A photocatalytic excited state quasi-reaction path self-adaptive enhanced sampling system and method

Abstract

The application relates to the field of photocatalysis and discloses a photocatalysis excited state quasi-reaction path self-adaptive enhancement sampling system and method, which comprises the following steps: introducing atomic structure parameters of a photocatalysis system containing a catalyst, a reaction substrate and a solvent and light excitation conditions, constructing a multi-electron excited state potential energy surface, selecting initial collective variables and setting initial enhancement sampling bias parameters; performing enhancement sampling according to a preset round, extracting electron density difference data to generate new collective variables every round, and dynamically adjusting bias distribution and trajectory allocation rules of the next round of sampling; and collecting a quasi-reaction path set after sampling convergence and outputting corresponding analysis data. Through the means of updating the collective variables based on the electron density difference data after each round of sampling, the dynamic electronic coupling characteristics of the excited state are dynamically included, and the sampling efficiency of a key reaction region and the completeness of the quasi-reaction path coverage are improved.

Description

Technical Field

[0001] This invention relates to the field of photocatalysis technology, and more specifically, to an adaptive enhanced sampling system and method for photocatalytic excited-state quasi-reaction pathways. Background Technology

[0002] As the application scenarios of photocatalysis technology continue to expand in clean energy conversion, pollutant degradation, and fine organic synthesis, the analysis of its microscopic reaction mechanisms has gradually become a core supporting link for the rational optimization design of photocatalytic materials. The core processes of photocatalytic reactions occur in the multi-electron excited state after molecules absorb photons, accompanied by complex microscopic behaviors such as non-adiabatic transitions, conical crossings, transient intermediate evolution, and competition among multiple reaction channels. Molecular simulation and enhanced sampling techniques are key tools for overcoming the limitations of conventional molecular dynamics timescales and capturing rare reaction events, and are now gradually being applied to the computational study of microscopic mechanisms in various homogeneous and heterogeneous photochemical systems.

[0003] Traditional excited-state enhanced sampling methods rely on pre-constructed static collective variable libraries, which can only select conventional structural variables such as bond length, bond angle, and dihedral angle as sampling criteria. This fails to adapt to the dynamic transient exciton coupling characteristics generated during carrier separation at the photocatalytic interface, easily leading to sampling biases in key reaction regions and omissions of reaction branches. Furthermore, the fixed sampling bias parameters and trajectory allocation rules cannot dynamically adapt to changes in the potential energy surface as the sampling process progresses, resulting in low sampling efficiency in excited-state transition and electronic-state intersection regions. This makes it difficult to fully cover all quasi-reaction pathways, significantly limiting the completeness and reliability of the calculation results and failing to meet the high-precision mechanistic analysis requirements of complex multiphase photocatalytic systems.

[0004] Therefore, it is necessary to design an adaptive enhanced sampling method for photocatalytic excited-state quasi-reaction pathways to address the problems of existing technologies, such as reliance on static preset libraries for collective variables, inability to incorporate excited-state dynamic electronic coupling characteristics in real time, insufficient rationality in the allocation of sampling resources, low sampling efficiency in key excited-state reaction regions due to incomplete coverage of quasi-reaction pathways, and difficulty in adapting to the analytical needs of complex systems. Summary of the Invention

[0005] In view of this, the present invention proposes an adaptive enhanced sampling system and method for photocatalytic excited-state quasi-reaction pathways, aiming to solve at least one of the problems in the background art.

[0006] In one aspect, this invention proposes an adaptive enhancement sampling method for photocatalytic excited-state quasi-reaction pathways, comprising: Step 1: Import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing the catalyst, reaction substrate and solvent. Construct a multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions. Select initial collective variables and set initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface. Step 2: Based on the initial enhanced sampling bias parameters, initiate multiple sets of parallel molecular dynamics simulation trajectories, perform the first round of enhanced sampling calculations, and obtain the atomic evolution trajectory and the distribution of the bias potential function that forms the first round of accumulation; Step 3: Divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to the evolution trajectories of all atoms after each round of enhanced sampling; Step 4: Based on the electron density difference data, new features associated with the reaction process are screened, new collective variables are generated according to the new features, and the current set of collective variables is updated according to the new collective variables. Step 5: Based on the updated set of collective variables and the accumulated bias potential function distribution, adjust the bias distribution and trajectory assignment rules for the next round of enhanced sampling, and execute the next round of enhanced sampling based on the adjusted bias distribution and trajectory assignment rules; repeat steps 3 to 5 until the sampling meets the preset convergence conditions. Step 6: Summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

[0007] Furthermore, the atomic structure parameters include the atomic coordinates, atom types, charge distribution, and molecular topology of the photocatalytic system; The photoexcitation conditions include the excitation light wavelength, light intensity, and polarization direction; The process of constructing a multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions, and selecting initial collective variables and setting initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface, includes: Based on time-dependent density functional theory, the energy gradients and transition dipole moments of multiple excited states are calculated to construct a multi-electron excited state energy surface, which includes the energy distribution of the ground state and at least three low-excited states. Select one or more of the interatomic distance, bond angle, or dihedral angle from the reaction coordinates of the potential energy surface as the initial collective variable; The initial enhanced sampling bias parameters include bias potential height, bias potential width, and bias potential deposition frequency.

[0008] Furthermore, based on the initial enhanced sampling bias parameters, multiple sets of parallel molecular dynamics simulation trajectories are initiated to perform the first round of enhanced sampling calculations, obtaining atomic evolution trajectories and the process of forming the first round of accumulated bias potential function distribution, including: Based on the preset number of parallel trajectories, the initial collective variable space is divided into several sub-regions, and each sub-region is assigned an independent molecular dynamics simulation trajectory. Gaussian bias potentials are deposited on each trajectory with the same bias potential height and width to form the first round of cumulative bias potential function distribution; The maximum and minimum values ​​of the bias potential function on each trajectory are monitored in real time. When the difference between the maximum and minimum values ​​exceeds the preset bias potential uniformity threshold, the first round of sampling is paused and the bias potential deposition frequency is adjusted.

