An accelerated molecular dynamics method and system based on dislocation loop habit plane change

Abstract

The application discloses an accelerated molecular dynamics method based on habit plane change of dislocation loop, comprising the following steps: recording the coordinates of each atom in the system when the last event occurs and the coordinates of each atom when the current detection is performed by using a global variable in LAMMPS, inputting all the atom coordinates as parameters into an algorithm for dislocation detection and identification, and searching the dislocations of the system at this time; after the dislocation searching operation is completed, returning the dislocation searching result of this time to the LAMMPS; whenever the LAMMPS obtains the dislocation searching result, updating the corresponding FLAG value according to the need; whenever the value of the FLAG is updated, it is judged as the occurrence of an event, and then the system enters the corresponding decorrelation stage and recycling; whenever the LAMMPS obtains the dislocation searching result, updating the corresponding FLAG value according to the need; whenever the value of the FLAG is updated, it is judged as the occurrence of an event, and then the system enters the corresponding decorrelation stage and recycling.

Description

TECHNICAL FIELD

[0001] The present application relates to the numerical simulation algorithm research technology for the formation mechanism research of gap dislocation loop (dislocation loop is a ring-shaped edge dislocation line in crystal), and particularly relates to an accelerated molecular dynamics method and system based on dislocation loop habit plane change. BACKGROUND

[0002] Accelerated molecular dynamics (AMD) is a way to extend the simulation time scale. Common accelerated molecular dynamics methods include parallel replica replication (PRD), temperature acceleration (TAD), and hyperdynamics (HD). Compared with direct molecular dynamics simulation, these AMD methods can increase the simulation time scale by several orders of magnitude, so that more important phenomena can be observed.

[0003] Although current molecular dynamics (MD) can achieve effective processing for large systems (10^6-10^8 atoms) with the help of parallel computing, for some smaller systems, it will be limited by inter-processor communication. Moreover, for many classical systems, their characteristics are long-term residence in a certain state, and occasional state transitions. Therefore, how to effectively use the parallel computer to extend the simulation experiment in time scale with high parallel efficiency becomes a key point. PRD is to replicate the system into multiple copies, and each copy is simulated using different processors. According to the transition state theory (TST), parallel running of multiple independent trajectories can give the exact dynamics evolution from one state to another. Therefore, PRD can effectively extend the time scale for the dynamics research of small and rare event systems. The process of using PRD to simulate the system is as follows:

[0004] Step one, replicate the current configuration of the system on M processors.

[0005] Step two, perform minimization to generate a reference configuration for transition checking.

[0006] Step three, perform phase-out on each processor, and go through the momentum randomization stage to eliminate the correlation with other copies.

[0007] Step four, check whether a state transition occurs in each copy every tblock time, and when a state transition occurs in one of the copies, it is defined as an event, and all other copies stop simulation.

[0008] Step five, the simulation time is advanced by tsum, that is, the total time of all copies from step three to the occurrence of the event.

[0009] Step six, start the decorrelation phase, the copy of the event continues to simulate a period of time (Δt corr ≥t corr ), if another event occurs in this period of time, it is considered to be a relevant event, and will be re-decorrelated.

[0010] Step seven, the copy that completes the decorrelation will become the new configuration of the system at this time, and return to step one.

[0011] In the above process, it can be found that if the system waits for a period of time before a new state transition occurs that is much longer than the time required for decorrelation, and the clock between each transition is much larger than the communication time between copies, then the entire process will be efficient and parallel.

[0012] But for the simulation of dislocation loop transition involving <1 0 0> ring to 1 / 2 <1 1 1> ring, the existing PRD simply judges whether an event occurs according to whether the displacement of an atom in the system reaches a threshold value, but in the simulation of dislocation loop transition, the diffusion of the dislocation loop itself will also cause the atoms to continuously displace, and the diffusion of the dislocation loop is meaningless for observing the phenomenon of dislocation loop transition. The phenomenon that really needs to be captured is the rotation of the dislocation loop rather than the diffusion, but the original PRD cannot distinguish between diffusion and rotation phenomena simply by the displacement of the atom. Therefore, in the experiment of simulating the dislocation loop using the original PRD, the accuracy of the simulation is greatly affected. For example, in the experiment of simulating a <1 0 0> ring of 25 size, the traditional MD method can see the phenomenon of transition to 1 / 2 <1 1 1> ring occur within 0.3-0.5 ns, while using the original PRD requires acceleration to tens of nanoseconds to see it. Even as the temperature rises, the dislocation loop diffusion is more active, and the atomic displacement caused by diffusion is more obvious, so that the original PRD not only reduces the simulation accuracy to a greater extent, but also seriously affects the speedup ratio. SUMMARY

[0013] The purpose of the present application is to solve the above technical problems existing in the simulation of dislocation loop transition in the prior art PRD.

