A method of molecular dynamics crystal structure modeling based on three-dimensional geometric models

Abstract

This invention discloses a molecular dynamics crystal structure modeling method based on a three-dimensional geometric model, belonging to the interdisciplinary field of molecular dynamics simulation and materials modeling. It aims to solve the problems of existing technologies, such as difficulty in handling complex multi-region geometric models, low efficiency in multi-material modeling, poor boundary accuracy, and insufficient automation. This method first reads a multi-region three-dimensional geometric model in OBJ format, analyzes and obtains independent geometric regions, and determines the overall boundary. Then, it uniformly performs atomic-scale scaling and centering on all regions. For each region, it matches corresponding crystal information, performs crystal cell expansion to generate atomic clouds, and performs preliminary screening. It then precisely filters atoms within each region through parallel spatial clipping. Finally, it merges the atomic information of all regions and outputs a molecular dynamics input file. This invention supports integrated modeling of complex structures using multiple materials, significantly improving modeling efficiency and boundary clipping accuracy, and is suitable for high-throughput molecular dynamics simulations of semiconductor devices, composite materials, and other systems.

Description

Technical Field

[0001] This invention relates to the intersection of molecular dynamics simulation technology and materials modeling, and in particular to a molecular dynamics crystal structure modeling method based on a three-dimensional geometric model. Background Technology

[0002] Molecular dynamics (MD) simulations have wide applications in materials science, semiconductors, and nanotechnology, and their accuracy is highly dependent on the accuracy of the initial crystal structure model. Traditional modeling methods typically use combinations of simple geometries (such as cubes and spheres) to construct multi-material composite structures, or rely on manually defined lattice positions to achieve atomic filling.

[0003] However, these methods have significant limitations. Existing tools struggle to handle multi-region geometric models with complex surfaces, porous structures, or irregular boundaries, resulting in insufficient fidelity in reproducing the microscopic morphology of real materials. For composite systems containing heterogeneous materials, each region must be modeled separately and then manually assembled, which easily introduces atomic overlap or gap errors, and the process is cumbersome and inefficient. The conversion from geometric models to atomic coordinates requires manual intervention, making it impossible to achieve integrated automatic generation of multiple materials and regions, thus limiting the application of high-throughput simulations. Furthermore, existing technologies lack effective solutions for automatically converting complex geometric models exported from mature 3D modeling software (such as Blender) into molecular dynamics models, resulting in low visualization of the modeling process, high user operation threshold, and difficulty in meeting the fine-grained modeling needs of complex devices (such as semiconductor devices and porous composite materials).

[0004] Currently, while existing technologies in this field can achieve basic crystal modeling, none have solved the core problems of automated analysis, material matching, parallel trimming, and integrated output for complex multi-region geometric models. Existing molecular dynamics preprocessing tools are generally more suitable for regular geometries or require additional scripts, and their ability to automatically fill and trim atoms for complex multi-region geometric models is limited. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a molecular dynamics crystal structure modeling method based on a three-dimensional geometric model. The aim is to achieve fully automated construction of the molecular dynamics atomic structure model from a complex three-dimensional geometric model, improve modeling accuracy, efficiency and geometric freedom, and adapt to the high-throughput simulation requirements of multi-material composite systems.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] A molecular dynamics crystal structure modeling method based on a three-dimensional geometric model includes the following steps:

[0008] S1: Model analysis steps: Read the 3D geometric model file in OBJ format, analyze each object / group in the file as an independent geometric region, extract the corresponding local vertex and triangle face information for each geometric region, and determine the overall boundary of all geometric regions;

[0009] S2: Scale unification step: Treat all geometric regions as a whole, calculate the unified scaling factor and overall offset based on the preset simulation box size, perform atomic scale scaling and center positioning on all geometric regions, and update the coordinates of each geometric region and the overall boundary.

[0010] S3: Regional Atom Generation and Trimming Steps: For each independent geometric region, perform the following operations sequentially:

[0011] S3.1 Obtain the crystal information file corresponding to this geometric region;

[0012] S3.2 Based on the lattice parameters in the crystal information file, crystal cell expansion is performed to generate an atomic cloud covering the geometric region, and preliminary screening is performed using a boundary box.

