Atomic scale MD-KMC parallel simulation unified modeling method and system

Abstract

The invention discloses an atomic scale MD-KMC parallel simulation unified modeling method and system. The method comprises the following steps: establishing an atomic scale unified architecture; carrying out molecular dynamics simulation based on the atomic scale unified architecture; and dynamics Monte Carlo simulation based on the atomic scale unified architecture. According to the method, decoupling between the MD software development module and the KMC software development module can be facilitated; the effect of optimizing one part and improving multiple parts is achieved, MD simulation and KMC simulation are supported, and meanwhile the large-scale simulation calculation level is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of high-performance computing and algorithm technology, is aimed at material computational simulation applications, and specifically relates to a parallel molecular dynamics program and a parallel dynamics Monte Carlo unified modeling technology, and in particular to an atomic scale MD-KMC parallel simulation unified modeling method and system. Background Art

[0002] The evolutionary behavior of materials is a multi-scale process, spanning multiple scales such as atoms, defect clusters, and macroscopic scales in space, and spanning femtoseconds to seconds or even years in time. Using computer simulation, especially supercomputer simulation, to study the evolutionary behavior of materials is a common method and a hot field in the current interdisciplinary research of materials science and computers. Among them, at the atomic scale, molecular dynamics (MD) and kinetic Monte Carlo (KMC) are two classic and important simulation methods at the atomic scale. At present, the time scale of classic MD simulation is basically only in the nanosecond level; through MD and KMC coupled simulation, the time scale of simulation can be extended to the second or year level. Since the use of MD and KMC for material simulation requires huge computing power and memory of computers, it is generally necessary to introduce large-scale parallel computing to realize distributed memory and improve computing speed.

[0003] The MD method is a computational method based on the classical Newton's laws of motion. It takes each particle as the research object. In the system composed of these particles, each particle moves under the action of the empirical potential field provided by other particles. In each simulation time step Δt, it is necessary to traverse the neighboring particles of each particle, calculate the interaction force between them through the potential function, and update the speed and position of the system particles according to the Newton's laws of motion through the force. The KMC method uses random sampling of system events to drive the system forward. For each system event sampling, it is necessary to calculate the probability of the system jumping over the potential barrier and reaching a new system state (i.e., transition probability), where the calculation of the transition probability can be obtained by calculating the system potential energy using the potential function. If both MD and KMC simulation methods use spatial decomposition for parallelization, the particle storage, parallel algorithm, parallel communication strategy, and potential function and other computing modules are common in implementation.

[0004] In order to enable the atomic-scale MD and KMC simulation methods to perform large-scale and longer-term simulations, facilitate the efficient transfer of parameters during coupled calculations between the two, and facilitate software development for large-scale parallelism, the present invention proposes a unified modeling scheme for MD and KMC parallel simulation programs. Summary of the invention

[0005] The present invention is provided to solve the above problems existing in the prior art. Therefore, a unified modeling method and system for atomic scale MD-KMC parallel simulation is needed, which uniformly considers the particle storage, parallel strategy, communication strategy and potential function calculation in the parallel MD and KMC simulation methods, and forms a unified architecture for atomic scale simulation. The solution can effectively reuse software modules and transfer parameters when the two software interact, supporting atomic scale material simulation to achieve high performance and high precision calculation on supercomputers.

[0006] The present invention provides an atomic scale MD-KMC parallel simulation unified modeling method, the method comprising:

[0007] Establishing a unified architecture at the atomic scale;

[0008] Molecular dynamics simulations based on the atomic-scale unified architecture;

[0009] Dynamical Monte Carlo simulations based on the atomic-scale unified architecture;

[0010] During the simulation, a three-dimensional neighbor forwarding communication module is used to synchronize the particle information of the simulation area and the ghost area, and data compression technology is introduced during the synchronization process;

[0011] Based on the unified architecture at the atomic scale, MD and KMC coupling simulations are carried out. After the simulation, based on the atomic coordinate data output by the simulation, defect and cluster analysis are performed on the simulation data through machine learning algorithms and cluster analysis algorithms to obtain the defect evolution trend of the simulation system and predict the long-term evolution behavior of the system.