[0014] To achieve the above purpose, on the one hand, the present application provides an accelerated molecular dynamics method based on the change of dislocation loop habit plane, which comprises the following steps:

[0015] The global variable in the Large-scale Atomic / Molecular Massively Parallel Simulator (LAMMPS) records the coordinates of each atom in the system at the last event occurrence and the coordinates of each atom at the current detection, and all atom coordinates are transmitted as parameters into the dislocation detection and identification algorithm, and the dislocation of the system at this time is searched; after the dislocation search operation is completed, the dislocation search result of this time is returned to the LAMMPS;

[0016] Whenever the LAMMPS obtains the dislocation search result, the corresponding FLAG value is updated as needed;

[0017] Whenever the value of the FLAG is updated, it is determined as an occurrence of an event, and then the system enters the corresponding decorrelation stage and recycling.

[0018] On the other hand, the application provides an accelerated molecular dynamics system based on dislocation loop habit plane change, comprising:

[0019] A first processing module is configured to record the coordinates of each atom in the system at the last event occurrence and the coordinates of each atom at the current detection by using a global variable in the Large-scale Atomic / Molecular Massively Parallel Simulator (LAMMPS), and all atom coordinates are transmitted as parameters into the dislocation detection and identification algorithm, and the dislocation of the system at this time is searched; after the dislocation search operation is completed, the dislocation search result of this time is returned to the LAMMPS;

[0020] A second processing module is configured to update the corresponding FLAG value as needed whenever the LAMMPS obtains the dislocation search result.

[0021] A third processing module is configured to enter the corresponding decorrelation stage and recycling whenever the value of the FLAG is updated.

[0022] The application is based on the implementation of PRD in LAMMPS, integrates the algorithm for automatic recognition and indexing of dislocations in crystal interfaces in the molecular dynamics simulation software OVITO, and realizes the detection of whether the Burgers vector of the current dislocation ring changes every time in the process of simulating the dislocation ring by using the PRD method, so as to accurately capture each transition phenomenon of the dislocation ring. Compared with the original PRD, the improvement of the application is in the judgment and capture of events. By analyzing the characteristics of the dislocation ring and some changes in the habit plane and other characteristics during the transition, the capture of each event by PRD is beneficial to the purpose of the whole dislocation ring transition. Considering that PRD itself is based on the principle of transition state theory, the accuracy of the accelerated simulation by the replica parallel method has been proved in various experiments in the past. Therefore, the improved PRD method of the application has certain support in principle and practicality for simulating dislocation rings. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A flowchart of an accelerated molecular dynamics method based on the change of the habit plane of the dislocation ring is provided for the implementation of the application.

[0024] Figure 2 A flowchart of the algorithm for dislocation detection and recognition in the method shown. Figure 1

[0025] Figure 3 Three basic stages of whether a flag event occurs in the method shown. Figure 1

[0026] Figure 4 Fine-grained events in the FLAG=1 state in the method shown. Figure 1

[0027] A schematic diagram of an accelerated molecular dynamics system structure based on the change of the habit plane of the dislocation ring is provided for the implementation of the application. Figure 5 DETAILED DESCRIPTION The specific examples of the application are described below in conjunction with the accompanying drawings of the specification:

[0028]

[0029] ​​​The core idea of the embodiment of the present application for the improved PRD is whether the judgment standard of the PRD for the event is reasonable. The original PRD only judges whether an event occurs according to whether the displacement of an atom in the system reaches a threshold value, however, in the experiment of simulating the dislocation loop transition, the diffusion of the dislocation loop itself also causes the atoms to continuously displace, and the diffusion of the dislocation loop is meaningless for observing the phenomenon of dislocation loop transition. The phenomenon that actually needs to be captured is the rotation of the dislocation loop rather than the diffusion, but the original PRD cannot distinguish the diffusion phenomenon from the rotation phenomenon only by the displacement of the atom.

[0030] The embodiment of the present application notices that the transition of the dislocation loop is often accompanied by the change of the habit plane, and the direction of the habit plane is related to the Burgers vector corresponding to the habit plane. This means that if the Burgers vector of the dislocation loop is extracted and analyzed, and the change of the Burgers vector is considered to affect the transition of the dislocation loop, we can use the change as the basis for the PRD to judge whether an event occurs, so as to make corresponding improvements to the PRD, so that the accelerated molecular dynamics method can be well applied to the experiment of simulating the dislocation loop.