[0009] Furthermore, the step of dividing the enhanced sampling process into several rounds according to a preset number of simulation steps, and extracting the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling, includes: The total number of simulation steps is divided into multiple rounds according to a preset fixed interval, and each round contains the same number of simulation steps; After each round, starting from the end configuration of each trajectory, the electron density distribution under that configuration is obtained by single-point energy calculation, and the difference is made with the electron density distribution at the initial moment to obtain electron density difference data. Statistical analysis was performed on the electron density difference data of all trajectories within the same round to obtain the average difference value and the standard deviation of the difference; When the average difference value is less than the preset difference noise threshold, it is determined that the current round of sampling has not generated effective electronic rearrangement information, and the current round is marked as an invalid round.

[0010] Furthermore, based on the electron density difference data, the process of filtering new features associated with the reaction process, generating new collective variables based on the new features, and updating the current set of collective variables based on the new features includes: Principal component analysis was performed on the effective electron density difference data of each round to extract the top-preset number of differential principal components ranked by variance contribution rate. Calculate the linear correlation coefficient between each difference principal component and the current collective variable. When the correlation coefficient between a certain difference principal component and all current collective variables is lower than the preset correlation coefficient threshold, the difference principal component is determined to be a new feature. Based on the atomic coordinate change pattern corresponding to the newly added feature, a new collective variable is generated, which is one of the following: interatomic distance, bond angle, or coordination number. The new collective variable is added to the current set of collective variables, forming an updated set of collective variables.

[0011] Furthermore, the process of adjusting the bias distribution and trajectory allocation rules for the next round of augmented sampling based on the updated set of collective variables and the accumulated bias potential function distribution, and executing the next round of augmented sampling based on the adjusted bias distribution and trajectory allocation rules, includes: Based on the changes in the dimensions of the updated collective variable set, the coverage of each trajectory in the expanded collective variable space is re-acquired; When the addition of a new collective variable leads to an increase in the dimension of the collective variable, the number of parallel trajectories is expanded according to the increase in dimension, and different initial positions are assigned to the new trajectories on the dimension of the new collective variable. The accumulated bias potential function distribution is projected and stretched along the direction of the newly added collective variable, and the bias potential deposition height is reduced. When the number of newly added collective variables is zero, the number of parallel trajectories and the bias potential parameter remain unchanged, and only the distribution of the starting positions of each trajectory in the original variable space is adjusted.

[0012] Furthermore, the process of repeatedly executing steps three to five until the sampling meets the preset convergence condition includes: After each cycle, the root mean square deviation of the free energy surface between the current cycle and the previous cycle is calculated as a convergence criterion. When the root mean square deviation is less than the preset convergence threshold and remains at the preset number of rounds, it is determined that the sampling meets the preset convergence condition, and the loop is terminated. When the root mean square deviation is greater than or equal to the convergence threshold, continue to execute the next round of the loop; When the number of iterations exceeds the preset maximum number of iterations, the loop stops, and the result of the last iteration is used as the final sampling result. At the same time, a warning sign indicating incomplete convergence is generated.

[0013] Furthermore, the process of summarizing all atomic evolution trajectories and classifying these trajectories to obtain a set of quasi-reaction pathways includes: Extract the initial and final configurations of each trajectory from all atomic evolution trajectories, and calculate the root mean square atomic displacements of the final configuration relative to the initial configuration. The trajectory is divided into reaction path and non-reaction path based on the root mean square of atomic displacement. The trajectory with the root mean square of atomic displacement greater than the preset reaction displacement threshold is classified as the reaction path, otherwise it is classified as the non-reaction path. Trajectories belonging to the reaction pathway are clustered according to the structural similarity of the final product, with each cluster corresponding to a quasi-reaction pathway. By averaging multiple trajectories in each category, a representative reaction coordinate sequence of the quasi-reaction path is obtained.

[0014] Furthermore, the process of outputting the analytical data corresponding to the set of quasi-reaction pathways includes: Output the free energy profile for each quasi-reaction path, including the barrier height along the reaction coordinate, reaction energy, and intermediate stabilization energy; The ratio of the number of samples for each quasi-reaction path to the total number of reaction trajectories is obtained as an estimate of the probability of occurrence of that quasi-reaction path. When the barrier height of a quasi-reaction path is lower than a preset low barrier threshold and the probability of its occurrence is greater than a preset probability threshold, the quasi-reaction path is marked as a primary reaction path; when the barrier height is higher than a preset high barrier threshold, the quasi-reaction path is marked as a secondary path. The output is sorted in order of priority for the main reaction path and the secondary path, and provides the corresponding atomic structure animation file for each path.

[0015] Compared with existing technologies, the advantages of this invention are as follows: This invention employs a round-based enhanced sampling execution method. After each round of sampling, it extracts electron density difference data based on atomic evolution trajectories, generates new collective variables, and dynamically updates the collective variable set. Then, based on the updated collective variables, it adjusts the bias distribution and trajectory allocation rules for the next round of sampling. This invention can incorporate the dynamic electronic coupling characteristics of excited states into the sampling basis without relying on a statically preset collective variable library. It effectively adapts to the complex characteristics of multi-electronic state coupling in photocatalytic excited states, improves the sampling efficiency of key regions such as excited state transition states and electronic state intersections, and completely captures multi-channel competing quasi-reaction paths, providing comprehensive and reliable computational results support for the microscopic mechanism analysis of photocatalytic reactions. Furthermore, this invention, through parallel trajectory scheduling and dynamic adaptation of bias potentials, ensures that sampling resources are tilted towards high-value reaction regions, improving overall computational efficiency while maintaining path coverage integrity and avoiding the waste of computational resources caused by invalid sampling. The entire process is highly automated, allowing researchers to complete the full analysis of multi-channel excited-state reaction pathways without manually adjusting complex sampling parameters, significantly lowering the barrier to entry. At the same time, the computational process is controllable, providing stable computational support for the rational design of photocatalytic materials.

