A method and system for predicting the dielectric constant of crystals based on a causal diffusion model

Abstract

This specification discloses a method and system for predicting the dielectric constant of a crystal based on a causal diffusion model, comprising: acquiring crystallographic information data of a material to be predicted; parsing the crystallographic information data of the material to be predicted to generate a normalized structure sequence; performing sequence encoding using a lightweight encoder based on the normalized structure sequence to obtain a guiding vector; using the guiding vector as input to a pre-trained diffusion model to obtain a normalized dielectric constant vector; performing denormalization on the normalized dielectric constant vector to generate a denormalized actual dielectric constant; and outputting the three-dimensional vector of the denormalized actual dielectric constant as the dielectric constant of the material to be predicted to obtain the crystal dielectric constant prediction result of the material to be predicted.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method and system for predicting the dielectric constant of crystals based on a causal diffusion model. Background Technology

[0002] The dielectric constant is a crucial physical quantity describing a material's polarization capability under an applied electric field, reflecting its intrinsic electrical response characteristics that influence energy storage and electromagnetic wave propagation. It is a fundamental design parameter for key electronic devices such as high-performance microwave ceramics, dielectric filters for mobile communication base stations, low-loss substrate materials (e.g., 5G / 6G RF front-ends), high-k dielectric films, advanced packaging materials, and energy storage capacitors, decisively influencing performance parameters such as resonant frequency, propagation velocity, capacitance density, and loss factor. In new energy and semiconductor materials, the dielectric constant is closely related to polarization behavior and is a key indicator affecting the interface stability of lithium-ion battery cathode materials, the carrier shielding capability of perovskite solar cells, and the energy level modulation of defects in wide-bandgap semiconductors. In fundamental materials science, the dielectric constant reflects the electronic structure and crystal symmetry of materials, serving as an important window for understanding crystal polarization mechanisms, phase transition behavior, and structure-property relationships. Therefore, studying and accurately predicting the dielectric constant has significant engineering and industrial implications.

[0003] In existing technologies, deep learning prediction methods and machine learning algorithms used for predicting the dielectric constant of crystals can achieve prediction to a certain extent. However, these methods lack sufficient ability to characterize the structure of complex inorganic materials, automatically extract features, and have low computational efficiency, resulting in unsatisfactory prediction results for the dielectric constant.

[0004] Therefore, a new method and system for predicting the dielectric constant of crystals is needed. Summary of the Invention

[0005] This specification provides a method and system for predicting the dielectric constant of crystals based on a causal diffusion model, to address the following technical problem: While existing deep learning prediction methods and machine learning algorithms can predict the dielectric constant of crystals, they lack sufficient ability to characterize the structure of complex inorganic materials, automatically extract features, and have low computational efficiency, resulting in unsatisfactory prediction results.

[0006] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:

[0007] This specification provides an embodiment of a method for predicting the dielectric constant of a crystal based on a causal diffusion model, including:

[0008] Obtain crystallographic information data of the material to be predicted;

[0009] The crystallographic information data of the material to be predicted is analyzed to generate a standardized structure sequence;

[0010] Based on the standardized structure sequence, a lightweight encoder is used to encode the sequence to obtain the guiding vector;

[0011] The guide vector is used as input to the pre-trained diffusion model to obtain the normalized dielectric constant vector;

[0012] The normalized dielectric constant vector is denormalized to generate the denormalized actual dielectric constant;

[0013] The three-dimensional vector of the inversely normalized actual dielectric constant is output as the dielectric constant of the material to be predicted, thereby obtaining the crystal dielectric constant prediction result of the material to be predicted.

[0014] This specification also provides an embodiment of a crystal dielectric constant prediction system based on a causal diffusion model, comprising:

[0015] The acquisition module acquires crystallographic information data of the material to be predicted;

[0016] The analysis module analyzes the crystallographic information data of the material to be predicted and generates a standardized structure sequence;

[0017] The encoding module, based on the standardized structure sequence, uses a lightweight encoder to perform sequence encoding to obtain the guiding vector;

[0018] The inference module uses the guiding vector as input to the pre-trained diffusion model to obtain a normalized dielectric constant vector.

