0052] To facilitate understanding, the following examples are provided for illustration, but they do not impose specific limitations on this embodiment. For instance, in a hydrogen molecule simulation scenario, the preset convergence condition set for the above-mentioned device is: the decrease in the expected energy value after correction is less than 1 × 10⁻⁶ in each of the five consecutive iterations. -4 Hartree, or the number of iterations reaches 200. The device mentioned above is currently in its 45th iteration. The expected energy value after correction in this round is -1.136 Hartree, compared to -1.135 Hartree in the previous round, a decrease of 0.001 Hartree. The decreases in the previous four rounds were 0.0015, 0.0012, 0.0011, and 0.0010 Hartree, respectively. The current round marks the 5th consecutive round with a decrease of less than 1.5 × 10. - ³ Hartree, but not all of them are less than 1×10 - 4 Hartree, therefore, does not meet the convergence condition, and the above device continues to execute the next round of iteration. After several rounds, in the 152nd iteration, the corrected energy expectation value is -1.137 Hartree, with a decrease of 2×10 for each of the five consecutive rounds. -5 3×10 -5 1×10 -5 2×10 -5 1×10 -5 Hartree, all are less than 1×10 -4 Hartree, if the preset convergence condition is met, the above device stops iterating and outputs the circuit parameters of the current 152nd round as the optimized preset circuit parameters.

[0053] This embodiment acquires molecular interaction information of the target molecular system and constructs a parameterized quantum circuit. It then obtains noise characteristic parameters and constructs a noise model, acquiring the current and initial circuit parameters. Based on these parameters, the system corrects the circuit's energy expectation value using the noise model, obtaining a corrected energy expectation value. The current circuit parameters are then updated based on this corrected energy expectation value. When the updated parameters meet a preset convergence condition, the optimized circuit parameters are obtained. Furthermore, during the actual optimization process, the system utilizes the noise characteristic parameters to construct a noise model, performing reverse correction of the energy expectation value on the noisy hardware output during parameter optimization. By introducing a noise model to correct the energy expectation value, this embodiment can compensate for the influence of hardware noise on the measurement results, improving the convergence stability of the variable quantum characteristic solver in the noisy environment of a neutral atom quantum computing platform.

[0054] Based on the first embodiment of this application, a second embodiment of this application is proposed. In the second embodiment, content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the parameter optimization method for the variable quantum feature solver of this application.

[0055] In this embodiment, the step of obtaining noise feature parameters and constructing a noise model based on the noise feature parameters includes: Step S21: Collect noise data through quantum process tomography, and obtain noise characteristic parameters based on the noise data; Step S22: Construct a noise model based on the noise characteristic parameters. The noise model is used to describe the deviation distribution of noise on the execution result of the parameterized quantum circuit.

[0056] It should be noted that the aforementioned quantum process tomography can be a measurement technique used to comprehensively characterize the performance of quantum operations or quantum hardware. Quantum process tomography involves executing a series of known standard input quantum states on the quantum hardware, performing a complete set of projection measurements under a measurement basis for each input state, collecting measurement data, and then using mathematical reconstruction algorithms (such as maximum likelihood estimation or linear inversion) to reconstruct the process matrix corresponding to the quantum operation. This process matrix comprehensively describes the execution effect of the operation on the actual hardware, including its deviation from the ideal operation and the influence of noise. The aforementioned noise data can be the raw measurement data collected through quantum process tomography. This data records the statistical information of the output results of the quantum hardware under various test conditions and serves as the basis for extracting noise characteristic parameters. The aforementioned deviation distribution can be the probability distribution characteristic of the deviation of the measurement result from the noise-free ideal value under given circuit parameters, as described by the aforementioned noise model. This distribution can be used to quantitatively predict the measurement offset direction and amplitude under specific parameter settings.

[0057] In its implementation, the aforementioned device acquires noise data from a neutral atom quantum computing platform using quantum process tomography (QT) when obtaining noise characteristic parameters. The device sends a set of standard test commands to the neutral atom quantum computing platform, which include predefined test quantum state preparation sequences and test quantum gate operation sequences. The device controls the neutral atom quantum computing platform to execute these test sequences sequentially, and collects the measurement results fed back by the platform after each execution. The device performs repeated acquisitions multiple times for different input quantum states and different measurement bases, collecting complete raw quantum process tomography data, i.e., noise data.

