Potential function determination method and device, storage medium and electronic device
Through constrained molecular dynamics simulation and Lagrangian multiplier method to optimize the potential function, the problem of low accuracy of potential function in the existing technology is solved, and accurate description of catalytic reactions and efficient path exploration are realized.
Patent Information
- Application Number
- CN202410468515.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-18
- Publication Date
- 2024-11-15
AI Technical Summary
In the prior art, the accuracy of the potential function is not high enough to accurately describe the characteristics of a catalytic reaction, especially when considering the catalyst dynamic effects and the electronic structure of chemical reactions.
Through a series of constrained molecular dynamics simulations and Lagrangian multiplier methods, the potential function is gradually trained and optimized, and the different points on the reaction coordinates are traversed, and the reaction path from the beginning to the end state is sampled to build an accurate machine learning potential function. .
The accurate description of catalytic reactions is achieved, the accuracy of the potential function is improved, the reaction paths including transition states can be explored efficiently, and the problem of insufficient accuracy in the existing technology is solved.
Smart Images

Figure CN118969106A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of computational chemistry and physics, and in particular, to a method and device for determining a potential function, a storage medium, and an electronic device. Background Art
[0002] In-situ experimental characterization technology has confirmed that catalysts undergo dynamic changes during catalytic reactions, and their dynamics affect the activity of the catalytic reaction. Therefore, static calculation methods that consider a single active site are not sufficient to describe the characteristics of the reaction. The complexity of the system poses a challenge to the reliable and efficient exploration of the configuration space in in-situ and operating simulations. In recent years, more and more research has focused on computational studies that consider the dynamic effects of catalysts. Studies have found that the entropy contribution brought about by the dynamic structure of catalysts such as nanoparticles has an impact on reaction activity, pathways, selectivity, etc.
[0003] Molecular dynamics is based on the laws of Newtonian mechanics to deduce the changes of system molecules and atoms over time within a certain time scale. Its simulation takes into account the dynamic evolution of the structure. The free energy calculated by statistical methods can be used to calculate the free energy of the catalytic reaction under the condition of considering the dynamics of the catalyst. Molecular dynamics based on the classical force field model can efficiently solve trajectory information and is widely used in the simulation of biological macromolecules such as protein systems, and a series of free energy calculation methods have been developed. However, since chemical reactions involve bond formation and breaking processes, the classical force field lacks electronic structure information and it is difficult to accurately describe the properties of chemical reactions. First-principles molecular dynamics is based on the principles of quantum mechanics, takes into account the properties of electronic structure, and can accurately describe chemical reactions. However, since this type of method requires iterative solution of the system energy and gradient, it consumes a lot of computing power, is difficult to expand to a longer time scale, and is limited to small systems.
[0004] The Machine Learning Potential (MLP) method fits high-precision molecular force fields based on artificial intelligence methods. Its emergence and development has greatly improved the spatiotemporal scale of molecular dynamics simulations with first-principles accuracy. By annotating data sets based on the Density Functional Theory (DFT) method, a deep neural network is used to map the local environment of each atom to the total energy of the system to train and fit high-dimensional potential energy surfaces. The application of this method significantly improves the efficiency of molecular simulations.
[0005] Static calculation methods only consider the most stable structure and ignore the impact of the dynamic evolution of the structure on the reaction free energy. Classical molecular dynamics lacks an accurate description of the electronic structure, so it is difficult to fully describe the properties of the reaction system, and its accuracy in simulating chemical reactions is insufficient. First-principles molecular dynamics consumes too much computing power and is difficult to expand to larger systems and longer time scales. Machine learning potential functions can solve the shortcomings of the first two, but their training requires full exploration of the reaction space. Conventional active learning methods are often based on molecular dynamics simulations of stable structures such as the initial and final states of the reaction at a certain scale. This type of method is difficult to obtain enough data for rare events such as reaction transition states, and it is difficult to efficiently construct training data sets, which leads to the lack of accuracy of the constructed potential function.
[0006] With regard to the problem that the potential function is not accurate enough in related technologies, no effective solution has been proposed so far.
[0007] Therefore, it is necessary to improve the related technology to overcome the above-mentioned defects in the related technology. Summary of the invention
[0008] The embodiments of the present application provide a method and device for determining a potential function, a storage medium, and an electronic device to at least solve the problem that the potential function is not sufficiently accurate.