[0012] Furthermore, the atomic scale unified architecture includes:

[0013] The parallel region partitioning module is configured as follows: given the simulation region size x, y, z and the number of parallel MPI processes n, where x, y, z are the number of lattices in each direction in the three-dimensional space, the parallel region partitioning is realized by using the MPI function, and each MPI process obtains the size and range of the region it simulates;

[0014] Particle storage data structure and neighbor particle search module, including:

[0015] The first initialization interface is configured to: given the size of a single MPI process simulation region (u, v, w) and the size of the ghost region used for communication (g, g, g), apply for particle memory, and initialize the particle position as required (such as placing the particle on the lattice point);

[0016] The neighbor particle search interface is configured to: given a particle information (such as a memory address) in the memory, return all its neighbor particles from the particle storage space;

[0017] The particle position update interface is configured to: update the particle position or velocity information in the particle storage according to the given new particle information;

[0018] Potential function calculation module, including:

[0019] The second initialization interface is configured to: read the potential function table file and initialize the potential function data;

[0020] The potential energy calculation interface is configured to: Given the distance and type between two atoms, return the force between particles or the potential energy contribution of the system according to the EAM potential and LJ potential calculation rules;

[0021] Three-dimensional neighbor communication module, including:

[0022] The synchronous ghost region interface is configured to: given the particle storage data structure and parallel region partitioning, send the particle information of the simulation region to the neighbor process through MPI message communication; at the same time, receive messages from the neighbor process to synchronize the ghost region;

[0023] The synchronous simulation area is configured as follows: given the particle storage data structure, parallel area division and ghost area information, the particle information of the ghost area is sent to the neighbor process through MPI message communication; at the same time, messages are received from the neighbor process to synchronize the simulation area.

[0024] Furthermore, the molecular dynamics simulation based on the atomic scale unified architecture specifically includes:

[0025] (1) Parallel initialization: Given the simulation region sizes x, y, z and the number of parallel MPI processes n, call the parallel region partitioning module to perform parallel initialization;

[0026] (2) Particle storage initialization: Based on the simulation area size (u, v, w) after the parallel area division in (1), and the ghost area size, the first initialization interface in the particle storage data structure and the neighbor particle search module is called to initialize the particle storage;

[0027] (3) Potential energy initialization: calling the second initialization interface in the potential function calculation module to initialize the potential function;

[0028] (4) Ghost area synchronization: calling the synchronization ghost area interface in the three-dimensional neighbor communication module to establish or synchronize particle information on the ghost area;

[0029] (5) Force calculation:

[0030] 1) Traverse all particles, for each particle i, call the particle storage data structure and the neighbor particle search interface in the neighbor particle search module to obtain its neighbor particles;

[0031] 2) For each neighbor particle j of particle i, the potential energy calculation interface in the potential function calculation module is called to calculate the force between particles i and j, and the force contributions are accumulated as the force of particle i;

[0032] (6) Based on the force on the particle, the particle storage data structure and the neighbor particle search module are called, and the speed and position of each particle are updated according to the rules of Newton's laws of motion;

[0033] (7) Simulation area synchronization: calling the synchronization simulation area interface in the three-dimensional neighbor communication module to synchronize the particle information on the simulation area;

[0034] (8) Return to (4) and use the unified architecture to calculate the next force until the given time step is reached.

[0035] Furthermore, the dynamic Monte Carlo simulation based on the atomic scale unified architecture specifically includes:

[0036] (1) Parallel initialization: Given the simulation region sizes x, y, z and the number of parallel MPI processes n, call the parallel region partitioning module to perform parallel initialization;

[0037] (2) Particle storage initialization: Based on the simulation area size (u, v, w) after the parallel area division in (1), and the ghost area size, the first initialization interface in the particle storage data structure and the neighbor particle search module is called to initialize the particle storage;

[0038] (3) Potential energy initialization: calling the second initialization interface in the potential function calculation module to initialize the potential function.

[0039] (4) Ghost area synchronization: calling the synchronization ghost area interface in the three-dimensional neighbor communication module to establish or synchronize particle information on the ghost area;

[0040] (5) Event search and potential energy calculation: The event search algorithm is used to perform KMC event search; the system potential energy required for event transition probability calculation during the search process is completed by calling the potential energy calculation interface in the potential function calculation module;

[0041] (6) Select KMC events;

[0042] (7) Execute the selected event, call the particle storage data structure and neighbor particle search module, update the particle position, and let the system reach a new state.