[0016] On the other hand, the present invention also proposes an adaptive enhancement sampling system for photocatalytic excited-state quasi-reaction pathways, applied to a photocatalytic excited-state quasi-reaction pathway adaptive enhancement sampling method, comprising: The system import module is used to import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing the catalyst, reaction substrate and solvent, and to construct a multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions; the system import module also selects initial collective variables and sets initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface. The parallel sampling module is used to initiate multiple sets of parallel molecular dynamics simulation trajectories based on the initial enhanced sampling bias parameters, perform the first round of enhanced sampling calculations, obtain atomic evolution trajectories and form the first round of accumulated bias potential function distribution; The round management module is used to divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling; The feature filtering module is used to filter new features associated with the reaction process based on electron density difference data; the feature filtering module is also used to generate new collective variables based on the new features, and update the current set of collective variables based on the new collective variables. An automatic adjustment module is used to adjust the bias distribution and trajectory allocation rules of the next round of enhanced sampling based on the updated set of collective variables and the accumulated bias potential function distribution; the automatic adjustment module also executes the next round of enhanced sampling based on the adjusted bias distribution and trajectory allocation rules; the automatic adjustment module also executes steps three to five in a loop until the sampling meets the preset convergence conditions; The path merging module is used to summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

[0017] It is understandable that the above-mentioned photocatalytic excited-state quasi-reaction pathway adaptive enhancement sampling system and method have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of an adaptive enhanced sampling method for photocatalytic excited-state quasi-reaction pathways, provided in an embodiment of the present invention.

[0019] Figure 2 This is a functional block diagram of a photocatalytic excited-state quasi-reaction path adaptive enhancement sampling system provided in an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] See Figure 1 As shown in some embodiments of this application, this embodiment provides an adaptive enhancement sampling method for photocatalytic excited-state quasi-reaction pathways, including the following steps: S100, import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing catalyst, reaction substrate and solvent, construct the multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions, select the initial collective variable and set the initial enhanced sampling bias parameter based on the multi-electron excited potential energy surface; In a specific application scenario, researchers can import pre-built photocatalytic architecture files into a computing workstation. For example, this could be a solvation system of CO2 and water molecules adsorbed on anatase titanium dioxide surface. This system contains complete coordinate information of catalyst atoms, reactant atoms, and solvent water molecules. Simultaneously, corresponding excitation light parameters, such as 365nm ultraviolet light, are input. This architecture and excitation parameters serve as input data for subsequent potential surface construction and sampling initialization. Quantum chemical calculations are then performed on the input system parameters to construct the multi-electron excited potential surface, and initial sampling criteria and parameters are selected.

[0022] The construction of the excited state energy surface can be calculated using time-dependent density functional theory. This calculation method has been validated in numerous photochemical systems and can accurately describe the energy distribution and electronic transition characteristics of low-excited states. After inputting the system architecture parameters and photoexcitation conditions, the calculation program performs single-point energy and energy gradient calculations on the ground state and multiple low-excited states of the system, obtaining the energy distribution and atomic force information under each electronic state, and constructing the multi-electron excited state energy surface.

[0023] Specifically, atomic structure parameters include the atomic coordinates, atom types, charge distribution, and molecular topology of the photocatalytic system; Photoexcitation conditions include the wavelength, intensity, and polarization direction of the excitation light; The process of constructing a multi-electron excited potential energy surface based on atomic structure parameters and photoexcitation conditions, and selecting initial collective variables and setting initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface, includes: Based on time-dependent density functional theory, the energy gradients and transition dipole moments of multiple excited states are calculated, and a multi-electron excited state energy surface is constructed. The multi-electron excited state energy surface includes the energy distribution of the ground state and at least three low-excited states. Select one or more of the interatomic distance, bond angle, or dihedral angle from the reaction coordinates of the potential energy surface as the initial collective variable; Initial enhancement sampling bias parameters include bias potential height, bias potential width, and bias potential deposition frequency.

[0024] For example, in the above-mentioned embodiment of titanium dioxide photocatalytic CO2 reduction, the constructed multi-electron excited potential energy surface includes the energy distribution of the ground state S0 and the first, second, and third excited states S1, S2, and S3, and the transition dipole moments between each excited state are calculated. The initial collective variables are selected as the CO bond length, OCO bond angle, and the perpendicular distance between the CO2 molecule and the catalyst surface as the initial sampling basis. The initial enhanced sampling bias parameters are set as bias potential height (determining the bias potential amplitude of each deposition), bias potential width (determining the coverage range of the bias potential in the collective variable space), and bias potential deposition frequency (determining the number of deposition intervals). These parameters together determine the bias potential application intensity and distribution density of the first round of enhanced sampling, which is used to drive the system to cross the reaction energy barrier along the initial collective variable direction.

[0025] S200. Based on the initial enhanced sampling bias parameters, multiple sets of parallel molecular dynamics simulation trajectories are initiated, and the first round of enhanced sampling calculations are performed to obtain the atomic evolution trajectory and the distribution of the bias potential function that forms the first round of accumulation. For example, based on the initially set bias parameters, multiple parallel molecular dynamics simulation trajectories can be started, bias potentials can be applied synchronously on each trajectory, the first round of enhanced sampling simulation can be executed, the evolution data of atomic positions over time can be recorded, and the bias potential distribution function of the first round can be accumulated.

[0026] Specifically, the process involves initiating multiple parallel molecular dynamics simulations to perform the first round of enhanced sampling calculations, obtaining atomic evolution trajectories, and forming the initial accumulated bias potential function distribution, including: Based on the preset number of parallel trajectories, the initial collective variable space is divided into several sub-regions, and each sub-region is assigned an independent molecular dynamics simulation trajectory. Gaussian bias potentials are deposited on each trajectory with the same bias potential height and width to form the first round of cumulative bias potential function distribution; The maximum and minimum values ​​of the bias potential function on each trajectory are monitored in real time. When the difference between the maximum and minimum values ​​exceeds the preset bias potential uniformity threshold, the first round of sampling is paused and the bias potential deposition frequency is adjusted.