[0009] According to one aspect of an embodiment of the present application, a method for determining a potential function is provided, comprising: a first training step: determining an initial structure of a target molecule and an Nth reaction coordinate point of each reaction coordinate, so that a first potential function performs a constrained molecular dynamics simulation based on the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether a target structure exists in the first simulation trajectory, wherein the target structure is a structure for which the first potential function cannot accurately predict atomic force information of the target structure; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function based on the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0010] Furthermore, after training the first potential function according to the target energy and target atomic force of the target structure, the method also includes: executing a second training step: obtaining the trained first potential function so that the trained first potential function performs constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a third simulation trajectory; when the target structure exists in the third simulation trajectory, determining the energy and atomic force of the target structure in the third simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the third simulation trajectory; looping through the second training step until the target structure does not exist in the fourth simulation trajectory simulated according to the initial structure and the Nth reaction coordinate point, and using the first potential function used to simulate the fourth simulation trajectory as the target potential function.
[0011] Furthermore, determining whether there is a target structure in the first simulation trajectory includes: determining multiple structures in the first simulation trajectory, and determining the atomic force and energy of each structure according to multiple second potential functions, wherein the second potential function is a potential function trained according to a training data set, and the multiple second potential functions include: the first potential function, the training data set is a data set for training the first potential function, and the structure is a structure in the first simulation trajectory; determining the maximum atomic force deviation of each structure according to the atomic force and energy of each structure; and determining the target structure among the multiple structures according to the maximum atomic force deviation of each structure.
[0012] Furthermore, after using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function, the method also includes: when performing molecular dynamics simulation of the target molecule according to the target potential function, applying constraint force to the target atom based on the Lagrange multiplier method, wherein the target atom is any atom constituting the target molecule; determining the Lagrange multiplier corresponding to each simulation step in the molecular dynamics simulation process; determining the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each simulation step; and determining the free energy change of the target molecule from the initial state to the final state according to the mean force potential.
[0013] Furthermore, determining the Lagrange multiplier corresponding to each simulation step in the molecular dynamics simulation includes: determining the velocity information of the target atom in each simulation step; determining the atomic coordinates of the target atom in each simulation step according to the velocity information; and determining the Lagrange multiplier according to the atomic coordinates.
[0014] Furthermore, the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point is determined according to the Lagrange multipliers corresponding to each step of the simulation process, including: determining the statistical average of multiple Lagrange multipliers according to the Lagrange multipliers corresponding to each step of the simulation process; and determining the statistical average as the mean force potential.
[0015] Further, applying a constraint force to the target atom based on the Lagrange multiplier method includes: applying a constraint force to the target atom based on the following formula: σ(q)=ξ(q)-ξ;where, is the Lagrangian, is the Lagrangian function, q is the atomic coordinate, is the velocity information of the target atom, λ is the Lagrange multiplier, σ(q) is the imposed constraint force, ξ is the target reaction coordinate point, and ξ(q) is the response to the generalized atomic coordinate q at the target reaction coordinate point.
[0016] According to another aspect of the embodiment of the present application, a device for determining a potential function is also provided, including: a first training module, used to perform a first training step: determine the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determine whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure for which the first potential function cannot accurately predict the atomic force information of the target structure; if the target structure exists in the first simulation trajectory, determine the energy and atomic force of the target structure in the first simulation trajectory, and train the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determine the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; a second training module, used to loop through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and use the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0017] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for determining the potential function when running.
[0018] According to another aspect of an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method for determining the potential function through the computer program.
[0019] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, wherein the computer program is configured to execute the above-mentioned method for determining the potential function when running.