[0019] The denormalization module denormalizes the normalized dielectric constant vector to generate the denormalized actual dielectric constant;

[0020] The output module outputs the three-dimensional vector of the inversely normalized actual dielectric constant as the dielectric constant of the material to be predicted, thereby obtaining the predicted crystal dielectric constant of the material to be predicted.

[0021] The crystal dielectric constant prediction method and system based on a causal diffusion model provided in this specification acquires crystallographic information data of the material to be predicted; parses the crystallographic information data of the material to be predicted to generate a standardized structure sequence; based on the standardized structure sequence, performs sequence encoding using a lightweight encoder to obtain a guiding vector; uses the guiding vector as input to a pre-trained diffusion model to obtain a normalized dielectric constant vector; denormalizes the normalized dielectric constant vector to generate a denormalized actual dielectric constant; outputs the three-dimensional vector of the denormalized actual dielectric constant as the dielectric constant of the material to be predicted to obtain the crystal dielectric constant prediction result of the material to be predicted. This method achieves comprehensive improvement in prediction accuracy, inference efficiency, computational cost, and applicability, and can significantly improve the efficiency of complex material screening and performance evaluation. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A schematic diagram of the system architecture for a crystal dielectric constant prediction method based on a causal diffusion model, provided for embodiments of this specification;

[0024] Figure 2 A schematic flowchart illustrating a method for predicting the dielectric constant of a crystal based on a causal diffusion model, provided in the embodiments of this specification;

[0025] Figure 3 A schematic diagram of the loss function results during the training process of the causal diffusion model provided in the embodiments of this specification;

[0026] Figure 4 A schematic diagram illustrating the prediction results during the training process of the causal diffusion model provided in the embodiments of this specification;

[0027] Figure 5 A schematic diagram of the framework structure of a crystal dielectric constant prediction method based on a causal diffusion model provided in the embodiments of this specification;

[0028] Figure 6 This is a schematic diagram of a crystal dielectric constant prediction system based on a causal diffusion model, provided as an embodiment of this specification. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0030] In crystalline materials, the dielectric constant is typically expressed as an anisotropic tensor along the x, y, and z directions of the unit cell, and is a core parameter determining the electromagnetic properties of the material in the radio frequency (RF), microwave, millimeter wave (mmWave), and optical frequency bands. Due to differences in the physical mechanisms such as atomic arrangement, bond polarization, and ion displacement in different crystal orientations, the three-dimensional dielectric constants often exhibit significant inequality, directly affecting the material's performance in devices.

[0031] In existing technologies, deep learning prediction methods use crystal structures as the core input. They obtain the chemical formula, dielectric constant measurements, and corresponding CIF files of microwave dielectric ceramics from literature and databases, and pair the CIF files with the dielectric constants to construct a training dataset. Subsequently, they extract the types, positions, and chemical bond information of unit cells from the CIF files, representing the crystal structure as crystal graph data composed of nodes (atoms) and edges (bonds), and perform Z-score normalization on the dielectric constant. In terms of model structure, a Crystal Graph Convolutional Neural Network (CGCNN) is used to perform multi-layer graph convolution and pooling on the crystal graph, automatically extracting the most useful structural features for dielectric constant prediction, and then performing scalar dielectric constant regression through fully connected layers. During training, an appropriate loss function and optimizer are selected, and the Optuna framework is used to automatically search and optimize hyperparameters such as learning rate, batch size, number of convolutional layers, and weight decay, ultimately achieving the prediction of the dielectric constant of microwave dielectric ceramics based on the input CIF. While this method achieves good prediction results for microwave dielectric ceramics by converting CIF files into crystal graph data and automatically extracting structural features using CGCNN, its technical approach has certain limitations. This scheme requires explicit mapping of the crystal structure, relying on manually setting hyperparameters such as neighborhood radii and bond connection rules. Once the graph structure is fixed, the model mainly learns features through local neighborhood aggregation, making it difficult to fully capture long-range structural correlations and overall symmetry information in complex unit cells. For crystal materials with large unit cells, complex compositions, or significant anisotropy, the receptive field expansion efficiency brought by the layer-by-layer graph convolution is limited, and prediction accuracy is easily affected by structural complexity. Furthermore, this patented method mainly targets the regression of a single scalar dielectric constant, focusing on fitting a certain direction or equivalent dielectric constant, and does not directly characterize the anisotropic response of the unit cell in the x, y, and z directions from the model architecture. Therefore, its expressive power is still insufficient when dealing with complex inorganic materials with significant direction dependence.