[0043] (8) Simulation area synchronization: calling the synchronization simulation area interface in the three-dimensional neighbor communication module to synchronize the particle information on the simulation area;

[0044] (9) Return to step (4) and perform the next event search and execution until the simulation end condition is reached.

[0045] Furthermore, in step (5) event search and potential energy calculation, when system potential energy calculation is required, the following steps are adopted:

[0046] 1) Initialize system energy E;

[0047] 2) Traverse all particles, for each particle i, call the particle storage data structure and the neighbor particle search interface in the neighbor particle search module to obtain its neighbor particles;

[0048] 3) For each neighbor particle j of particle i, the potential energy calculation interface in the potential function calculation module is called to calculate the potential energy contribution between particles i and j, and the potential energy contributions are accumulated as the potential energy contribution of particle i to the system.

[0049] 4) The potential energy contribution of particle i to the system is added to E; after traversing all particles, E is the potential energy of the system.

[0050] The present invention also provides an atomic scale MD-KMC parallel simulation unified modeling system, the system comprising:

[0051] Atomic scale unified architecture;

[0052] A molecular dynamics simulation unit, configured to perform molecular dynamics simulation based on the atomic scale unified architecture;

[0053] A kinetic Monte Carlo simulation unit is configured to perform a kinetic Monte Carlo simulation based on the atomic scale unified architecture;

[0054] The simulation synchronization unit is configured to synchronize particle information between the simulation area and the ghost area using a three-dimensional neighbor forwarding communication module during the simulation process, and introduce data compression technology during the synchronization process;

[0055] The simulation prediction unit is configured to carry out MD and KMC coupled simulations based on a unified atomic-scale architecture. After the simulation, based on the atomic coordinate data output by the simulation, the simulation data is subjected to defect and cluster analysis through machine learning algorithms and cluster analysis algorithms to obtain the defect evolution trend of the simulation system and predict the long-term evolution behavior of the system.

[0056] The atomic scale MD-KMC parallel simulation unified modeling method and system according to each solution of the present invention have at least the following technical effects:

[0057] (1) Unified modeling methods facilitate software development and performance optimization. Particle storage, parallel and communication algorithms, and potential functions are separated into separate modules to facilitate the decoupling of MD and KMC software development modules. When optimizing performance on a supercomputing architecture, the effect can be directly applied to the upper-level MD and KMC software at the same time, achieving the effect of "one-place optimization, two-place improvement", supporting MD and KMC simulations to reach large-scale levels at the same time.

[0058] (2) The unified modeling method facilitates efficient data exchange between modules. Traditional MD and KMC simulation software are basically designed and developed independently. When the software interacts, files are generally used for data transfer (that is, after the MD calculation is completed, the result is saved as a file, and then the KMC program reads the file for subsequent calculations. We call this "loose coupling"). This method can achieve efficient tight coupling simulation between software, and can directly exchange data based on memory, which can greatly improve the efficiency and performance of data exchange. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the embodiments of the invention. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0060] Figure 1 The figure shows a schematic diagram of parallel region division and ghost region (taking a two-dimensional system as an example). The 3×3 grid is the parallel region division. The number of parallel MPI processes in the example is 9, and the parallel region division of 9 MPI processes is realized by using the MPI function. The dotted area in the figure is the ghost region of the middle MPI process. The process establishes an area with a width of g at the process boundary to store copies of particles on neighboring processes for subsequent particle calculations. The solid colored area in the middle is the simulation area. This part of the data needs to be sent to the neighboring process to establish the ghost region of the neighboring process. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0062] The embodiment of the present invention provides an atomic scale MD-KMC parallel simulation unified modeling method, as shown in the attached Figure 1 As shown, the method comprises the following steps:

[0063] (1) Establishment of a unified atomic-scale architecture: including the necessary modules for unified MD and KMC modeling, such as particle storage, parallel domain decomposition, three-dimensional neighbor communication, potential function calculation, etc.

[0064] (2) Molecular dynamics simulation based on unified architecture: Call the relevant modules of the "atomic scale unified architecture" to perform parallel molecular dynamics simulation.