[0020] Through the embodiment of the present application, a first training step is performed: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure for which the first potential function cannot accurately predict the atomic force information of the target structure; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function based on the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function. That is, in the embodiment of the present application, in the process of constructing a data set, different reaction coordinate points on the same reaction coordinate are traversed to achieve sampling of each point on the reaction coordinate from the initial state to the final state, including the transition state, to achieve efficient exploration of rare events, and to construct an accurate machine learning potential function, thereby solving the problem of insufficient accuracy of the potential function in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The exemplary embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0022] Figure 1 It is a hardware structure block diagram of a computer terminal of the method for determining the potential function of an embodiment of the present application;
[0023] Figure 2 is a flow chart of a method for determining a potential function according to an embodiment of the present application;
[0024] Figure 3 is a flow chart of a method for determining a potential function according to an embodiment of the present application;
[0025] Figure 4a It is Pt 22 Convergence of the machine learning potential function of cluster-catalyzed O2 dissociation reaction with the number of rounds (I);
[0026] Figure 4b It is Pt 22 Convergence of the machine learning potential function of cluster-catalyzed O2 dissociation reaction with the number of rounds (Part 2);
[0027] Figure 4c It is Pt 22 Training errors of machine learning potential functions for cluster-catalyzed O2 dissociation reactions in different reaction coordinates;
[0028] Figure 5a It is a schematic diagram of the distribution of the data set along the reaction coordinate for the training potential function of O2 dissociation reaction under clusters of different sizes;
[0029] Figure 5b It is a sampling structure distribution diagram obtained by reducing the dimensionality of the molecular structure description;
[0030] Figure 6a is the average force corresponding to each point of the reaction coordinate at different temperatures calculated;
[0031] Figure 6b It is the calculated change in free energy of O2 dissociation reaction corresponding to each point of the reaction coordinate at different temperatures;
[0032] Figure 7 It is a structural block diagram of a device for determining a potential function according to an embodiment of the present application. DETAILED DESCRIPTION
[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0034] It should be noted that the terms and "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0035] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 is a hardware structure block diagram of a computer terminal for determining a potential function according to an embodiment of the present application. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown in the figure) processor 102 (processor 102 may include but is not limited to a microprocessor (Microprocessor Unit, referred to as MPU) or a programmable logic device (Programmable logic device, referred to as PLD)) and a memory 104 for storing data. In an exemplary embodiment, the computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Equivalent functions or comparisons shown Figure 1 A different configuration with more features is shown.
[0036] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for determining the potential function in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, which is equivalent to implementing the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.
[0037] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0038] In this embodiment, a method for determining a potential function is provided. Figure 2 : is a flow chart of a method for determining a potential function according to an embodiment of the present application, the flow chart comprising the following steps:
[0039] Step S202, the first training step: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory.
[0040] Step S204, looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0041] For example, for the target reaction coordinate, determine the initial structure of the target molecule and the first reaction coordinate point of the target reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the first reaction coordinate point to obtain a first simulation trajectory; determine whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; if the target structure exists in the first simulation trajectory, determine the energy and atomic force of the target structure in the first simulation trajectory, and train the first potential function based on the energy and atomic force of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determine a second reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the second reaction coordinate point to obtain a second simulation trajectory;
[0042] Determine whether the target structure exists in the second simulation trajectory; if the target structure exists in the first simulation trajectory, determine the energy and atomic forces of the target structure in the first simulation trajectory, and train the first potential function according to the energy and atomic forces of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determine the third reaction coordinate point so that the first potential function performs constrained molecular dynamics simulation according to the initial structure and the third reaction coordinate point to obtain a simulation trajectory, and so on, until it is determined whether the target structure exists in the simulation trajectory corresponding to the last reaction coordinate point. If the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, determine the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0043] Through the above steps, the first training step is performed: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; and looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function. That is, in the embodiment of the present application, in the process of constructing a data set, different reaction coordinate points on the same reaction coordinate are traversed to achieve sampling of each point on the reaction coordinate from the initial state to the final state, including the transition state, to achieve efficient exploration of rare events, and to construct an accurate machine learning potential function, thereby solving the problem of insufficient accuracy of the potential function in the prior art.
[0044] Optionally, after training the first potential function according to the target energy and target atomic force of the target structure, the following steps need to be performed: perform the second training step: obtain the trained first potential function, so that the trained first potential function performs constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a third simulation trajectory; when the target structure exists in the third simulation trajectory, determine the energy and atomic force of the target structure in the third simulation trajectory, and train the first potential function according to the energy and atomic force of the target structure in the third simulation trajectory; loop the second training step until the target structure does not exist in the fourth simulation trajectory simulated according to the initial structure and the Nth reaction coordinate point, and use the first potential function used to simulate the fourth simulation trajectory as the target potential function.
[0045] That is to say, for each reaction coordinate point, after performing the constrained molecular dynamics simulation for the first time according to the initial structure and the Nth reaction coordinate point, it is determined whether the target structure exists in the simulation trajectory obtained for the first time. If the target structure exists, the first potential function is trained according to the energy and atomic forces of the target structure in the third simulation trajectory obtained for the first time; after the training, after performing the constrained molecular dynamics simulation for the second time according to the initial structure and the Nth reaction coordinate point according to the trained first potential function, it is determined whether the target structure exists in the simulation trajectory obtained for the second time. If the target structure exists, the first potential function is trained again according to the energy and atomic forces of the target structure in the third simulation trajectory obtained for the second time; after the training, after performing the constrained molecular dynamics simulation for the third time according to the initial structure and the Nth reaction coordinate point according to the trained first potential function, it is determined whether the target structure exists in the simulation trajectory obtained for the third time. If the target structure does not exist, the first potential function after the second training is used as the target potential function.