[0032] Machine learning prediction methods use "material descriptors" as an intermediary to first map the composition and properties of materials into high-dimensional feature vectors, and then use machine learning models to perform dielectric constant regression. This method obtains the dielectric constant and basic material characteristics (such as element types, valence electron configuration, electronegativity, etc.) of sample materials through experiments or databases. Based on this, it constructs various types of descriptors, including configuration entropy ΔS_con derived from the configuration entropy formula, electronic structure attributes reflecting the proportion of s, p, d, and f valence electrons, electronegativity-related characteristics characterizing the degree of ionic compounds, and elemental statistical properties obtained by applying various statistical methods (such as weighted average, extreme values, standard deviation, etc.) to various elemental properties, resulting in a total of 909 candidate descriptors. Subsequently, a genetic algorithm is used to iteratively screen these high-dimensional descriptor spaces, using the coefficient of determination R² and mean square error... The difference between MSE and RMSE are used as fitness indicators for selection, crossover, mutation, and other operations. These operations converge to obtain several principal descriptors that are most sensitive to the dielectric constant (such as whether it is a d-block metal, atomic weight, number of electrons in the p-block, polarizability, Wigner-Seitz electron cloud density, electronegativity, etc.), and the remaining descriptors are regarded as auxiliary features. In terms of model architecture, a hierarchical neural network (HNN) composed of multiple sub-networks is adopted. Each sub-network receives different combinations of principal and other descriptors, and the intermediate outputs are passed and integrated layer by layer to realize the nonlinear mapping model of the dielectric constant of the material, thereby balancing prediction accuracy and generalization ability under small dataset conditions.

[0033] Current research on dielectric constant prediction still heavily relies on first-principles calculations, especially ab initio approaches based on Density Functional Theory (DFT). DFT calculates the dielectric response by solving the electronic structure of a material, and is currently the most classic and widely used theoretical prediction method. A typical workflow includes: establishing a unit cell model in quantum chemistry software based on the material's CIF structure, followed by structural optimization, band structure and density of states calculations, and then using linear response theory to obtain the dielectric tensor of the system in different directions. This method improves the generalization performance of traditional neural networks under small sample conditions, but it requires a large number of material-related physical parameters, such as polarizability, Wigner-Seitz electron cloud density, and electronegativity. These parameters often require costly experimental or quantitative calculations. When dealing with new materials or complex systems that have not yet been thoroughly characterized, the unavailability of these prerequisite features directly limits the applicability of the model. Meanwhile, this descriptor system, which is mainly constructed and statistically processed, does not directly utilize the sequence or local spatial information of the crystal structure itself. There is a strong correlation and redundancy among the features. Although the genetic algorithm can reduce dimensionality to a certain extent, the search process itself has a large computational cost and it is difficult to guarantee that the optimal or near-optimal feature combination is found in the high-dimensional space. As a result, the efficiency and accuracy of the overall scheme face bottlenecks when it is extended to larger material libraries or more complex structures.

[0034] The advantages of DFT lie in its solid theoretical foundation, wide applicability, and ability to simultaneously provide contributions to electronic and ionic polarization, making it widely used in dielectric research of high-performance ceramics, dielectric thin films, and semiconductor nanocrystals. However, DFT computation is extremely expensive, often taking hours to weeks in real-world material systems. For example, in a typical transition metal oxide system (such as Li4V3FeO10, 18 atoms), even on a 32-core research-grade CPU, structural optimization typically takes 4-12 hours, and a single self-consistent field (SCF) calculation takes 0.5-2 hours. Linear response (DFPT) calculations for obtaining the triaxial dielectric constant are even more expensive, taking 10-36 hours under standard settings, and potentially extending to days or even weeks when using higher-precision hybrid functionals. With increases in cell size, atomic number, and component complexity (such as defect structures, doped systems, or supercell models), these costs will increase exponentially. Furthermore, DFT results are highly sensitive to a range of settings, including the choice of exchange-correlated functional, pseudopotential type and accuracy, plane wave cutoff energy, k-point grid density, and initial magnetic guess. Significant differences in results often arise between different software platforms (VASP, QE, ABINIT, etc.) and with different parameter configurations. For tasks requiring batch screening of hundreds or thousands of structures, high-throughput crystal structure searches, defect state calculations, or spatial exploration of complex multi-component materials, the high computational cost and parameter sensitivity of DFT create a natural bottleneck, making it difficult to support large-scale material discovery and screening processes.