[0013] S3.3 performs parallel spatial trimming on the initially screened atomic cloud to select atoms located within the geometric region;

[0014] S4: Output Merging Step: Merge the atomic coordinates and atomic type information retained in all geometric regions to obtain the molecular dynamics crystal structure model corresponding to the multi-region three-dimensional geometric model, and output the molecular dynamics input file containing the atomic information of the model and the size of the simulation box.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] 1. It supports any number and shape of independent geometric regions, each of which can be filled with different crystalline materials, enabling integrated modeling of composite materials and solving the problem of difficult modeling of complex multi-region and multi-material structures in existing technologies;

[0017] 2. By employing multi-process / multi-threaded parallel pruning technology, the efficiency of atom screening is significantly improved, supporting ultra-large-scale simulation systems (atom count ≥ 10). 6 The construction of ) solves the problem of low efficiency in large-scale system modeling in existing technologies;

[0018] 3. A combination of ray casting and the Möller-Trumbore algorithm is used to accurately solve for the intersection points of rays and triangular faces based on analytical geometric equations, with a floating-point comparison tolerance of 1×10⁻⁶. -8This method enables high-precision determination of the internal and external positions of atoms on complex curved surfaces and irregular boundaries. It directly determines the positional relationship between atoms and continuous triangular meshes through analytical geometric calculations, and its theoretical accuracy is guaranteed by both floating-point tolerance and mesh geometric accuracy.

[0019] 4. This invention can be implemented using Python, C++, or other programming languages. It supports open formats such as OBJ / CIF and is compatible with mainstream 3D modeling software such as Blender and SolidWorks. Users can customize the material mapping relationship of regions through a visual interactive interface, reducing the operational threshold.

[0020] 5. The entire process from importing multi-region geometric models to outputting molecular dynamics input files is automated, avoiding errors such as atomic overlap and gaps caused by manual intervention, thus improving modeling accuracy and significantly increasing modeling efficiency;

[0021] 6. It supports customizable simulation box size, vacuum layer thickness, and parallel processing parameters, adapting to different types of molecular dynamics simulation needs and possessing greater versatility. Attached Figure Description

[0022] Figure 1 : Flowchart of the method of this invention.

[0023] Figure 2 3D geometric model of a high electron mobility transistor (HEMT) device.

[0024] Figure 3 : XYZ format model of the internal atomic structure of HEMT device.

[0025] Figure 4 : 3D geometric model of the hatch.

[0026] Figure 5 The internal atomic structure of the hatch is represented by an xyz format model. Detailed Implementation

[0027] In the specific implementation process, the modeling method mentioned in this invention includes the following steps: multi-region model analysis; simulated box scale transformation; multi-material atomic cloud generation; parallel spatial pruning; and multi-structure fusion output.

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] See Figure 1This invention discloses a molecular dynamics crystal structure modeling method based on a three-dimensional geometric model, comprising the following steps:

[0030] S1: Read the OBJ format 3D geometric model file, parse and extract feature identifiers line by line: extract vertex identifiers to record global vertex coordinates; extract object or group identifiers to divide independent geometric regions and record region names; extract face definition identifiers to obtain the vertex indices of polygon faces. For the extracted vertex indices, regardless of whether the face definition contains texture coordinates or normal data (such as v / vt / vn), the pure vertex index is extracted using a delimiter and converted from the default starting index (1-based) of the OBJ global file to the zero-based index (0-based) processed internally by the program; when encountering a polygon face with 4 or more vertices, a fan triangulation algorithm is used to split it into multiple triangular patches based on the first vertex. For free meshes that do not contain object or group identifiers, "default_region" is automatically assigned as the name. Subsequently, for each independent or default geometric region, a mapping table is established from the global vertex index to the local vertex list of that region, generating an array of vertex coordinates and an array of triangular patch indices for that region. The maximum and minimum values ​​of each region's vertices in the X, Y, and Z directions are extracted as local boundaries, and the overall boundary of the entire geometric model is determined based on the maximum and minimum values ​​of all local vertices.

[0031] S2: Treat all geometric regions as a whole and scale and center them uniformly.

[0032] S2.1 The thickness of the vacuum layer is calculated independently along each axis, using the following formula:

[0033]

[0034] in, Representing the three axes: X, Y, and Z. The preset scaling factor (in this embodiment, 25% of the axial dimensions of the simulated box are extracted as the relative thickness, i.e.) ), To simulate the corresponding axial dimensions of the box, The minimum absolute thickness is set (20.0 Å in this embodiment). The maximum value of the two is taken as the final vacuum layer thickness.