[0065] (3) Dynamic Monte Carlo simulation based on unified architecture: Call the relevant modules of "atomic scale unified architecture" to perform parallel dynamic Monte Carlo simulation;

[0066] (4) During the simulation process, a three-dimensional neighbor forwarding communication module is used to synchronize the particle information of the simulation area and the ghost area, and data compression technology is introduced during the synchronization process;

[0067] (5) Based on the unified atomic-scale architecture, MD and KMC coupling simulations are carried out. After the simulation is completed, based on the atomic coordinate data output by the simulation, the simulation data is subjected to defect and cluster analysis through machine learning algorithms and cluster analysis algorithms to obtain the defect evolution trend of the simulation system and predict the long-term evolution behavior of the system.

[0068] In this embodiment, the established atomic scale unified architecture is specifically divided into four sub-modules, each of which is relatively independent. The specific implementation methods of each sub-module are described below:

[0069] (1) Parallel region partitioning module: Given the simulation region size x, y, z (x, y, z are the number of lattices in each direction in three-dimensional space) and the number of parallel MPI processes n, use the MPI function to implement parallel region partitioning, and each MPI process can obtain the size and range of the region it simulates. (See Appendix Figure 1 (shown)

[0070] (2) Particle storage data structure and neighbor particle search module:

[0071] 1) First initialization interface: Given a single MPI process simulation area size (u, v, w) and a ghost area size for communication (g, g, g), apply for particle memory and initialize particle positions as required (such as placing particles on lattice points). This part of the particle storage rules can be implemented using public technologies, such as the neighbor list data structure or cell list data structure used by software such as lammps.

[0072] 2) Neighbor particle search interface: Given a particle information in memory (such as a memory address), return all its neighbor particles from the particle storage space.

[0073] 3) Update particle position interface: based on the given new particle information, update the particle position or speed and other information in the particle storage.

[0074] (3) Potential function calculation module:

[0075] 1) Second initialization interface: read the potential function table file and initialize the potential function data.

[0076] 2) Potential energy calculation interface: Given information such as the distance and type between two atoms, the force between particles or the potential energy contribution of the system is returned according to calculation rules such as EAM potential and LJ potential.

[0077] (4) Three-dimensional Neighborhood Communication Module:

[0078] Synchronous ghost region interface: Given the particle storage data structure (such as memory address) and parallel region division, send the particle information of the simulation region to the neighbor process through MPI message communication; at the same time, receive messages from the neighbor process to synchronize the ghost region. (See Appendix Figure 1 (shown)

[0079] Synchronous simulation area: Given the particle storage data structure (such as memory address), parallel area division and ghost area information (such as memory address), send the particle information of the ghost area to the neighbor process through MPI message communication; at the same time, receive messages from the neighbor process to synchronize the simulation area. Figure 1 shown).

[0080] In this embodiment, the parallel molecular dynamics simulation based on the unified architecture specifically includes the following steps:

[0081] (1) Parallel initialization: Given the simulation region size x, y, z (x, y, z are the number of lattices in each direction in three-dimensional space) and the number of parallel MPI processes n, call the "Parallel Region Partitioning" module to perform parallel initialization.

[0082] (2) Particle storage initialization: Based on the size of the simulation area (u, v, w) after the parallel area division in (1), and the size of the ghost area (width is g), call the "first initialization interface" in the "particle storage data structure and neighbor particle search" module to initialize the particle storage.

[0083] (3) Potential energy initialization: Call the "Second Initialization Interface" in the "Potential Function Calculation" module to initialize the potential function.

[0084] (4) Ghost area synchronization: Call the "Synchronize ghost area interface" in the "3D neighbor communication" module to establish or synchronize particle information on the ghost area.

[0085] (5) Force calculation:

[0086] 1) Traverse all particles, and for each particle i, call the "neighbor particle search interface" in the "particle storage data structure and neighbor particle search" module to obtain its neighbor particles;

[0087] 2) For each neighbor particle j of particle i, call the "Potential Energy Calculation Interface" in the "Potential Function Calculation" module to calculate the force between particles i and j, and add up the force contributions as the force of particle i.

[0088] (6) Based on the force applied to the particle, the “particle storage data structure and neighbor particle search” module is called, and the speed and position of each particle are updated according to the rules of Newton’s laws of motion.

[0089] (7)Simulation area synchronization: Call the “Synchronous Simulation Area Interface” in the “3D Neighbor Communication” module to synchronize the particle information on the simulation area.

[0090] (8) Return to (4) and use the unified architecture to calculate the next force until the given time step is reached.

