Method for automatically analyzing atomic structure of amorphous material on multiple scales

Abstract

The present application relates to the technical field of computers, and relates to a kind of amorphous material atomic structure multiscale automatic analysis method. Including: step S1: with reactant as initial unit, based on first principle molecular dynamics simulation obtains the cluster unit of reaction product;Step S2: the cluster unit is constructed, and its cell parameters and atomic coordinates are optimized using first principle calculation, to obtain cell unit;Step S3: the cell unit is stacked in space, and its cell parameters and atomic coordinates are optimized using first principle calculation, to obtain supercell unit;Step S4: the supercell unit is simulated by first principle molecular dynamics, to obtain accurate machine learning potential function;Step S5: the supercell unit is constructed by cell expansion, and the molecular dynamics simulation is carried out using accurate machine learning potential function, to obtain amorphous material atomic structure;Step S6: steps S1 to S5 are packaged into automation workflow. The method is efficient, and the error is small.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a multi-scale automatic analysis method for atomic structure of amorphous material. BACKGROUND

[0002] Amorphous materials exhibit unique physical and chemical properties due to the lack of long-range ordered structure, and have broad application prospects in new energy, aerospace and other fields. However, the complex atomic arrangement characteristics of amorphous materials also make the structural analysis extremely challenging. At present, the main methods for analyzing the atomic structure of amorphous materials include reverse Monte Carlo (RMC) simulation and high-temperature annealing molecular dynamics (MD) simulation.

[0003] Reverse Monte Carlo simulation is to obtain the pair distribution function (PDF) data from experiments, and then randomly adjust the atomic coordinates to match the experimental results by combining Monte Carlo algorithm. However, this method has significant limitations: first, RMC is an empirical algorithm, and the generated atomic structure of amorphous material has not been verified by first principle or dynamic simulation, resulting in lack of theoretical support for the rationality of the structure; second, this method cannot capture the dynamic behavior in the amorphous formation process, making it difficult to explain the relationship between local order and macroscopic performance of amorphous.

[0004] Although high-temperature annealing molecular dynamics simulation can overcome some of the above problems of reverse Monte Carlo simulation, it is limited by the constraint of the initial crystal structure, and the MD simulation results often deviate from the experimental observation of the atomic structure characteristics of amorphous materials. For example, traditional MD simulation requires setting a crystal or ordered structure as the starting point, and the complex dynamic behavior of amorphous formation process (such as lattice distortion and defect diffusion) cannot be fully characterized by the conventional simulation time scale (usually <100 picoseconds).

[0005] In summary, reverse Monte Carlo simulation cannot directly analyze new materials without experimental data because it relies on experimental PDF data; and high-temperature annealing molecular dynamics simulation is limited by the initial crystal structure, making it difficult to obtain the experimental synthesized amorphous structure. The deficiencies of the prior art can be summarized as the following key problems in the field of material theoretical calculation:

[0006] (1) Conflict between computational efficiency and scale: traditional first principle method is only suitable for small scale system (such as less than 200 atoms), and it is difficult to directly simulate large scale amorphous system (usually more than 1000 atoms), resulting in lack of representativeness and rationality of the analysis results.

[0007] (2) Kinetic time scale limitation: Amorphous formation process involves complex kinetic behavior, and conventional molecular dynamics simulation is difficult to span a sufficient time scale (> 100 picoseconds) to capture the details of structural evolution.

[0008] (3) Potential function precision limitation: Empirical potential functions (such as Lennard-Jones, Born-Mayer, embedded atom model EAM, etc.) cannot accurately describe the complex interactions in multi-component systems (such as amorphous alloys), resulting in significant structural prediction errors.

[0009] (4) Inefficient human intervention: Existing methods rely on human intervention for data transfer and parameter adjustment, resulting in error accumulation and a process that takes weeks.

[0010] In addition, existing solutions mostly use single-scale simulation (such as relying only on experimental data) or manual intervention for data transfer (such as manually adjusting parameters), which are inefficient, error-prone, and difficult to achieve automation in analyzing amorphous material atomic structure at atomic scale. In the prior art, there is no method that combines the high precision of first principles with the efficiency of machine learning potential functions, while achieving full automation of analyzing amorphous structure from atomic clusters to large-scale structures. SUMMARY

[0011] The technical problem to be solved by the present application is that existing methods for analyzing amorphous material atomic structure are inefficient, error-prone, and difficult to achieve automation in analyzing amorphous material atomic structure at atomic scale. To overcome the defects of the prior art, the present application provides a multi-scale automated analysis method for amorphous material atomic structure.

[0012] The present application provides a multi-scale automated analysis method for amorphous material atomic structure, comprising the following steps:

[0013] Step S1: Taking the reactants forming the amorphous material as the initial unit, and obtaining the cluster unit generated by the reactants based on first-principle molecular dynamics simulation;

[0014] Step S2: Based on the cluster unit, optimize its cell parameters and atomic coordinates based on first-principle calculation to obtain a cell unit;

[0015] Step S3: Stack the cell unit in space, optimize its cell parameters and atomic coordinates based on first-principle calculation to obtain a supercell unit;

[0016] Step S4: Perform first-principle molecular dynamics simulation on the supercell unit, and sample the data set from the simulation trajectory, train the data set to obtain an accurate machine learning potential function;

[0017] Step S5: cell expansion construction is performed on the supercell unit, and a machine learning potential function driven molecular dynamics simulation is performed based on the machine learning potential function, to obtain an amorphous material atomic structure;

[0018] Step S6: encapsulating the steps S1 to S5 into an automatic workflow for automatically analyzing the amorphous material atomic structure.

[0019] Compared with the prior art, the present application has the following advantages: the present application constructs a multi-scale modeling system of chemical reaction -> cluster unit -> unit cell -> supercell unit -> amorphous material atomic structure, deeply integrates machine learning algorithm and automatic calculation process, combines first principle calculation, high-precision machine learning potential function and automatic workflow engine, and systematically solves the long-standing cross-scale modeling bottleneck and precision-efficiency balance problem in amorphous material atomic structure analysis. The technical scheme of the present application improves the precision of amorphous material atomic structure analysis, and has important significance for promoting the application of amorphous material science and engineering.

[0020] In one possible implementation, the step S1 specifically includes the following steps:

[0021] Step S1.1: constructing a minimum cluster unit of reactants, the minimum cluster unit being the smallest molecular composition in the reactants;

[0022] Step S1.2: determining the reaction ratio between each minimum cluster unit, constructing each minimum cluster unit in the same unit cell, the vacuum layer thickness in each direction being greater than 1 nm, and the atomic spacing between adjacent minimum cluster units being greater than 1 nm, to form a cluster reaction model;

[0023] Step S1.3: performing first principle molecular dynamics simulation on the cluster reaction model, using the NVT ensemble of constant particle number, volume and temperature, and the simulation time being greater than 3 picoseconds, to generate a cluster unit.

[0024] In the above scheme, the first principle MD is suitable for theoretical calculation of small system (<100 atoms) and short time scale (<100 ps), can realize high-precision simulation of chemical reaction process without relying on empirical parameters, can dynamically capture atomic trajectory and energy evolution, and thus accurately characterize bond breaking / forming events in the reaction path, and obtain chemical reaction products in a real environment, so that the cluster unit generated based on the first principle molecular dynamics simulation is accurate.

