A method for calculating phase field parameters combining artificial intelligence and multi-scale calculation

Abstract

The application relates to a phase field method parameter calculation method combining artificial intelligence and multi-scale calculation, which comprises the following steps: calculating effective Hamilton method input parameters of a material; constructing an effective Hamilton model of the material; performing Monte Carlo simulation to obtain material information of the material under different doping concentrations and in a limited temperature range; obtaining Landau energy coefficients of different doping concentrations by using a neural network and a PSO algorithm; fitting the Landau energy coefficients by using a least square method to obtain a function of the Landau energy coefficients with respect to the doping concentration, and optimizing the function by using a Bayesian optimization method; performing DFT calculation to obtain various energies of a new material to be designed, and obtaining input parameters of a material phase field by fitting; and performing phase field method simulation of the material to obtain physical performance of the new material to be designed under a target condition; the method realizes multi-scale simulation under microscopes, lattices, mesoscopes and macroscopes, and connects parameters of different calculation methods under different scales by using artificial intelligence, so that the calculation result is more reliable and accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of phase field method parameter fitting, and particularly to a phase field method parameter calculation method combining artificial intelligence and multi-scale calculation. BACKGROUND

[0002] The phase field method is a calculation method for simulating and predicting material performance based on the Ginzburg-Landau theory by solving the TDGL equation. In the phase field method, phase transition thermodynamics and the accompanying structural evolution are achieved by selecting a free energy density functional that depends on conservative and non-conservative phase field order parameters. Through the dependence of the order parameters on space, the phase field of non-uniform composition and structure can be determined, and the phase field dynamics and its structural morphology can be simulated. The morphology and structure can be obtained through the spatial distribution of the order parameters. In the phase field simulation of ferroelectric materials, the polarization vector is used as the order parameter.

[0003] In the traditional phase field method, the phase field input parameters are fitted by experimentally measuring the electric hysteresis loop and the dielectric coefficient of the material at different temperatures. This method has the following defects: low efficiency, low accuracy and low reliability. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a phase field method parameter calculation method combining artificial intelligence and multi-scale calculation, which can efficiently and quickly realize the simulation of the material, and the fitting result is more accurate and reliable.

[0005] The technical solution adopted by the present application is a phase field method parameter calculation method combining artificial intelligence and multi-scale calculation, which comprises the following steps:

[0006] S1, obtaining the crystal structure, lattice constant and potential function of the new material to be designed, using the first principle to perform high-throughput calculation to obtain the effective Hamiltonian input parameters of the material;

[0007] S2, based on the first principle, taking the strain and polarization soft film of the material as the degrees of freedom, constructing an effective Hamiltonian model of the material;

[0008] S3, according to the parameters obtained in step S1, using Monte Carlo simulation to simulate the effective Hamiltonian model; through simulation, the average structure of the material under different doping concentrations is obtained, and the material information in a plurality of doping concentrations and a limited temperature range is obtained according to the average structure under different doping concentrations, the material information including the Curie temperature point, the polarization intensity and the dielectric constant;

[0009] S4, based on the material information obtained in step S3, the material information is optimized by using a neural network to obtain optimized material information; the optimized material information is used as data to be fitted, and the PSO algorithm is used for fitting to obtain Landau energy coefficients of different doping concentrations;

[0010] S5, using the least square method, taking the doping concentration as the independent variable, fitting the Landau energy coefficient obtained in step S4 to obtain a function of the Landau energy coefficient with respect to the doping concentration;

[0011] S6, using the Bayesian optimization method to optimize the function with respect to the doping concentration obtained in step S5 to obtain an optimized function with respect to the doping concentration;

[0012] S7, using VASP software to perform DFT calculation to obtain the coupling energy, elastic energy and gradient energy of the new material to be designed;

[0013] S8, based on the energies obtained in step S7, the input parameters of the material phase field are obtained by fitting, the input parameters of the material phase field including the elastic coefficient tensor c ijkl , the gradient coefficient tensor G ijkl1 , and the electrostriction tensor Q ijkl1 ;

[0014] S9, according to the obtained optimized function with respect to the doping concentration and the input parameters of the material phase field, performing phase field simulation of the material to obtain the physical performance of the new material to be designed under the target condition;

[0015] S10, based on the physical performance obtained in step S8, designing a new material with the physical performance.