[0046] Furthermore, determining whether there is a target structure in the first simulation trajectory includes: determining multiple structures in the first simulation trajectory, and determining the atomic force and energy of each structure according to multiple second potential functions, wherein the second potential function is a potential function trained according to a training data set, and the multiple second potential functions include: the first potential function, the training data set is a data set for training the first potential function, and the structure is a structure in the first simulation trajectory; determining the maximum atomic force deviation of each structure according to the atomic force and energy of each structure; and determining the target structure among the multiple structures according to the maximum atomic force deviation of each structure.
[0047] In an embodiment of the present application, there are multiple potential functions trained. Therefore, for the structures collected in the simulation trajectory generated above, other trained machine learning potential functions (equivalent to the second potential function in the above embodiment) are used to calculate the forces on all atoms in all structures.
[0048] When it is determined that the current machine learning potential function (equivalent to the first potential function in the above embodiment) cannot accurately describe the structure, first principles calculations are performed on the target structure to obtain energy information and atomic force information of the structure, and the obtained structure and the energy information and atomic force information corresponding to the structure are added to the training data set to update the training data set.
[0049] Clustering is performed based on the atomic local environment descriptors of the structure, and the structure with the sparsest distribution (equivalent to the target structure mentioned above) is selected, annotated by first-principles calculations, and added to the training data set.
[0050] Furthermore, after using the first potential function for simulating the simulation trajectory corresponding to the last reaction coordinate point as the target potential function, it also includes: when performing molecular dynamics simulation of the target molecule according to the target potential function, applying constraint force to the target atom based on the Lagrange multiplier method, wherein the target atom is any atom constituting the target molecule; determining the Lagrange multiplier corresponding to each simulation step in the molecular dynamics simulation process; determining the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each simulation step; and determining the free energy change of the target molecule from the initial state to the final state according to the mean force potential.
[0051] Specifically, the velocity information of the target atom in each simulation step is determined; the atomic coordinates of the target atom in each simulation step are determined according to the velocity information; and the Lagrange multiplier is determined according to the atomic coordinates.
[0052] In the embodiment of the present application, in each step of the molecular dynamics simulation process, the atomic coordinates are first updated according to Newton's laws of motion combined with the calculated velocity information, and then the Lagrange multiplier λ is solved by the updated reaction coordinates, and the velocity and atomic coordinates are updated again according to λ. The Lagrange multiplier for simulating a time scale is obtained through molecular dynamics simulation.
[0053] Furthermore, the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point is determined according to the Lagrange multipliers corresponding to each step of the simulation process, including: determining the statistical average of multiple Lagrange multipliers according to the Lagrange multipliers corresponding to each step of the simulation process; and determining the statistical average as the mean force potential.
[0054] According to the derivation and proof of Michael Sprik and Giovanni Ciccotti, the statistical average of this Lagrange multiplier is the mean force potential under the reaction coordinate. Therefore, the statistical average of the Lagrange multiplier can be calculated based on the Lagrange multiplier determined above, and then the above-mentioned mean force potential can be determined.
[0055] Further, applying a constraint force to the target atom based on the Lagrange multiplier method includes: applying a constraint force to the target atom based on the following formula: σ(q)=ξ(q)-ξ;where, is the Lagrangian, q is the atomic coordinate, is the velocity information of the target atom, λ is the Lagrange multiplier, σ(q) is the imposed constraint force, ξ is the target reaction coordinate point, and ξ(q) is the response to the generalized atomic coordinate q at the target reaction coordinate point.
[0056] In the embodiment of the present application, the user provides the initial or final state structure of the reaction system supported by the machine learning potential function, a reaction coordinate and its range required for calculating the reaction free energy, and the information of the multiple temperature conditions required to be calculated, and generates multiple processes in parallel at different temperatures to manage the molecular dynamics simulation tasks of each temperature, and the continuous operation target in each task is the constraint molecular dynamics simulation of the reaction coordinate from the initial state to the final state (or from the final state to the initial state, depending on the initial structure and reaction coordinate given by the user), and each simulation runs a certain time scale. Constrained molecular dynamics uses the SHAKE algorithm to impose constraints, and the coordinates of the target atoms are controlled by the Lagrange multiplier method to avoid their deviation from the set constraint target, and its formula is as follows:
[0057]
[0058] σ(q)=ξ(q)-ξ;
[0059] in, is the Lagrangian, q is the atomic coordinate, λ is the Lagrangian multiplier, σ is the imposed constraint, ξ(q) is the geometric parameter related to the atomic coordinate, and ξ is the ξ(q) corresponding to the reaction coordinate that needs to be fixed.