[0025] In one possible implementation, the step S2 specifically includes the following steps:

[0026] Step S2.1: extending the cluster unit into a periodic cluster unit, the vacuum layer thickness of the periodic cluster unit being not more than 1 nm;

[0027] Step S2.2: Geometric optimization of the unit cell parameters and atomic coordinates of the periodic cluster unit based on first-principles calculation, to obtain a unit cell.

[0028] In the above scheme, for the unit cell with periodic characteristics, the first-principles calculation can be performed without empirical parameters for high-precision simulation calculation, so that the first-principles calculation can simulate an accurate unit cell.

[0029] In one possible implementation, in step S2.2, the specific parameters for optimizing the unit cell parameters and atomic coordinates of the periodic cluster unit are: variable unit cell shape and symmetry, and external pressure of 1-1000 atmospheres.

[0030] In one possible implementation, the step S3 specifically includes the following steps:

[0031] Step S3.1: Stacking the unit cells in a close-packed manner to form a supercell, and the size of the supercell is at least 2 times larger than the unit cell along the XYZ three directions.

[0032] Step S3.2: Geometric optimization of the unit cell parameters and atomic coordinates of the supercell based on first-principles calculation, to obtain a supercell unit.

[0033] In one possible implementation, in step S3.2, the specific parameters for optimizing the unit cell parameters and atomic coordinates of the supercell are: variable unit cell shape and symmetry, and external pressure of 1-1000 atmospheres.

[0034] In one possible implementation, the step S4 specifically includes the following steps:

[0035] Step S4.1: First-principles molecular dynamics simulation of the supercell unit, using the NVT ensemble with constant particle number, volume and temperature, and the simulation time is greater than 3 picoseconds.

[0036] Step S4.2: Extracting the atomic coordinates, unit cell parameters, structure energy, atomic force, and unit cell stress data in the simulation trajectory of the first-principles molecular dynamics simulation in step S4.1, and constructing a high-dimensional training data set.

[0037] Step S4.3: Training a machine learning potential function based on the high-dimensional training data set, and the determination coefficients between the observation data and the test set data of each physical quantity in the prediction of the structure energy, atomic force and unit cell stress are all higher than 0.98, to obtain a trained machine learning potential function.

[0038] In one possible implementation, the step S5 specifically includes the following steps:

[0039] Step S5.1: cell expansion construction is performed on the supercell unit, which is expanded by at least 2 times in three directions of XYZ, to obtain an initial amorphous material atomic structure;

[0040] Step S5.2: the initial amorphous material atomic structure is subjected to machine learning potential function driven molecular dynamics simulation by using the accurate machine learning potential function obtained in step S4, to obtain a final amorphous material atomic structure.

[0041] In the above scheme, the high-precision machine learning potential function constructed based on the first-principle calculation provides a reliable energy-force field description for the large-scale long-range disordered structure calculation of amorphous materials. When dealing with such super-large systems containing thousands of atoms or more, the machine learning potential function driven molecular dynamics simulation can significantly improve the calculation efficiency while reasonably describing the atomic scale bonding and relaxation process. This method effectively balances the calculation accuracy and resource consumption while ensuring the physical rationality of the structure evolution, so that the reliability of the amorphous material atomic structure prediction reaches the experimental characterization level.

[0042] In a possible implementation, step S5.2 specifically includes the following processes:

[0043] Step S5.2.1: the initial amorphous material atomic structure is subjected to machine learning potential function driven molecular dynamics simulation, and an NPT ensemble with constant particle number, pressure and temperature is selected, the pressure is limited to 1000-10000 atmospheres, the temperature is greater than 300 Kelvin, and the simulation time is greater than 50 picoseconds, to obtain a pressure-balanced amorphous material atomic structure;

[0044] Step S5.2.2: the pressure-balanced amorphous material atomic structure obtained in step S5.2.1 is subjected to machine learning potential function driven molecular dynamics simulation for temperature rising and falling, and an NVT ensemble is selected, the temperature range is 300-400 Kelvin, and the temperature rising / dropping rate is greater than 1 picosecond / Kelvin, to obtain a final amorphous material atomic structure.

[0045] In a possible implementation, the step S6 specifically includes the following steps:

[0046] Step S6.1: encapsulating the step S1 into a reactant input module and a cluster unit module;

[0047] Step S6.2: encapsulating the step S2 into a unit cell module;

[0048] Step S6.3: encapsulating the step S3 into a supercell unit module;

[0049] Step S6.4: encapsulating the step S4 into a machine learning potential function training module;

[0050] Step S6.5: encapsulating the step S5 as an amorphous material atomic structure simulation module;

[0051] Step S6.6: integrating each module obtained from the step S6.1 to step S6.5.

[0052] The beneficial effects of the present application are:

[0053] 1. The present application constructs a multi-scale modeling framework of chemical reaction principle → cluster unit → unit cell → supercell → amorphous material atomic structure, and drives the structure evolution through chemical reaction kinetics mechanism (such as diffusion, nucleation, phase transition). Compared with the existing method, the accuracy and rationality of the amorphous structure are improved and guaranteed.

[0054] 2. The present application adopts an accurately trained machine learning potential function, which can accurately and efficiently simulate complex amorphous materials (such as amorphous materials containing more than 5 elements), and has the advantages of accuracy and efficiency compared with the existing solutions.

[0055] 3. The present application adopts the analytical method of chemical reaction principle between reactants → cluster unit → unit cell unit → supercell unit → amorphous material atomic structure, which can capture atomic rearrangement mechanism (such as local order → long-range disorder transition), defect diffusion path and phase transition critical point, and observe the specific process of the formation of amorphous material atomic structure. Compared with the existing solutions, it has the advantage of accurately restoring the dynamic amorphous material atomic structure.

[0056] 4. The analytical method starting from the chemical reaction principle between reactants in the present application can predict the amorphous materials formed by various reactants, and is suitable for complex systems containing metal / non-metal, main group / transition group elements (such as amorphous high-entropy alloy). Compared with the existing solutions, it has the advantages of good robustness and prediction ability.

[0057] 5. The present application realizes the full-process automation from the input of reactants to the output of amorphous material atomic structure through code algorithm encapsulation and modular workflow engine. Compared with the existing solutions, it can improve the efficiency of amorphous material structure analysis and reduce the cost of amorphous material research and development. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 The technical scheme flow chart of the amorphous material atomic structure multi-scale automatic analysis method of the present application;

[0059] Figure 2 The reactant schematic diagram in embodiment 1 of the present application;

[0060] Figure 3 The cluster unit schematic diagram in embodiment 1 of the present application;

[0061] Figure 4A schematic diagram of a unit cell in Example 1 of the present application;

[0062] Figure 5 A schematic diagram of a supercell unit in Example 1 of the present application;

[0063] Figure 6 A schematic diagram of the prediction accuracy of the structure energy of the machine learning potential function in Example 1 of the present application;

[0064] Figure 7 A schematic diagram of the prediction accuracy of the atomic force of the machine learning potential function in Example 1 of the present application;

[0065] Figure 8 A schematic diagram of the prediction accuracy of the unit cell stress of the machine learning potential function in Example 1 of the present application;

[0066] Figure 9 A schematic diagram of the initial amorphous structure in Example 1 of the present application;

[0067] Figure 10 A schematic diagram of the atomic structure of the final amorphous material in Example 1 of the present application;

[0068] Figure 11 A module packaging automation flowchart in the method for automatically analyzing the atomic structure of amorphous materials at multiple scales of the present application;

[0069] Figure 12 A comparison chart of the theoretical pair distribution function and the experimental sample pair distribution function of the atomic structure of the final amorphous material in Example 1 of the present application;

[0070] Figure 13 A schematic diagram of the atomic structure of the amorphous material obtained in Example 2 of the present application;

[0071] Figure 14 A comparison of the theoretical pair distribution function and the experimental sample pair distribution function of the atomic structure of the final amorphous material in Example 2 of the present application;

[0072] Figure 15 A schematic diagram of the atomic structure of the amorphous material obtained in Example 3 of the present application. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below. It should be noted that the following embodiments are only used to illustrate the implementation method and typical parameters of the present application, and do not limit the parameter range described in the present application, and reasonable changes derived therefrom are still within the protection scope of the claims of the present application.