[0016] The beneficial effects of the present application are: the above-mentioned phase field parameter calculation method combining artificial intelligence and multi-scale calculation is used, the input parameters of the phase field method are obtained by the calculation method, which is more efficient and faster than the traditional method of obtaining parameters by experiment; the present application combines various calculation methods to realize multi-scale simulation under micro, lattice, mesoscopic and macroscopic, and connects the parameters of different calculation methods under different scales by using artificial intelligence, realizes the conversion between different calculation methods, and the calculation result is more reliable and accurate compared with single calculation method; in the parameter fitting process, the artificial intelligence is used to repeat multiple times fitting, and the least square method and the Bayesian optimization are used to optimize the fitting result, further reducing the calculation error, thereby ensuring the accuracy and reliability of the calculation.

[0017] As preferred, the specific process of step S1 includes the following steps:

[0018] S1.1, obtaining the crystal structure, lattice constant and potential function of a new material to be designed from a database;

[0019] S1.2, performing a slight perturbation on the atomic positions in the crystal structure according to the crystal structure in the equilibrium state, and performing a high-throughput calculation on the crystal structure to obtain the polarization and energy of the material;

[0020] S1.3, fitting the polarization and energy of the material obtained in step S1.2 using the least squares method to obtain the correspondence between the ferroelectric polarization and the energy, and according to the correspondence, the input parameters of the effective Hamilton method of the material can be obtained.

[0021] As preferred, in step S4, the specific process of optimizing the material information using the neural network includes the following steps:

[0022] S4.11, taking the polarization intensity calculated in step S3 as the input of the neural network, and taking the Landau energy coefficient as the weight, and using the formula:

[0023]

[0024] to calculate the ideal dielectric constant, the ideal dielectric constant being the output of the neural network; wherein ε0 represents the vacuum dielectric constant, ε c represents the relative dielectric constant, F represents the Landau energy, P i and P j both represent polarization components;

[0025] S4.12, performing error calculation on the ideal dielectric constant calculated in step S4.11 and the dielectric constant calculated in step S3 to obtain a dielectric constant error value, and establishing a dielectric constant error function about the dielectric constant error value, and performing deep learning on the neural network according to the dielectric constant error function until the error value of the dielectric constant error function reaches a minimum, at which time the ideal dielectric constant output by the neural network is the optimized dielectric constant data;

[0026] S4.13, taking the dielectric constant calculated in step S3 as the input of the neural network, and taking the Landau energy coefficient as the weight, and using the formula:

[0027]

[0028] to calculate the ideal polarization intensity, the ideal polarization intensity being the output of the neural network; wherein ε0 represents the vacuum dielectric constant, ε c represents the relative dielectric constant, F represents the Landau energy, P i and P j both represent polarization components;

[0029] S4.14, error calculation is performed on the ideal polarization intensity calculated in step S4.13 and the polarization intensity calculated in step S3 to obtain a polarization intensity error value, and a polarization intensity error function about the polarization intensity error value is established, and deep learning is performed on the neural network according to the polarization intensity error function until the error value of the polarization intensity error function reaches a minimum, at which time the ideal polarization intensity output by the neural network is the optimized polarization intensity.

[0030] As preferred, in step S4, the specific process of obtaining the Landau energy coefficient of different doping concentrations by fitting using the PSO algorithm includes the following steps:

[0031] S4.21, according to the corresponding relationship of the polarization components of each phase of the material, the Landau energy expression is simplified by the total polarization intensity P to obtain a simplified Landau energy expression; the Landau energy expression is: F Landau = α ij P i P j + α ijkl P i P j P k P l ; wherein, α ij and α ijkl represent Landau energy coefficients, P i , P j , P k , P l all represent polarization components;

[0032] S4.22, according to the simplified Landau energy expression, determine the coefficients to be fitted;

[0033] S4.23, set the PSO algorithm parameters, including the number of coefficients to be fitted, the number of particles, the minimum value of the coefficients to be fitted, the maximum value of the coefficients to be fitted, and the maximum number of iterations;

[0034] S4.24, initialize the parameters in step S4.23;

[0035] S4.25, at each temperature, use the expression obtained by simplifying in step S4.21 and the initialized coefficients obtained in step S2.24 to calculate the minimum value of the Landau energy of different phases, and the phase with the minimum energy is the actual phase fitted at that temperature, and the polarization intensity corresponding to the minimum energy is fitted to obtain the dielectric constant according to the polarization intensity corresponding to the minimum energy;

[0036] S4.26. Compare the polarization intensity and dielectric constant obtained from step S2.25 with the optimized polarization intensity and dielectric constant obtained from step S3, and then calculate the root mean square error, the expression of which is: Where P and ε represent the polarization and dielectric constant obtained by fitting in step S2.25, respectively; P0 and ε0 represent the optimized polarization intensity and optimized dielectric constant, respectively; N is the number of data points; A, B, and C are weighting coefficients; phase_fit(yes=0, no=1) indicates that the phase is satisfied.