[0058] After acquiring the aforementioned noise data, the device inputs the data into a preset quantum process reconstruction algorithm. This algorithm uses mathematical optimization to reconstruct the matrix representation of the actual quantum process from the noise data. Then, it compares this actual quantum process matrix with the ideal quantum process matrix item by item, calculates the difference between them, and decomposes the intensity coefficients of each noise source from the difference. The device outputs the extracted intensity coefficients of each noise source as noise characteristic parameters. The device then substitutes these intensity coefficients into a preset model function expression, which correlates the noise characteristic parameters with the deviation distribution of the measurement results. The device determines the parameters in the model function through mathematical fitting or parameter calibration, thereby obtaining a complete and computable noise model. This noise model can receive input circuit parameter values ​​and output an estimate of the deviation distribution of the measurement results relative to the noise-free ideal value under those parameter conditions. The deviation distribution established by the noise model includes information on both the direction of the deviation (i.e., whether the measured value is too high or too low) and the magnitude (i.e., the degree of deviation), enabling subsequent correction operations to make differentiated corrections for different parameter conditions.

[0059] To facilitate understanding, the following explanation uses examples, but does not limit the scope of this embodiment. For instance, the device described above performs quantum process tomography on a neutral atom quantum computing platform, sequentially sending 10 different test quantum state preparation instructions to the platform. Each test state is measured 100 times, resulting in 1000 sets of raw measurement data collected as noise data. The device then inputs this noise data into a quantum process reconstruction algorithm. This algorithm reconstructs the actual quantum operation process matrix from the raw data using maximum likelihood estimation. This matrix is ​​then compared with the ideal process matrix, extracting parameters such as a laser phase noise intensity of 0.012 rad², an atomic leakage rate of 0.6%, and a Rydberg state decay rate of 0.0012 μs. - ¹ Noise characteristic parameters. Subsequently, the device substitutes these three noise intensity parameters into a predefined comprehensive noise function. This function is expressed as a weighted superposition of the noise sources. After determining the weighting coefficients through fitting, a complete noise model is constructed. This noise model can predict the deviation distribution of the measurement results under the input circuit parameter values. For example, when the parameter θ is 1.2 radians, the predicted measurement value is approximately 0.03 Hartree higher than the ideal value without noise, and the standard deviation of this deviation is approximately 0.005 Hartree.

[0060] Further, in this embodiment, the step of correcting the energy expectation value of the parameterized quantum circuit based on the initial circuit parameters and the current circuit parameters using the noise model to obtain the corrected energy expectation value includes: Step S41: Measure the expected energy value of the target molecular system using the parameterized quantum circuit to obtain the initial expected energy value; Step S42: Based on the initial energy expectation value, calculate the noise offset under the current noise conditions using the noise model; Step S43: Determine the weighting coefficient of the noise offset based on the initial circuit parameters and the current circuit parameters; Step S44: Compensate and correct the initial energy expectation value according to the weighting coefficient and the noise offset to obtain the corrected energy expectation value.

[0061] It should be noted that the aforementioned initial energy expectation value can be the target molecular system energy value obtained by statistically averaging the measurement results after the parameterized quantum circuit is first run on the neutral atom quantum computing platform. This initial energy expectation value includes the bias component introduced by hardware noise and has not yet undergone noise correction processing; therefore, it cannot be directly used as the effective energy estimate for parameter updates in this iteration. The aforementioned noise offset can be an estimate of the direction and magnitude of the deviation of the measurement result from the noise-free ideal value under the current noise conditions, calculated by the aforementioned device using a noise model. This noise offset reflects the quantitative impact of hardware noise on the measurement results under the current parameter settings. The aforementioned weighting coefficient can be a value between 0 and 1, used to scale and adjust the noise offset before applying it to the initial energy expectation value to achieve compensation correction of the measured value. The value of the weighting coefficient determines the degree of correction; when the weighting coefficient is 1, it indicates that the correction is performed entirely according to the offset calculated by the noise model; when the weighting coefficient is 0, it indicates that no correction is performed on the measured value.

[0062] In its implementation, the aforementioned device sends the circuit parameters of its currently held parameterized quantum circuit to a neutral atom quantum computing platform and controls the platform to run the parameterized quantum circuit on a reconfigurable atom array. After each running cycle, the device collects the projected measurement results returned by the platform under each measurement basis. After statistical averaging of the results from multiple runs, the device obtains the initial expected energy value of the target molecular system under the current circuit parameters. The device then temporarily stores this initial expected energy value in its memory as the raw measurement data for this iteration.