[0031] According to the theory proposed by Frank, the dislocation can be identified by the Burgers loop, that is, given a closed path C, the Burgers vector b of the closed dislocation defect is defined as the line integral of the elastic displacement field:

[0032]

[0033] Where, U e is the elastic displacement vector field under the space configuration, and the closed path C is the path from atom to atom in the space.

[0034] According to some characteristics of the dislocation loop and simulation, when the dislocation loop transitions, the phenomenon of rotation of the dislocation loop often occurs, at this time, the habit plane of the dislocation loop also changes with the rotation. A <1 0 0> dislocation loop can change the habit plane and cause the Burgers vector to change by means of gap rearrangement, and finally make the whole dislocation loop rotate to a 1 / 2 <1 1 1> loop.

[0035] According to the algorithm proposed by OVITO for the automatic identification and indexing of dislocations in crystal interfaces, the most important thing to achieve automatic dislocation identification through Burgers vectors is to ensure that the closed path C from atom to atom mentioned in equation (1) is selected so that it exactly includes a single dislocation. In addition, it is necessary to assume an ideal perfect lattice (a lattice is a spatial framework representing the regular arrangement of atoms in a crystal), and let its atom-to-atom path be ΔX. The atom-to-atom path in the closed path C that constitutes the actual Burgers vector is set as Δx. We need to map the actual path to the ideal path, that is, to complete the mapping of Δx→ΔX, so as to show the incompatible displacement caused by the dislocation and determine its Burgers vector, and finally achieve the function of automatic dislocation identification.

[0036] This invention employs a suitable shortest path search algorithm to ensure that each vector Δx in the Burgers loop connects to its two nearest neighbors (NN). This invention also uses a dislocation identification algorithm to obtain the most accurate solution for irreducible loops, thus ensuring that the algorithm can identify each individual dislocation segment. By calculating and dividing the crystal into good and bad regions, and ensuring that both conditions are met, a suitable Burgers loop and each indivisible vector Δx within the loop can be obtained. The algorithm then uses these vectors to construct triangular mesh surfaces, sequentially scanning these mesh surfaces along the loop and continuously calculating the current dislocation segment. Finally, the algorithm finds each distinct dislocation segment in the crystal and integrates those that can be connected, transforming them into a continuous dislocation line.

[0037] Figure 1 This is a schematic diagram of an accelerated molecular dynamics method based on the change of dislocation loop habit plane, provided for the implementation of this invention. Figure 1 As shown, the method includes steps S101-S103:

[0038] The logic of the original PRD method is as follows: In LAMMPS, the user first specifies the detection interval, decorrelation time, and atom offset of the PRD. Then, the PRD will record the position coordinates of each atom in the entire system at fixed intervals according to the user-defined parameters, and compare them with the atom coordinates recorded when the previous event occurred. If the offset of any atom in the system exceeds the user-defined threshold at this time, it is considered that an event has occurred. Since the PRD uses a parallel simulation method with multiple replicas, multiple replicas may detect the occurrence of the event at the same time. In this case, the PRD will randomly select a replica for decorrelation, and finally repeat the entire process.

[0039] This invention improves the event determination logic of the PRD in LAMMPS. The improvement steps include:

[0040] Step S101: The coordinates of each atom in the system at the time of the last event and the coordinates of each atom at the current detection time are recorded using global variables in the large-scale atomic and molecular parallel simulator LAMMPS. All atom coordinates are passed as parameters to the algorithm for dislocation detection and identification, and dislocations in the system at this time are searched. After the dislocation search operation is completed, the result of this dislocation search is returned to LAMMPS.

[0041] Specifically, firstly, based on the atomic coordinates recorded by LAMMPS, the Delaunay triangulation algorithm is used to decompose the atomic space into many tetrahedra. Then, for each edge of each tetrahedron, an attempt is made to map the atom to the ideal atomic path Δx→ΔX. If all six edges of a tetrahedron can complete this mapping, it is defined as a good tetrahedron; otherwise, it is considered a bad tetrahedron. After this classification of all tetrahedra, the system space can be divided into good and bad regions, with the bad regions containing crystal defects. Next, Burgers loops are constructed based on the edges of the tetrahedra in the bad regions. To ensure that the Burgers loops are irreducible loops containing a single dislocation, a recursive algorithm is used to select the loop with increasing length. When the calculated Burgers vector is not 0, an irreducible loop is determined. After scanning all bad regions in the system space in this way, all dislocation segments can be obtained. Finally, these dislocations are categorized according to Burgers vectors, the total length of different dislocation segments is calculated, and the results are sent back to LAMMPS for event determination. The specific process is as follows: Figure 2 As shown.