[0074] It is to be understood that the endpoints of the ranges specified in this disclosure are too be understood as substantially the endpoints of the ranges, and that the description in sections below identifies many embodiments that fall within the scope of the disclosure. Moreover, all numerical values are "approximate", meaning that the term "comprising" is used in the appended claims as well as in the body of the specification. Any numerical value, however, can include fractions of the recited value, ranges deriving from the recited value, and ranges larger than the recited value, as if each and every narrower numerical range is expressly and specifically recited herein.

[0075] Unless otherwise defined, all terms, symbols and other scientific terminology used herein are to be interpreted in accordance with their ordinary meaning within the technical field, unless a differently specific meaning is given herein. In some cases, terms that are commonly understood by those skilled in the art are defined herein for the sake of clarity or convenience, and such definitions should not be interpreted as indicating a significant difference from the common understanding. The technical methods described or referenced herein are generally well understood by those skilled in the art and are employed by conventional methods.

[0076] The technical solutions of the present application will be described in detail below in conjunction with the drawings and embodiments of the present application. It should be particularly noted that the specific forms of the structures in the embodiments described in the present specification are only exemplary descriptions, and are intended to facilitate understanding of the core technical concepts of the present application. The protection scope of the amorphous material atomic structure multi-scale automatic analysis method based on first principles and machine learning proposed by the present application is not limited to the specific structural forms explicitly described in the embodiments. Other embodiments obtained by conventional technical means improvement, parameter adjustment or equivalent replacement based on the prior art knowledge of those skilled in the art should be included in the protection scope of the present application.

[0077] The amorphous material atomic structure multi-scale automatic analysis method of the present application comprises the following steps:

[0078] Step S1: Taking the reactants forming the amorphous material and their stoichiometric ratio as the initial object, the chemical reaction process between the reactants is simulated based on first principles molecular dynamics, and a cluster unit with short-range order long-range disorder and non-periodic arrangement characteristics formed by the reaction of the reactants is obtained;

[0079] Step S2: Based on the cluster unit, the unit cell parameters and atomic coordinates of the cluster unit are geometrically optimized based on first principles calculation, and a unit cell unit with short-range order long-range order and periodic arrangement characteristics is obtained;

[0080] Step S3: The unit cell unit is expanded and stacked in space, and the unit cell parameters and atomic coordinates are geometrically optimized based on first principles calculation, and a supercell unit with short-range order long-range order and periodic arrangement characteristics is obtained;

[0081] Step S4: performing first-principles molecular dynamics simulation on the supercell unit for a simulation time not less than 10 picoseconds to obtain a first-principles molecular dynamics simulation trajectory. By extracting atomic coordinates, cell parameters, structure energy, atomic force, and cell stress data in the simulation trajectory, a high-dimensional training data set is constructed, which is used as a training set, a validation set, and a test set to train a machine learning potential function, and a machine learning potential function with high-precision prediction capability is obtained;

[0082] Step S5: periodic cell expansion is performed on the basis of the supercell unit to obtain an initial amorphous structure in a mechanical metastable state without structure optimization. Then, based on the trained machine learning potential function, molecular dynamics simulation driven by the machine learning potential function is performed, and finally the atomic structure of the amorphous material with long-range disorder characteristics is obtained;

[0083] Step S6: encapsulating steps S1 to S5 into an automated workflow for automatically analyzing the atomic structure of the amorphous material.

[0084] In steps S1-S6, the specific software package for first-principles calculation and first-principles molecular dynamics simulation can be: when a commercial software VASP is selected, a plane wave basis set is selected, a pseudo potential is set to PBE-PAW pseudo potential, and a cutoff energy is set to 1.3 times the maximum cutoff energy in the pseudo potential file; when an open source software CP2K is selected, a MOLOPT-DZVP Gaussian basis set is selected, a pseudo potential is set to GTH-PBE pseudo potential, and a cutoff energy CUTOFF and REAL_CUTOFF are set to 600 Ry and 60 Ry or more, respectively; when a domestic software ABACUS is selected, a numerical orbital DZP basis set is selected, a basis set cutoff radius is selected to be 8 angstroms, a pseudo potential is set to Dojo-NC-SR pseudo potential, and a real space cutoff energy is set to 100 Ry or more.

[0085] The step S1 specifically includes the following steps:

[0086] Step S1.1: constructing a minimum cluster unit of a reactant, the minimum cluster unit being the smallest molecular composition in the reactant;

[0087] Step S1.2: determining the reaction ratio between each minimum cluster unit, and constructing each minimum cluster unit in step S1.1 in the same cell, with a vacuum layer thickness of more than 1 nm in each direction and an atomic spacing between adjacent minimum cluster units of more than 1 nm, to form a cluster reaction model;

[0088] Step S1.3: performing first-principles molecular dynamics simulation on the cluster reaction model in step S1.2 using an NVT ensemble with constant particle number, volume, and temperature, and selecting PBE or R 2SCAN, dispersion described by D3 or D4, self-consistent field energy convergence criterion less than 1x10 -5 eV, without K-point grid, simulation temperature set to 400K, time step set to 2 femtoseconds, temperature control time set to 20-50 times the time step, simulation duration greater than 3 picoseconds, to obtain a cluster unit with short-range order long-range disorder non-periodic arrangement characteristics.

[0089] The step S2 specifically comprises the following steps:

[0090] Step S2.1: placing the cluster unit in step S1.3 in a periodic boundary box, requiring the vacuum layer thickness to be no more than 1 nm, to obtain a periodic cluster unit;

[0091] Step S2.2: geometric optimization of the unit cell parameters and atomic coordinates of the periodic cluster unit in step S2.1 based on first-principles calculation, to obtain a unit cell unit with short-range order long-range order periodic arrangement characteristics.

[0092] In step S2.2, the specific parameters for geometric optimization of the unit cell parameters and atomic coordinates of the periodic cluster unit are: variable unit cell shape and symmetry, exchange-correlation functional selected as PBE or R 2 SCAN, dispersion described by D3 or D4, self-consistent field energy convergence criterion less than 1x10 -5 eV, atomic force convergence criterion less than 1x10 -3 eV / Å, using a uniform K-point grid of 2x2x2, external pressure set to 1-1000 atmospheres.

[0093] The step S3 specifically comprises the following steps:

[0094] Step S3.1: stacking the unit cell unit in step S2.2 in a dense stacking manner to form a supercell, the size of the supercell being at least 2 times larger than the unit cell unit in XYZ three directions;

[0095] Step S3.2: geometric optimization of the unit cell parameters and atomic coordinates of the supercell in step S3.1 based on first-principles, to obtain a supercell unit with short-range order long-range order periodic arrangement characteristics.

