Quantum state manipulation quality quantification verification method for optical quantum chip

Abstract

The present application belongs to the technical field of quantum computing, and is used for solving the problem that the prior art cannot distinguish which time periods of photon events in the collection window truly reflect the steady-state manipulation quality of the target state and which time periods are disturbed by the transition state. Specifically, the present application is a quantum state manipulation quality quantification verification method for optical quantum chips, which comprises an acquisition step, a first processing step, a template generation step, a second processing step, and a reconstruction step. The present application decomposes the collection window into multiple continuous sub-windows along the time axis and establishes a normalized count rate matrix. The first characteristic quantity is used to represent the smoothness of the latter half, and the generation mode of the proxy steady state is dynamically determined according to the smoothness. The difference between the proxy steady state and the initial state is used to construct an intra-frame self-generated ideal evolution direction template, so that the deviation of the actual evolution direction of each sub-window from the template is quantitatively described. Finally, the continuous contribution weight is assigned according to the comparison relationship between the deviation and the decision tolerance, and the weighted average reconstruction is performed.

Description

Technical Field

[0001] This invention belongs to the field of quantum computing technology, specifically a method for quantizing and verifying the quality of quantum state manipulation in optical quantum chips. Background Technology

[0002] In the process of verifying the control quality of optical quantum chips, a typical existing approach is to classify all single-photon detection data from all output ports acquired within a preset acquisition window after the target control command is issued as belonging to the target quantum state corresponding to the command, and calculate the fidelity or perform state tomography reconstruction based on the overall data, thus serving as the quantitative basis for the control quality of the target state. The implicit premise of this approach is that all photon events within the acquisition window correspond to the same stable and effective chip state. However, optical quantum chips typically contain multiple types of adjustment units, and the time constants of electro-optic response and thermo-optic response differ by orders of magnitude.

[0003] When multiple control units work together to complete a state switch, the physical activation time of the control command on each unit is not strictly synchronized, and some slower-responding units will experience a longer transition process. Therefore, the data in the early stage of the acquisition window actually contains photon events corresponding to intermediate state combinations where "some units are in place and some units are not in place." This transition state contamination may only manifest as random fluctuations in fidelity in a single measurement, but in closed-loop calibration or continuous rapid switching verification scenarios, the calibration algorithm will use the contaminated data as feedback, which may gradually drive the control parameters to deviate from the correct steady-state value, resulting in a more skewed result after calibration. The underlying reason for the above situation is that the existing verification logic mechanically classifies the acquired data according to the time window relationship to the target command, without considering the substantial differences that the on-chip physical states experienced at different times within the window may have.

[0004] Therefore, how to distinguish which photon events within the acquisition window truly reflect the steady-state control quality of the target state and which periods are affected by transitional state interference is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a quantum state manipulation quality quantization verification method for optical quantum chips, which solves the problem that existing technologies cannot distinguish which photon events within the acquisition window truly reflect the steady-state manipulation quality of the target state and which periods are affected by transition state interference.

[0006] The technical problem to be solved by this invention is: how to provide a quantitative verification method for quantum state manipulation quality of optical quantum chips that can distinguish which time periods of photon events within the acquisition window truly reflect the steady-state manipulation quality of the target state and which time periods are affected by transition state interference.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for quantitative verification of quantum state manipulation mass for optical quantum chips includes the following steps:

[0009] Acquisition steps: Acquire single-photon detection events from multiple output ports of the photonic quantum chip within a preset acquisition window after the target control command is issued, and generate a normalized count rate matrix based on the photon counts of each output port within each time sub-window;

[0010] First processing step: Based on the normalized count rate matrix, extract the first feature quantity that characterizes the stability of the data in the second half of the acquisition window, and the second feature quantity sequence that characterizes the evolution trend of photon distribution between adjacent time sub-windows;

[0011] Template generation steps: Based on the first feature quantity, determine the surrogate steady-state distribution from the data in the second half of the acquisition window, and generate the ideal evolution direction template corresponding to the acquisition window based on the difference between the surrogate steady-state distribution and the data in the first half of the acquisition window;

[0012] The second processing step is to determine the decision tolerance based on the first feature quantity, and compare each second feature quantity with the ideal evolution direction template to obtain the deviation measure corresponding to each time sub-window.

[0013] Reconstruction steps: Based on the comparison between the deviation measure and the decision tolerance of each time sub-window, a continuous contribution weight is assigned to each time sub-window, and the normalized count rate matrix is ​​weighted and averaged according to the contribution weight to obtain the reconstructed photon distribution data for verification.