[0035] Based on this, the embodiments of this specification provide a method for predicting the dielectric constant of crystals based on a causal diffusion model. Through a new structural representation method and lightweight deep model design, the dielectric constant of complex crystal materials can be predicted quickly while maintaining accuracy. This makes up for the shortcomings of existing technologies in terms of structural characterization ability, automatic feature extraction ability and computational efficiency, and provides a more efficient and universal solution for high-throughput screening and application research of complex material systems.

[0036] Figure 1 This diagram illustrates the system architecture of a crystal dielectric constant prediction method based on a causal diffusion model, as provided in the embodiments of this specification. Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0037] Terminal devices 101, 102, and 103 interact with server 105 via network 104 to receive or send messages, etc. Various client applications can be installed on terminal devices 101, 102, and 103, such as dedicated programs for predicting the dielectric constant of crystals based on a causal diffusion model.

[0038] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various dedicated or general-purpose electronic devices, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module.

[0039] Server 105 can be a server that provides various services, such as a backend server that provides services to client applications installed on terminal devices 101, 102, and 103. For example, the server can perform crystal dielectric constant prediction based on a causal diffusion model so that the crystal dielectric constant prediction results based on the causal diffusion model can be displayed on terminal device servers 101, 102, and 103.

[0040] Server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services), or as a single software program or software module.

[0041] Figure 2 This is a flowchart illustrating a method for predicting the dielectric constant of a crystal based on a causal diffusion model, provided in an embodiment of this specification. From a programming perspective, the execution entity of the process can be a program hosted on an application server or application terminal. It is understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the method for predicting the dielectric constant of a crystal includes:

[0042] Step S201: Obtain crystallographic information data of the material to be predicted.

[0043] In the embodiments of this specification, the crystallographic information data of the material to be predicted is CIF file data, which includes: unit cell parameters, atom types, atomic fraction coordinates, and space group information.

[0044] CIF (Crystallographic Information File) files are standardized file formats for storing crystal structure data. In this application, only the unit cell parameters, atom types, atomic fraction coordinates, and space group information are used as inputs for subsequent crystal dielectric constant prediction. It is not necessary to consider other physical properties of the material to be predicted, such as energy, electron cloud density, and polarizability, nor is it necessary to rely on pre-calculated descriptors or external database information. Therefore, it is more convenient to predict the crystal dielectric constant.

[0045] Step S203: Analyze the crystallographic information data of the material to be predicted and generate a standardized structure sequence.

[0046] After obtaining the crystallographic information data of the material to be predicted, further analysis is required to generate a standardized sequence.

[0047] In the embodiments of this specification, the step of parsing the crystallographic information data of the material to be predicted and generating a normalized structure sequence specifically includes:

[0048] Extract lattice geometry parameters, space group information, and atomic coordinates from the crystallographic information data of the material to be predicted;

[0049] Structure tags are inserted between the lattice geometry parameters, space group information, and atomic coordinates to generate the standardized structure sequence.

[0050] Lattice geometry parameters are extracted from the CIF file to determine the basic scale and shape of the crystal in three-dimensional space. Specifically, the extracted lattice geometry parameters include: cell edge lengths a, b, and c, crystal axis angles α, β, and γ, and cell volume.

[0051] In the embodiments of this specification, the structural label includes: <boa> 、 <bou> 、 <boc> 、 <eoc>,in, <boa>The symbols boU, boC, and eoC represent the beginning of coordinates and the starting point of coordinates, respectively.

[0052] The presence of structural tags ensures that the symmetry, atomic configuration, and geometric parameters of the crystal possess stable structural boundaries within the sequence. This approach not only makes the structural representation compatible with the Transformer's self-attention mechanism but also avoids problems such as "non-unique adjacency definitions" in crystal mapping, fundamentally improving the generalization ability and scalability of structural inputs.

[0053] In the embodiments described in this specification, the standardized sequence obtained after the above operations is a sequence consisting only of characters.

[0054] Step S205: Based on the standardized structure sequence, a lightweight encoder is used to encode the sequence to obtain the guiding vector.