[0027] Specifically, the multidimensional space formed by the initial collective variables can be uniformly divided into sub-regions corresponding to the number of parallel trajectories, typically 4 to 16, depending on the number of available computational cores. Each trajectory initiates the simulation from the center of its corresponding sub-region. This ensures that different trajectories start from different regions of the collective variable space, avoiding multiple trajectories from getting trapped in the same local region. Each sub-region is assigned an independent molecular dynamics simulation trajectory, and each trajectory runs independently within its own sub-region for enhanced sampling computation. During the simulation, each trajectory deposits a Gaussian bias potential at a fixed time step. As the simulation progresses, the deposited Gaussian bias potential gradually accumulates, forming the first-round accumulated bias potential function distribution. This distribution records the historical frequency information of each trajectory being accessed in the collective variable space. To ensure uniform filling of the bias potential in the collective variable space and avoid over-biasing in some areas while under-biasing in others, this embodiment monitors the maximum and minimum values ​​of the bias potential function on each trajectory in real time during the first round of sampling. When the difference between the two exceeds a preset threshold, it indicates that the uniformity of the bias potential distribution is insufficient at the current deposition frequency. At this time, the first round of sampling is paused and the bias potential deposition frequency is adjusted (e.g., reducing the deposition frequency to slow down the accumulation rate of the bias potential in the visited areas), so that the subsequent deposited bias potential can fill the entire collective variable space more uniformly and avoid simulation distortion caused by excessive local bias.

[0028] S300. Divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling; For example, the number of simulation steps for each round of sampling can be preset. Sampling is paused after a fixed number of simulation steps are completed. The trajectory data generated in this round is analyzed to extract the corresponding electron density difference information for subsequent new feature mining.

[0029] Specifically, the enhanced sampling process is divided into several rounds according to a preset number of simulation steps, and the electron density difference data corresponding to the evolution trajectories of all atoms after each round of enhanced sampling are extracted, including: The total number of simulation steps is divided into multiple rounds according to a preset fixed interval, and each round contains the same number of simulation steps; After each round, starting from the end configuration of each trajectory, the electron density distribution under that configuration is obtained by single-point energy calculation, and the difference is made with the electron density distribution at the initial moment to obtain electron density difference data. Statistical analysis was performed on the electron density difference data of all trajectories within the same round to obtain the average difference value and the standard deviation of the difference; When the average difference value is less than the preset difference noise threshold, it is determined that the current round of sampling has not generated effective electronic rearrangement information, and the current round is marked as an invalid round.

[0030] Specifically, to enhance the real-time sensing of changes in the electronic structure of the system during the sampling process, this embodiment introduces electron density difference data as a criterion for the reaction progress. The total number of simulation steps is divided into multiple rounds at preset fixed intervals, such as every 5,000 steps or every 10,000 steps. Each round contains the same number of simulation steps, which facilitates data comparison and statistical analysis between rounds.

[0031] Specifically, after each round, starting from the final configuration of each trajectory, single-point energy calculations are used to obtain the electron density distribution under that configuration. These single-point energy calculations are static electronic structure calculations, without geometric optimization or molecular dynamics evolution; they only obtain the electron density corresponding to the current nuclear configuration. The electron density distribution is then subtracted point-by-point from the electron density distribution at the initial moment (i.e., when the system is in the reactant ground state configuration) to obtain electron density difference data. This difference data reflects the redistribution of the system's electron density in three-dimensional space after this round of enhanced sampling, and is a direct basis for determining whether chemical bond breaking, formation, and charge transfer have occurred. Statistical analysis is performed on the electron density difference data of all trajectories within the same round to obtain the average difference value and the standard deviation of the difference. When the average difference value is less than the preset difference noise threshold, it indicates that the electron density changes of each trajectory in the current round have not significantly exceeded the computational noise level. The system has not generated effective electron rearrangement information (i.e., no effective reaction-related information) in this round. At this time, the round is marked as an invalid round and excluded in the subsequent new feature screening to avoid the erroneous expansion of collective variables due to interference from invalid data.

[0032] S400: Based on electron density difference data, new features associated with the reaction process are screened, new collective variables are generated based on the new features, and the current set of collective variables is updated based on the new collective variables. For example, the electron density difference data of effective rounds can be analyzed to uncover reaction-related features that are not covered by existing collective variables, and these features can be transformed into new collective variables to supplement the current variable set.

[0033] Specifically, the process of screening new features associated with the reaction process based on electron density difference data, generating new collective variables based on these new features, and updating the current set of collective variables based on these new features includes: Principal component analysis was performed on the effective electron density difference data of each round to extract the top-preset number of differential principal components ranked by variance contribution rate. Calculate the linear correlation coefficient between each difference principal component and the current collective variable. When the correlation coefficient between a certain difference principal component and all current collective variables is lower than the preset correlation coefficient threshold, the difference principal component is determined to be a new feature. Based on the atomic coordinate change pattern corresponding to the newly added feature, a new collective variable is generated. The new collective variable is one of the following: interatomic distance, bond angle, or coordination number. The new collective variable is added to the current set of collective variables, forming an updated set of collective variables.