[0014] The present invention has the following beneficial effects:

[0015] 1. By decomposing the acquisition window into multiple continuous sub-windows along the time axis and establishing a normalized count rate matrix, the first feature is used to characterize the stability of the latter half, and this stability dynamically determines the generation method of the surrogate steady state. Then, the difference between the surrogate steady state and the initial state is used to construct an intra-frame self-generated ideal evolution direction template, so that the deviation of the actual evolution direction of each sub-window from this template is quantitatively described. Finally, based on the comparison between the deviation and the decision tolerance, continuous contribution weights are assigned and weighted average reconstruction is performed. This mechanism ensures that the initial portion of the window, where some units are not yet in place, is reconstructed. The transitional photon events are given lower weights because their evolution direction deviates from the ideal path, while the photon events approaching steady state in the later stage are given higher weights because their evolution direction is consistent with the template. The two are separated during the weighted averaging process. The contribution of transient deviation in the reconstructed output distribution is suppressed while the steady-state component is retained. The contamination component in the input data of the closed-loop calibration system is reduced accordingly. The parameter correction direction is no longer deviated due to transitional interference. The fluctuation of fidelity with the number of iterations in continuous multi-frame verification is constrained within the preset tolerance range. The convergence state of the feedback loop is no longer diverged due to data misalignment.

[0016] 2. The decision tolerance is set as the sum of the hardware's inherent minimum resolution angle and the first feature multiplied by a preset conversion coefficient, so that the decision criterion dynamically fluctuates with the actual fluctuation level of the second half of the current frame: when the stability level of the second half is low, the decision tolerance is automatically widened to avoid misjudging normal evolutionary fluctuations caused by noise as deviations; when the stability level of the second half is high, the decision tolerance is automatically tightened to ensure that the sensitivity to deviations is not weakened due to the reduction of the noise floor. This dynamic tolerance mechanism ensures that the threshold for measuring the degree of deviation maintains effective discriminative power under different signal-to-noise ratio conditions, avoiding the dilemma of increased false alarm rate in low signal-to-noise ratio scenarios and increased false negative rate in high signal-to-noise ratio scenarios with a fixed threshold. The determination results of the degree of deviation of each sub-window are still comparable when the noise floor changes.

[0017] 3. A continuous weighting mapping function is used to replace the binarized elimination or retention decision. This allows the variation of the deviation measure within the range of zero to the decision tolerance to be smoothly mapped to a continuous decay of weights between 1 and 1 / 2. After the deviation measure exceeds the decision tolerance, the weights continue to asymptotically approach zero as the deviation increases. This continuous processing avoids abrupt changes in the statistical sample size caused by hard threshold truncation. This ensures that the reconstructed photon distribution data for verification retains the effective information of the window with a small deviation, and does not introduce artificial distribution distortion due to weight jumps. The sum of each component of the weighted average output distribution remains normalized and can be directly input into the fidelity calculation module without additional normalization processing. The data stream is not interrupted during processing.

[0018] 4. The sample variance of the deviation measure of each window in the current frame is mapped to a confidence factor. This factor, together with the evolutionary activity of the current frame, is used to drive the adjustment of the sub-window segmentation in the next frame: when the confidence is high and the activity is high, the number of sub-windows is increased to improve the sampling granularity of transient features; when the confidence is low or the activity is low, the number of sub-windows is reduced to avoid shot noise introduced by oversampling. This feedback mechanism, combined with the single-step hysteresis adjustment strategy, makes the segmentation change smoothly between consecutive frames and is limited to the preset upper and lower limits. This verifies that the system's temporal resolution adapts to the fluctuation of data quality, while avoiding parameter oscillation due to hysteresis protection, ensuring that usable temporal resolution can be maintained under different signal quality conditions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method according to Embodiment 1 of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The idealized premise of the quantum chip control quality verification scenario involved in this solution is that after the target control command is issued, all adjustment units on the chip participating in state switching synchronously complete physical positioning, and all photon events at the output ports within the acquisition window correspond to the same stable control state. However, at the physical implementation level, there is an order of magnitude difference between the physical time constants of electro-optic response and thermo-optic response. The software simultaneity of commands during multi-unit collaborative adjustment is not equivalent to the synchronicity of hardware activation.

[0023] Existing verification techniques archive acquired data uniformly according to target control commands, ignoring the substantial differences that photons may experience in manipulation combinations within the chip at different times within a window. This leads to two problems. First, when some units with high thermal inertia are still stuck in the response process of the previous state, the data from the beginning of the current acquisition window contains photon events corresponding to intermediate state combinations. These events cannot be distinguished from steady-state events in fidelity calculations, resulting in traces of the switching process being mixed into the verification results of a single instance. Second, in closed-loop calibration scenarios, the calibration algorithm uses data containing transitional state contamination as feedback, which is equivalent to using the deviation from steady-state properties to correct the steady-state control parameters. After multiple iterations, the control quantity drifts in the wrong direction, exhibiting an abnormal trend of decreasing fidelity with increasing calibration counts.