[0037] S4.27. Iterate the root mean square error obtained in step S4.26 until the root mean square error is reduced to the expected value. The corresponding fitted polarization intensity and dielectric constant are the Landau parameter values. Attached Figure Description

[0038] Figure 1 This is a flowchart of a phase-field method parameter calculation method combining artificial intelligence and multi-scale computation according to the present invention.

[0039] Figure 2 This is a linear graph of the optimized doping concentration using BaTiO3 as an example in this invention; where Figure (a) shows the relationship between α0 and Ba 2+ Concentration relationship graph, where graph (b) represents T C with Ba 2+ Concentration relationship graph. Detailed Implementation

[0040] The invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can implement it based on the description. The scope of protection of the invention is not limited to these specific embodiments.

[0041] This invention relates to a method for calculating phase-field parameters that combines artificial intelligence and multi-scale computation. The method includes the following steps:

[0042] S1. Obtain the crystal structure, lattice constant, and potential function of the new material to be designed, and use first-principles calculations to perform high-throughput calculations to obtain the effective Hamiltonian input parameters of the material.

[0043] The specific process of step S1 includes:

[0044] S1.1 Obtain the crystal structure, lattice constant, and potential function of the new material to be designed from literature or databases;

[0045] S1.2. Based on the crystal structure in equilibrium, the atomic positions in the crystal structure are perturbed one by one, and high-throughput calculations are performed on the crystal structure to obtain the polarization and energy of the material.

[0046] S1.3. The polarization and energy of the material obtained in step S1.2 are fitted using the least squares method to obtain the correspondence between ferroelectric polarization and energy. Based on the correspondence, the input parameters of the effective Hamiltonian method for the material can be obtained.

[0047] S2. Based on first principles, an effective Hamiltonian model of the material is constructed, using the material strain and polarization-related soft film as degrees of freedom;

[0048] S3. Based on the parameters obtained in step S1, the effective Hamiltonian model is simulated using Monte Carlo simulation; the average structure of the material under different doping concentrations is obtained through simulation, and material information under multiple doping concentrations and within a limited temperature range is obtained based on the average structure under different doping concentrations, including Curie temperature, polarization intensity and dielectric constant.

[0049] S4. Based on the material information obtained in step S3, the material information is optimized using a neural network to obtain optimized material information; the optimized material information is used as the data to be fitted, and the Landau energy coefficients of different doping concentrations are obtained by fitting using the PSO algorithm.

[0050] Furthermore, in step S4, the specific process of optimizing the material information using a neural network includes:

[0051] S4.11. Using the polarization intensity calculated in step S3 as the input to the neural network, and the Landau energy coefficient as the weight, the formula is applied.

[0052]

[0053] The ideal dielectric constant is calculated, and this ideal dielectric constant is the output of the neural network; where ε0 represents the vacuum dielectric constant, ε c P represents the relative permittivity, F represents the Landau energy, and P represents the relative permittivity. i and P j Both represent polarization components;

[0054] S4.12. Perform error calculation between the ideal dielectric constant calculated in step S4.11 and the dielectric constant calculated in step S3 to obtain the dielectric constant error value, and establish a dielectric constant error function based on the dielectric constant error value. Perform deep learning on the neural network based on the dielectric constant error function until the error value of the obtained dielectric constant error function reaches the minimum. At this time, the ideal dielectric constant output by the neural network is the optimized dielectric constant data.

[0055] S4.13. Using the dielectric constant calculated in step S3 as the input to the neural network and the Landau energy coefficient as the weight, the formula is applied.