[0060] Optionally, after determining the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each step of the simulation process, the method further includes: determining an error value in the molecular dynamics simulation process according to a preset strategy; determining a magnitude relationship between the error value and a preset threshold; and determining whether to determine the free energy change of the target molecule from the initial state to the final state according to the mean force potential based on the magnitude relationship.
[0061] Further, determining whether to determine the free energy change of the target molecule from the initial state to the final state according to the mean force potential is determined based on the size relationship, including: when the size relationship indicates that the error value is greater than the preset threshold, determining the free energy change of the target molecule from the initial state to the final state according to the mean force potential; when the size relationship indicates that the error value is less than or equal to the preset threshold, determining that the molecular dynamics simulation has not converged, and continuing to execute the molecular dynamics simulation.
[0062] In the embodiment of the present application, after each molecular dynamics simulation, a block average strategy is used to perform statistical averaging and error estimation. If the error is greater than the acceptance threshold, it is considered that there is no convergence, and the calculation is automatically continued until convergence; otherwise, the calculated statistical average and the corresponding error are saved, and the trajectory end structure is used as the initial, and a new molecular dynamics simulation task is generated with the next reaction coordinate as the target value, until the statistical average of the Lagrange multipliers corresponding to each required reaction coordinate at the temperature is obtained.
[0063] At each temperature, the mean force potential is definite integrated along the reaction coordinate axis to obtain the free energy change from the initial state to each point on the reaction coordinate:
[0064]
[0065] Where r0 is the initial reaction coordinate, r is the current reaction coordinate, and W(x) is the corresponding mean force potential under the reaction coordinate.
[0066] The change of ΔG from the initial state to the final state is obtained at different temperatures, and the reaction energy barrier ΔG at different temperatures can be determined from this. ≠ (i.e., the maximum value of ΔG) and the corresponding reaction free energy change Δ r G (i.e., ΔG of the final state relative to the initial state).
[0067] Obviously, the embodiments described above are only some embodiments of the present application, not all embodiments. In order to better understand the method for determining the potential function, the above process is described below in conjunction with the embodiments, but it is not intended to limit the technical solutions of the embodiments of the present application, specifically:
[0068] In an alternative embodiment, Figure 3 is a flow chart of a method for determining a potential function according to an embodiment of the present application, Figure 3 The workflow for training machine learning potential functions and accelerating the calculation of catalytic reaction free energy by combining the mean force potential method is described in the paper. It includes the following three main steps:
[0069] Step 1: Model the structure and construct an initial data set. Obtain the initial structural model of the catalyst from the literature or database, and add the adsorption of the initial or final reactant molecules to the structural model. Then, obtain the pre-equilibrium trajectory through a short molecular dynamics simulation or Monte Carlo simulation, extract some structures from it, perform first-principles calculations and annotations, obtain the corresponding potential energy and gradient (i.e., atomic force), and construct an initial training data set for training the machine learning potential function.
[0070] Step 2: Training the machine learning potential function. This includes a loop of four sub-steps, namely, training multiple different machine learning potential functions from the current data set, and then using one of them to perform constrained molecular dynamics simulations based on the given initial structure and set reaction coordinates. At the same time, using these machine learning potential functions as references, screen out structures with large maximum prediction force deviations from the obtained trajectories, cluster them in combination with atomic local environment descriptors, select the most sparsely distributed structures, perform first-principles calculation annotations, and add them to the training data set. The above steps include two loop iterations. First, if the current reaction coordinates have been fully sampled, that is, there are no new structures, a new reaction coordinate is set to deviate from the current settings to achieve sampling along the reaction coordinate axis; second, if all grid points set on the reaction coordinate axis have been fully sampled, the data set is complete, and a relatively complete machine learning potential function can be obtained at this time.