[0024] If the aforementioned problem of the mixture of transitional and steady-state states is not resolved, the closed-loop calibration system will continuously lose its ability to determine the purity of the current verification data. The proportion of transitional events in a single frame fluctuates with the switching frequency and thermal time constant. Under rapid continuous switching conditions, its proportion may exceed that of steady-state events, directly leading to non-physical solutions in the reconstructed density matrix or a fidelity statistical variance exceeding the acceptable range. Furthermore, after the calibration algorithm corrects the parameters based on the contaminated data, the transitional characteristics in the next acquisition will change accordingly. An uncertain coupling relationship will be formed between the contamination mode and the control parameters, making the correspondence between the verification results and the actual control quality untraceable. This systematic feedback inaccuracy will result in a lack of reliable benchmark references for the quality assessment of the quantum chip in multi-task continuous verification scenarios, affecting the effectiveness of chip parameter tuning and process evaluation.

[0025] Example 1: As Figure 1 As shown, the method for quantitative verification of quantum state manipulation mass for optical quantum chips includes the following steps:

[0026] Acquisition steps: Acquire single-photon detection events from multiple output ports of the photonic quantum chip within a preset acquisition window after the target control command is issued, and generate a normalized count rate matrix based on the photon counts of each output port within each time sub-window;

[0027] Before the specific execution of the acquisition step, the quantum chip is already in the data acquisition stage after the target control command is issued. In this embodiment, a single-photon detector array is arranged at N output ports of the quantum chip, where N is a positive integer greater than 1. Each detector is connected to a time-to-digital converter to record the arrival event of each photon relative to the time when the target control command is issued. The core task of step S1, which is to obtain the timestamps, is to convert the above continuous sequence of photon arrival timestamps into a normalized count rate matrix that can be used for subsequent quality quantification analysis.

[0028] Specifically, the total duration T of the acquisition window is set. The value of T is determined to be no less than three times the maximum thermal time constant among all adjustment units on the quantum chip, ensuring that the latter half of the acquisition window is sufficient to cover the complete transition process of each adjustment unit from the issuance of the control command to basic convergence. If the chip does not provide time constant parameters for each adjustment unit, a pre-scan procedure is allowed before formal verification: the output data under the same control command is repeatedly acquired with an increasing T value. After each acquisition, the mean of the sample standard deviation of the normalized count rate of each output port in the latter half of the acquisition window is calculated. When the mean is less than one percent, the scan is stopped, and T at this time is the engineering usable value. After T is determined, the acquisition window is... Divide the time axis evenly into M adjacent and non-overlapping time sub-windows, each sub-window having a width of [missing value]. M is a positive integer greater than or equal to 4, and its initial value can be set to 8; the time sub-windows are numbered sequentially according to the time order. The time interval corresponding to the kth sub-window is .

[0029] For the k-th time sub-window and the i-th output port, where The total number of photon events falling within this time interval is counted and recorded as the original photon count. Based on this, the normalized count rate is calculated. When the sum of the original photon counts of all N output ports within the k-th time sub-window is greater than zero, according to... The calculation is performed by dividing the original count of the port within the sub-window by the sum of the original counts of all ports within the sub-window. After this normalization process, the sum of the normalized count rates of all ports within the sub-window at any given time is always equal to 1. This eliminates the influence of common-mode fluctuations in light source intensity along the time axis on subsequent analysis. When the sum of the original photon counts of all N output ports within the k-th time sub-window is zero, meaning no photon event is detected within that time interval, the time sub-window is marked as an empty window, and its count rate vector is normalized. Each component is assigned a value That is, using equally probable distributions as neutral fill values ​​to ensure the numerical stability of subsequent matrix operations; the N-dimensional weight vectors of all M time sub-windows are arranged column-wise to form a matrix of size... Normalized count rate matrix ,in .

[0030] For ease of understanding, a simplified numerical example is given here; assume that the photonic quantum chip has N=4 output ports, and the acquisition window is divided into M=6 time sub-windows; within the 3rd sub-window, the original photon counts detected by the 4 ports are as follows: , , , Then the sum of the original counts within that sub-window is Therefore, the normalized count rates of each port are respectively , , , If no photons are detected at any port within the 5th sub-window, this sub-window is marked as an empty window, and each component of its normalized count rate vector is assigned a value of 1 / 4 = 0.25. After all sub-windows have been processed, the normalized count rate matrix... This means that the evolution trajectory of photon distribution over time within the acquisition window is fully recorded.

[0031] It is worth noting that the value of M is not fixed, but dynamically adjusted based on the data quality assessment results of the previous acquisition window. The specific adjustment mechanism will be explained in detail in subsequent steps. However, during initial acquisition or system cold start, the value of M is 8. In addition, the empty window fill value in the above normalization process... The basis for this setting is that when there are no photon events in a certain time sub-window, the window does not provide any effective information. Filling it with an equal probability distribution can naturally dilute the contribution of the window in the subsequent weighted average processing, while avoiding numerical anomalies that may be caused by the zero vector. Thus, step S1 completes the complete transformation from the original photon arrival events to the normalized count rate matrix, providing standardized data input for feature extraction in the subsequent first processing step.