[0096] In step S3.2, the specific parameters for geometric optimization of the unit cell parameters and atomic coordinates of the supercell are: variable unit cell shape and symmetry, exchange-correlation functional selected as PBE or R 2 SCAN, dispersion described by D3 or D4, self-consistent field energy convergence criterion less than 1x10 -5 eV, atomic force convergence criterion less than 1x10 -3eV / Å, without K-point mesh, external pressure is set to 1-1000 atm.

[0097] The step S4 specifically comprises the following steps:

[0098] Step S4.1: Perform first-principles molecular dynamics simulation on the supercell unit in step S3.2, using NVT ensemble with constant particle number, volume and temperature, exchange-correlation functional selected as PBE or R 2 SCAN, set D3 or D4 to describe dispersion effect, self-consistent field energy convergence criterion is less than 1x10 -5 eV, without K-point mesh, simulation temperature is set to 400K, time step is set to 2 femtoseconds, temperature control time is set to 20-50 times of the time step, simulation time is not less than 10 picoseconds, and first-principles molecular dynamics simulation trajectory is obtained;

[0099] Step S4.2: Extract atomic coordinates, cell parameters, structure energy, atomic force and cell stress data in the simulation trajectory of the first-principles molecular dynamics simulation in step S4.1, obtain first-principles dynamics simulation sampling results, and construct a high-dimensional training data set.

[0100] Step S4.3: Divide the high-dimensional training data set in step S4.2 into training set, validation set and test set according to the ratio of 75:5:10, train artificial neural network (such as graph neural network) based machine learning potential function, and obtain machine learning potential function with high precision prediction ability after 150-200 training cycles. When predicting structure energy, atomic force and cell stress, the determination coefficients between the observation data of each physical quantity and the test set data are all higher than 0.98.

[0101] In step S4.3, the specific parameters for training artificial neural network (such as graph neural network) based machine learning potential function are as follows: when using graph neural network MACE, the graph neural network selects 128 equivalent features, two message passing layers and three related orders, the angle resolution is set to 1, the cutoff radius is set to 6Å, the radial basis function is set to 8, and the training weights of structure energy, atomic force and cell stress are set to 1, 10 and 100 respectively.

[0102] The step S5 specifically comprises the following steps:

[0103] Step S5.1: Perform cell expansion on the supercell unit in step S3.2 to obtain an initial amorphous material atomic structure;

[0104] Step S5.2: performing machine learning potential function driven molecular dynamics simulation on the initial amorphous material atomic structure in step S5.1 using the trained accurate machine learning potential function in step S4.3 to obtain the atomic structure of the amorphous material.

[0105] Step S5.2 specifically comprises the following processes:

[0106] Step S5.2.1: performing constant particle number, pressure and temperature machine learning potential function driven molecular dynamics simulation on the initial amorphous material atomic structure in step S5.1 using NPT ensemble, setting the pressure to 1000-10000 atmospheres, setting the simulation temperature to 300K, setting the time step to 1 femtosecond, setting the temperature control time to 20-50 times the time step, setting the pressure control time to 100 times the time step, and setting the simulation time to be greater than 50 picoseconds to obtain the atomic structure of the pressure balanced amorphous material;

[0107] Step S5.2.2: performing machine learning potential function driven molecular dynamics simulation on the pressure balanced amorphous material atomic structure obtained in step S5.2.1 by increasing and decreasing the temperature using NVT ensemble, setting the temperature range to 300-400K, setting the temperature increasing / decreasing rate to be greater than 1 picosecond / K, setting the time step to 2 femtoseconds, setting the temperature control time to 20-50 times the time step, and setting the simulation time to be greater than 200 picoseconds.

[0108] Step S6: encapsulating steps S1 to S5 into an automated workflow for automatically analyzing the atomic structure of the amorphous material. If a Python module is used to encapsulate the automated workflow, step S6 specifically comprises the following steps:

[0109] Step S6.1: encapsulating steps S1.1 to S1.3 into an automated reactant input Python module and a cluster unit generation Python module; the specific steps are as follows:

[0110] Step S6.1.1: user interface, user inputs the molecular formula of the reactant (such as Li3PO4, ZrCl4) and the stoichiometric ratio of the two (such as 1:2), and calls the Python code to generate the cluster unit of the reactant in step S1.1, with the conditions set as described in step S1.1;

[0111] Step S6.1.2: cluster reaction unit construction, calling the Python code to construct the cluster unit in step S6.1.1 in the same unit cell, selecting the atomic spacing between adjacent clusters to be greater than 1 nm, obtaining the cluster reaction unit in step S1.2, and setting the conditions as described in step S1.2;

[0112] Step S6.1.3: Call ASE Python interface to perform first-principles molecular dynamics simulation on the cluster reaction unit of step S6.1.2, call ASE Python to get the cluster reaction products, condition settings as described in step S1.3.

[0113] Step S6.2: Encapsulate steps S2.1 to S2.2 into an automated unit cell unit optimization Python module; the specific steps are:

[0114] Step S6.2.1: Call ASE Python interface to perform cell expansion on the cluster reaction products of step S6.1.3 to obtain a periodic cluster unit, condition settings as described in step S2.1;

[0115] Step S6.2.2: Call ASE Python interface to optimize the cell parameters and atomic coordinates of the periodic cluster unit of step S6.2.1 to obtain a unit cell unit, condition settings as described in step S2.2.

[0116] Step S6.3: Encapsulate steps S3.1 to S3.2 into an automated supercell unit construction Python module; the specific steps are:

[0117] Step S6.3.1: Call ASE Python interface to perform cell expansion on the unit cell unit of step S6.1.3 to obtain a supercell structure, condition settings as described in step S3.1;

[0118] Step S6.3.2: Call ASE Python interface to optimize the cell parameters and atomic coordinates of the supercell structure of step S6.3.1 to obtain a supercell unit, condition settings as described in step S3.2.

[0119] Step S6.4: Encapsulate steps S4.1 to S4.3 into an automated machine learning potential function training Python module; the specific steps are:

[0120] Step S6.4.1: Call ASE Python interface to perform first-principles molecular dynamics simulation on the supercell unit of step S6.1.3, condition settings as described in step S4.1;

[0121] Step S6.4.2: Call ASE Python interface to extract atomic coordinates, cell parameters, structure energy, atomic force and cell stress data from the first-principles molecular dynamics simulation trajectory of step S6.4.1 to an extxyz format file to obtain first-principles sampling results, construct a high-dimensional training data set, condition settings as described in step S4.2;

[0122] Step S6.4.3: Through the Python interface, the first-principles sampling results of step S6.4.2 are optimized using the machine learning potential function to obtain an accurate machine learning potential function. The conditions are set as described in step S4.3.

[0123] Step S6.5: Encapsulate steps S5.1 to S5.2 into an automated amorphous material atomic structure simulation Python module; the specific steps are:

[0124] Step S6.5.1: Call the ASE Python interface to expand the supercell in step S6.1.3 to obtain the initial amorphous structure. The conditions are set as described in step S5.1.

[0125] Step S6.5.2: Call the ASE Python interface to perform a molecular dynamics simulation driven by a machine learning potential function on the initial amorphous structure in step S6.5.1. After the simulation, the final atomic structure of the amorphous material is obtained. The conditions are set as described in step S5.2.