[0071] Step 3: Calculation of free energy of catalytic elementary reactions. Combined with the machine learning potential function obtained in the above steps, machine learning accelerated constrained molecular dynamics simulations are performed at different temperatures and different reaction coordinates. The obtained data is then post-processed and analyzed. For example, the solid-liquid phase transition of the catalyst at different temperatures can be analyzed by the fluctuations of the atomic coordinates on the structural trajectory, and the average force potential obtained by constrained molecular dynamics at different reaction coordinate points at the same temperature can be integrated to obtain the free energy change curve from the initial state to the transition state to the final state at that temperature.
[0072] Figure 4a , Figure 4b , Figure 4c It is Pt 22 Convergence of the machine learning potential function of cluster-catalyzed O2 dissociation reaction with the number of rounds and the training error under different reaction coordinates.
[0073] like Figure 4a As shown in the figure, the proportion of accurately predicted, candidate structures, and discarded structures to all sampled structures evolves with the number of training iterations. In round 0, most structures are in the candidate area, and a few are even discarded because they do not conform to the laws of physicochemistry. As the number of rounds increases, the proportion of structures that can be accurately predicted continues to rise. By round 8, most structures can be accurately predicted.
[0074] like Figure 4b As shown in the figure, with the number of training iterations, the distribution probability density obtained by kernel density approximation analysis of the maximum force prediction deviation of several potential functions is shown in Figure 2. In the 0th round, most structures are distributed in With the increase of rounds, the overall distribution shifts to the left, and by the 8th round, almost all
[0075] like Figure 4cAs shown, the trained potential function is used to set different values (1.4, 1.8, 3.0, 3.8, unit: ) to perform constrained molecular dynamics simulations, extract 500 structures for each value, perform first-principles calculations and annotate the energy and force, and compare the energy and force predicted by machine learning. It can be seen from the figure that there is no obvious systematic deviation in the prediction. The RMSE of energy prediction is 6.84, 7.16, 9.09, and 10.42 meV / atom, respectively, while the RMSE of force prediction is 176.15, 186.76, and 238.27, respectively.
[0076] Figure 5a The training data set of the potential function for the O2 dissociation reaction under clusters of different sizes is distributed along the reaction coordinate. It can be seen that in the training data sets of the three systems Pt22, Pt27 and Pt36, the OO distance reaction coordinate There are distributions between them, and the overall distribution is relatively uniform; Figure 5b It is a sampling structure distribution map obtained by dimensionality reduction of the molecular structure description, and the approximate distribution area is drawn using kernel density approximation. The dark and light colors correspond to the cases where constrained molecular dynamics sampling is used and not used, respectively. When constrained molecular dynamics sampling is not used, the distribution of points is more concentrated, and the structure distribution is more extensive after constrained molecular dynamics sampling.
[0077] Figure 6a and Figure 6b It is the calculated average force corresponding to each point of the reaction coordinate at different temperatures and the integral obtained by the O2 dissociation reaction free energy change, with the initial state as a reference; the temperature range includes 400, 500, 600, 700, 750, 800, 900, 1000K. The horizontal axis is the distance between OO atoms, Figure 6a The vertical axis is the average force, Figure 6b The vertical axis is the free energy change.
[0078] Figure 6a The average force is presented first From about 0 (initial state) to the minimum value (about Department, Then it rises to 0 (about After the transition state) it continues to rise to a maximum value (the distribution varies according to temperature), and then gradually decays back to around 0 (approximately final state).
[0079] Figure 6b The free energy change curves at different temperatures are obtained by integrating the distance between OO atoms at each point in the left figure. The overall curves show a single peak, with the initial and final states as minimum points and the transition state as maximum point. The absolute values are all referenced to The free energy at this point is set to 0eV. At high temperatures, the free energy change from the initial state to the final state converges to about -1.5eV, and at low temperatures it is at -2.8eV, and the temperatures in between show a gradual upward trend. However, the free energy change from the initial state to the transition state does not show a significant change, and is around 0.2eV.
[0080] In the embodiment of the present application, a computational workflow for calculating the free energy of the catalytic reaction by molecular dynamics simulation and statistical averaging is constructed, and the influence of the dynamic structure of the catalyst on the reaction is taken into account during the sampling process. By constructing a machine learning potential function in a data-driven manner based on the first-principles calculation data, molecular dynamics simulations with first-principles accuracy are efficiently performed. Combining constrained molecular dynamics and mean force potential methods, in the process of constructing a data set, different points on the same reaction coordinate are traversed to achieve sampling of each point on the reaction coordinate from the initial state to the final state, including the transition state, to achieve efficient exploration of rare events, and to efficiently construct a machine learning potential function describing the elementary reaction process. The workflow is constructed to automatically submit constrained molecular dynamics simulation tasks, which realizes the automated traversal of each point on the reaction coordinate in parallel at different temperatures, and automatically realizes convergence judgment and data processing, and calculates the free energy of elementary reactions.