[0063] Subsequently, based on the initial expected energy value, the device calculates the noise offset under the current noise conditions using a noise model. The device reads the current circuit parameter values ​​from its memory and passes these values ​​as input parameters to the noise model. The noise model then calculates the expected noise offset under the current parameter conditions based on its internally established deviation distribution function. The noise model outputs a numerical value with a positive or negative sign, where a positive sign indicates that the noise causes the measured value to be higher than the ideal value without noise, and a negative sign indicates that the noise causes the measured value to be lower than the ideal value without noise.

[0064] After acquiring the noise offset, the device determines the weighting coefficient of the noise offset based on the initial circuit parameters and the current circuit parameters. The device first calculates the degree of difference between the current circuit parameters and the initial circuit parameters, which can be obtained by calculating the vector distance between them. The device compares the calculated vector distance with a preset distance threshold and determines the value of the weighting coefficient based on the comparison result. The weighting coefficient is positively correlated with the vector distance; that is, when the deviation of the current circuit parameters from the initial circuit parameters is small, the weighting coefficient is also relatively small, and the correction range of the offset calculated by the noise model for the measured value is limited. When the deviation of the current circuit parameters from the initial circuit parameters is large, the weighting coefficient approaches 1, and the offset calculated by the noise model is fully used for the correction of the measured value. The device subtracts the product of the weighting coefficient and the noise offset from the initial energy expectation value to obtain the corrected energy expectation value.

[0065] To facilitate understanding, the following explanation uses examples, but does not impose specific limitations on this embodiment. For example, in a hydrogen molecule simulation scenario, assume the initial circuit parameter is 0.5 radians and the current circuit parameter is 1.2 radians. After running the parameterized quantum circuit on the neutral atom quantum computing platform, the device measures an initial energy expectation of -1.10 Hartree. The device inputs the current circuit parameter of 1.2 radians into the noise model, and the model outputs a current noise offset of +0.05 Hartree, indicating that the noise causes the measured value to be 0.05 Hartree higher than the ideal value. The device calculates that the absolute value of the difference between the current circuit parameter of 1.2 radians and the initial circuit parameter of 0.5 radians is 0.7 radians. This difference exceeds the preset linear threshold of 0.1 radians, therefore the device determines the weighting coefficient to be a value close to 1, such as 0.95. The above equipment performs calculations according to the correction formula: Corrected expected energy value = Initial expected energy value - Weighting coefficient × Noise offset = -1.10 - 0.95 × 0.05 = -1.1475 Hartree, which is -1.148 Hartree after taking the effective value. This correction result is used as the final energy estimate for this iteration.

[0066] Based on the first and second embodiments of this application, a third embodiment of this application is proposed. In this third embodiment, content that is the same as or similar to the first and second embodiments described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the parameter optimization method for the variable quantum feature solver of this application.

[0067] In this embodiment, the step of updating the current circuit parameters according to the corrected energy expectation value, and obtaining the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition, includes: Step S51: Calculate the gradient value of the parameterized quantum circuit based on the corrected energy expectation value; Step S52: Update the current circuit parameters according to the gradient value to obtain the updated circuit parameters; Step S53: When the updated circuit parameters meet the preset convergence conditions, the optimized preset circuit parameters are obtained.

[0068] It should be noted that the aforementioned gradient values ​​can be vectors composed of the partial derivatives of the corrected energy expectation value with respect to each adjustable parameter in the parameterized quantum circuit. These gradient values ​​characterize the rate of change of the energy expectation value along each parameter dimension at the current circuit parameter position. During the parameter optimization process of the variable quantum characteristic solver, these gradient values ​​indicate the direction and magnitude of parameter adjustment, enabling the energy expectation value to gradually approach the true ground state energy of the target molecular system. The optimized preset circuit parameters can be the optimal circuit parameter values ​​output by the device after the iterative optimization process meets the preset convergence conditions. This set of parameters ensures that the quantum trial state prepared by the parameterized quantum circuit is closest to the true ground state wavefunction of the target molecular system.

[0069] In a specific implementation, after obtaining the corrected expected energy value, the device reads the current circuit parameters from the memory and sequentially applies a small perturbation increment to each adjustable parameter to generate a set of perturbed circuit parameters. Subsequently, following the same process as obtaining the corrected expected energy value, the device obtains the corrected expected energy value corresponding to each perturbed circuit parameter. The device divides the difference between each perturbed corrected expected energy value and the original corrected expected energy value by the corresponding perturbation amount to obtain an approximate value of the partial derivative of each adjustable parameter at that point. The set of partial derivatives of all adjustable parameters constitutes the gradient value.