[0032] First processing step: Based on the normalized count rate matrix, extract the first feature quantity that characterizes the stability of the data in the second half of the acquisition window, and the second feature quantity sequence that characterizes the evolution trend of photon distribution between adjacent time sub-windows;

[0033] The first processing step includes two parallel feature extraction tasks, targeting the photon distribution stability in the latter half of the acquisition window and the evolution trend between adjacent sub-windows, respectively.

[0034] For a regulating unit with a first-order inertial or overdamped response characteristic, the latter half of the response will inevitably be closer to the steady-state value than the first half. The optical quantum chip regulating unit targeted in this embodiment belongs to the above type, so this assumption holds true.

[0035] Regarding the extraction of the first feature, it is first necessary to determine the range of the time sub-window covered by the second half; assuming the acquisition window is divided into M time sub-windows, and the starting index L of the second half is... Given, among which For floor operations; for example, when M=8, L=6, meaning the 6th, 7th, and 8th sub-windows constitute the second half; for each output port i, obtain the normalized count rate value of that port in each sub-window of the second half. Calculate the sample standard deviation of these values. When the number of non-empty child windows in the second half is no less than 2, The calculation uses the usual definition of sample standard deviation; if the number of non-empty sub-windows is less than 2, the effective standard deviation cannot be calculated. In this case, the first feature is directly set to 1, representing the maximum uncertainty; for all N ports... The arithmetic mean is taken, and the resulting mean is the first characteristic quantity, denoted as . The smaller the value of this quantity, the smaller the fluctuation of the count rate of each port in the second half over time, and the photon distribution has approached convergence; the larger the value, the more significant the residual drift in the second half.

[0036] The extraction of the second feature sequence focuses on the direction of change in photon distribution between adjacent time sub-windows; for each pair of adjacent sub-windows k-1 and k ( ), and the normalized count rate vector of the next window Subtract the previous window The difference vector is obtained. The components of this vector describe the increase or decrease in the counting rate of each output port between two adjacent sub-windows; calculate the magnitude of this difference vector. This involves a preset minimum threshold. Preset minimum threshold The calibration is performed once during the system initialization phase. The specific process is as follows: In an idle state where the chip has no adjustment unit operation and the light source maintains stable output, 100 frames of data are continuously acquired; for each frame, the difference vector magnitude between adjacent windows is calculated as described above; the upper quintile (i.e., the 95th quintile) of the set of difference vector magnitudes for all frames and all adjacent windows is taken as... The value of ; if the value is less than 10 -7 Then take 10 -7 As a lower limit for engineering; when the module length is greater than... When the difference vector is divided by its magnitude and normalized, a unit vector is obtained. This indicates that a significant change has occurred in the photon distribution during that period. This is called the evolution direction vector of the adjacent window pair; when the magnitude is not greater than When the distribution difference between two consecutive sub-windows is negligible or entirely due to rounding errors, the difference vector does not contain effective evolutionary information. The corresponding evolutionary direction vector is then marked as an invalid vector and represented by an all-zero vector. The evolutionary direction vectors of all adjacent window pairs are arranged in chronological order, thus forming the second feature sequence. .

[0037] It should be noted that the preset minimum threshold is... The calibration can be completed when the chip is idle, that is, when no adjustment unit is activated, multiple frames of data are acquired, the magnitude of the difference vector between each adjacent window is calculated, and the upper quantile of the statistical distribution of these magnitudes is taken as the standard value. This calibration only needs to be performed once during the system initialization phase; after the two feature quantities are extracted, the first feature quantity... The second feature sequence will be used to determine the stability of the second half in the subsequent template generation step, and to dynamically determine the decision tolerance in the second processing step; the second feature sequence will be used to compare with the subsequently generated ideal evolution direction template to quantify the degree of deviation from the ideal path at each time period.

[0038] Template generation steps: Based on the first feature quantity, determine the surrogate steady-state distribution from the data in the second half of the acquisition window, and generate the ideal evolution direction template corresponding to the acquisition window based on the difference between the surrogate steady-state distribution and the data in the first half of the acquisition window;

[0039] The core task of the template generation step is to construct a reference benchmark, namely the ideal evolution direction template, from the data of the acquisition window itself. This template does not depend on any external calibration or theoretical calculation and is generated entirely by the data of the second half of the current frame. The whole step is divided into two stages: determining the surrogate steady-state distribution and generating the ideal evolution direction template.