[0081] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.
[0082] In the present embodiment, a device for determining a potential function is also provided, and the device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made are omitted. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and contemplated.
[0083] Figure 7 is a structural block diagram of a device for determining a potential function according to an embodiment of the present application, the device comprising:
[0084] The first training module 72 is used to perform the first training step: determine the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determine whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information of the target structure cannot be accurately predicted by the first potential function; if the target structure exists in the first simulation trajectory, determine the energy and atomic force of the target structure in the first simulation trajectory, and train the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determine the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory;
[0085] The second training module 74 is used to loop through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and use the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0086] Through the present application, a first training step is performed: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure for which the first potential function cannot accurately predict the atomic force information of the target structure; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function based on the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation based on the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function. That is, in the embodiment of the present application, in the process of constructing a data set, different reaction coordinate points on the same reaction coordinate are traversed to achieve sampling of each point on the reaction coordinate from the initial state to the final state, including the transition state, to achieve efficient exploration of rare events, and to construct an accurate machine learning potential function, thereby solving the problem of insufficient accuracy of the potential function in the prior art.
[0087] Furthermore, the first training module 72 is used to execute the second training step: obtaining the trained first potential function so that the trained first potential function performs constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a third simulation trajectory; when the target structure exists in the third simulation trajectory, determining the energy and atomic forces of the target structure in the third simulation trajectory, and training the first potential function according to the energy and atomic forces of the target structure in the third simulation trajectory; looping the execution of the second training step until the target structure does not exist in the fourth simulation trajectory simulated according to the initial structure and the Nth reaction coordinate point, and using the first potential function used to simulate the fourth simulation trajectory as the target potential function.
[0088] Furthermore, the first training module 72 is used to determine multiple structures in the first simulation trajectory, and to determine the atomic force and energy of each structure according to multiple second potential functions, wherein the second potential function is a potential function trained according to a training data set, and the multiple second potential functions include: the first potential function, the training data set is a data set for training the first potential function, and the structure is a structure in the first simulation trajectory; determine the maximum atomic force deviation of each structure according to the atomic force and energy of each structure; and determine the target structure among the multiple structures according to the maximum atomic force deviation of each structure.
[0089] Furthermore, the above-mentioned device also includes a calculation module, which is used to apply constraint force to the target atom based on the Lagrange multiplier method when performing molecular dynamics simulation of the target molecule according to the target potential function, wherein the target atom is any atom that constitutes the target molecule; determine the Lagrange multiplier corresponding to each simulation step in the molecular dynamics simulation process; determine the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each simulation step; and determine the free energy change of the target molecule from the initial state to the final state according to the mean force potential.
[0090] Furthermore, a calculation module is used to determine the velocity information of the target atom in each simulation step; determine the atomic coordinates of the target atom in each simulation step according to the velocity information; and determine the Lagrange multiplier according to the atomic coordinates.
[0091] Furthermore, a calculation module is used to determine a statistical average of multiple Lagrange multipliers according to the Lagrange multipliers corresponding to each step of the simulation process; and determine the statistical average as the mean force potential.
[0092] Furthermore, the calculation module is used to apply a constraint force to the target atom based on the following formula: in, is the Lagrangian, is the Lagrangian function, q is the atomic coordinate, is the velocity information of the target atom, λ is the Lagrange multiplier, σ(q) is the imposed constraint force, ξ is the target reaction coordinate point, and ξ(q) is the response to the generalized atomic coordinate q at the target reaction coordinate point.
[0093] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:
[0094] S1, executing the first training step: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory;
[0095] S2, looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0096] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.
[0097] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.
[0098] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
[0099] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:
[0100] S1, executing the first training step: determining the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determining whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; in the case where the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; in the case where the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory;
[0101] S2, looping through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function.
[0102] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.
[0103] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.
[0104] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.
[0105] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.
Claims
1. A method for determining a potential function, characterized in that: include: Execute the first training step: determine the initial structure of the target molecule and the Nth reaction coordinate point of each reaction coordinate, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a first simulation trajectory, wherein N is a positive integer; determine whether there is a target structure in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; if the target structure exists in the first simulation trajectory, determine the energy and atomic force of the target structure in the first simulation trajectory, and train the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determine the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point to obtain a second simulation trajectory; The first training step is executed in a loop until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point is used as the target potential function.