[0035] S2.2 Calculate the globally uniform scaling factor based on the overall model dimensions and the simulated box dimensions. The formula for calculating the overall axial dimensions of the model is:

[0036]

[0037] in, and The maximum and minimum coordinate values ​​of all vertices of the model along the corresponding axes.

[0038] The model can be computed using spatial formulas:

[0039]

[0040] Calculation of independent scaling factors for each axis:

[0041]

[0042] Global uniform scaling factor calculation:

[0043]

[0044] The minimum value of the independent scaling factors for each axis is taken as the global unified scaling factor to ensure that the model is scaled proportionally without exceeding the available space.

[0045] S2.3 calculates the offset vector that centers the overall model, performing atomic-scale scaling and centering on all geometric regions.

[0046] Calculation of the geometric center of the scaled model:

[0047]

[0048] Simulated box-center calculation:

[0049]

[0050] Global offset vector calculation:

[0051]

[0052] Final transformation of vertex coordinates: Apply the above scaling factor and offset vector to all vertices within each geometric region, using the following formula:

[0053]

[0054] in, These are the axial coordinates corresponding to the original vertex. These are the final vertex coordinates after scaling and centering. After the coordinate transformation is complete, the local and global boundaries of each local geometric region are updated synchronously.

[0055] S3: For each individual geometric region, perform the following operations in sequence:

[0056] S3.1 Obtain the crystal information file corresponding to the geometric region, and use the material analysis library to read and extract the Cartesian lattice matrix, unit cell atomic coordinates, and element type of the unit cell.

[0057] S3.2 Based on the lattice parameters in the crystal information file, crystal cell expansion is performed to generate an atomic cloud covering the geometric region.

[0058] Let the three basis vectors of the unit cell be... The lattice matrix in Cartesian coordinates is (3×3 matrix), calculate its inverse matrix. The updated local Cartesian minimum and maximum coordinates (Region_min and Region_max) of the current region are transformed into approximate ranges in a fractional coordinate system. The coordinate transformation formula is as follows:

[0059]

[0060] In the formula, Let Cartesian coordinate column vectors be used. It is a column vector of fractional coordinates.

[0061] The upper and lower limits of the fractional coordinates of the region boundary obtained by the above transformation are respectively , To avoid boundary truncation errors, a preset number of buffer cells are added to each dimension of the fractional coordinate range. The formula for calculating the cell expansion start and end indices is as follows:

[0062]

[0063]

[0064] In the formula, The three dimensions representing the unit cell ( ), The preset number of buffer cells is (preferably 3 in this embodiment).

[0065] After determining the start and end indices, an initial atomic cloud covering the region is generated through vector translation. The formula for calculating the translated atomic coordinates is as follows:

[0066]

[0067] In the formula, For the first intracellular Cartesian coordinates of 1 atom; These are the cell translation indices in three dimensions, with values ​​ranging from... , , ; The coordinates of the newly generated atom in the global Cartesian coordinate system.

[0068] The local Cartesian bounding box of the current region is extended outwards by a preset Cartesian buffer distance (e.g., 10.0 Å) in six directions to generate an initial screening bounding box. By comparing coordinate ranges, external atoms that are clearly not within the specified bounding box are quickly eliminated. Simultaneously, an overlap check is performed on the generated atomic clouds; the relevant formula for overlap determination is as follows:

[0069] Calculation of Euclidean distance between atoms:

[0070] Overlap detection threshold calculation:

[0071] When the condition is met When an atom is identified as overlapping, it is automatically discarded, and only one is retained. In the formula, This is a preset proportional coefficient. The distance is the Euclidean distance (norm 2). and Atoms With atoms The corresponding atomic radius, and These are the spatial coordinates of the two atoms.

[0072] S3.3 performs parallel space pruning on the initially screened atomic cloud. The number of working units N (e.g., the number of CPU cores) is obtained, and the atomic cloud is divided into multiple batches using the formula: Batch Size = max(1, Total Number of Atoms to be Judged / / (N × 4)). The atomic batches with their original indices are then distributed to each working unit for independent computation using a process pool or thread pool.