[0040] The determination of the steady-state distribution of the proxy is based on the first characteristic quantity. With the noise threshold The comparison result is used as the criterion. Noise floor threshold. The value is obtained in advance through calibration: Multiple frames of data are collected in an idle state where the chip has no adjustment unit operation; the first feature quantity of each frame is calculated as described above; and the arithmetic mean of the first feature quantities of all frames plus three times the standard deviation is taken as the first feature quantity. And this value is not less than 0.003; when When the data in the latter half of the acquisition window is determined to be sufficiently stable, the normalized count rate vectors of all non-empty time sub-windows in the latter half are taken as the arithmetic mean, and this average is used as the surrogate steady-state distribution. ;when When this occurs, it indicates that residual drift still exists in the latter half. In this case, the first non-empty sub-window is selected from the sub-windows of the latter half in chronological order from back to front, and the normalized count rate vector of this sub-window is directly used as the surrogate steady-state distribution. If all time sub-windows within the entire acquisition window are empty, the verification process will terminate and a message indicating no valid photon events will be output; Initial state distribution of the first half. Take the normalized count rate vector of the first non-empty time sub-window within the acquisition window.

[0041] After the surrogate steady-state distribution is determined, the global difference vector is calculated. And find its modulus. This involves two preset thresholds; the first preset threshold... The preset minimum threshold value is the same as that used in the second feature extraction mentioned above, representing the smallest resolvable magnitude of effective change; the second preset threshold is 10 times the first preset threshold, i.e. This value corresponds to a typical rounding error of 10 times in double-precision floating-point arithmetic; when At that time, the global difference vector is normalized by dividing it by its magnitude to obtain the ideal evolution direction template. .

[0042] when If no effective change is observed in the photon distribution within the acquisition window, the deviation measurement is no longer performed. Instead, the contribution weights of all time sub-windows are set to an equal value of 1, and the arithmetic mean of the normalized count rate vectors of all windows is directly taken as the mean. At the same time, the confidence factor Set it to 1 (because all windows are oriented in the same direction), and then proceed to the next frame for processing.

[0043] when and The absolute value of the difference is less than When it reaches 1%, it is directly judged as ≤ In cases where there is no effective evolution, the process is considered to have jumped to the weighted equalization stage and will not proceed to the next stage. The low-confidence flag branch; this boundary protection rule has higher priority than the three-segment threshold division, and it is only applied when the modulus is... The absolute value of the difference is less than The data is triggered within the 1% neighborhood. The width of this neighborhood is equivalent to the extended boundary of the rounding error in floating-point operations in numerical calculations. When the modulus falls within this range, the original data does not have the statistical significance to distinguish between ineffective evolution and low-confidence evolution. Unifying them into the ineffective evolution branch can avoid branch conflicts and eliminate frequent jumps.

[0044] like Although the modulus is greater than the first preset threshold, its value is still within the noise-sensitive range. At this point, the reliability of the ideal evolution direction template obtained by normalization is low, and the system generates... Simultaneously, an additional low-confidence flag signal is generated, which will be used to penalize the confidence factor in subsequent reconstruction steps; if Then the template direction has statistical reliability and no additional labels are added; taking the scenario of N=4 ports and M=6 child windows as an example: if the arithmetic mean vector of each non-empty window in the second half Initial state ,but Its modulus length is approximately 0.48, which is much larger than... After normalization, an ideal evolutionary direction template with a clear direction is obtained; this template records the global change direction required from the initial distribution to the surrogate steady state, and will be used as a geometric reference to determine whether the evolutionary path in each period is normal.

[0045] The second processing step is to determine the decision tolerance based on the first feature quantity, and compare each second feature quantity with the ideal evolution direction template to obtain the deviation measure corresponding to each time sub-window.

[0046] The second processing step involves two levels of computation: determining the decision tolerance and calculating the deviation measure of each time sub-window.

[0047] Judgment tolerance The value of is not a fixed constant, but is dynamically adjusted according to the stability of the latter half of the acquisition window; its calculation method is as follows: first, obtain the pre-calibrated hardware-inherent minimum resolution angle. The unit is radians. This value reflects the minimum angular resolution achievable by the time-to-digital converter under the combined effects of jitter and photon shot noise. A typical calibration method is to acquire multiple frames of data while the chip is idle, calculate the deviation angle between the evolution direction vector of each adjacent window and the zero vector, and take twice the average deviation angle as the mean. ; then With the first characteristic Multiply by the preset conversion factor The sum of the products, i.e. Preset conversion coefficient The value ranges from 0.5 to 1.5, and in this embodiment, it is set to 1.0; this coefficient determines the contribution weight of the intra-frame stability coefficient ξ to the decision tolerance: mean It contributes an additional tolerance increment of 0.01 radians; if If the value is too large, the tolerance will be excessively broadened with noise, which may cause the deviation measurement to be unable to distinguish between valid deviation and noise; if If the value is too small, normal fluctuations under the influence of noise may be misjudged as valid deviations; 1.0 is a compromise between the two; the physical meaning of this formula is that the larger the fluctuation in the second half, the stronger the disturbance to the current frame as a whole. The criteria for judging the consistency of the evolution direction should be appropriately relaxed in order to avoid misjudging small deviations caused by noise as valid differences.