[0091] In this embodiment, the parallel dynamics Monte Carlo simulation based on the unified architecture specifically includes the following steps:

[0092] (1) Parallel initialization: Given the simulation region size x, y, z (x, y, z are the number of lattices in each direction in three-dimensional space) and the number of parallel MPI processes n, call the "Parallel Region Partitioning" module to perform parallel initialization.

[0093] (2) Particle storage initialization: Based on the size of the simulation area (u, v, w) after the parallel area division in (1), and the size of the ghost area (width is g), call the "first initialization interface" in the "particle storage data structure and neighbor particle search" module to initialize the particle storage.

[0094] (3) Potential energy initialization: Call the "Second Initialization Interface" in the "Potential Function Calculation" module to initialize the potential function.

[0095] (4) Ghost area synchronization: Call the "Synchronize ghost area interface" in the "3D neighbor communication" module to establish or synchronize particle information on the ghost area.

[0096] (5) Event search and potential energy calculation: Use event search algorithms (such as the dimer method) to perform KMC event search. During the search, if system potential energy calculation is required, use the following steps:

[0097] 1) Initialize system energy E;

[0098] 2) Traverse all particles, and for each particle i, call the "neighbor particle search interface" in the "particle storage data structure and neighbor particle search" module to obtain its neighbor particles;

[0099] 3) For each neighbor particle j of particle i, call the "Potential Energy Calculation Interface" in the "Potential Function Calculation" module to calculate the potential energy contribution between particles i and j, and add up the potential energy contributions as the potential energy contribution of particle i to the system.

[0100] 4) The potential energy contribution of particle i to the system is added to E; after traversing all particles, E is the potential energy of the system.

[0101] (6) Select KMC events.

[0102] (7) Execute the selected event and call the "particle storage data structure and neighbor particle search" module to update the particle position and allow the system to reach a new state.

[0103] (8)Simulation area synchronization: Call the “Synchronous Simulation Area Interface” in the “3D Neighbor Communication” module to synchronize the particle information on the simulation area.

[0104] (9) Return to step (4) and perform the next event search and execution until the simulation end condition is reached.

[0105] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0106] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A unified modeling method for atomic scale MD-KMC parallel simulation, characterized in that: The method comprises: Establishing a unified architecture at the atomic scale; Molecular dynamics simulations based on the atomic-scale unified architecture; Dynamical Monte Carlo simulations based on the atomic-scale unified architecture; During the simulation, a three-dimensional neighbor forwarding communication module is used to synchronize the particle information of the simulation area and the ghost area, and data compression technology is introduced during the synchronization process; Based on the unified architecture at the atomic scale, MD and KMC coupling simulations are carried out. After the simulation, based on the atomic coordinate data output by the simulation, defect and cluster analysis are performed on the simulation data through machine learning algorithms and cluster analysis algorithms to obtain the defect evolution trend of the simulation system and predict the long-term evolution behavior of the system.

2. The method according to claim 1, characterized in that The atomic scale unified architecture includes: The parallel region partitioning module is configured as follows: given the simulation region size x, y, z and the number of parallel MPI processes n, where x, y, z are the number of lattices in each direction in the three-dimensional space, the parallel region partitioning is realized by using the MPI function, and each MPI process obtains the size and range of the region it simulates; Particle storage data structure and neighbor particle search module, including: The first initialization interface is configured to: given the size of a single MPI process simulation region (u, v, w) and the size of the ghost region used for communication (g, g, g), apply for particle memory and initialize the particle position as required; The neighbor particle search interface is configured to: given a particle information in the memory, return all its neighbor particles from the particle storage space; The particle position update interface is configured to: update the particle position or velocity information in the particle storage according to the given new particle information; Potential function calculation module, including: The second initialization interface is configured to: read the potential function table file and initialize the potential function data. The potential energy calculation interface is configured to: Given the distance and type between two atoms, return the force between particles or the potential energy contribution of the system according to the EAM potential and LJ potential calculation rules; Three-dimensional neighbor communication module, including: The synchronous ghost region interface is configured to: given the particle storage data structure and parallel region partitioning, send the particle information of the simulation region to the neighbor process through MPI message communication; at the same time, receive messages from the neighbor process to synchronize the ghost region; The synchronous simulation area is configured as follows: given the particle storage data structure, parallel area division and ghost area information, the particle information of the ghost area is sent to the neighbor process through MPI message communication; at the same time, messages are received from the neighbor process to synchronize the simulation area.