[0056]

[0057] The ideal polarization intensity is calculated, and this ideal polarization intensity is the output of the neural network; where ε0 represents the vacuum permittivity, ε c P represents the relative permittivity, F represents the Landau energy, and P represents the relative permittivity. i and P j Both represent polarization components;

[0058] S4.14. Calculate the error between the ideal polarization intensity calculated in step S4.13 and the polarization intensity calculated in step S3 to obtain the polarization intensity error value, and establish a polarization intensity error function based on the polarization intensity error value. Perform deep learning on the neural network based on the polarization intensity error function until the error value of the obtained polarization intensity error function reaches the minimum. At this time, the ideal polarization intensity output by the neural network is the optimized polarization intensity.

[0059] Furthermore, in step S4, the specific process of obtaining the Landau energy coefficients for different doping concentrations using the PSO algorithm includes:

[0060] S4.21. Based on the corresponding relationship of polarization components of each phase of the material, simplify the Landau energy expression with the total polarization intensity P to obtain the simplified Landau energy expression.

[0061] Specifically, the corresponding relationships of the polarization components of each phase in the material are as follows: R: p1 = p2 = p3 ≠ 0; 0: p1 = p2 ≠ 0, p3 = 0 or p1 = p3 ≠ 0, p2 = 0 or p2 = p3 ≠ 0, p1 = 0. T: p1 ≠ 0, p2 = p3 = 0 or p2 ≠ 0, p1 = p3 = 0 or p3 ≠ 0, p1 = p2 = 0. C: p1 = p2 = p3 = 0; Taking BaTiO3 as an example, the Landau energy expression is:

[0062] F Landau =α ij P i P j +α ijkl P i P j P k P l

[0063] Where α ij and α ijkl P is the Landau energy coefficient. i P j Pk P l These are the polarization components, and their expansion is an eighth-order polynomial:

[0064] F=α1(P1 2 +P2 2 +P3 2 )+α 11 (P1 4 +P2 4 +P3 4 )+α 12 (P1 2 P2 2 +P2 2 P3 2 +P3 2 P1 2 )+α 111 (P1 6 +P2 6 +P3 6 )+α 112 [(P1 4 (P2 2 +P3 2 )+P2 4 (P1 2 +P3 2 )+P3 4 (P1 2 +P2 2 )]+α 123 P1 2 P2 2 P3 2 +α 1111 (P1 8 +P2 8 +P3 8 )+α 1112 [(P1 6 (P2 2 +P3 2 )+P2 6 (P1 2 +P3 2 )+P3 6 (P1 2 +P2 2 )]+α 1123 P1 2 P2 2 P3 2 (P1 2 +P2 2 +P3 2 )

[0065] The simplified expressions for each phase are:

[0066] R:

[0067]

[0068] O:

[0069] T: F = a0(TT) C )P 2 +α 11 P 4 +α 111 P 6 +α 1111 P 8

[0070] C:F=0

[0071] S4.22. Determine the coefficients to be fitted, i.e.: a0, T C α 11 α 12 α 111 α 112 α 123 α 1111 α 1112 α 1122 and α 1123 ;

[0072] S4.23. Set the PSO algorithm parameters, including dim (the number of coefficients to be fitted) and pop (the number of particles).

[0073] lb (minimum value of coefficients), ub (maximum value of coefficients), and max_iter (maximum number of iterations);

[0074] S4.24. Initialize the parameters in step S4.23;

[0075] S4.25. At each temperature, the minimum Landau energy of different phases is calculated using the simplified expression obtained in step S4.21 and the initialization coefficients obtained in step S2.24. The phase with the minimum energy is the actual phase fitted at that temperature. At the same time, the polarization intensity corresponding to the minimum energy is fitted, and the dielectric constant is fitted based on the polarization intensity corresponding to the minimum energy.

[0076] S4.26. Compare the polarization intensity and dielectric constant obtained from step S2.25 with the optimized polarization intensity and dielectric constant obtained from step S3, and then calculate the root mean square error, the expression of which is: Where P and ε represent the polarization and dielectric constant obtained by fitting in step S2.25, respectively; P0 and ε0 represent the optimized polarization intensity and optimized dielectric constant, respectively; N is the number of data points; A, B, and C are weighting coefficients; phase_fit(yes=0, no=1) indicates that the phase is satisfied.

[0077] S4.27. Iterate the root mean square error obtained in step S4.26 until the root mean square error is reduced to the expected value. The corresponding fitted polarization intensity and dielectric constant are the Landau parameter values.

[0078] S5. Using the least squares method, with the doping concentration as the independent variable, fit the Landau energy coefficient obtained in step S4 in the form of a first or second quadratic function to obtain the Landau energy coefficient as a function of the doping concentration.