[0126] Step S6.6: Integrate the modules obtained from steps S6.1 to S6.5.

[0127] The packaging method of the automated workflow of the present invention can also be other computer module packaging forms, such as using programming languages ​​​​such as C / C++, Ruby, Perl or Java for development and packaging, and combined with program packaging, containerized deployment, virtual machines or serverless architectures for expansion, which will not be repeated here.

[0128] The following provides specific automated analysis methods for different types of amorphous materials. The following embodiments all use Python modules to encapsulate automated workflows. The detailed steps are consistent with the specific description of step S6 above. In the embodiments, they are simply referred to as modules and will not be repeated. Example 1

[0129] like Figure 1 As shown, a multi-scale automated analysis method for the atomic structure of an amorphous material in this embodiment includes the following steps:

[0130] In this embodiment, the specific software package for the first-principles calculation and first-principles molecular dynamics simulation is the open source software CP2K, the basis set is the MOLOPT-DZVP Gaussian basis set, the pseudopotential is set to the GTH-PBE pseudopotential, and the cutoff energies CUTOFF and REAL_CUTOFF are set to 800Ry and 60Ry, respectively.

[0131] First, proceed to step S1, as Figure 2As shown, the reactants are selected as Li3PO4 and ZrCl4, and the reaction ratio is selected as 1:2. Then, the acquisition of the cluster unit in the step S1 corresponding to the reaction system specifically includes the following steps:

[0132] Step S1.1: constructing the minimum cluster units of the reactants Li3PO4 and ZrCl4;

[0133] Step S1.2: establishing a cluster reaction model according to the reaction ratio of 1:2, constructing each of the minimum cluster units in step S1.1 in the same unit cell, the vacuum layer thickness in each direction is 1.5 nm, and the atomic spacing between adjacent minimum cluster units is 1.5 nm, to form a cluster reaction model;

[0134] Step S1.3: performing first-principle molecular dynamics simulation on the cluster reaction model in step S1.2, using the NVT ensemble of constant particle number, volume and temperature, selecting R 2 SCAN for exchange-correlation functional, setting D3 description for dispersion effect, setting the self-consistent field energy convergence standard to be less than 1×10 -5 eV, not using K-point grid, setting the simulation temperature to 400K, setting the time step to 2 femtoseconds, setting the temperature control time to 30 times the time step, and setting the simulation time to 10 picoseconds, to obtain a cluster unit with short-range ordered long-range disordered aperiodic arrangement characteristics, as shown in Figure 3 .

[0135] Then, step S2 is performed, specifically including the following steps:

[0136] Step S2.1: placing the cluster unit in step S1.3 in a periodic boundary box, requiring the vacuum layer thickness to be not more than 1 nm, to obtain a periodic cluster unit;

[0137] Step S2.2: performing geometric optimization of the unit cell parameters and atomic coordinates of the periodic cluster unit in step S2.1 based on first-principle calculation, setting the unit cell shape and symmetry to be variable, selecting R 2 SCAN for exchange-correlation functional, setting D3 description for dispersion effect, setting the self-consistent field energy convergence standard to be less than 1×10 -5 eV, setting the atomic force convergence standard to be less than 1×10 -3 eV / Å, using a 2×2×2 uniform K-point grid, setting the external pressure to 100 atmospheres, to obtain a unit cell unit with short-range ordered long-range ordered periodic arrangement characteristics, as shown in Figure 4 .

[0138] Then, step S3 is performed, specifically including the following steps:

[0139] Step S3.1: The unit cells in step S2.2 are stacked in a dense manner to form a supercell, and the size of the supercell is enlarged by 2 times in the XYZ three directions relative to the unit cells;

[0140] Step S3.2: Geometric optimization of the supercell in step S3.1 is performed based on first principles, the cell shape and symmetry are variable, the exchange-correlation functional is selected as R 2 SCAN, the dispersion effect is described by D3, the self-consistent field energy convergence criterion is less than 1x10 -5 eV, the atomic force convergence criterion is less than 1x10 -3 eV / Å, the K-point grid is not used, and the external pressure is set to 100 atmospheres, obtaining a supercell unit with periodic arrangement characteristics of short-range order and long-range order, as shown in Figure 5 .

[0141] Then step S4 is performed, specifically including the following steps:

[0142] Step S4.1: First-principles molecular dynamics simulation is performed on the supercell unit in step S3.2, using the NVT ensemble with constant particle number, volume and temperature, and the exchange-correlation functional is selected as R 2 SCAN, the dispersion effect is described by D3, the self-consistent field energy convergence criterion is less than 1x10 -5 eV, the K-point grid is not used, the simulation temperature is set to 400K, the time step is set to 2 femtoseconds, the temperature control time is set to 30 times the time step, and the simulation time is set to 20 picoseconds, obtaining a first-principles molecular dynamics simulation trajectory;

[0143] Step S4.2: The atomic coordinates, cell parameters, structure energy, atomic force, and cell stress data in the simulation trajectory of the first-principles molecular dynamics simulation in step S4.1 are obtained, and a high-dimensional training dataset is constructed;

[0144] Step S4.3: The high-dimensional training dataset in step S4.2 is divided into a training set, a validation set and a test set according to a ratio of 75:5:10, and a graph neural network-based machine learning potential function MACE is trained, the training parameters are set as follows: the graph neural network selects 128 equivalent features, two layers of message passing layers and three related orders, the angle resolution is set to 1, the cutoff radius is set to 6Å, the radial basis function is set to 8, and the training weights of the structure energy, atomic force and cell stress are set to 1, 10 and 100 respectively. After 150 training cycles, a machine learning potential function with high precision prediction ability is obtained, which can predict the structure energy, atomic force and cell stress, as shown in Figures 6-8As shown in the table, the RMSE errors of the structural energy, atomic force and lattice strain predicted by the machine learning potential function are 0.2 meV / atom, 9.2 meV / angstrom and 364 bar, respectively, and the determination coefficients between the observed data and the test set data are all higher than 0.98, and finally a training-accurate machine learning potential function is obtained.

[0145] Then step S5 is performed, specifically including the following steps:

[0146] Step S5.1: cell expansion construction is performed on the supercell unit in step S3.2, at least expanded by 2 times along the XYZ three directions, to obtain an initial amorphous material atomic structure, as shown in Figure 9 ;

[0147] Step S5.2: the initial amorphous material atomic structure in step S5.1 is subjected to machine learning potential function driven molecular dynamics simulation, using the training-accurate machine learning potential function in step S4.3, as shown in Figure 10 , to obtain an atomic structure of the amorphous material, and the machine learning potential function driven molecular dynamics simulation specifically includes the following processes:

[0148] Step S5.2.1: the initial amorphous material atomic structure in step S5.1 is subjected to machine learning potential function driven molecular dynamics simulation with constant particle number, pressure and temperature, using NPT ensemble, with the pressure limited to 3000 atmospheres, the simulation temperature set to 300K, the time step set to 1 femtosecond, the temperature control time set to 30 times the time step, the pressure control time set to 100 times the time step, and the simulation time set to 100 picoseconds, to obtain an atomic structure of the pressure-balanced amorphous material;

[0149] Step S5.2.2: the atomic structure of the pressure-balanced amorphous material obtained in step S5.2.1 is subjected to machine learning potential function driven molecular dynamics simulation with temperature rising and falling, using NVT ensemble, with the temperature range of 300-400K, the rising / falling rate set to 2 picoseconds / K, the time step set to 2 femtoseconds, the temperature control time set to 30 times the time step, and the simulation time set to 400 picoseconds, to obtain a final atomic structure of the amorphous material.