[0073] Within each working unit, high-precision inside / outside determination based on ray casting and the Möller-Trumbore algorithm is performed. First, a rigorous local bounding box check is performed again to eliminate atoms outside the region boundary. Then, starting from the Cartesian coordinates of the atom to be determined, rays are uniformly emitted along the positive X-axis of the global coordinate system (i.e., direction vector [1.0, 0.0, 0.0]). All triangular faces in the geometric region are traversed. For each triangular face, the determinant (det) of the product of the ray direction vector and the two side vectors (edge1, edge2) of the triangular face is calculated. A floating-point comparison tolerance ε (set to 1×10 in this embodiment) is introduced. -8When |det| < ε, the ray is determined to be parallel to or coplanar with the triangular face, and the face is skipped directly to solve the fault tolerance problem of degenerate scene; when |det| ≥ ε, the centroid coordinates (u, v) of the intersection point and the ray parameter t are calculated. If 0.0 ≤ u ≤ 1.0 and 0.0 ≤ v and u + v ≤ 1.0 are satisfied, and t ≥ -ε is further satisfied (allowing a small offset of the intersection point behind the ray starting point to handle the floating-point error case where the atom just falls on the surface), then a valid intersection point is determined to be generated. After traversing all faces, if the total number of valid intersection points is odd, the atom is determined to be inside the geometric mesh and the corresponding atom type and coordinates are retained; if it is even, it is determined to be outside and discarded.

[0074] After the parallel spatial trimming is completed, overlapping atom culling logic is performed on the retained atoms to eliminate interface errors introduced by multi-material splicing. The specific steps are as follows:

[0075] (1) Calculate the atoms to be determined With adjacent atoms Euclidean distance between The formula is ,in and The coordinates of the two atoms are Cartesian coordinates.

[0076] (2) If satisfied If the two atoms overlap, then it is determined that the two atoms overlap; among them, and These are the van der Waals radii for the corresponding atom types. This is the overlap determination ratio, representing the actual chemical bond contraction effect at the compatibility interface. The value range is limited to between 0.6 and 0.85 (preferably set to 0.75 in this embodiment).

[0077] (3) For the geometric region currently being processed, the atoms retained within it are compared with other atoms in the same region, as well as with atoms generated from other geometric regions that have been processed previously. If cross-regional atomic overlap occurs, the preset priority weights of each geometric region will be read, and atoms with higher priority in their respective regions will always be retained, while atoms with lower priority in their respective regions will be removed; if two overlapping atoms belong to the same priority region, the atom closer to the geometric center of that region will be retained.

[0078] S4: After processing all geometric regions, merge the element type lists and coordinate arrays of the atoms retained in all regions. During the merging process, execute the logic for eliminating overlapping atoms across regions: based on preset region priority rules and distance determination formulas, eliminate overlapping atoms at cross-region boundaries. Create a molecular dynamics input file (e.g., XYZ format file) according to the preset file name. Write the total number of atoms to be retained and the preset simulation box size in the file header. Subsequent lines output the element type of each atom and its corresponding X, Y, and Z Cartesian coordinates one by one, finally generating a molecular dynamics crystal structure model file containing complex multi-region structures.

[0079] Example 1

[0080] Taking the three-dimensional multi-region structure of β-Ga2O3 high electron mobility transistor (HEMT) as an example (geometric model see...) Figure 2 The specific implementation steps are as follows:

[0081] 1. For example Figure 2 As shown, in 3D modeling software (such as Blender), based on the structure of the HEMT device (such as... Figure 2 (As shown) Construct a geometric model. Assign a unique name to each geometric region to distinguish the type of material it is filled with, and perform a model compliance check.

[0082] 2. Export the completed HEMT device model as an OBJ format file. To achieve optimal resolution, this embodiment specifies the following model export guidelines: during export, set the Z-axis to the upward axis to conform to the conventional orientation of semiconductor devices; check the "Triangulate Mesh" option; and check the "Object Group" option to generate labels with material prefixes. Furthermore, based on the fault-tolerance mechanism in step S1 of this invention, the system can still successfully model the device even if the above guidelines are not strictly followed: for example, if "Triangulate Mesh" is not checked, the system will automatically convert it into triangles that support ray projection using a sector triangulation algorithm in the background; if the user does not check "Object Group," the system will automatically merge the entire device into a "default region" for the user to uniformly fill with material; if the 3D software exports units other than Å, the scaling factor calculation mechanism in step S2 will adaptively convert it to the correct molecular dynamics simulation scale.

[0083] 3. Read the exported OBJ file and parse the defined objects / groups as independent geometric regions. Extract the local vertex and triangle information for each region and establish a mapping relationship from global vertices to local vertices of the region. In this process, the overall boundary of the entire HEMT model is determined.