[0055] The standardized structure sequence obtained through the above steps needs to be further encoded.

[0056] In the embodiments of this specification, the step of using a lightweight encoder to encode the sequence based on the standardized structure sequence to obtain the guiding vector specifically includes:

[0057] After converting the standardized structure sequence into an index sequence, the index sequence is truncated or supplemented according to a preset length;

[0058] The truncated or supplemented index sequence is input into the Tiny Llama encoder for sequence encoding to obtain the guiding vector.

[0059] The normalized structure sequence is transformed into an index sequence, which can be achieved using a character-level tokenizer. This tokenizer maps the characters or tags in the normalized structure sequence to unique integer codes, and then truncates or pads them according to a preset length. Specifically, if the normalized structure sequence is longer than the preset length, it is truncated to ensure that the input sequence length meets the requirements of subsequent inputs; if the normalized structure sequence is shorter than the preset length, it is padded. Through these operations, the sequence length input to the Tiny Llama encoder is kept consistent, i.e., a uniform input dimension is maintained.

[0060] A character-level word segmenter is a word segmentation method that divides text into individual characters as the smallest processing unit. In this embodiment, the specific type of character-level word segmenter adopts existing technology and will not be described in detail here.

[0061] In the embodiments described in this specification, the preset length is 1024. In practice, the preset length can be determined according to the business scenario in specific embodiments.

[0062] Generally speaking, regardless of whether the crystal structure is simple or relatively complex, the number of atoms in the crystal does not exceed 250. Therefore, a preset length of 1024 can meet the requirements for predicting the dielectric constant of most crystals. Of course, for extremely complex crystal structures, the preset length can be set to a larger value.

[0063] In the embodiments of this specification, the step of inputting the truncated or supplemented index sequence into a Tiny Llama encoder for sequence encoding to obtain the guiding vector specifically includes:

[0064] The truncated or supplemented index sequence is input into the Tiny Llama encoder. The Tiny Llama encoder's Transformer, through its attention layer, focuses on the interaction between the lattice parameters, atomic arrangement, and space group symbols in the truncated or supplemented index sequence, generating high-quality structural features.

[0065] The hidden states at effective locations in the high-quality structural features are masked average pooling to obtain the guiding vector;

[0066] in,

[0067] The mask average pooling uses the following formula:

[0068] ;

[0069] in,

[0070] The hidden layer vector representing the effective position of each token in the high-quality structural features;

[0071] A mask representing the implicit state of the effective location in the high-quality structural feature;

[0072] This represents the i-th valid position in the high-quality structural feature;

[0073] The first of the high-quality structural features L One valid location;

[0074] cond represents the guiding vector.

[0075] In the embodiments of this specification, a Transformer self-attention layer is introduced into the Llama encoder, enabling the model to effectively capture long-range dependencies in sequences of thousands of characters with a very small parameter scale, achieving holistic modeling of structural elements such as lattice parameters, space group information, and atomic arrangement. The Transformer self-attention layer can compute multiple attention heads to capture different types of feature relationships. Simultaneously, it preserves the sequence order information of the index sequence through rotational position encoding. Finally, through feature interaction, it learns the complex relationships between lattice parameters, atomic arrangement, and space group symbols, thereby obtaining high-quality structural features. These high-quality structural features focus on the features of the index sequence itself, while also paying attention to the complex relationships between lattice parameters, atomic arrangement, and space group symbols. The Transformer can simultaneously focus on the relationship between any two positions in the structure, thus significantly enhancing the model's ability to express complex unit cells and long-range structural relationships.

[0076] To obtain a consistent characterization of the entire crystal structure, it is necessary to further pool the features of the non-filled positions in the high-quality structural features, that is, to perform masked average pooling on the hidden states of the effective positions in the high-quality structural features. During masked average pooling, the effective token positions are represented by an attention mask mechanism, and then average pooling is performed, that is, the hidden states of the effective token positions are averaged to obtain the guiding vector, which is a representation vector with fixed dimensions.

[0077] Step S207: Use the guiding vector as input to the pre-trained diffusion model to obtain the normalized dielectric constant vector.