[0034] Specifically, electron density difference data typically exhibits high dimensionality, containing tens of thousands of grid points in space. Directly using it as a collective variable would lead to the curse of dimensionality. Therefore, this embodiment employs principal component analysis to reduce the dimensionality of the effective rounds of electron density difference data, extracting the top 3 to 5 differential principal components (DPCs) based on their variance contribution rate. These DPCs represent several independent patterns of the most significant electron density variations. To determine whether these DPCs are already included in the current set of collective variables, the linear correlation coefficients between each DPC and all current collective variables are calculated. When the correlation coefficients between a DPC and all current collective variables are below a preset threshold (e.g., 0.3), it indicates that the atomic motion pattern represented by that DPC cannot be represented by existing collective variables, and the nuclear motion pattern corresponding to that DPC has not been fully represented; thus, it is identified as a new feature. Based on the atomic coordinate change patterns corresponding to the new features—that is, by analyzing the load distribution of the DPCs in atomic coordinate space to determine which atomic relative motions contribute to the main variance of the DPC—new collective variables are generated. The newly generated collective variable can be one of the following: interatomic distance (describing the bond length change between two specific atoms), bond angle (describing the angular change between three atoms), or coordination number (describing the change in the number of neighboring atoms around an atom). The specific type is determined according to the specific nuclear motion mode corresponding to the principal component of the difference. The new collective variable is added to the current set of collective variables to form an updated set of collective variables, enabling subsequent rounds of enhanced sampling to simultaneously apply biases to more reaction degrees of freedom in the expanded collective variable space.

[0035] S500. Based on the updated set of collective variables and the cumulative distribution of bias potential functions, adjust the bias distribution and trajectory allocation rules for the next round of enhanced sampling, and execute the next round of enhanced sampling based on the adjusted bias distribution and trajectory allocation rules; repeat steps three to five until the sampling meets the preset convergence conditions. For example, the sampling configuration for the next round can be adapted and adjusted based on the updated set of collective variables, including the distribution of the bias potential and the number and location of parallel trajectories. After the adjustment is completed, the next round of sampling is started, and so on iteratively until the sampling results reach the convergence criterion.

[0036] Specifically, based on the updated set of collective variables and the cumulative distribution of the bias potential function, the bias distribution and trajectory assignment rules for the next round of augmented sampling are adjusted, and the next round of augmented sampling is executed based on the adjusted bias distribution and trajectory assignment rules, including: Based on the changes in the dimensions of the updated collective variable set, the coverage of each trajectory in the expanded collective variable space is re-acquired; When the addition of a new collective variable leads to an increase in the dimension of the collective variable, the number of parallel trajectories is expanded according to the increase in dimension, and different initial positions are assigned to the new trajectories on the dimension of the new collective variable. The accumulated bias potential function distribution is projected and stretched along the direction of the newly added collective variable, and the bias potential deposition height is reduced. When the number of newly added collective variables is zero, the number of parallel trajectories and the bias potential parameter remain unchanged, and only the distribution of the starting positions of each trajectory in the original variable space is adjusted.

[0037] Specifically, when the collective variable set is expanded, the bias distribution and trajectory allocation rules of the enhanced sampling need to be adjusted accordingly to adapt to the new variable space dimension. This embodiment first re-evaluates the coverage of each existing trajectory in the expanded collective variable space based on the dimensional changes of the updated collective variable set—that is, the number of newly added collective variables. When the addition of new collective variables leads to an increase in the collective variable dimension, the original number of parallel trajectories may not be sufficient to fully cover the higher-dimensional variable space. In this case, the number of parallel trajectories is expanded according to the dimension increase ratio (for example, when the collective variable dimension increases from 2 to 3, the number of trajectories is expanded to 1.5 or 2 times the original number), and different initial positions are assigned to the new trajectories in the new collective variable dimension, so that the new trajectories have different starting values ​​in the new dimension, thereby accelerating the exploration of the new dimension space. Simultaneously, the accumulated bias potential function distribution from the previous round is projected and stretched along the direction of the new collective variable, so that the existing bias potential information can continue to play a role in the expanded space, and the bias potential deposition height is appropriately reduced to slow down the accumulation rate of the bias potential in the new dimension space, avoiding the obstacle of free exploration in the new dimension due to excessively strong bias potential. When the number of newly added collective variables is zero (i.e. no new features are selected in the current round), it indicates that the current set of collective variables can effectively describe the reaction process. At this time, the number of parallel trajectories and the bias potential parameter remain unchanged. Only the distribution of the starting positions of each trajectory in the original variable space is adjusted (e.g., the trajectories are redistributed to the insufficiently covered areas in the variable space) to further deepen the sampling coverage of the existing variable space, shifting it to the insufficiently sampled areas and improving the uniformity of spatial coverage.

[0038] Specifically, the process of repeatedly executing steps three through five until the sampling meets the preset convergence condition includes: After each cycle, the root mean square deviation of the free energy surface between the current cycle and the previous cycle is calculated as a convergence criterion. When the root mean square deviation is less than the preset convergence threshold and remains at the preset number of rounds, the sampling is determined to meet the preset convergence condition, and the loop is terminated. When the root mean square deviation is greater than or equal to the convergence threshold, continue to execute the next round of the loop; When the number of iterations exceeds the preset maximum number of iterations, the loop stops, and the result of the last iteration is used as the final sampling result. At the same time, a warning sign indicating incomplete convergence is generated.

[0039] Specifically, the termination condition for enhanced sampling needs to balance sampling sufficiency and computational resource efficiency. In this embodiment, after each iteration, the free energy surface is reconstructed based on the accumulated bias potential function distribution of the current iteration. The root mean square deviation (RMS) of the free energy surface between the current and previous iterations is calculated. This RMS deviation reflects the magnitude of change in the free energy estimate between adjacent iterations and is an objective quantitative indicator of whether the sampling has converged. When the RMS deviation is less than a preset convergence threshold and remains at a preset number of iterations (e.g., 3 consecutive iterations), it indicates that the free energy surface has basically stabilized, and new sampling no longer significantly changes the free energy estimate. At this point, the sampling is deemed to meet the preset convergence condition, and the loop terminates. When the RMS deviation is greater than or equal to the convergence threshold, it indicates that the free energy surface has not yet converged, and the next iteration continues. In addition, to prevent infinite loops caused by the high complexity of the potential energy surface, this embodiment also presets a maximum limit on the number of cycles (such as 50 or 100 cycles). If the number of cycles reaches the preset maximum limit and still fails to converge, the simulation will be forcibly stopped, and the result of the last cycle will be used as the final sampling result. The current result will be output and a warning sign will be generated to remind researchers that the sampling of the system has not yet fully converged and may require further verification or extended sampling.