[0048] The deviation metric is calculated separately for each region of the time sub-window; for each non-empty sub-window in the first half that is located before the starting index L of the second half, its corresponding evolution direction vector is extracted. ( );when When it is an effective unit vector, calculate its relationship with the ideal evolution direction template. inner product Since both are unit vectors, the inner product value is the cosine of the angle between them. To prevent floating-point rounding errors from causing the input of the inverse cosine operation to exceed the domain, the inner product result needs to be clamped to... Within the interval, the inverse cosine operation is then performed to obtain the included angle value. The included angle value serves as a measure of the deviation of that time sub-window; when When a vector is marked as invalid, the deviation metric for that window is directly set to... (Right angle), representing that their evolution directions are completely unrelated; taking the aforementioned scenario with N=4 ports and a large difference between the initial state and the surrogate steady state as an example, if the evolution direction vector of a certain front window Ideal template The inner product of the two is approximately 0.99, corresponding to an angle of only about 0.14 radians, indicating that the evolutionary path during this period is highly consistent with the ideal direction.

[0049] The deviation measurement assignment rule for each non-empty sub-window in the second half depends on the first feature. With the noise threshold The comparison results; when When the latter half of the data is deemed sufficiently stable, all windows are uniformly assigned a deviation measure of zero; when When the number of non-empty sub-windows in the second half is zero, the deviation metric is set to zero. If the number of non-empty sub-windows in the second half is one, the mapping process terminates, and there are no other non-empty windows that need to be assigned values. If the number of non-empty sub-windows in the second half is greater than one, the deviation metric is calculated for the remaining non-empty sub-windows except the last one as follows: For each such window, its normalized count rate vector is calculated. With the last non-empty window vector Manhattan distance Find the maximum distance in all such windows. ,like ,make ;like If all such windows have the same normalized count rate vector as the last non-empty window, then the deviation metric will be uniformly set to 1. .

[0050] After the above operations are completed, each time sub-window obtains a deterministic deviation measure. This value will be directly input into subsequent reconstruction steps for calculating contribution weights.

[0051] If the number of valid evolution direction vectors in the preceding non-empty window is zero, then the frame is marked as "no valid evolution information," and this information is included in the reconstruction step. The default value is 0.1 (rather than based on) (Calculation) to indicate that the current frame data lacks credible transient evolution evidence.

[0052] Reconstruction steps: Based on the comparison between the deviation measure and the decision tolerance of each time sub-window, a continuous contribution weight is assigned to each time sub-window, and the normalized count rate matrix is ​​weighted and averaged according to the contribution weight to obtain the reconstructed photon distribution data for verification.

[0053] The reconstruction step targets the deviation measurement sequence output by the second processing step. and judgment tolerance The goal is to fuse the data from each time window into a single photon distribution estimate that eliminates transition state interference through continuous weighting.

[0054] For each non-empty time sub-window k, define its contribution weight. The calculation method; when hour, ;when At that time, the weight value is determined by the ratio. It is obtained through mapping using a monotonically decreasing function; specifically, a Lorentz-type mapping relationship is used: This mapping ensures that when the deviation metric is exactly equal to the decision tolerance, the weight is halved; when the deviation metric is much greater than the decision tolerance, the weight approaches zero; when the deviation metric is between zero and the decision tolerance, the weight smoothly varies between halved and one; empty windows do not participate in subsequent weighted calculations, and their weights are directly assigned to zero; taking the aforementioned case of M=8 and L=6 as an example, if the preceding window ( The deviation measures are respectively , , , Judgment tolerance The weights of each window are as follows: , , , The second half of the window ( )because The weights are all 1.

[0055] After the weights are assigned, the normalized count matrix is ​​processed. Perform a weighted average; the numerator is the normalized count rate vector for each time sub-window. According to the corresponding weight The weighted sum, with the sum of all weights in the denominator, is divided element-wise to obtain the reconstructed photon distribution vector for verification. The sum of all components of this vector remains 1, and it can be directly input into the downstream quality quantization module.

[0056] Based on this, a confidence factor is also calculated. This is used to quantify the reliability of the overall data quality of the current frame; it extracts the deviation metric of all non-empty sub-windows in the preceding segment. Calculate the sample variance within the range of non-empty windows. If the number of non-empty windows within this range is less than 2, then Values This is equivalent to assuming that the included angle has a moderate dispersion between zero and the decision tolerance; if a low-confidence marker signal is generated in the template generation step, then additional [information is added] on top of this. As a penalty term; then the variance is mapped to a confidence factor: The value of this factor ranges from 0 to 1. The larger the variance, the lower the factor value, indicating a poorer consistency in the evolution direction across different time periods of the frame, and a corresponding decrease in the reference value of the reconstruction results; Angle value The unit is radians, and its sample variance is... The unit is the square of radians; this scheme calculates... When Treating it as a dimensionless numerical value for calculation is reasonable because the definition of radians as an angle is essentially the ratio of arc length to radius, which is a dimensionless derived quantity. In numerical calculations, radian values ​​can be directly used as real numbers in arithmetic operations. Therefore, there is no computational barrier between the square of radians and the pure numerical value 1 at the calculation level. This approach conforms to the general convention in engineering numerical calculations, that is, the angle is treated as dimensionless in algebraic operations.