[0078] In the embodiments of this specification, the step of using the guiding vector as input to a pre-trained diffusion model to obtain a normalized dielectric constant vector specifically includes:

[0079] Using the guiding vector as the conditional input to the pre-trained diffusion model, the initial noise vector and the diffusion time step are fed into the diffusion regression head network;

[0080] The diffusion regression head network uses time embedding and a multi-layer fully connected network to perform noise prediction using a noise prediction formula, and obtains the noise prediction result.

[0081] Based on the noise prediction results, the noise input of the previous diffusion time step is updated using a preset diffusion inverse process formula.

[0082] After multiple iterations, the normalized dielectric constant vector is obtained.

[0083] In the embodiments described in this specification, the noise prediction formula is:

[0084] ;

[0085] in,

[0086] This indicates the noise prediction result;

[0087] The noise represents the diffusion time step t;

[0088] Indicates the diffusion time step;

[0089] Indicates model parameters;

[0090] Represents the guiding vector;

[0091] The preset formula for the reverse diffusion process is:

[0092] ;

[0093] in,

[0094] This represents the noise of the previous diffusion time step.

[0095] In other words, in the embodiments of this specification, if the diffusion time step is t, then the previous diffusion time step is t-1.

[0096] In the embodiments of this specification, when training the pre-trained diffusion model, the training and validation sets are sourced from a crystal database. Specifically, the type and source of the crystal database are not limited. To ensure the stability of the training results of the pre-trained diffusion model, the crystal data in the training and test sets must be complete, including cell parameters, atom types, atomic fractional coordinates, and space group information.

[0097] Simultaneously, an early stopping mechanism is employed to prevent overfitting during the training of the diffusion model. Using R... 2 R reaches its maximum value within 10 cycles 2 The value stops increasing, marking the end of diffusion model training. In the embodiments of this specification, the pre-trained diffusion model is obtained after 115 training cycles. During model training, the loss function used is MAE, i.e., Mean Absolute Error. See details... Figure 3 and Figure 4 .

[0098] In actual testing, this invention relies solely on the CIF structure of the material. The performance metrics on the validation set are R² = 0.8703 and MAE = 16.7465, which are basically comparable to the accuracy of existing deep learning methods that rely on a large number of physical quantities or complex graph structures. However, it does not require the calculation of high-cost descriptors such as electron cloud density, polarizability, and electronegativity, nor does it require crystal mapping. This makes the model more adaptable and generalizable when facing unknown materials, low-symmetry structures, and complex large unit cells.

[0099] It should be noted that, in the embodiments of this specification, the causal diffusion model refers to a causal-diffusion model, specifically including a lightweight encoder and a pre-trained diffusion model. The lightweight encoder is used to sequence encode the normalized sequence to obtain a guiding vector; the pre-trained diffusion model is used to take the guiding vector as input and then output a normalized dielectric constant vector.

[0100] Step S209: Denormalize the normalized dielectric constant vector to generate the denormalized actual dielectric constant.

[0101] To improve the stability of the crystal dielectric constant predictions provided in the embodiments of this specification, it is further necessary to denormalize the normalized dielectric constant vector, thereby restoring the normalized dielectric constant vector to the same physical dimensions and numerical range as the original data. The denormalized three-dimensional vector can be directly compared and verified with the dielectric constant obtained by traditional first-principles calculations or experimental measurements.

[0102] In the embodiments of this specification, the step of denormalizing the normalized dielectric constant vector to generate the denormalized actual dielectric constant specifically includes:

[0103] Based on the standard deviation vector and mean vector of the training data, the normalized dielectric constant vector is denormalized using an inverse transformation formula to generate the denormalized actual dielectric constant.

[0104] In the embodiments described in this specification, the inverse transform formula is:

[0105] ;

[0106] This represents the actual dielectric constant after inversion;

[0107] This represents the normalized dielectric constant vector;

[0108] This represents the standard deviation vector of the training data;

[0109] This represents the mean vector of the training data.

[0110] Step S211: Output the three-dimensional vector of the denormalized actual dielectric constant as the dielectric constant of the material to be predicted, and obtain the crystal dielectric constant prediction result of the material to be predicted.

[0111] The denormalized three-dimensional vector is used as the dielectric constant of the material's unit cell in the x, y, and z directions. , , The output is directly called by downstream microwave device design, material screening and performance evaluation modules, thus completing the entire process from crystal structure input to polycrystalline dielectric constant prediction.