3. The method according to claim 2, characterized in that The molecular dynamics simulation based on the atomic scale unified architecture specifically includes: (1) Parallel initialization: given the simulation region sizes x, y, z and the number of parallel MPI processes n, the parallel region partitioning module is called to perform parallel initialization; (2) Particle storage initialization: Based on the simulation area size (u, v, w) after the parallel area division in (1), and the ghost area size, the first initialization interface in the particle storage data structure and the neighbor particle search module is called to initialize the particle storage; (3) Potential energy initialization: calling the second initialization interface in the potential function calculation module to initialize the potential function; (4) Ghost area synchronization: calling the synchronization ghost area interface in the three-dimensional neighbor communication module to establish or synchronize particle information on the ghost area; (5) Force calculation: 1) Traverse all particles, for each particle i, call the particle storage data structure and the neighbor particle search interface in the neighbor particle search module to obtain its neighbor particles; 2) For each neighbor particle j of particle i, the potential energy calculation interface in the potential function calculation module is called to calculate the force between particles i and j, and the force contributions are accumulated as the force of particle i; (6) based on the force on the particle, calling the particle storage data structure and the neighbor particle search module to update the speed and position of each particle; (7) Simulation area synchronization: calling the synchronization simulation area interface in the three-dimensional neighbor communication module to synchronize the particle information on the simulation area; (8) Return to (4) and use the unified architecture to calculate the next force until the given time step is reached.

4. The method according to claim 2, characterized in that: The dynamic Monte Carlo simulation based on the atomic scale unified architecture specifically includes: (1) Parallel initialization: Given the simulation region sizes x, y, z and the number of parallel MPI processes n, call the parallel region partitioning module to perform parallel initialization; (2) Particle storage initialization: Based on the simulation area size (u, v, w) after the parallel area division in (1), and the ghost area size, the first initialization interface in the particle storage data structure and the neighbor particle search module is called to initialize the particle storage; (3) Potential energy initialization: calling the second initialization interface in the potential function calculation module to initialize the potential function; (4) Ghost area synchronization: calling the synchronization ghost area interface in the three-dimensional neighbor communication module to establish or synchronize particle information on the ghost area; (5) Event search and potential energy calculation: Use event search algorithm to perform KMC event search; (6) Select KMC events; (7) Execute the selected event, call the particle storage data structure and neighbor particle search module, update the particle position, and let the system reach a new state. (8) Simulation area synchronization: calling the synchronization simulation area interface in the three-dimensional neighbor communication module to synchronize the particle information on the simulation area; (9) Return to step (4) and perform the next event search and execution until the simulation end condition is reached; In step (5) event search and potential energy calculation, when system potential energy calculation is required, the following steps are adopted: 1) Initialize the system energy count; 2) Traverse all particles, for each particle i, call the particle storage data structure and the neighbor particle search interface in the neighbor particle search module to obtain its neighbor particles; 3) For each neighbor particle j of particle i, the potential energy calculation interface in the potential function calculation module is called to calculate the potential energy contribution between particles i and j, and the potential energy contributions are accumulated as the potential energy contribution of particle i to the system. 4) The potential energy contribution of particle i to the system is added to E; after traversing all particles, E is the potential energy of the system.

5. An atomic scale MD-KMC parallel simulation unified modeling system, characterized in that: The system comprises: Atomic scale unified architecture; A molecular dynamics simulation unit, configured to perform molecular dynamics simulation based on the atomic scale unified architecture; A kinetic Monte Carlo simulation unit is configured to perform a kinetic Monte Carlo simulation based on the atomic scale unified architecture; The simulation synchronization unit is configured to synchronize particle information between the simulation area and the ghost area using a three-dimensional neighbor forwarding communication module during the simulation process, and introduce data compression technology during the synchronization process; The simulation prediction unit is configured to carry out MD and KMC coupled simulations based on a unified atomic-scale architecture. After the simulation, based on the atomic coordinate data output by the simulation, the simulation data is subjected to defect and cluster analysis through machine learning algorithms and cluster analysis algorithms to obtain the defect evolution trend of the simulation system and predict the long-term evolution behavior of the system.

Images