[0057] In output At the same time, calculate the effective window ratio. ,in This represents the number of empty windows; if If it is below 0.3, then Additional multiplication by a coefficient (Right now This is to reduce the confidence output in scenarios with low statistics.

[0058] This scheme decomposes the acquisition window into multiple continuous sub-windows along the time axis and establishes a normalized count rate matrix. It dynamically determines the generation method of the surrogate steady state using the first feature representing the stability of the latter half. Then, it constructs an intra-frame self-generated ideal evolution direction template based on the difference between this surrogate steady state and the initial state. This allows the deviation of the actual evolution direction of each sub-window from the template to be quantitatively described. Furthermore, based on the comparison between the deviation and the decision tolerance, a continuous contribution weight is assigned to each sub-window, and a weighted average reconstruction is performed. Thus, the transition from some units not yet in place in the first half of the window... Transient photon events are given lower weights because their evolution direction deviates from the ideal path, while photon events approaching steady state in the latter part of the window are given higher weights because their evolution direction is consistent with the template. The two are separated during the weighted averaging process, the contribution of transient deviation in the output distribution is suppressed, the contribution of steady-state component is retained, the contamination component in the input data of the closed-loop calibration system is reduced accordingly, the parameter correction direction is no longer deviated by transient interference, the fluctuation of fidelity with the number of iterations in continuous multi-frame verification is constrained within the preset tolerance range, and the convergence state of the feedback loop no longer diverges due to data misalignment.

[0059] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0060] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for quantitative verification of quantum state manipulation mass for optical quantum chips, characterized in that, Includes the following steps: Acquisition steps: Acquire single-photon detection events from multiple output ports of the photonic quantum chip within a preset acquisition window after the target control command is issued, and generate a normalized count rate matrix based on the photon counts of each output port within each time sub-window; First processing step: Based on the normalized count rate matrix, extract the first feature quantity characterizing the stability of the data in the second half of the acquisition window, and the second feature quantity sequence characterizing the evolution trend of photon distribution between adjacent time sub-windows; Template generation steps: Based on the first feature quantity, determine the surrogate steady-state distribution from the data in the second half of the acquisition window, and generate the ideal evolution direction template corresponding to the acquisition window based on the difference between the surrogate steady-state distribution and the data in the first half of the acquisition window; The second processing step is to determine the decision tolerance based on the first feature quantity, and compare each of the second feature quantities with the ideal evolution direction template to obtain the deviation measure corresponding to each time sub-window. Reconstruction steps: Based on the comparison between the deviation metric and the decision tolerance of each time sub-window, assign continuous contribution weights to each time sub-window, and perform weighted averaging on the normalized count rate matrix according to the contribution weights to obtain the reconstructed photon distribution data for verification.

2. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 1, characterized in that, The specific steps for obtaining the normalized count rate matrix include: uniformly dividing the preset acquisition window along the time axis into a preset number of adjacent and non-overlapping time sub-windows; for the original photon count of each output port in each time sub-window, dividing the original photon count by the sum of the original photon counts of all output ports in the time sub-window to obtain the normalized count rate of the output port in the time sub-window; when the sum of the original photon counts of all output ports in a time sub-window is zero, marking the time sub-window as an empty window, and assigning the normalized count rate of each output port in the time sub-window the reciprocal of the total number of output ports; and constructing the normalized count rate matrix from the normalized count rates of each output port in each time sub-window.

3. The method for quantitative verification of quantum state manipulation quality for optical quantum chips according to claim 1, characterized in that, The specific steps for extracting the first feature quantity characterizing the stability of the data in the second half of the acquisition window include: determining multiple time sub-windows located in the second half from all time sub-windows of the acquisition window; for each output port, calculating the sample standard deviation of the normalized count rate of the output port in each time sub-window of the second half; taking the average of the sample standard deviations of all output ports, and using the obtained average as the first feature quantity, wherein the smaller the value of the first feature quantity, the more stable the photon distribution in the second half of the acquisition window.

4. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 1, characterized in that, The specific steps for extracting the second feature sequence characterizing the evolution trend of photon distribution between adjacent time sub-windows include: for each pair of adjacent time sub-windows, subtracting the normalized count rate vector of the previous time sub-window from the normalized count rate vector of the subsequent time sub-window to obtain a difference vector; calculating the magnitude of the difference vector; when the magnitude is greater than a preset minimum threshold, dividing the difference vector by the magnitude to obtain the evolution direction vector corresponding to the adjacent window pair; when the magnitude is not greater than the preset minimum threshold, marking the evolution direction vector corresponding to the adjacent window pair as an invalid vector; the evolution direction vectors corresponding to each adjacent window pair constitute the second feature sequence in chronological order.