[0070] After obtaining the gradient values, the device acquires a preset optimization algorithm. This algorithm calculates the step size and direction for adjusting the current circuit parameters in each dimension based on the gradient values ​​corresponding to the current circuit parameters and the historical gradient information from the previous iteration, according to its internal update rules. The device then superimposes the current circuit parameters with the adjusted values ​​to generate updated circuit parameters. After each update operation, the device saves the updated circuit parameters to memory and replaces the previous current circuit parameters as the current circuit parameters for the next iteration.

[0071] Next, the device determines whether the updated circuit parameters meet the preset convergence conditions. These preset convergence conditions include multiple criteria: the decrease in the corrected energy expectation value over multiple consecutive iterations is less than a preset energy change threshold; or the magnitude of the gradient value of the current circuit parameters is less than a preset gradient threshold; or the current iteration round reaches a preset maximum number of iterations. The device checks each criterion one by one, and when any one of the criteria is met, the device determines that the updated circuit parameters meet the preset convergence conditions. When the preset convergence conditions are met, the device stops iterative updating and outputs the currently held updated circuit parameters as the optimized preset circuit parameters.

[0072] To facilitate understanding, the following explanation uses examples, but does not limit the scope of this embodiment. For instance, in a hydrogen molecule simulation scenario, the parameterized quantum circuit includes one adjustable parameter. The current circuit parameter is 1.2 radians, and the corrected expected energy value is -1.148 Hartree. The device applies a positive perturbation of 0.01 radians to this adjustable parameter, obtaining the corrected expected energy value corresponding to the perturbated circuit parameter of 1.21 radians as -1.145 Hartree. Based on this, the gradient value at this parameter is calculated as (-1.145 + 1.148) / 0.01 = 0.3 Hartree / radians. The device inputs this gradient value into a preset Adam optimizer. The optimizer, combined with historical momentum information, calculates the adjustment amount of the current parameter as -0.03 radians. Subtracting 0.03 radians from 1.2 radians yields the updated circuit parameter of 1.17 radians. The device stores this updated circuit parameter as the current circuit parameter. The aforementioned device further checks the convergence conditions: the current iteration is the 20th iteration, which has not yet reached the maximum number of iterations of 200; the gradient value of 0.3 Hartree / radians is greater than the preset gradient threshold of 0.01 Hartree / radians; the corrected energy expectation value of -1.148 Hartree decreases by 0.003 Hartree compared to -1.145 Hartree in the previous round, which is greater than the preset energy change threshold of 0.001 Hartree. Therefore, the aforementioned device determines that none of the convergence conditions are currently met and will continue to execute the next iteration.

[0073] Further, the step of calculating the gradient value of the parameterized quantum circuit based on the corrected energy expectation value includes: Step S511: Based on the corrected energy expectation value, calculate the initial gradient value of the current circuit parameters using a random perturbation approximation algorithm; Step S512: Calculate the gradient deviation of the current circuit parameters under the current noise environment based on the noise model; Step S513: Correct the initial gradient value according to the gradient deviation to obtain the corrected gradient value.

[0074] It should be noted that the aforementioned random perturbation approximation algorithm can be a type of numerical method that estimates the gradient by applying random perturbations to circuit parameters and observing changes in the objective function value, without relying on parameter shift rules. Unlike parameter shift rules, which require performing two circuit measurements for each parameter, the aforementioned random perturbation approximation algorithm applies a random perturbation vector to all circuit parameters simultaneously in each iteration. It only needs to measure the objective function values ​​before and after the perturbation to calculate an approximate estimate of the gradient, and its gradient computation complexity is independent of the number of circuit parameters. The aforementioned gradient bias can be the deviation component predicted by the noise model of the aforementioned device, caused by noise factors under the current circuit parameter conditions, that deviates the gradient estimate from the noise-free true gradient value. The aforementioned corrected gradient value can be the gradient estimate obtained by subtracting the aforementioned gradient bias from the aforementioned initial gradient value.

[0075] In its implementation, after obtaining the corrected expected energy value, the device first reads the numerical vector of the current circuit parameters from the memory. Then, it generates a random perturbation vector with the same dimension as the current circuit parameters according to a preset random distribution. Each component of this random perturbation vector is independent and follows a distribution with a mean of zero. The device superimposes the current circuit parameters with the random perturbation vector to generate a positive perturbation parameter vector, and subtracts the random perturbation vector from the current circuit parameters to generate a negative perturbation parameter vector. The device obtains the corrected expected energy value corresponding to each of the positive and negative perturbation parameter vectors, and then calculates the initial gradient value according to the following formula: the initial gradient value equals the corrected expected energy value under positive perturbation minus the corrected expected energy value under negative perturbation, divided by twice the random perturbation vector. The device calculates the initial gradient value vector for the current circuit parameters simultaneously according to the above formula for all parameter dimensions.