[0040] S600: Summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

[0041] For example, after the sampling cycle terminates, all atomic evolution trajectories generated in all rounds are summarized, the trajectories are classified and clustered, the entire quasi-reaction path network is sorted out, and the corresponding kinetic and thermodynamic analysis data are output.

[0042] Specifically, all atomic evolution trajectories are compiled and categorized to obtain a set of quasi-reaction pathways, including: Extract the initial and final configurations of each trajectory from all atomic evolution trajectories, and calculate the root mean square atomic displacements of the final configuration relative to the initial configuration. The trajectory is divided into reaction path and non-reaction path based on the root mean square of atomic displacement. The trajectory with the root mean square of atomic displacement greater than the preset reaction displacement threshold is classified as the reaction path, otherwise it is classified as the non-reaction path. Trajectories belonging to the reaction pathway are clustered according to the structural similarity of the final product, with each cluster corresponding to a quasi-reaction pathway. By averaging multiple trajectories in each category, a representative reaction coordinate sequence of the quasi-reaction path is obtained.

[0043] Specifically, all atomic evolution trajectories contain a large number of ineffective trajectories where no chemical reaction occurs (i.e., trajectories where the system oscillates within a local region of the potential energy surface but does not cross the energy barrier) and effective trajectories where a reaction occurs. To extract quasi-reaction paths from the mixed trajectories, this embodiment first extracts the initial configuration (the configuration at the start of the trajectory) and the final configuration (the configuration at the final moment of the trajectory) from each trajectory. The root mean square value of all atomic displacements relative to the initial configuration is calculated, which comprehensively reflects the overall displacement amplitude of the entire trajectory in nuclear coordinate space. Based on the root mean square of atomic displacements, the trajectories are divided into reactive paths and non-reactive paths—trajectories with a root mean square of atomic displacements greater than a preset reaction displacement threshold are classified as reactive paths (indicating that the system has undergone a significant configurational change in this trajectory and crossed the energy barrier), otherwise they are classified as non-reactive paths (indicating that the system has not crossed the energy barrier). Trajectories belonging to the reaction path are clustered according to the structural similarity of the final product, for example, by using the atomic coordinate overlap matrix of the final configuration or molecular fingerprint similarity as a distance metric. Hierarchical clustering or K-means clustering algorithms are used for merging, and each cluster corresponds to a quasi-reaction path (i.e., a set of trajectories starting from the same reactant and reaching the same product branch). Frame-by-frame averaging is performed on multiple trajectories in each cluster to obtain a representative reaction coordinate sequence for that quasi-reaction path, which is used for subsequent energy barrier calculation and path analysis. This sequence describes the average evolution process from reactant to product along the path.

[0044] Specifically, the process of outputting analytical data corresponding to the set of quasi-reaction pathways includes: Output the free energy profile for each quasi-reaction path, including the barrier height along the reaction coordinate, reaction energy, and intermediate stabilization energy; The ratio of the number of samples for each quasi-reaction path to the total number of reaction trajectories is obtained as an estimate of the probability of occurrence of that quasi-reaction path. When the barrier height of a quasi-reaction path is lower than a preset low barrier threshold and the probability of its occurrence is greater than a preset probability threshold, the quasi-reaction path is marked as a primary reaction path; when the barrier height is higher than a preset high barrier threshold, the quasi-reaction path is marked as a secondary path. The output is sorted in order of priority for the main reaction path and the secondary path, and provides the corresponding atomic structure animation file for each path.

[0045] For example, for each quasi-reaction path, the free energy change curve along the reaction coordinates is calculated based on the free energy surface, and key parameters such as barrier height, reaction heat, and intermediate energy are extracted. Simultaneously, the number of sampling trajectories corresponding to each path is counted, and its proportion among all reaction trajectories is calculated as the probability of occurrence for that path. Based on the barrier height and probability of occurrence, the paths are divided into primary and secondary reaction paths, sorted by priority, and output. Animations of the atomic structure evolution for each path are also generated, allowing researchers to visually observe the reaction process.

[0046] Specifically, the analytical data output of the quasi-reaction pathway set needs to provide users with clear and prioritized reaction information. This application outputs the free energy profile for each quasi-reaction pathway, including the barrier height along the reaction coordinates (energy difference from reactant to transition state), reaction energy (energy difference of product relative to reactant), and intermediate stabilization energy (energy difference of intermediate relative to adjacent transition state if intermediate state exists). These data are core thermodynamic parameters for evaluating the feasibility and efficiency of photocatalytic reactions. Simultaneously, the ratio of the number of samples for each quasi-reaction pathway to the total number of reaction trajectories is obtained as an estimate of the occurrence probability of that quasi-reaction pathway. This estimate reflects the relative competition intensity of different reaction branches under the same conditions. When the barrier height of a quasi-reaction pathway is lower than a preset low barrier threshold (e.g., 5 kcal / mol) and the occurrence probability is greater than a preset probability threshold (e.g., 10%), the quasi-reaction pathway is marked as the main reaction pathway, indicating that this pathway has the fastest reaction rate and the highest branch ratio under the reaction conditions. When the potential barrier height exceeds a preset high barrier threshold (e.g., 20 kcal / mol), the quasi-reaction path is marked as a secondary path, indicating that the path does not have a competitive advantage thermodynamically or kinetically. The output is sorted in order of priority for primary reaction paths and subsequent secondary paths, and includes an atomic structure animation file for each path, allowing users to visually observe the details of atomic motion along each quasi-reaction path.