[0042] To better define the criteria for PRD (Programming Decision) event determination, we divided the system into three phases, such as... Figure 3 As shown:

[0043] The system contains only <1 0 0> dislocation cycles.

[0044] The system contains both <1 0 0> and 1 / 2 <1 1 1> dislocation loops.

[0045] The system contains only 1 / 2 <1 1 1> dislocation cycles.

[0046] These three stages correspond to three different values ​​of the FLAG set in LAMMPS. Each time LAMMPS obtains the result of a bit error retrieval, it updates the corresponding FLAG value as needed. The FLAG values ​​are 0, 1, and 2, corresponding to the three stages mentioned above.

[0047] In step S102, at the beginning of the simulation, the system is initially configured to have only <1 0 0> dislocation loops, meaning that the Burgers vectors in the system only have the <1 0 0> direction. Therefore, the FLAG value is set to 0. As the simulation progresses and the PRD is periodically detected, when a 1 / 2 <1 1 1> Burgers vector appears in the system, the FLAG value is set to 1, and the system waits for the <1 0 0> dislocation loop to completely transform into a 1 / 2 <1 1 1> loop. Finally, if the <1 0 0> loop in the system completely transforms into a 1 / 2 <1 1 1> loop, according to the dislocation detection algorithm mentioned in step S101, only a 1 / 2 <1 1 1> Burgers vector remains, so the FLAG value will be updated to 2. As for PRD, apart from the case where FLAG changes from 1 back to 0 (this change represents the dislocation loop returning from an intermediate state of a transition to the initial state, which can be understood as failing to cross the potential barrier required for the transition, equivalent to a failed attempt, and has no positive significance for observing the phenomenon of a complete transition).

[0048] In step S103, each time the value of FLAG is updated, it is determined as an event occurring, and then the system enters the corresponding decorrelation stage and re-loop.

[0049] To further ensure that each event detected by the PRD is conducive to the transition from the <1 0 0> ring to the 1 / 2 <1 1 1> ring, thereby achieving better acceleration, in the FLAG=1 stage, we also statistically analyze the proportion of the 1 / 2 <1 1 1> Burger vectors to the total Burger vectors of the system, thus refining the FLAG=1 stage. For example... Figure 3 As shown, during each detection, the proportion of the Burgers vector of 1 / 2 <11 1> of the system detected in the current replica is compared with the proportion at the time of the previous event. If it is higher than the previous proportion, considering the nature of dislocation loop transformation, it indicates that the habituation plane of the dislocation loop in the system has changed to a greater extent due to rotation, and therefore can be considered a state more likely to undergo a complete transformation. Based on this, we also use this situation as a criterion for PRD to determine whether an event has occurred, thus avoiding the problem that the simulation system may remain in the FLAG=1 stage for a long time, resulting in unsatisfactory acceleration.

[0050] Accordingly, embodiments of the present invention provide an accelerated molecular dynamics system based on the variation of the dislocation loop habit surface. For example... Figure 4 As shown, the system includes:

[0051] The first processing module 41 is used to record the coordinates of each atom in the system when the last event occurred and the coordinates of each atom at the current detection time using global variables in the large-scale atomic and molecular parallel simulator LAMMPS. It also passes all the atom coordinates as parameters to the algorithm for dislocation detection and identification, and searches for dislocations in the system at this time. After completing the dislocation search operation, it returns the result of the dislocation search to LAMMPS.

[0052] The second processing module 42 is used to update the corresponding FLAG value as needed whenever LAMMPS obtains the result of the bit error retrieval.

[0053] The third processing module 43 is used to enter the corresponding decorrelation stage and recycle whenever the value of FLAG is updated.