5. The method for quantitative verification of quantum state manipulation quality for optical quantum chips according to claim 1, characterized in that, The specific steps for determining the surrogate steady-state distribution from the data in the latter half of the acquisition window based on the first feature value include: when the first feature value is not greater than the noise floor threshold, taking the arithmetic mean of the normalized count rate vectors of each non-empty time sub-window in the latter half of the acquisition window, and using the obtained arithmetic mean as the surrogate steady-state distribution; when the first feature value is greater than the noise floor threshold, selecting the first non-empty time sub-window in the latter half of the acquisition window in chronological order from back to front, and using the normalized count rate vector of the non-empty time sub-window as the surrogate steady-state distribution; when all time sub-windows in the acquisition window are empty windows, terminating the verification process and outputting a prompt that there are no valid photon events.

6. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 5, characterized in that, The specific steps for generating the ideal evolution direction template based on the difference between the surrogate steady-state distribution and the initial segment data of the acquisition window include: using the normalized count rate vector of the first non-empty time sub-window within the acquisition window as the initial state distribution; calculating the difference between the surrogate steady-state distribution and the initial state distribution to obtain a global difference vector; calculating the magnitude of the global difference vector; when the magnitude is greater than a first preset threshold, dividing the global difference vector by the magnitude to obtain the ideal evolution direction template; when the magnitude is not greater than the first preset threshold, determining that the photon distribution within the acquisition window has not undergone effective change, setting the contribution weights of all time sub-windows to equal values, and skipping subsequent comparison and reconstruction processing.

7. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 6, characterized in that, When the magnitude of the global difference vector is greater than the first preset threshold but not greater than the second preset threshold, the global difference vector is divided by the magnitude to obtain the ideal evolution direction template, and a marker signal is generated to indicate that the confidence of the ideal evolution direction template is low. The second preset threshold is greater than the first preset threshold.

8. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 1, characterized in that, The specific method for determining the decision tolerance based on the first feature is as follows: obtain the pre-calibrated hardware-inherent minimum resolution angle; sum the hardware-inherent minimum resolution angle with the first feature by multiplying by a preset conversion coefficient, and use the summation result as the decision tolerance; the decision tolerance increases with the increase of the first feature and decreases with the decrease of the first feature.

9. The method for quantitative verification of quantum state manipulation quality for optical quantum chips according to claim 1, characterized in that, The specific steps for comparing each of the second feature quantities with the ideal evolution direction template to obtain the deviation measure corresponding to each time sub-window include: extracting the evolution direction vectors corresponding to each non-empty time sub-window located before the start position of the second half of the acquisition window; for each evolution direction vector, when the evolution direction vector is a valid vector, calculating the inner product of the evolution direction vector and the ideal evolution direction template, and performing an inverse cosine operation on the inner product to obtain the corresponding angle value, and using the angle value as the deviation measure of the time sub-window; when the evolution direction vector is an invalid vector, setting the deviation measure of the time sub-window as... Right angle value; For each non-empty time sub-window in the second half of the acquisition window, when the first feature value is not greater than the noise floor threshold, the deviation measure of each time sub-window is set to zero; when the first feature value is greater than the noise floor threshold, the deviation measure of the last non-empty time sub-window in the second half is set to zero, and the deviation measures of the remaining non-empty time sub-windows in the second half, except for the last one, are set to a mapping value between zero and the decision tolerance. The mapping value increases as the difference between the normalized count rate vector of the time sub-window and the normalized count rate vector of the last non-empty time sub-window increases.

10. The method for quantizing and verifying the quality of quantum state manipulation for optical quantum chips according to claim 1, characterized in that, The specific method for assigning continuous contribution weights to each time sub-window based on the comparison between the deviation metric of each time sub-window and the decision tolerance is as follows: For each non-empty time sub-window, when the deviation metric of the time sub-window is equal to zero, its contribution weight is assigned the maximum value of 1; when the deviation metric of the time sub-window is greater than zero, the contribution weight is calculated based on the ratio of the deviation metric to the decision tolerance. The contribution weight decreases monotonically as the ratio increases, and when the deviation metric is equal to the decision tolerance, the contribution weight is half. For empty windows, their contribution weight is set to zero; the specific method for obtaining the reconstructed verification photon distribution data by weighted averaging of the normalized count rate matrix based on the contribution weight is as follows: multiply the normalized count rate vector of each time sub-window by the corresponding contribution weight, sum them, and then divide by the sum of the contribution weights to obtain the reconstructed verification photon distribution data.

Images