[0084] 4. Set the target molecular dynamics simulation box size to 600 Å × 600 Å × 600 Å. Treat all geometric regions as a whole, and calculate a uniform scaling factor and overall offset based on the set simulation box size and the overall boundary of the HEMT model. According to claim 3, ensure that the thickness of the vacuum layer around the model is not less than 25% or 20 Å of the simulation box size. Perform atomic-scale scaling and centering on all regions of the HEMT model, update the coordinates and local boundaries of each region, and redefine the overall boundary of the entire model.

[0085] 5. Assign a corresponding crystal information file for each region. Based on the lattice parameters of the specified crystal file, expand the crystal cells within the simulation space of each region. According to claim 4, add three additional cells outside the overall boundary to ensure complete coverage, generating an atomic cloud covering the region. Subsequently, use the expanded bounding box to perform preliminary screening on the generated atomic cloud, quickly eliminating atoms clearly located outside the region. Perform precise parallel spatial pruning for each region. First, perform bounding box checks to quickly eliminate external atoms. Then, use ray casting combined with the Möller-Trumbore algorithm to determine the internal and external positions of each atom in the atomic cloud. According to claims 5 and 6, divide the atomic cloud into multiple batches, implement multi-process parallel processing through a process pool, and each working unit independently executes ray casting to delete atoms located outside the region according to the parity rule. Figure 3 The internal atomic structure of the HEMT device, preserved after trimming, is shown.

[0086] 6. Merge all regions, retaining the atomic coordinates and atom type information after cropping. Output the merged atomic data as a molecular dynamics input file, such as an XYZ format file. Figure 3 As shown. According to claim 8, the output file includes a file header (recording the total number of atoms and the three-dimensional dimensions of the simulation box) and a row of atomic data (containing element types and precise coordinates in Cartesian coordinates) for use in subsequent molecular dynamics simulations.

[0087] Example 2

[0088] Taking a hatch assembly with complex mechanical structure, multiple curved surfaces, and precision assembly features as an example (geometric model see...), Figure 4 The generated atomic model can be found here. Figure 5 This verifies the ability of this method to handle asymmetric, highly complex geometries and multi-material integrated modeling. The specific implementation steps are as follows: 1. (The rest of the text appears to be incomplete and requires further context.) Figure 4 As shown, a hatch model with intricate mechanical features is constructed in Blender. This model exhibits significant irregularities, including a spherical valve core, a non-homogeneous valve cover with reinforcing ribs, cylindrical connecting rods, and a complex hinge rotation mechanism.

[0089] 2. Execute steps S1 and S2. The program automatically reads the OBJ file and identifies each region. During the analysis process, for the asymmetric center of gravity position of the hatch assembly, the model is automatically centered by calculating the overall boundary box and combining it with the offset vector calculation method described in step S2. The simulation box is set as a large-scale system of 600Å × 600Å × 600Å. According to step S3, the uniform scaling factor is calculated. Since there are many suspended and recessed areas in the hatch structure, the program calculates the vacuum layer thickness axially (taking a 25% scale) to ensure that the model does not unexpectedly overlap with the periodic boundary in the simulation space.

[0090] 3. Specify different CIF files for different regions.

[0091] 4. For the alternating concave and convex curved surfaces, dense hinge gaps, and hollow internal pipes in the hatch assembly, perform parallel spatial trimming in step S3.3.

[0092] 5. Merge the atomic data from each region to generate the final atomic structure as follows: Figure 5 As shown.

[0093] This embodiment demonstrates that the present invention is not only applicable to layered semiconductor devices, but also capable of handling macro-micro cross-scale modeling scenarios with complex topological structures, porous and hollow features, and the need for precision assembly of multiple materials, providing a solution for molecular dynamics simulation of engineering-level complex structures.