[0054] This invention aims to extract features such as the characteristics of dislocation loops and changes in habit surface during transitions by analyzing them, so that the PRD's capture of each event is conducive to the transformation of the entire dislocation loop.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An accelerated molecular dynamics method based on dislocation loop habit plane change, characterized in that: a global variable in a large-scale atom-molecule parallel simulator LAMMPS is used to record the coordinates of each atom in the system at the time of the last event and the coordinates of each atom at the time of the current detection, and all atomic coordinates are transmitted as parameters into a dislocation detection and identification algorithm, and the dislocations of the system at this time are searched; after completing the dislocation search operation, the results of this dislocation search are returned to the LAMMPS; each time the LAMMPS obtains the results of the dislocation search, the corresponding FLAG value is updated as needed; The above steps include: filling the atomic space with Delaunay triangulation algorithm to decompose it into a plurality of tetrahedrons according to the atomic coordinates recorded by LAMMPS; then trying to perform atom-to-atom ideal path for each edge of each tetrahedron If the mapping can be completed for all the six edges of a tetrahedron, the tetrahedron is defined as a good tetrahedron, otherwise it is considered as a bad tetrahedron; after classifying all the tetrahedrons, the system space is divided into good regions and bad regions, wherein the bad regions contain crystal defects; then the Burgers loop is constructed according to each edge of the tetrahedrons in the bad regions, and a recursive algorithm is used to select the length in an increasing manner, and when the calculated Burgers vector is not 0, an irreducible loop is determined; after scanning all the bad regions in the system space, all the dislocation segments at present are obtained; finally, the dislocations are classified according to the Burgers vector, the total length of different dislocation segments is calculated, and the result is returned to LAMMPS. the step of updating the corresponding FLAG value as needed comprises: the initial configuration of the system is only <1 0 0> dislocation loop, so the FLAG at this time is 0; when 1 / 2 <1 1 1> Burgers vector appears in the system, the FLAG of the system at this time is set to 1, and the occurrence of the <1 0 0> dislocation loop completely changing into 1 / 2 <1 1 1> loop is started to be waited for; if the <1 0 0> loop in the system completely changes into 1 / 2 <1 1 1> loop, according to the dislocation detection algorithm, only 1 / 2 <1 1 1> Burgers vector is left at this time, so the FLAG value will be updated to 2; each time the value of the FLAG is updated, it is determined as an occurrence of an event, and then the system enters the corresponding decorrelation stage and recirculation. in the stage of FLAG = 1, the proportion of 1 / 2 <1 1 1> Burgers vector in the total Burgers vector of the system is counted and analyzed; if it is higher than the last proportion, it is determined that the dislocation loop habit plane in the system rotates and changes.

2. The method of claim 1, wherein, comprise:

3. An accelerated molecular dynamics system based on dislocation loop habit plane change, characterized by, a first processing module for using a global variable in a large-scale atom-molecule parallel simulator LAMMPS to record the coordinates of each atom in the system at the time of the last event and the coordinates of each atom at the time of the current detection, and transmitting all atomic coordinates as parameters into a dislocation detection and identification algorithm, and searching the dislocations of the system at this time; after completing the dislocation search operation, the results of this dislocation search are returned to the LAMMPS; a second processing module for updating the corresponding FLAG value as needed each time the LAMMPS obtains the results of the dislocation search; The first processing module is specifically used for filling and decomposing the atomic space into a plurality of tetrahedrons by using a Delaunay triangulation algorithm according to atomic coordinates recorded by LAMMPS; then attempting to perform atom-to-atom ideal path on each edge of each tetrahedron If the mapping can be completed on six edges of a tetrahedron, the tetrahedron is defined as a good tetrahedron, otherwise, the tetrahedron is defined as a bad tetrahedron; after classifying all the tetrahedrons, the system space is divided into a good region and a bad region, wherein the bad region contains crystal defects; then constructing a Burgers loop according to each edge of the tetrahedron in the bad region, and selecting the length in an increasing manner by using a recursive algorithm, when the calculated Burgers vector is not 0, an irreducible loop is determined; after scanning all the bad regions in the system space, all the dislocation segments are obtained; finally, the dislocations are classified according to the Burgers vector, the total length of different dislocation segments is calculated, and the result is returned to LAMMPS. the step of updating the corresponding FLAG value as needed comprises: the initial configuration of the system is only <1 0 0> dislocation loop, so the FLAG at this time is 0; when 1 / 2 <1 1 1> Burgers vector appears in the system, the FLAG of the system at this time is set to 1, and the occurrence of the <1 0 0> dislocation loop completely changing into 1 / 2 <1 1 1> loop is started to be waited for; if the <1 0 0> loop in the system completely changes into 1 / 2 <1 1 1> loop, according to the dislocation detection algorithm, only 1 / 2 <1 1 1> Burgers vector is left at this time, so the FLAG value will be updated to 2; ​ The third processing module is configured to enter a corresponding de-correlation stage and recirculation whenever the value of the FLAG is updated.

4. The system of claim 3, wherein, In the stage of FLAG=1, the proportion of Burgers vector of 1 / 2<1 1 1> in the total Burgers vector of the system is counted and analyzed; if it is higher than the last time, it is determined that the habit plane of the dislocation loop in the system is rotated and changed.

Images