[0112] To understand the crystal dielectric constant prediction method based on the causal diffusion model provided in the embodiments of this specification, the embodiments of this specification also provide a framework structure for the crystal dielectric constant prediction method based on the causal diffusion model. Figure 5 This is a schematic diagram illustrating the framework of a crystal dielectric constant prediction method based on a causal diffusion model, provided in the embodiments of this specification. Figure 5 As shown in the embodiments of this specification, the crystal dielectric constant prediction method based on a causal diffusion model, after obtaining the CIF file, performs structural analysis and normalized sequence generation, then sequence encoding, and then inference through a diffusion model to obtain a normalized dielectric constant vector; the normalized dielectric constant vector is then post-processed, i.e., denormalized, to generate the denormalized actual dielectric constant, and finally outputs the crystal dielectric constant prediction result. , , .

[0113] The crystal dielectric constant prediction method and system based on a causal diffusion model provided in this specification acquires crystallographic information data of the material to be predicted; parses the crystallographic information data of the material to be predicted to generate a standardized structure sequence; based on the standardized structure sequence, performs sequence encoding using a lightweight encoder to obtain a guiding vector; uses the guiding vector as input to a pre-trained diffusion model to obtain a normalized dielectric constant vector; denormalizes the normalized dielectric constant vector to generate a denormalized actual dielectric constant; outputs the three-dimensional vector of the denormalized actual dielectric constant as the dielectric constant of the material to be predicted to obtain the crystal dielectric constant prediction result of the material to be predicted. This method achieves comprehensive improvement in prediction accuracy, inference efficiency, computational cost, and applicability, and can significantly improve the efficiency of complex material screening and performance evaluation.

[0114] The above describes in detail a method for predicting the dielectric constant of crystals based on a causal diffusion model. Correspondingly, this specification also provides a system for predicting the dielectric constant of crystals based on a causal diffusion model, such as... Figure 6 As shown. Figure 6 This is a schematic diagram of a crystal dielectric constant prediction system based on a causal diffusion model, provided in an embodiment of this specification. The crystal dielectric constant prediction system includes:

[0115] Module 601 acquires crystallographic information data of the material to be predicted;

[0116] The analysis module 603 analyzes the crystallographic information data of the material to be predicted and generates a standardized structure sequence;

[0117] The encoding module 605, based on the standardized structure sequence, uses a lightweight encoder to perform sequence encoding to obtain the guiding vector;

[0118] The inference module 607 uses the guiding vector as input to the pre-trained diffusion model to obtain a normalized dielectric constant vector.

[0119] The denormalization module 609 denormalizes the normalized dielectric constant vector to generate the denormalized actual dielectric constant.

[0120] The output module 611 outputs the three-dimensional vector of the inversely normalized actual dielectric constant as the dielectric constant of the material to be predicted, thereby obtaining the crystal dielectric constant prediction result of the material to be predicted.

[0121] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0122] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for apparatus, electronic devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0123] The apparatus, electronic device, and non-volatile computer storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, electronic device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device, and non-volatile computer storage medium will not be repeated here.

[0124] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0125] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0126] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0127] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0128] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0130] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0131] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0132] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0133] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0134] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0135] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0136] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside on local and remote computer storage media, including storage devices.

[0137] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0138] The above description is merely an embodiment of this specification and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.< / boa> < / eoc> < / boc> < / bou> < / boa>

Claims

1. A method for predicting the dielectric constant of a crystal based on a causal diffusion model, characterized in that, The crystal dielectric constant prediction method comprises: obtaining crystallographic information data of a material to be predicted; parsing the crystallographic information data of the material to be predicted to generate a standardized structure sequence; based on the standardized structure sequence, using a lightweight encoder for sequence encoding to obtain a guide vector; inputting the guide vector into a pre-trained diffusion model to obtain a normalized dielectric constant vector; de-normalizing the normalized dielectric constant vector to generate a de-normalized actual dielectric constant; outputting the three-dimensional vector of the de-normalized actual dielectric constant as the dielectric constant of the material to be predicted to obtain the crystal dielectric constant prediction result of the material to be predicted.

2. The crystal dielectric constant prediction method of claim 1, wherein, The crystallographic information data of the material to be predicted is CIF file data, and the crystallographic information data of the material to be predicted includes unit cell parameters, atomic species, atomic fraction coordinates and space group information.