[0076] After obtaining the initial gradient value, the device inputs the current circuit parameter values ​​into the noise model. The noise model then calculates the impact of noise on the gradient estimate under the current parameter conditions based on its internally established deviation distribution function. The calculation logic for gradient deviation in the noise model differs from that for energy expectation deviation; the gradient deviation is obtained by taking the partial derivative of the noise model's influence on the energy expectation deviation with respect to the parameters. The device subtracts the gradient deviation from the initial gradient value to obtain the corrected gradient value. The corrected gradient value eliminates the systematic gradient estimation bias introduced by hardware noise, making subsequent parameter updates more accurate.

[0077] Further, the step of updating the current circuit parameters based on the gradient value to obtain the updated circuit parameters includes: Step S521: Based on the noise model, determine the learning rate corresponding to the current circuit parameters, wherein the learning rate is used to control the update step size of the current circuit parameters; Step S522: Calculate the first-order momentum term based on the gradient value, and smooth the gradient value based on the first-order momentum term to obtain the smoothed gradient value; Step S523: Based on the noise model and the number of iterations of the current circuit parameters, dynamically adjust the learning rate to obtain the adjusted learning rate; Step S524: Update the current circuit parameters based on the smoothed gradient value and the adjusted learning rate to obtain the updated circuit parameters.

[0078] It should be noted that the learning rate mentioned above can be a scaling factor used to control the magnitude of the adjustment of the current circuit parameters along the gradient direction in each iteration. In the parameter optimization process of the variable quantum feature solver, the gradient value indicates the direction of parameter adjustment, while the learning rate determines the step size along that direction. An excessively large learning rate may cause oscillations or even divergence in the optimization process, while an excessively small learning rate will result in excessively slow convergence. The first-order momentum term mentioned above can be a statistic obtained by performing an exponentially weighted moving average of historical gradient information, used to smooth random fluctuations in gradient values ​​and make the parameter update direction more stable. The number of iterations mentioned above can be the total number of update rounds performed by the device since the start of parameter optimization iterations, and this value increases with each iteration.

[0079] In its implementation, after acquiring the corrected gradient value, the device inputs the current circuit parameter values ​​into the noise model. The noise model then calculates the degree of noise influence on the objective function under the current parameter conditions based on its internally established deviation distribution function. The device determines the learning rate based on the degree of noise influence: when the noise influence is high, the learning rate is set to a smaller value to avoid noise-driven parameter updates leading to optimization path deviation; when the noise influence is low, the learning rate is set to a larger value to accelerate parameter convergence towards the optimal direction. The noise model outputs a learning rate within a preset minimum to maximum range, and the device selects the corresponding value within the linear mapping range based on the specific quantization result of the noise influence at the current circuit parameters.

[0080] While acquiring the learning rate, the device reads the historical first-order momentum term from the memory, which was set to zero during initialization. The device then weights and sums the historical first-order momentum term from the previous iteration with the corrected gradient value of the current iteration using preset weighting coefficients to obtain the updated first-order momentum term. This updated first-order momentum term is the first-order momentum term for the current iteration. The device then smooths the corrected gradient value based on this first-order momentum term: outputting the smoothed gradient value instead of the original corrected gradient value for subsequent parameter updates. Through this smoothing process, random fluctuations in the gradient value are effectively suppressed, resulting in a more stable and reliable parameter update direction.

[0081] Subsequently, the device dynamically adjusts the learning rate based on the noise model and the number of iterations of the current circuit parameters. The device obtains the number of iterations completed and calculates an attenuation coefficient based on this number. As the number of iterations increases, the attenuation coefficient gradually decreases. The device multiplies the learning rate by the attenuation coefficient to obtain the adjusted learning rate, resulting in a gradual reduction in the step size of parameter updates in later iterations. When calculating the attenuation coefficient, the device uses the noise level at the current circuit parameters output by the noise model as a weighting factor: when the noise level is high, the rate attenuation coefficient decreases with the number of iterations, thus reducing the learning rate more quickly and suppressing noise interference with subsequent update steps; when the noise level is low, the rate attenuation coefficient decreases with the number of iterations, keeping the learning rate at an optimal level to maintain convergence speed.