Claims

1. A molecular dynamics crystal structure modeling method based on a three-dimensional geometric model, characterized in that, Includes the following steps: S1: Model analysis steps: Read the 3D geometric model file in OBJ format, analyze each object / group in the file as an independent geometric region, extract the corresponding local vertex and triangle face information for each geometric region, and determine the overall boundary of all geometric regions; S2: Scale unification step: Treat all geometric regions as a whole, calculate the unified scaling factor and overall offset based on the preset simulation box size, perform atomic scale scaling and center positioning on all geometric regions, and update the coordinates of each geometric region and the overall boundary. S3: Regional Atom Generation and Trimming Steps: For each independent geometric region, perform the following operations sequentially: S3.1 Obtain the crystal information file corresponding to this geometric region; S3.2 Based on the lattice parameters in the crystal information file, crystal cell expansion is performed to generate an atomic cloud covering the geometric region, and preliminary screening is performed using a boundary box. S3.3 performs parallel spatial trimming on the initially screened atomic cloud to select atoms located within the geometric region; S4 Merge Output Steps: Merge the atomic coordinates and atomic type information retained in all geometric regions to obtain the molecular dynamics crystal structure model corresponding to the multi-region three-dimensional geometric model, and output the molecular dynamics input file containing the atomic information of the model and the size of the simulation box.

2. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, In step S1, when parsing independent geometric regions, the regions defined in the file are identified, the polygon faces are triangulated, and a mapping relationship is established from global vertices to local vertices of the regions.

3. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, In step S2, when calculating the unified scaling factor, the independent scaling factor for each axis is first calculated based on the overall size of the model and the size of the simulation box, and the minimum value of the independent scaling factor for each axis is taken as the unified scaling factor. When calculating the available space, the vacuum layer thickness is calculated independently for each axis. The vacuum layer thickness for each axis is taken as the larger of the preset ratio of the simulation box size for that axis and the preset minimum absolute thickness. The preset ratio and the preset minimum absolute thickness can be adjusted according to the actual simulation requirements.

4. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, In step S3.2, during crystal cell expansion, the Cartesian coordinate boundary of the corresponding geometric region is converted into an approximate range in the fractional coordinate system. An additional number of buffer cells are added to the upper and lower limits of each dimension of this range to ensure that the atomic cloud completely encloses the geometric region. Then, the Cartesian boundary box of the geometric region is extended outward by a predetermined distance along each axis to form a preliminary screening boundary box. The generated atomic cloud is quickly screened using this preliminary screening boundary box to remove atoms that are obviously located outside the boundary box.

5. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, In step S3.3, during parallel space pruning, atoms outside the region are first quickly excluded by bounding box inspection, and then the remaining atomic clouds to be determined are divided into multiple batches. Multi-threaded / multi-process parallel computing is achieved through thread pool or process pool, and each parallel work unit independently performs the determination of the internal and external positions of atoms.

6. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 5, characterized in that, The determination of the atom's position as inside or outside is achieved using a ray casting method combined with the Möller-Trumbore algorithm. A ray is emitted from the coordinate position of the atom to be determined along a preset fixed direction in the global coordinate system. The spatial relationship between the ray and the triangular face is determined by calculating the determinant of the mixed product formed by the two side vectors of the triangular face and the ray direction vector. When the absolute value of the mixed product determinant is less than a preset floating-point comparison tolerance, the ray is determined to be parallel to the triangular face or in the same plane, and the triangular face is ignored. For non-parallel intersections, the ray intersection parameters are calculated and the validity of the intersection point is verified by combining the floating-point comparison tolerance. Finally, the total number of valid intersection points is counted, and the atom is determined to be inside or outside the geometric region according to the parity rule.

7. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, Step S3.3 includes overlapping atom removal logic, specifically characterized by: calculating the atoms to be determined. With adjacent atoms Euclidean distance between If satisfied If the two atoms overlap, then it is determined that the two atoms overlap; among them, and These are the van der Waals radii for the corresponding atom types. The overlap determination ratio coefficient ranges from 0.6 to 0.

85. Regarding the execution timing and retention rules, after the parallel pruning of the current geometric region is completed, the retained atoms are compared with the atoms generated by other processed geometric regions. If cross-regional overlap occurs, the atoms with higher priority in the region are retained and the atoms with lower priority in the region are removed according to the preset geometric region priority weight.

8. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, The molecular dynamics input file output in step S4 has a header that records the total number of atoms and the three-dimensional dimensions of the simulation box, and an atom data line that includes the element type and the atomic coordinates in the Cartesian coordinate system.

9. The molecular dynamics crystal structure modeling method based on a three-dimensional geometric model according to claim 1, characterized in that, It also includes user interaction configuration steps: it supports specifying a corresponding crystal information file for each geometric region, multiple geometric regions can share the same crystal information file, or multiple crystal information files can be called in turn in a preset order for region filling. Configure the number of processes / threads for parallel processing, and customize the name and format of the output file.

Images