2. The method for determining the potential function according to claim 1, characterized in that: After training the first potential function according to the target energy and target atomic force of the target structure, the method further includes: Perform a second training step: obtain a trained first potential function, so that the trained first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point to obtain a third simulation trajectory; when the target structure exists in the third simulation trajectory, determine the energy and atomic force of the target structure in the third simulation trajectory, and train the first potential function according to the energy and atomic force of the target structure in the third simulation trajectory; The second training step is executed in a loop until the target structure does not exist in the fourth simulation trajectory obtained by simulating the initial structure and the Nth reaction coordinate point, and the first potential function used to simulate the fourth simulation trajectory is used as the target potential function.
3. The method for determining the potential function according to claim 1, characterized in that: Determining whether a target structure exists in the first simulation trajectory comprises: Determine a plurality of structures in the first simulation trajectory, and determine the atomic forces and energies of each structure respectively according to a plurality of second potential functions, wherein the second potential function is a potential function trained according to a training data set, and the plurality of second potential functions include: the first potential function, the training data set is a data set for training the first potential function, and the structure is a structure in the first simulation trajectory; Determining the maximum atomic force deviation of each structure according to the atomic forces and energies of each structure; The target structure is determined among the plurality of structures according to a maximum atomic force deviation of each of the structures.
4. The method for determining the potential function according to claim 1, characterized in that: After using the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point as the target potential function, the method further includes: In the case of performing a molecular dynamics simulation of the target molecule according to the target potential function, applying a constraint force to a target atom based on a Lagrange multiplier method, wherein the target atom is any atom constituting the target molecule; Determine the Lagrange multiplier corresponding to each step of the molecular dynamics simulation process; Determine the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each step of the simulation process; The free energy change of the target molecule from an initial state to a final state is determined based on the mean force potential.
5. The method for determining the potential function according to claim 4, characterized in that: Determining the Lagrange multiplier corresponding to each step of the molecular dynamics simulation process includes: Determining the velocity information of the target atom during each simulation step; Determining the atomic coordinates of the target atom in each simulation process according to the velocity information; The Lagrange multiplier is determined according to the atomic coordinates.
6. The method for determining the potential function according to claim 4, characterized in that: Determining the mean force potential of the molecular dynamics simulation process at the target reaction coordinate point according to the Lagrange multiplier corresponding to each step of the simulation process, including: Determine a statistical average of a plurality of Lagrange multipliers according to the Lagrange multipliers corresponding to each step of the simulation process; The statistical mean is determined as the mean force potential.
7. The method for determining a potential function according to claim 4, characterized in that: Apply constraints to target atoms based on the Lagrange multiplier method, including: The constraint force is applied to the target atom based on the following formula: in, is the Lagrangian, is the Lagrangian function, q is the atomic coordinate, is the velocity information of the target atom, λ is the Lagrange multiplier, σ(q) is the imposed constraint force, ξ is the target reaction coordinate point, and ξ(q) is the response to the generalized atomic coordinate q at the target reaction coordinate point.
8. A device for determining a potential function, characterized in that: include: A first training module is used to perform a first training step: determining an initial structure of a target molecule and an Nth reaction coordinate point of each reaction coordinate, so that a first potential function performs a constrained molecular dynamics simulation according to the initial structure and the Nth reaction coordinate point, so as to obtain a first simulation trajectory, wherein N is a positive integer; determining whether a target structure exists in the first simulation trajectory, wherein the target structure is a structure whose atomic force information cannot be accurately predicted by the first potential function; if the target structure exists in the first simulation trajectory, determining the energy and atomic force of the target structure in the first simulation trajectory, and training the first potential function according to the energy and atomic force of the target structure in the first simulation trajectory; if the target structure does not exist in the first simulation trajectory, determining the N+1th reaction coordinate point, so that the first potential function performs a constrained molecular dynamics simulation according to the initial structure and the N+1th reaction coordinate point, so as to obtain a second simulation trajectory; The second training module is used to loop through the first training step until the target structure does not exist in the simulation trajectory corresponding to the last reaction coordinate point, and the first potential function used to simulate the simulation trajectory corresponding to the last reaction coordinate point is used as the target potential function.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 7 when executed.
10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 7 through the computer program.
Citation Information
Cited By
Information processing system, information processing method, and program
WO2026168196A1