[0082] Finally, the device updates the current circuit parameters based on the smoothed gradient values ​​and the adjusted learning rate. The device multiplies the adjusted learning rate by the corresponding components of the smoothed gradient values ​​one by one to obtain the adjustment amount for each parameter dimension in the current round. The current circuit parameters are then superimposed with these adjustment amounts to generate the updated circuit parameters.

[0083] To facilitate understanding, the following examples are provided for illustration, but they do not impose specific limitations on this embodiment. For instance, in a lithium hydride molecule simulation scenario, the parameterized quantum circuit contains four adjustable parameters. The current circuit parameters are (1.2, 0.8, 0.5, 0.3), and the corrected gradient values ​​are (0.48, -0.24, 0.315, -0.98). The device inputs the current circuit parameters into a noise model, and the model determines a learning rate of 0.02 based on the noise sensitivity of the current parameter position. The device reads the first-order momentum term from the memory as (0.2, -0.1, 0.15, -0.4). Using a preset momentum weight coefficient of 0.9, it multiplies the historical momentum term by 0.9 to obtain (0.18, -0.09, 0.135, -0.36). It then multiplies the current corrected gradient value by 0.1 to obtain (0.048, -0.024, 0.0315, -0.098). The sum of these two values ​​yields the first-order momentum term for the current iteration as (0.228, -0.114, 0.1665, -0.458), which is used as the smoothed gradient value. Subsequently, the device obtains the current iteration number as 150 and multiplies the base learning rate of 0.02 by the iteration decay coefficient of 0.8 to obtain an adjusted learning rate of 0.016. The device multiplies the adjusted learning rate of 0.016 with the smoothed gradient values ​​(0.228, -0.114, 0.1665, -0.458) component by component to obtain the adjustment amount (0.0036, -0.0018, 0.0027, -0.0073). The current circuit parameters are then superimposed with the adjustment amount to obtain the updated circuit parameters (1.2036, 0.7982, 0.5027, 0.2927).

[0084] Furthermore, the step of correcting the energy expectation value of the parameterized quantum circuit using the noise model includes: Step S44: Input the initial circuit parameters into the noise model to obtain the initial noise offset.

[0085] Step S45: Input the current circuit parameters into the noise model to obtain the current noise offset.

[0086] Step S46: Correct the energy expectation value based on the difference between the current noise offset and the initial noise offset to obtain the corrected energy expectation value.

[0087] It should be explained that the aforementioned noise offset can be an estimate output by the noise model based on the input circuit parameters, used to quantify the degree of interference of hardware noise on the energy measurement results under the current parameters.

[0088] The aforementioned current circuit parameters can be the specific values ​​of the adjustable parameters currently used by the parameterized mimicry circuit in each iteration of optimization.

[0089] The aforementioned expected energy value can be the average energy output of the target molecular system measured after running the parameterized mimicry circuit under the current circuit parameters. The aforementioned correction can be a process of subtracting the interference component caused by noise from the measured value to restore or approximate the true energy value under noise-free conditions.

[0090] In its implementation, before starting iterative optimization, the device inputs the generated initial circuit parameters into a pre-built noise model. Based on the input parameter values, the noise model outputs a corresponding initial noise offset, reflecting the degree of influence of hardware noise on the energy measurement results at the optimization starting point. Subsequently, the device inputs the current circuit parameters into the same noise model to obtain the current noise offset, which reflects the degree of influence of hardware noise on the energy measurement results in the current iteration step. The device compares the current noise offset with the initial noise offset and calculates the difference. A positive difference indicates that the current noise interference has increased compared to the initial value; a negative difference indicates that the current noise interference has decreased compared to the initial value. Next, the device obtains the expected energy value measured by running a parameterized mimicry circuit under the current circuit parameters and corrects this expected energy value based on the aforementioned difference. The correction method is as follows: if the current noise offset is greater than the initial noise offset, the difference is subtracted from the measured expected energy value to eliminate the increase in noise compared to the initial state; if the current noise offset is less than the initial noise offset, the absolute value of the difference is added to the measured expected energy value to compensate for the deviation caused by the reduction in noise interference. After the device completes the correction, the corrected expected energy value is obtained. This corrected value serves as the basis for evaluating the quality of circuit parameters during subsequent optimization iterations.