[0079] S6. To further reduce the error function and make the Landau energy coefficient function more accurate, the function of doping concentration obtained in step S5 is optimized using Bayesian optimization to obtain the optimized function of doping concentration; taking BaTiO3 as an example, as follows... Figure 2 As shown, a linear graph of the function of the optimized doping concentration is presented;

[0080] S7. Use VASP software to perform DFT calculations to obtain the coupling energy, elastic energy, and gradient energy of the new material to be designed.

[0081] S8. Based on the energies obtained in step S7, the input parameters of the material phase field are obtained by fitting. The input parameters of the material phase field include the elastic coefficient tensor c. ijkl Gradient coefficient tensor G ijkl1 and the electrostriction tensor Q ijkl1 Its specific expression is as follows:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] Among them, P i It is polarization, Ω C It is the volume of the unit cell, e is the elementary charge, and Z is the volume of the unit cell. j It is the Born effective charge tensor, ω j It is the weighting coefficient, μ iRepresents atomic displacement, ε ij This indicates the total strain. It is the strain generated by electrostriction; in the model, let equal to ε ij (Without external strain and under free relaxation), the electrostriction coefficient Q can be obtained. ijkl γ is the domain wall energy, E DW and E bulk f represents the total energy of the supercell with and without domain walls, respectively, and S represents the area; grad Let f represent the gradient energy, and in the model let f grad The gradient coefficient G can be obtained by assuming the domain wall energy (where the polarization is uniform within a single domain) is equal to the domain wall energy. ijkl ;

[0088] S9. Based on the optimized function of doping concentration and the input parameters of the material phase field, perform phase field simulation of the material to obtain the physical properties of the new material to be designed under the target conditions.

[0089] S10. Based on the physical properties obtained in step S8, design a new material with those physical properties.

[0090] The present invention employs a phase-field method parameter calculation method that combines artificial intelligence and multi-scale computation. This method obtains the input parameters of the phase-field method through computation, which is more efficient and faster than the traditional method of obtaining parameters through experiments. The present invention combines multiple computational methods to realize multi-scale simulation at the microscopic, lattice, mesoscopic, and macroscopic levels. It also uses artificial intelligence to connect the parameters of different computational methods at different scales, realizing the conversion between different computational methods. Compared with a single computational method, the calculation results are more reliable and accurate. In the parameter fitting process, the present invention uses artificial intelligence to repeatedly fit the parameters and uses least squares method and Bayesian optimization to optimize the fitting results, further reducing the calculation error and thus ensuring the accuracy and reliability of the calculation.

Claims

1. A method for calculating phase field method parameters combining artificial intelligence and multiscale computation, characterized in that: The method comprises the following steps: S1, obtaining the crystal structure, lattice constant and potential function of the new material to be designed, performing high-throughput calculation by using the first principle to obtain the effective Hamilton method input parameters of the material; S2, based on the first principle, taking the strain and polarization soft film of the material as the degrees of freedom, an effective Hamilton model of the material is constructed; S3, according to the parameters obtained in step S1, the effective Hamilton model is simulated by using Monte Carlo simulation; the average structure of the material under different doping concentrations is obtained through simulation, and the material information in a plurality of doping concentrations and a limited temperature range is obtained according to the average structure under different doping concentrations, the material information includes Curie temperature point, polarization strength and dielectric constant; S4, based on the material information obtained in step S3, the material information is optimized by using a neural network to obtain optimized material information; the optimized material information is used as the data to be fitted, and the Landau energy coefficient of different doping concentrations is obtained by fitting using PSO algorithm; S5, using the least square method, taking the doping concentration as the independent variable, the Landau energy coefficient obtained in step S4 is fitted to obtain the function of the Landau energy coefficient about the doping concentration; S6, the function about the doping concentration obtained in step S5 is optimized by using Bayesian optimization method to obtain the optimized function about the doping concentration; S7, using VASP software to perform DFT calculation to obtain the coupling energy, elastic energy and gradient energy of the new material to be designed; S8、based on each energy obtained in step S7, input parameters of a material phase field are obtained by fitting, the input parameters of the material phase field including an elastic coefficient tensor c ijkl , a gradient coefficient tensor , and an electrostriction tensor S9, according to the obtained optimized function about the doping concentration and the input parameters of the material phase field, the phase field method simulation of the material is carried out to obtain the physical performance of the new material to be designed under the target condition; S10, based on the physical performance obtained in step S8, a new material with the physical performance is designed.