[0150] Finally, step S6 is performed, specifically including the following steps, as shown in Figure 11 ;

[0151] Step S6.1: encapsulating the step S1 as a reactant input module and a cluster unit module;

[0152] Step S6.2: encapsulating the step S2 as a unit cell module;

[0153] Step S6.3: encapsulating the step S3 as a supercell unit module;

[0154] Step S6.4: encapsulating the step S4 as a machine learning potential function training module;

[0155] Step S6.5: encapsulating the step S5 as an amorphous material atomic structure simulation module;

[0156] Step S6.6: integrating each module obtained from the step S6.1 to the step S6.5.

[0157] To further illustrate the accuracy of the amorphous material atomic structure analysis method of the present application, as shown in FIG. 6, the pair distribution function of the amorphous material atomic structure obtained in the present embodiment is calculated, and the result obtained by the present application is basically consistent with the pair distribution function tested in the experiment, which verifies the accuracy of the multi-scale automatic analysis method of the amorphous material atomic structure in the present application. Figure 12 Embodiment 2

[0158] The multi-scale automatic analysis method of the amorphous material atomic structure in the present embodiment comprises the following steps:

[0159] In the present embodiment, the specific software package of the first-principle calculation and the first-principle molecular dynamics simulation is selected as the commercial software VASP, the basis set is selected as the plane wave basis set, the pseudo-potential is set as the PBE-PAW pseudo-potential, and the cutoff energy is set as 1.3 times of the maximum cutoff energy in the pseudo-potential file.

[0160] Firstly, in the step S1, the reactants are selected as Li2SO4 and ZrCl4, and the reaction ratio is selected as 1:2, so that the acquisition of the cluster unit in the step S1 corresponding to the reaction system specifically comprises the following steps:

[0161] Step S1.1: constructing the minimum cluster units Li2SO4 and ZrCl4 of the reactants;

[0162] Step S1.2: establishing a cluster reaction model according to the reaction ratio of 1:2, constructing each minimum cluster unit in step S1.1 in the same unit cell, the vacuum layer thickness in each direction is 1.5 nm, and the atomic spacing between adjacent minimum cluster units is 1.5 nm, to form a cluster reaction model;

[0163] Step S1.3: performing first-principle molecular dynamics simulation on the cluster reaction model in step S1.2, adopting the NVT ensemble of constant particle number, volume and temperature, selecting R 2 SCAN for exchange-correlation functional, setting D3 for dispersion interaction, and setting the self-consistent field energy convergence standard to be less than 1×10 -5 ​eV, without using K-point grid, simulation temperature is set to 400 K, time step is set to 2 femtoseconds, temperature control time is set to 30 times of the time step, simulation time is set to 10 picoseconds, to obtain the cluster unit with short-range order and long-range disorder and non-periodic arrangement characteristics.

[0164] Then is step S2, specifically comprising the following steps:

[0165] Step S2.1: placing the cluster unit in step S1.3 in a periodic boundary box, requiring that the vacuum layer thickness is not more than 1 nm, to obtain a periodic cluster unit;

[0166] Step S2.2: geometrically optimizing the unit cell parameters and atomic coordinates of the periodic cluster unit in step S2.1 based on first-principles calculation, setting the unit cell shape and symmetry to be variable, selecting R 2 SCAN, setting D3 to describe dispersion effect, self-consistent field energy convergence standard is less than 1×10 -5 eV, atomic force convergence standard is less than 1×10 -3 eV / Å, using 2×2×2 uniform K-point grid, setting external pressure to 100 atmospheres, to obtain the unit cell unit with short-range order and long-range order and periodic arrangement characteristics.

[0167] Then is step S3, specifically comprising the following steps:

[0168] Step S3.1: stacking the unit cell unit in step S2.2 in a dense stacking manner to form a supercell, the size of the supercell is enlarged by 2 times in XYZ three directions relative to the unit cell unit;

[0169] Step S3.2: geometrically optimizing the unit cell parameters and atomic coordinates of the supercell in step S3.1 based on first-principles, setting the unit cell shape and symmetry to be variable, selecting R 2 SCAN, D3 to describe dispersion effect, self-consistent field energy convergence standard is less than 1×10 -5 eV, atomic force convergence standard is less than 1×10 -3 eV / Å, without using K-point grid, setting external pressure to 100 atmospheres, to obtain the supercell unit with short-range order and long-range order and periodic arrangement characteristics.

[0170] Then is step S4, specifically comprising the following steps:

[0171] Step S4.1: performing first-principles molecular dynamics simulation on the supercell unit in step S3.2, using NVT ensemble with constant particle number, volume and temperature, selecting R 2SCAN, set D3 to describe the dispersion effect, the self-consistent field energy convergence standard is less than 1*10 -5 eV, without K-point grid, the simulation temperature is set to 400K, the time step is set to 2 femtoseconds, the temperature control time is set to 30 times of the time step, the simulation time is set to 20 picoseconds, and the first-principles molecular dynamics simulation trajectory is obtained;

[0172] Step S4.2: taking the atomic coordinates, cell parameters, structure energy, atomic force, and cell stress data in the first-principles molecular dynamics simulation trajectory in step S4.1, obtaining the first-principles dynamics simulation sampling result, and constructing a high-dimensional training data set;

[0173] Step S4.3: dividing the high-dimensional training data set in step S4.2 into a training set, a validation set, and a test set according to a ratio of 75:5:10, training the graph neural network-based machine learning potential function MACE, setting the training parameters as follows: selecting 128 equivalent features for the graph neural network, two layers of message passing layers, and three related orders, setting the angle resolution to 1, the cutoff radius to 6 Å, the radial basis function to 8, and the training weights of the structure energy, atomic force, and cell stress to 1, 10, and 100, respectively, obtaining a machine learning potential function with high precision prediction ability after 150 training cycles, and when predicting the structure energy, atomic force, and cell stress, the RMSE errors of the structure energy, atomic force, and lattice stress predicted by the machine learning potential function are 0.3 meV / atom, 10.1 meV / Å, and 301 bar, respectively, and the determination coefficients between the observed data of each physical quantity and the test set data are all higher than 0.98. Finally, a training-accurate machine learning potential function is obtained.