[0091] To facilitate understanding, the following examples are provided for illustration, but they do not impose specific limitations on this embodiment. For instance, suppose that in a molecular simulation task, the initial noise offset output after the initial circuit parameters are input into the noise model is 0.02 Hartree. During the iteration process, in a certain round of optimization, the current noise offset output after the current circuit parameters are input into the noise model is 0.07 Hartree, and the difference between the current noise offset and the initial noise offset is 0.05 Hartree. The energy expectation value measured by the device in this round is -1.05 Hartree. Then, the device subtracts 0.05 Hartree from the energy expectation value to obtain the corrected energy expectation value of -1.10 Hartree. As another example, if in another iteration the current noise offset is 0.01 Hartree, the difference between it and the initial noise offset of 0.02 Hartree is -0.01 Hartree, and the measured energy expectation value is -1.08 Hartree, then the device adds 0.01 Hartree to the energy expectation value to obtain the corrected energy expectation value of -1.07 Hartree. In this way, the device can dynamically correct the impact of noise on energy measurement in each iteration, thus advancing the optimization process in a more accurate direction.

[0092] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the parameter optimization method of the variable quantum feature solver of this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0093] This application also provides a parameter optimization device for a variational quantum characteristic solver; please refer to [reference needed]. Figure 4 The device includes: The circuit construction module 10 is used to acquire molecular interaction information of the target molecular system and construct a parameterized quantum circuit based on the molecular interaction information. The noise modeling module 20 is used to acquire noise characteristic parameters and construct a noise model based on the noise characteristic parameters. The parameter acquisition module 30 is used to acquire the current circuit parameters and initial circuit parameters of the parameterized quantum circuit; The energy correction module 40 is used to correct the energy expectation value of the parameterized quantum circuit according to the initial circuit parameters and the current circuit parameters through the noise model, so as to obtain the corrected energy expectation value. The parameter optimization module 50 is used to update the current circuit parameters according to the corrected energy expectation value, and to obtain the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition.

[0094] The parameter optimization device for the variable quantum characteristic solver provided in this application employs the parameter optimization method for the variable quantum characteristic solver in the above embodiments, and can solve the technical problem of how to improve the convergence stability of parameter optimization of the variable quantum characteristic solver in the noisy environment of a neutral atom quantum computing platform. Compared with the prior art, the beneficial effects of the parameter optimization device for the variable quantum characteristic solver provided in this application are the same as the beneficial effects of the parameter optimization method for the variable quantum characteristic solver provided in the above embodiments, and other technical features in the parameter optimization device for the variable quantum characteristic solver are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0095] This application provides a parameter optimization device for a variable quantum characteristic solver, which is applied to a neutral atom quantum computing platform; The parameter optimization device for the variable quantum feature solver includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the parameter optimization method of the variable quantum feature solver in Embodiment 1 above.

[0096] The following is for reference. Figure 5 This document illustrates a schematic diagram of a parameter optimization device suitable for implementing the variable quantum feature solver in the embodiments of this application. The parameter optimization device for the variable quantum feature solver in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The parameter optimization device for the variable quantum feature solver shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0097] like Figure 5As shown, the parameter optimization device of the variable quantum characteristic solver may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the parameter optimization device of the variable quantum characteristic solver. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the parameter optimization device of the variational quantum characteristic solver to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows parameter optimization devices for the variational quantum characteristic solver with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0098] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0099] The parameter optimization device for the variable quantum characteristic solver provided in this application employs the parameter optimization method for the variable quantum characteristic solver in the above embodiments, and can solve the technical problem of how to improve the convergence stability of parameter optimization of the variable quantum characteristic solver in the noisy environment of a neutral atom quantum computing platform. Compared with the prior art, the beneficial effects of the parameter optimization device for the variable quantum characteristic solver provided in this application are the same as the beneficial effects of the parameter optimization method for the variable quantum characteristic solver provided in the above embodiments, and other technical features in the parameter optimization device for the variable quantum characteristic solver are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0100] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0102] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the parameter optimization method for the component quantum feature solver as described above.

[0103] The computer program product provided in this application can solve the technical problem of how to improve the convergence stability of parameter optimization of the variable quantum characteristic solver in the noisy environment of the neutral atom quantum computing platform. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the parameter optimization method of the variable quantum characteristic solver provided in the above embodiments, and will not be repeated here.