2. The method of claim 1, wherein the method is characterized by: The specific process of step S1 comprises the following steps: S1.1, obtaining the crystal structure, lattice constant and potential function of the new material to be designed from the database; S1.2, according to the crystal structure under the equilibrium state, the atomic positions in the crystal structure are perturbed one by one, and the crystal structure is calculated to obtain the polarization and energy of the material; S1.3, the polarization and energy of the material obtained in step S1.2 are fitted by using the least square method to obtain the corresponding relationship between the ferroelectric polarization and the energy, and the input parameters of the effective Hamilton method of the material can be obtained according to the corresponding relationship.

3. The method of claim 2, wherein: In step S4, the specific process of optimizing the material information by using the neural network comprises the following steps: S4.11, taking the polarization intensity calculated in step S3 as the input of the neural network, and taking the Landau energy coefficient as the weight, and using the formula: to calculate the ideal dielectric constant, which is the output of the neural network; wherein ε0 represents the vacuum dielectric constant, ε c represents the relative dielectric constant, F represents the Landau energy, P i and P j both represent polarization components; S4.12, the ideal dielectric constant calculated in step S4.11 and the dielectric constant calculated in step S3 are calculated to obtain the dielectric constant error value, and a dielectric constant error function about the dielectric constant error value is established, the neural network is learned according to the dielectric constant error function until the error value of the dielectric constant error function reaches the minimum, at this time the ideal dielectric constant output by the neural network is the optimized dielectric constant data; S4.13, taking the dielectric constant calculated in step S3 as input of the neural network and the Landau energy coefficient as weight, using the formula: The ideal polarization intensity is calculated, which is the output of the neural network; wherein ε0 represents the vacuum dielectric constant, ε c represents the relative dielectric constant, F represents the Landau energy, P i and P j all represent polarization components; S4.14, calculating the error between the ideal polarization intensity calculated in step S4.13 and the polarization intensity calculated in step S3 to obtain a polarization intensity error value, and establishing a polarization intensity error function about the polarization intensity error value, and performing deep learning on the neural network according to the polarization intensity error function until the error value of the polarization intensity error function reaches a minimum, at which time the ideal polarization intensity output by the neural network is the optimized polarization intensity.

4. The method of claim 3, wherein: In step S4, the specific process of obtaining the Landau energy coefficient of different doping concentrations by fitting using the PSO algorithm includes the following steps: S4.21, according to the corresponding relationship of the polarization component of each phase of the material, using the total polarization intensity P to simplify the Landau energy expression, obtaining the simplified Landau energy expression; the Landau energy expression is: F Landau = α ij P i P j + α ijkl P i P j P k P l ; wherein, α ij and α ijkl represent Landau energy coefficient, P i , P j , P k , P l all represent polarization component; S4.22, determining the coefficients to be fitted according to the simplified Landau energy expression; S4.23, setting the PSO algorithm parameters, including the number of coefficients to be fitted, the number of particles, the minimum value of the coefficients to be fitted, the maximum value of the coefficients to be fitted, and the maximum number of iterations; S4.24, initializing the parameters in step S4.23; S4.25, at each temperature, using the expression obtained by simplifying in step S4.21 and the initialized coefficients obtained in step S2.24 to calculate the minimum value of the Landau energy of different phases, and the phase with the minimum energy is the actual phase obtained by fitting at this temperature, and the polarization intensity corresponding to the minimum energy is obtained by fitting, and the dielectric constant is fitted according to the polarization intensity corresponding to the minimum energy; S4.26, comparing the polarization intensity and dielectric constant obtained by fitting in step S2.25 with the optimized polarization intensity and dielectric constant in step S3, and then calculating the root mean square error, which is expressed as: wherein P, ε represent the polarization and dielectric constant obtained by fitting in step S2.25, P0, ε0 represent the optimized polarization intensity and optimized dielectric constant, N is the number of data points, A, B, C are weight coefficients; phase_fit(yes=0, no=1) represents the phase meets; S4.27, continuously iterating the root mean square error obtained in step S4.26 until the root mean square error is reduced to the expected value, at which time the polarization intensity and dielectric constant obtained by fitting correspond to the Landau parameter value.

Images