[0047] Understandably, this invention constructs a multi-electron excited potential energy surface based on atomic structure parameters and photoexcitation conditions, providing an accurate energy landscape foundation for enhanced sampling. By dividing the sampling process into rounds with a preset number of simulation steps, each round extracts electron density difference data and uses this data to filter new features and generate new collective variables. This allows the collective variable set to automatically expand as the reaction progresses during sampling, avoiding sampling failures caused by incomplete preset collective variables. By adaptively adjusting the bias distribution and trajectory allocation rules based on the updated collective variable set, it achieves smooth migration of the bias potential in the expanded variable space and dynamic optimization of trajectory resources. Real-time monitoring and convergence determination of the root mean square deviation of the free energy surface ensures the thermodynamic reliability of the sampling results. Through the classification and multi-path analysis output of all trajectories, it provides users with complete reaction information including free energy profiles, occurrence probabilities, and primary and secondary path divisions. This invention, without prior knowledge of the reaction coordinates, can adaptively discover and accurately characterize multiple quasi-reaction paths of the photocatalytic excited state, significantly improving the efficiency and accuracy of research on complex photocatalytic reaction mechanisms. This embodiment can incorporate the dynamic electronic coupling characteristics of the excited state into the sampling basis without relying on a statically preset collective variable library. It effectively adapts to the complex characteristics of multi-electronic state coupling in photocatalytic excited states, improves the sampling efficiency of key regions such as excited state transition states and electronic state crossovers, and completely captures the quasi-reaction pathways of multi-channel competition, providing comprehensive and reliable calculation results support for the microscopic mechanism analysis of photocatalytic reactions.

[0048] Based on a general inventive concept, the present invention also provides an adaptive enhancement sampling system for photocatalytic excited-state quasi-reaction pathways, used to implement the above-described method embodiments.

[0049] like Figure 2 As shown, an adaptive enhancement sampling system for photocatalytic excited-state quasi-reaction pathways includes: The system import module is used to import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing the catalyst, reaction substrate and solvent, and to construct the multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions. The system import module also selects the initial collective variables and sets the initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface. The parallel sampling module is used to initiate multiple sets of parallel molecular dynamics simulation trajectories based on the initial enhanced sampling bias parameters, perform the first round of enhanced sampling calculations, obtain the atomic evolution trajectory and form the distribution of the bias potential function accumulated in the first round; The round management module is used to divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling; The feature filtering module is used to filter new features associated with the reaction process based on electron density difference data; the feature filtering module is also used to generate new collective variables based on the new features, and update the current set of collective variables based on the new collective variables. An automatic adjustment module is used to adjust the bias distribution and trajectory allocation rules of the next round of enhanced sampling based on the updated set of collective variables and the accumulated bias potential function distribution; the automatic adjustment module also executes the next round of enhanced sampling based on the adjusted bias distribution and trajectory allocation rules; the automatic adjustment module also executes steps three to five in a loop until the sampling meets the preset convergence conditions; The path merging module is used to summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

[0050] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0051] It is understandable that an adaptive enhanced sampling system and method for photocatalytic excited-state quasi-reaction pathways would have the same beneficial effects, and will not be elaborated further here.

[0052] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An adaptive enhanced sampling method for photocatalytic excited-state quasi-reaction pathways, characterized in that, include: Step 1: Import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing the catalyst, reaction substrate and solvent. Construct a multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions. Select initial collective variables and set initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface. Step 2: Based on the initial enhanced sampling bias parameters, initiate multiple sets of parallel molecular dynamics simulation trajectories, perform the first round of enhanced sampling calculations, and obtain the atomic evolution trajectory and the distribution of the bias potential function that forms the first round of accumulation; Step 3: Divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to the evolution trajectories of all atoms after each round of enhanced sampling; Step 4: Based on the electron density difference data, new features associated with the reaction process are screened, new collective variables are generated according to the new features, and the current set of collective variables is updated according to the new collective variables. Step 5: Based on the updated set of collective variables and the cumulative distribution of bias potential functions, adjust the bias distribution and trajectory allocation rules for the next round of augmented sampling, and execute the next round of augmented sampling based on the adjusted bias distribution and trajectory allocation rules. Repeat steps three through five until the sampling meets the preset convergence condition; Step 6: Summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

2. The method according to claim 1, characterized in that, The atomic structure parameters include the atomic coordinates, atom types, charge distribution, and molecular topology of the photocatalytic system. The photoexcitation conditions include the excitation light wavelength, light intensity, and polarization direction; The process of constructing a multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions, and selecting initial collective variables and setting initial enhanced sampling bias parameters based on the multi-electron excited potential energy surface, includes: Based on time-dependent density functional theory, the energy gradients and transition dipole moments of multiple excited states are calculated to construct a multi-electron excited state energy surface, which includes the energy distribution of the ground state and at least three low-excited states. Select one or more of the interatomic distance, bond angle, or dihedral angle from the reaction coordinates of the potential energy surface as the initial collective variable; The initial enhanced sampling bias parameters include bias potential height, bias potential width, and bias potential deposition frequency.

3. The method according to claim 2, characterized in that, Based on the initial enhanced sampling bias parameters, multiple sets of parallel molecular dynamics simulation trajectories are initiated to perform the first round of enhanced sampling calculations, obtaining atomic evolution trajectories and the process of forming the first round of accumulated bias potential function distribution, including: Based on the preset number of parallel trajectories, the initial collective variable space is divided into several sub-regions, and each sub-region is assigned an independent molecular dynamics simulation trajectory. Gaussian bias potentials are deposited on each trajectory with the same bias potential height and width to form the first round of cumulative bias potential function distribution; The maximum and minimum values ​​of the bias potential function on each trajectory are monitored in real time. When the difference between the maximum and minimum values ​​exceeds the preset bias potential uniformity threshold, the first round of sampling is paused and the bias potential deposition frequency is adjusted.

4. The method according to claim 3, characterized in that, The enhanced sampling process is divided into several rounds according to a preset number of simulation steps, and the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling is extracted, including: The total number of simulation steps is divided into multiple rounds according to a preset fixed interval, and each round contains the same number of simulation steps; After each round, starting from the end configuration of each trajectory, the electron density distribution under that configuration is obtained by single-point energy calculation, and the difference is made with the electron density distribution at the initial moment to obtain electron density difference data. Statistical analysis was performed on the electron density difference data of all trajectories within the same round to obtain the average difference value and the standard deviation of the difference; When the average difference value is less than the preset difference noise threshold, it is determined that the current round of sampling has not generated effective electronic rearrangement information, and the current round is marked as an invalid round.