[0104] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A parameter optimization method for a variational quantum characteristic solver, applied to a neutral atom quantum computing platform, characterized in that, The method includes: Obtain molecular interaction information of the target molecular system, and construct a parameterized quantum circuit based on the molecular interaction information; Obtain noise characteristic parameters and construct a noise model based on the noise characteristic parameters; Obtain the current circuit parameters and initial circuit parameters of the parameterized quantum circuit; Based on the initial circuit parameters and the current circuit parameters, the energy expectation value of the parameterized quantum circuit is corrected using the noise model to obtain the corrected energy expectation value; The current circuit parameters are updated based on the corrected energy expectation value, and the optimized preset circuit parameters are obtained when the updated current circuit parameters meet the preset convergence condition.

2. The method as described in claim 1, characterized in that, The steps of obtaining noise feature parameters and constructing a noise model based on the noise feature parameters include: Noise data is acquired through quantum process tomography, and noise characteristic parameters are obtained based on the noise data; A noise model is constructed based on the noise characteristic parameters. The noise model is used to describe the deviation distribution of noise on the execution result of the parameterized quantum circuit.

3. The method as described in claim 1, characterized in that, The step of correcting the energy expectation value of the parameterized quantum circuit based on the initial circuit parameters and the current circuit parameters, and obtaining the corrected energy expectation value, includes: The initial energy expectation value is obtained by measuring the energy expectation value of the target molecular system through the parameterized quantum circuit. Based on the initial energy expectation value, the noise offset under the current noise conditions is calculated using the noise model; The weighting coefficient of the noise offset is determined based on the initial circuit parameters and the current circuit parameters; The initial energy expectation value is compensated and corrected based on the weighting coefficient and the noise offset to obtain the corrected energy expectation value.

4. The method as described in claim 1, characterized in that, The step of updating the current circuit parameters based on the corrected energy expectation value, and obtaining the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition, includes: The gradient value of the parameterized quantum circuit is calculated based on the corrected energy expectation value; The current circuit parameters are updated based on the gradient value to obtain the updated circuit parameters; When the updated circuit parameters meet the preset convergence conditions, the optimized preset circuit parameters are obtained.

5. The method as described in claim 4, characterized in that, The step of calculating the gradient value of the parameterized quantum circuit based on the corrected energy expectation value includes: Based on the corrected energy expectation value, the initial gradient value of the current circuit parameters is calculated using a random perturbation approximation algorithm; Based on the noise model, calculate the gradient deviation of the current circuit parameters under the current noise environment; The initial gradient value is corrected based on the gradient deviation to obtain the corrected gradient value.

6. The method as described in claim 4, characterized in that, The step of updating the current circuit parameters based on the gradient value to obtain the updated circuit parameters includes: Based on the noise model, the learning rate corresponding to the current circuit parameters is determined, and the learning rate is used to control the update step size of the current circuit parameters; The first-order momentum term is calculated based on the gradient value, and the gradient value is smoothed based on the first-order momentum term to obtain the smoothed gradient value. The learning rate is dynamically adjusted based on the noise model and the number of iterations of the current circuit parameters to obtain the adjusted learning rate. The current circuit parameters are updated based on the smoothed gradient value and the adjusted learning rate to obtain the updated circuit parameters.

7. The method as described in claim 1, characterized in that, The step of correcting the energy expectation value of the parameterized quantum circuit using the noise model includes: The initial circuit parameters are input into the noise model to obtain the initial noise offset; The current circuit parameters are input into the noise model to obtain the current noise offset; The energy expectation value is corrected based on the difference between the current noise offset and the initial noise offset to obtain the corrected energy expectation value.

8. A parameter optimization device for a variational quantum characteristic solver, characterized in that, The device includes: A circuit construction module is used to acquire molecular interaction information of the target molecular system and construct a parameterized quantum circuit based on the molecular interaction information. The noise modeling module is used to acquire noise characteristic parameters and construct a noise model based on the noise characteristic parameters. The parameter acquisition module is used to acquire the current circuit parameters and initial circuit parameters of the parameterized quantum circuit. An energy correction module is used to correct the energy expectation value of the parameterized quantum circuit based on the initial circuit parameters and the current circuit parameters, using the noise model, to obtain the corrected energy expectation value. The parameter optimization module is used to update the current circuit parameters according to the corrected energy expectation value, and to obtain the optimized preset circuit parameters when the updated current circuit parameters meet the preset convergence condition.

9. A parameter optimization device for a variational quantum characteristic solver, characterized in that, The device is used in a neutral atom quantum computing platform; The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the parameter optimization method for the variable quantum feature solver as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the parameter optimization method for the variable quantum feature solver as described in any one of claims 1 to 7.

Images