[0174] Then step S5, specifically comprising the following steps:

[0175] Step S5.1: expanding the supercell unit in step S3.2 to construct an initial amorphous material atomic structure, which is at least expanded by 2 times in XYZ three directions;

[0176] Step S5.2: performing machine learning potential function-driven molecular dynamics simulation on the initial amorphous material atomic structure in step S5.1 using the training-accurate machine learning potential function in step S4.3, to obtain the atomic structure of the amorphous material, as shown in Figure 13 The machine learning potential function-driven molecular dynamics simulation specifically includes the following processes:

[0177] Step S5.2.1: constant particle number, pressure and temperature machine learning potential function driven molecular dynamics simulation is performed on the initial amorphous material atomic structure described in step S5.1, using an NPT ensemble, the pressure is limited to 3000 atmospheres, the simulation temperature is set to 300K, the time step is set to 1 femtosecond, the temperature control time is set to 30 times the time step, the pressure control time is set to 100 times the time step, and the simulation time is set to 100 picoseconds, to obtain the atomic structure of the pressure balanced amorphous material;

[0178] Step S5.2.2: temperature increasing and decreasing machine learning potential function driven molecular dynamics simulation is performed on the pressure balanced amorphous material atomic structure obtained in step S5.2.1, using an NVT ensemble, the temperature range is 300-400K, the temperature increasing / decreasing rate is set to 2 picoseconds / K, the time step is set to 2 femtoseconds, the temperature control time is set to 30 times the time step, and the simulation time is set to 400 picoseconds, to obtain the final atomic structure of the amorphous material, as shown in Figure 13

[0179] Finally, step S6 includes the following steps:

[0180] Step S6.1: encapsulating the step S1 as a reactant input module and a cluster unit module;

[0181] Step S6.2: encapsulating the step S2 as a unit cell module;

[0182] Step S6.3: encapsulating the step S3 as a supercell unit module;

[0183] Step S6.4: encapsulating the step S4 as a machine learning potential function training module;

[0184] Step S6.5: encapsulating the step S5 as an amorphous material atomic structure simulation module;

[0185] Step S6.6: integrating each module obtained from the step S6.1 to step S6.5.

[0186] To further illustrate the accuracy of the amorphous material atomic structure analysis method of the present application, the pair distribution function of the amorphous material atomic structure obtained in the present embodiment is calculated, as shown in Figure 14 The results obtained by the present application are basically consistent with the pair distribution function tested in the experiment, verifying the accuracy of the amorphous material atomic structure multi-scale automated analysis method in the present application. Embodiment 3

[0187] The amorphous material atomic structure multi-scale automated analysis method in the present embodiment includes the following steps:

[0188] ​In the present embodiment, the specific software package of the first-principles calculation and the first-principles molecular dynamics simulation is selected as the commercial software VASP, the basis set is selected as the plane wave basis set, the pseudo potential is set as the PBE-PAW pseudo potential, and the cut-off energy is set as 1.3 times the maximum cut-off energy in the pseudo potential file.

[0189] Firstly, in step S1, the reactants are selected as LiCl and ZrCl4, and the reaction ratio is selected as 2:1. Then, the acquisition of the cluster unit in step S1 includes the following steps:

[0190] Step S1.1: constructing the minimum cluster units of the reactants LiCl and ZrCl4;

[0191] Step S1.2: establishing a cluster reaction model according to the reaction ratio of 2:1, constructing each minimum cluster unit in step S1.1 in the same unit cell, setting the vacuum layer thickness in each direction as 1.5 nm, and setting the atomic spacing between adjacent minimum cluster units as 1.5 nm to form a cluster reaction model;

[0192] Step S1.3: performing first-principles molecular dynamics simulation on the cluster reaction model in step S1.2, adopting the NVT ensemble of constant particle number, volume and temperature, selecting R 2 SCAN as the exchange-correlation functional, setting D3 description for dispersion interaction, setting the self-consistent field energy convergence standard less than 1×10 -5 eV, not using the K-point grid, setting the simulation temperature as 400K, setting the time step as 2 femtoseconds, setting the temperature control time as 30 times the time step, and setting the simulation time length as 10 picoseconds to obtain a cluster unit with short-range ordered long-range disordered non-periodic arrangement characteristics.

[0193] Then, step S2 includes the following steps:

[0194] Step S2.1: placing the cluster unit in step S1.3 in a periodic boundary box, and setting the vacuum layer thickness to be not more than 1 nm to obtain a periodic cluster unit;

[0195] Step S2.2: performing geometric optimization of the cell parameters and atomic coordinates of the periodic cluster unit in step S2.1 based on first-principles calculation, setting the cell shape and symmetry to be variable, selecting R 2 SCAN as the exchange-correlation functional, setting D3 description for dispersion interaction, setting the self-consistent field energy convergence standard less than 1×10 -5 eV, setting the atomic force convergence standard less than 1×10 -3 eV / Å, adopting a 2×2×2 uniform K-point grid, and setting the external pressure as 100 atmospheres to obtain a periodic arrangement characteristic of the cell unit with short-range order and long-range order.

[0196] Then is step S3, specifically comprising the following steps:

[0197] Step S3.1: The unit cells in step S2.2 are stacked in a dense manner to form a supercell, and the size of the supercell is enlarged by 2 times in three directions XYZ relative to the unit cell;

[0198] Step S3.2: Geometric optimization of the supercell in step S3.1 is performed based on first principles, the cell shape and symmetry are variable, the exchange-correlation functional is selected as R 2 SCAN, D3 describes the dispersion effect, the self-consistent field energy convergence standard is less than 1×10 -5 eV, the atomic force convergence standard is less than 1×10 -3 eV / Å, without K-point grid, the external pressure is set to 100 atmospheres, and the supercell unit with periodic arrangement characteristics of short-range order and long-range order is obtained.

[0199] Then is step S4, specifically comprising the following steps:

[0200] Step S4.1: First-principles molecular dynamics simulation is performed on the supercell unit in step S3.2, using the NVT ensemble with constant particle number, volume and temperature, the exchange-correlation functional is selected as R 2 SCAN, D3 describes the dispersion effect, the self-consistent field energy convergence standard is less than 1×10 -5 eV, without K-point grid, the simulation temperature is set to 400K, the time step is set to 2 femtoseconds, the temperature control time is set to 30 times the time step, the simulation time is set to 20 picoseconds, and the first-principles molecular dynamics simulation trajectory is obtained;

[0201] Step S4.2: The atomic coordinates, cell parameters, structure energy, atomic force, and cell stress data in the simulation trajectory of the first-principles molecular dynamics simulation in step S4.1 are obtained, the first-principles dynamics simulation sampling results are obtained, and a high-dimensional training data set is constructed;

[0202] Step S4.3: The high-dimensional training data set described in step S4.2 is divided into a training set, a validation set and a test set in a ratio of 75:5:10, and a graph neural network-based machine learning potential function MACE is trained, with the training parameters set as follows: 128 equal-variant features, two layers of message passing layers and three relevant orders of the graph neural network, an angle resolution of 1, a cutoff radius of 6 Å, 8 radial basis functions, and training weights of 1, 10 and 100 for the structure energy, atomic force and cell stress, respectively. After 150 training cycles, a machine learning potential function with high precision prediction capability is obtained, which has a structure energy, atomic force and lattice strain RMSE error of 0.3 meV / atom, 9.5 meV / Å and 350 bar, respectively, in predicting the structure energy, atomic force and lattice strain, and the determination coefficients between the observed data and the test set data are all higher than 0.98. Finally, a training-accurate machine learning potential function is obtained.

[0203] Then, step S5 is performed, which specifically includes the following steps:

[0204] Step S5.1: The supercell unit described in step S3.2 is expanded to obtain an initial amorphous material atomic structure;

[0205] Step S5.2: The initial amorphous material atomic structure described in step S5.1 is subjected to machine learning potential function-driven molecular dynamics simulation using the training-accurate machine learning potential function described in step S4.3, to obtain an atomic structure of the amorphous material, as shown in FIG. 5B. Figure 15 The machine learning potential function-driven molecular dynamics simulation specifically includes the following processes:

[0206] Step S5.2.1: The initial amorphous material atomic structure described in step S5.1 is subjected to machine learning potential function-driven molecular dynamics simulation with constant particle number, pressure and temperature, using an NPT ensemble, with the pressure limited to 3000 atmospheres, the simulation temperature set to 300 K, the time step set to 1 femtosecond, the temperature control time set to 30 times the time step, the pressure control time set to 100 times the time step, and the simulation duration set to 100 picoseconds, to obtain a pressure-balanced atomic structure of the amorphous material.