3. The crystal dielectric constant prediction method of claim 1, wherein, The parsing of the crystallographic information data of the material to be predicted to generate a standardized structure sequence specifically comprises: extracting lattice geometry parameters, space group information and atomic coordinates from the crystallographic information data of the material to be predicted; inserting structure labels between the lattice geometry parameters, space group information and atomic coordinates to generate the standardized structure sequence.

4. The crystal dielectric constant prediction method of claim 1, wherein, Based on the standardized structure sequence, the lightweight encoder is used for sequence encoding to obtain a guide vector, which specifically comprises: After converting the standardized structure sequence into an index sequence, the index sequence is truncated or supplemented according to a preset length; inputting the index sequence obtained by truncation or supplementation into a Tiny Llama encoder for sequence encoding to obtain the guide vector.

5. The crystal dielectric constant prediction method of claim 4, wherein, The inputting of the index sequence obtained by truncation or supplementation into the Tiny Llama encoder for sequence encoding to obtain the guide vector specifically comprises: inputting the index sequence obtained by truncation or supplementation into the Tiny Llama encoder, and the Transformer self-attention layer of the Tiny Llama encoder focuses on the mutual influence of lattice parameters, atomic arrangement and space group symbols in the index sequence obtained by truncation or supplementation to generate high-quality structure features; mask average pooling is performed on the hidden states of the effective positions in the high-quality structure features to obtain the guide vector; wherein, the mask average pooling uses the following formula: ； wherein, a hidden layer vector for each token representing an effective position in the high-quality structural feature; a mask representing the implicit states of valid positions in the high-quality structure features; represents the i-th significant position in the high quality structure feature; representing a first effective position in said high quality structural feature; L representing a first effective position in said high quality structural feature; cond represents the guide vector.

6. The crystal dielectric constant prediction method of claim 1, wherein, The guide vector is input into a pre-trained diffusion model to obtain a normalized dielectric constant vector, which specifically comprises: inputting the guide vector as a conditional input of the pre-trained diffusion model, and inputting an initial noise vector and a diffusion time step into a diffusion regression head network; the diffusion regression head network uses a noise prediction formula to predict noise by time embedding and a multi-layer fully connected network to obtain a noise prediction result; based on the noise prediction result, a preset diffusion inverse process formula is used to update the noise input of the previous diffusion time step of the diffusion time step; after multiple iterations, the normalized dielectric constant vector is obtained.

7. The crystal dielectric constant prediction method of claim 6, wherein, The noise prediction formula is: ； wherein, represents the noise prediction result; represents the noise of the diffusion time step t; denotes a diffusion time step; denotes the model parameters; Represents the guiding vector; the preset diffusion inverse process formula is: ； wherein, represents the noise of the previous diffusion time step of said diffusion time step.

8. The crystal dielectric constant prediction method of claim 1, wherein, The inverse normalization of the normalized dielectric constant vector generates an inverse normalized actual dielectric constant, specifically comprising: Based on the standard deviation vector of the training data and the mean vector of the training data, using an inverse transformation formula, the normalized dielectric constant vector is inversely normalized to generate the inverse normalized actual dielectric constant.

9. The crystal dielectric constant prediction method of claim 8, wherein, The inverse transformation formula is: ； denotes the actual dielectric constant of the de-normalization; denotes the normalized permittivity vector; a standard deviation vector representing the training data; denotes the mean vector of the training data.

10. A system for predicting the dielectric constant of a crystal based on a causal diffusion model, the system comprising: The crystal dielectric constant prediction system comprises: An acquisition module acquires crystallographic information data of a material to be predicted; An analysis module analyzes the crystallographic information data of the material to be predicted to generate a standardized structure sequence; An encoding module uses a lightweight encoder to perform sequence encoding based on the standardized structure sequence to obtain a guide vector; An inference module inputs the guide vector into a pre-trained diffusion model to obtain a normalized dielectric constant vector; An inverse normalization module inversely normalizes the normalized dielectric constant vector to generate an inverse normalized actual dielectric constant; An output module outputs a three-dimensional vector of the inverse normalized actual dielectric constant as the dielectric constant of the material to be predicted to obtain a crystal dielectric constant prediction result of the material to be predicted.

Images