5. The method according to claim 4, characterized in that, The process of filtering new features associated with the reaction process based on the electron density difference data, generating new collective variables based on the new features, and updating the current set of collective variables based on the new features includes: Principal component analysis was performed on the effective electron density difference data of each round to extract the top-preset number of differential principal components ranked by variance contribution rate. Calculate the linear correlation coefficient between each difference principal component and the current collective variable. When the correlation coefficient between a certain difference principal component and all current collective variables is lower than the preset correlation coefficient threshold, the difference principal component is determined to be a new feature. Based on the atomic coordinate change pattern corresponding to the newly added feature, a new collective variable is generated, which is one of the following: interatomic distance, bond angle, or coordination number. The new collective variable is added to the current set of collective variables, forming an updated set of collective variables.

6. The method according to claim 5, characterized in that, The process of adjusting the bias distribution and trajectory allocation rules for the next round of augmented sampling based on the updated set of collective variables and the accumulated bias potential function distribution, and then executing the next round of augmented sampling based on the adjusted bias distribution and trajectory allocation rules, includes: Based on the changes in the dimensions of the updated collective variable set, the coverage of each trajectory in the expanded collective variable space is re-acquired; When the addition of a new collective variable leads to an increase in the dimension of the collective variable, the number of parallel trajectories is expanded according to the increase in dimension, and different initial positions are assigned to the new trajectories on the dimension of the new collective variable. The accumulated bias potential function distribution is projected and stretched along the direction of the newly added collective variable, and the bias potential deposition height is reduced. When the number of newly added collective variables is zero, the number of parallel trajectories and the bias potential parameter remain unchanged, and only the distribution of the starting positions of each trajectory in the original variable space is adjusted.

7. The method according to claim 6, characterized in that, The process of repeatedly executing steps three to five until the sampling meets the preset convergence condition includes: After each cycle, the root mean square deviation of the free energy surface between the current cycle and the previous cycle is calculated as a convergence criterion. When the root mean square deviation is less than the preset convergence threshold and remains at the preset number of rounds, it is determined that the sampling meets the preset convergence condition, and the loop is terminated. When the root mean square deviation is greater than or equal to the convergence threshold, continue to execute the next round of the loop; When the number of iterations exceeds the preset maximum number of iterations, the loop stops, and the result of the last iteration is used as the final sampling result. At the same time, a warning sign indicating incomplete convergence is generated.

8. The method according to claim 7, characterized in that, The process of summarizing all atomic evolution trajectories and classifying these trajectories to obtain a set of quasi-reaction pathways includes: Extract the initial and final configurations of each trajectory from all atomic evolution trajectories, and calculate the root mean square atomic displacements of the final configuration relative to the initial configuration. The trajectory is divided into reaction path and non-reaction path based on the root mean square of atomic displacement. The trajectory with the root mean square of atomic displacement greater than the preset reaction displacement threshold is classified as the reaction path, otherwise it is classified as the non-reaction path. Trajectories belonging to the reaction pathway are clustered according to the structural similarity of the final product, with each cluster corresponding to a quasi-reaction pathway. By averaging multiple trajectories in each category, a representative reaction coordinate sequence of the quasi-reaction path is obtained.

9. The method according to claim 8, characterized in that, The process of outputting the analytical data corresponding to the set of quasi-reaction pathways includes: Output the free energy profile for each quasi-reaction path, including the barrier height along the reaction coordinate, reaction energy, and intermediate stabilization energy; The ratio of the number of samples for each quasi-reaction path to the total number of reaction trajectories is obtained as an estimate of the probability of occurrence of that quasi-reaction path. When the barrier height of a quasi-reaction path is lower than a preset low barrier threshold and the probability of its occurrence is greater than a preset probability threshold, the quasi-reaction path is marked as a primary reaction path; when the barrier height is higher than a preset high barrier threshold, the quasi-reaction path is marked as a secondary path. The output is sorted in order of priority for the main reaction path and the secondary path, and provides the corresponding atomic structure animation file for each path.

10. A photocatalytic excited-state quasi-reaction pathway adaptive enhancement sampling system, applied to a photocatalytic excited-state quasi-reaction pathway adaptive enhancement sampling method as described in any one of claims 1-9, characterized in that, include: The system import module is used to import the atomic structure parameters and photoexcitation conditions of the photocatalytic system containing the catalyst, reaction substrate and solvent, and to construct the multi-electron excited potential energy surface based on the atomic structure parameters and photoexcitation conditions. The system import module also selects initial collective variables and sets initial enhanced sampling bias parameters based on the multi-electron excitation potential energy surface. The parallel sampling module is used to initiate multiple sets of parallel molecular dynamics simulation trajectories based on the initial enhanced sampling bias parameters, perform the first round of enhanced sampling calculations, obtain atomic evolution trajectories and form the first round of accumulated bias potential function distribution; The round management module is used to divide the enhanced sampling process into several rounds according to the preset number of simulation steps, and extract the electron density difference data corresponding to all atomic evolution trajectories after each round of enhanced sampling; The feature filtering module is used to filter new features associated with the reaction process based on electron density difference data; the feature filtering module is also used to generate new collective variables based on the new features, and update the current set of collective variables based on the new collective variables. An automatic adjustment module is used to adjust the bias distribution and trajectory allocation rules of the next round of enhanced sampling based on the updated set of collective variables and the accumulated bias potential function distribution; the automatic adjustment module also executes the next round of enhanced sampling based on the adjusted bias distribution and trajectory allocation rules; the automatic adjustment module also executes steps three to five in a loop until the sampling meets the preset convergence conditions; The path merging module is used to summarize all atomic evolution trajectories, classify the trajectories to obtain a set of quasi-reaction paths, and output the analysis data corresponding to the set of quasi-reaction paths.

Images