[0207] Step S5.2.2: The pressure-balanced atomic structure of the amorphous material obtained in step S5.2.1 is subjected to machine learning potential function-driven molecular dynamics simulation with temperature rising and falling, using an NVT ensemble, with the temperature ranging from 300 K to 400 K, the rising / falling rate set to 2 picoseconds / K, the time step set to 2 femtoseconds, the temperature control time set to 30 times the time step, and the simulation duration set to 400 picoseconds, to obtain a final atomic structure of the amorphous material, as shown in FIG. 5C. Figure 15are shown.

[0208] Finally, step S6 includes the following steps:

[0209] Step S6.1: encapsulating the step S1 as a reactant input module and a cluster cell module;

[0210] Step S6.2: encapsulating the step S2 as a unit cell module;

[0211] Step S6.3: encapsulating the step S3 as a supercell cell module;

[0212] Step S6.4: encapsulating the step S4 as a machine learning potential function training module;

[0213] Step S6.5: encapsulating the step S5 as an amorphous material atomic structure simulation module;

[0214] Step S6.6: integrating each module obtained from the step S6.1 to the step S6.5.

[0215] The above describes the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any changes or replacements easily thought of by those skilled in the art within the technical range disclosed by the present application should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for multi-scale automated analysis of amorphous material atomic structure, comprising: The method comprises the following steps: ​ Step S1: taking a reaction substance forming an amorphous material as an initial unit, and obtaining a cluster unit generated by the reaction substance based on first-principle molecular dynamics simulation; Step S2: taking the cluster unit as a basis, optimizing the unit cell parameters and atomic coordinates of the cluster unit based on first-principle calculation, and obtaining a unit cell; Step S3: stacking the unit cell in space, optimizing the unit cell parameters and atomic coordinates of the unit cell based on first-principle calculation, and obtaining a supercell unit; Step S4: performing first-principle molecular dynamics simulation on the supercell unit, sampling data sets from the simulation trajectory, training and obtaining an accurate machine learning potential function using the data sets; Step S5: expanding the supercell unit to construct, and performing molecular dynamics simulation driven by the machine learning potential function based on the machine learning potential function, to obtain an atomic structure of the amorphous material; Step S6: encapsulating steps S1 to S5 into an automated workflow for automatically analyzing the atomic structure of the amorphous material.

2. The method of claim 1, wherein the method is characterized by, The step S1 specifically comprises the following steps: Step S1.1: constructing a minimum cluster unit of the reaction substance, the minimum cluster unit being the smallest molecular composition in the reaction substance; Step S1.2: determining the reaction ratio between each minimum cluster unit, constructing each minimum cluster unit in the same unit cell, the vacuum layer thickness in each direction being greater than 1 nm, and the atomic spacing between adjacent minimum cluster units being greater than 1 nm, to form a cluster reaction model; Step S1.3: performing first-principle molecular dynamics simulation on the cluster reaction model, using an NVT ensemble with constant particle number, volume and temperature, and the simulation time being greater than 3 picoseconds, to generate a cluster unit.

3. The method of claim 1, wherein the method is characterized by, The step S2 specifically comprises the following steps: Step S2.1: extending the cluster unit to a periodic cluster unit, the vacuum layer thickness of the periodic cluster unit being not more than 1 nm; Step S2.2: performing geometric optimization of the unit cell parameters and atomic coordinates of the periodic cluster unit based on first-principle calculation, to obtain a unit cell.

4. The method of claim 3, wherein the method is characterized by, In step S2.2, the specific parameters for optimizing the unit cell parameters and atomic coordinates of the periodic cluster unit are: the unit cell shape and symmetry are variable, and the external pressure is 1-1000 atmospheres.

5. The method of claim 1, wherein the method is characterized by, The step S3 specifically comprises the following steps: Step S3.1: stacking the unit cell in a dense stacking manner to form a supercell, the size of the supercell being at least 2 times larger than the unit cell along the XYZ three directions; Step S3.2: performing geometric optimization of the unit cell parameters and atomic coordinates of the supercell based on first-principle calculation, to obtain a supercell unit.

6. The method of claim 5, wherein the method is characterized by, In step S3.2, the specific parameters for optimizing the unit cell parameters and atomic coordinates of the supercell are: the unit cell shape and symmetry are variable, and the external pressure is 1-1000 atmospheres.

7. The method of claim 1, wherein the method is characterized by, The step S4 specifically comprises the following steps: Step S4.1: performing first-principle molecular dynamics simulation on the supercell unit, using an NVT ensemble with constant particle number, volume and temperature, and the simulation time being greater than 3 picoseconds; Step S4.2: Extracting atomic coordinates, cell parameters, structure energy, atomic forces, and cell stress data from the simulation trajectory of the first-principles molecular dynamics simulation in step S4.1, and constructing a high-dimensional training dataset; Step S4.3: Training a machine learning potential function based on the high-dimensional training dataset, wherein the determination coefficients between the observed data of each physical quantity and the test set data in predicting the structure energy, atomic forces, and cell stress are all higher than 0.98, and an accurate machine learning potential function is obtained.

8. The method of claim 1, wherein the method is characterized by, The step S5 specifically comprises the following steps: Step S5.1: Cell expansion and construction of the supercell unit, which is at least expanded by 2 times in the XYZ three directions, to obtain an initial amorphous material atomic structure; Step S5.2: Performing machine learning potential function-driven molecular dynamics simulation on the initial amorphous material atomic structure using the accurate machine learning potential function obtained in step S4 to obtain a final amorphous material atomic structure.

9. The method of claim 8, wherein the method further comprises: determining a plurality of atomic structures of the amorphous material; and determining a plurality of atomic structures of the amorphous material. Step S5.2 specifically comprises the following processes: Step S5.2.1: Performing machine learning potential function-driven molecular dynamics simulation on the initial amorphous material atomic structure, selecting a constant particle number, pressure, and temperature NPT ensemble, limiting the pressure to 1000-10000 atmospheres and the temperature to greater than 300 Kelvin, and simulating for a duration of greater than 50 picoseconds to obtain a pressure-balanced amorphous material atomic structure; Step S5.2.2: Performing machine learning potential function-driven molecular dynamics simulation on the pressure-balanced amorphous material atomic structure obtained in step S5.2.1 by increasing and decreasing the temperature, selecting an NVT ensemble, and setting the temperature range to 300-400 Kelvin and the temperature increasing / decreasing rate to greater than 1 picosecond / Kelvin, to obtain a final amorphous material atomic structure.

10. The method of any one of claims 1-9, wherein the method is for atomistic structure multiscale automated resolution of amorphous materials. The step S6 specifically comprises the following steps: Step S6.1: Packaging the step S1 as a reactant input module and a cluster unit module; Step S6.2: Packaging the step S2 as a cell unit module; Step S6.3: Packaging the step S3 as a supercell unit module; Step S6.4: Packaging the step S4 as a machine learning potential function training module; Step S6.5: Packaging the step S5 as an amorphous material atomic structure simulation module; Step S6.6: Integrating each module obtained from steps S6.1 to S6.5.

Images