Methods for predicting lithium diffusivity

Abstract

To provide a method capable of predicting lithium diffusivity with high precision.SOLUTION: A method for predicting lithium diffusivity causes a computer to execute the steps of: acquiring crystal structure information for a lithium-containing oxide from a structure database; generating X-ray diffraction pattern data for the crystal structure according to the acquired information; generating electron state density data for the crystal structure according to the acquired information; and evaluating lithium diffusivity using machine learning according to the generated X-ray diffraction pattern data and the electron state density data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a method for predicting lithium diffusivity. [Background technology]

[0002] Machine learning techniques are used to predict material properties.

[0003] Non-Patent Document 1 discloses a technology that uses XRD pattern data obtained from crystal structure data as feature quantities and utilizes machine learning techniques to predict the merits and demerits of various material properties with high accuracy. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Advanced Intelligent Systems, Volume4, Issue7, July 2022 Summary of the Invention [Problem to be solved by the invention]

[0005] For example, it is desirable to be able to predict the diffusivity of lithium with high accuracy as a property of lithium-containing oxides used as electrode materials in lithium-ion batteries.

[0006] An object of the present disclosure is to provide a method capable of predicting lithium diffusivity with high accuracy. [Means for solving the problem]

[0007] The present inventors have found that the above problems can be solved by the following means. <Aspect 1> A method for predicting lithium diffusivity, comprising: Obtaining information on the crystal structure of the lithium-containing oxide from a structure database; generating X-ray diffraction pattern data of the crystal structure based on the acquired information; generating electronic density of states data for the crystal structure based on the acquired information; and Using the generated X-ray diffraction pattern data and the generated electron density of states data, the diffusivity of lithium is evaluated by machine learning. [Effects of the Invention]

[0008] According to the present disclosure, a method for predicting lithium diffusivity with high accuracy can be provided. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a flow chart illustrating the disclosed method for predicting lithium diffusivity. [Figure 2] FIG. 2 is a graph showing the relationship between the number of crystal structure files and the logarithm of the lithium diffusion coefficient, obtained based on molecular dynamics calculations. [Figure 3] FIG. 3 is a graph showing the accuracy rate of the lithium diffusivity prediction at each threshold value. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present disclosure will be described in detail. Note that the present disclosure is not limited to the following embodiments, and various modifications can be made within the scope of the present disclosure.

[0011] <<Method for predicting lithium diffusivity>> The disclosed method for predicting lithium diffusivity causes a computer to execute processing including the following steps: acquiring information on the crystal structure of a lithium-containing oxide from a structure database, generating X-ray diffraction pattern data of the crystal structure based on the acquired information, generating electron density of states data of the crystal structure based on the acquired information, and evaluating the lithium diffusivity by machine learning using the generated X-ray diffraction pattern data and electron density of states data.

[0012] The present inventors have found that when the diffusivity of lithium in the crystal structure of a lithium-containing oxide is predicted by machine learning using only X-ray diffraction pattern data as a feature, the prediction accuracy is low.

[0013] In this regard, the present inventors have discovered that by using electron density of states data as a feature in addition to X-ray diffraction pattern data in the crystal structure of a lithium-containing oxide, it is possible to predict the diffusivity of lithium with high accuracy through machine learning. Without intending to be bound by any theory, the reason for this is presumed to be as follows. That is, it is believed that the diffusivity of lithium depends not only on the positional relationship of atomic arrangements and the like in the crystal structure, but also on the electronic state. Therefore, it is believed that the accuracy of predicting the diffusivity of lithium can be improved by adding electron density of states data as information expressing the diffusibility of lithium.

[0014] The method of the present disclosure for predicting lithium diffusivity will be described below with reference to the flowchart shown in Figure 1. Note that Figure 1 is a flowchart illustrating the method of the present disclosure for predicting lithium diffusivity.

[0015] As shown in FIG. 1, in step 101, the computer executes a process of acquiring information on the crystal structure of the lithium-containing oxide from a structure database.

[0016] Examples of structural databases include, but are not limited to, public databases on the Internet such as the Crystallography Open Database and the Materials Project.

[0017] The method for obtaining the crystal structure information is not particularly limited, and an example is a method of downloading the desired information from the above-mentioned structure database.

[0018] 1, in step 102, the computer executes a process of generating X-ray diffraction (XRD) pattern data of the crystal structure based on the acquired information. The method of generating the XRD pattern data is not particularly limited, but an example is a method of calculating XRD pattern data (horizontal axis: 2θ, vertical axis: intensity) from a crystal structure file using a function such as pymatgen, an open source Python library.

[0019] 1, in step 103, the computer executes a process of generating density of states (DOS) data of the crystal structure based on the acquired information. The method of generating DOS data is not particularly limited, but an example is a method of calculating DOS data (horizontal axis: energy, vertical axis: density of states) from a crystal structure file using first-principles calculation software such as VASP.

[0020] It should be noted that, in the context of this disclosure, "electronic density of states (DOS)" refers to the number of eigenstates per unit magnitude of energy. The electronic density of states is also called the density of states.

[0021] The computer may perform principal component analysis on the generated XRD patterns and DOS data to obtain principal components for each data. The principal components obtained by the principal component analysis for each data are used as features to perform the process of step 104 described below, thereby enabling more accurate prediction of lithium diffusivity. In the context of the present disclosure, "principal component analysis" refers to a type of "unsupervised learning" in machine learning, a statistical analysis method that aggregates data with multiple variables to create principal components.

[0022] As shown in FIG. 1, in step 104, the computer executes a process for evaluating the diffusivity of lithium by machine learning using the generated XRD pattern data and DOS data.

[0023] The computer may execute the above process using, as features, principal components obtained by principal component analysis performed on the XRD pattern data and the DOS data. Feature quantities such as electronegativity may also be used in combination.

[0024] An example of a method for evaluating lithium diffusivity using machine learning is a method for classifying lithium diffusivity using machine learning, particularly a method for classifying lithium diffusivity as either high or low. A method for classifying lithium diffusivity is based on "supervised learning," a type of machine learning, particularly a method that uses the machine learning algorithm SVM (support vector machine), which is a function of the Python open source library scikit-learn. "SVM" is one type of "supervised learning" in machine learning, and refers to a method of learning correct data and classifying unknown data based on the results.

[0025] Lithium-containing oxides can be used, for example, as electrode active materials for lithium-ion batteries. Therefore, the method of the present disclosure is particularly effective when selecting materials that can be used as electrode active materials for lithium-ion batteries. For example, the method of the present disclosure can be used to examine the applicability of lithium-containing oxides predicted to have high lithium diffusivity as electrode active materials for lithium-ion batteries. Furthermore, the method of the present disclosure can be used to examine the applicability of materials having a crystal structure similar to that of lithium-containing oxides predicted to have high lithium diffusivity as electrode active materials for lithium-ion batteries. [Example]

[0026] Example Using the XRD pattern data and the main components of the DOS data in the crystal structure of a given lithium-containing oxide as features, the machine learning algorithm SVM, which is a function of scikit-learn, an open source Python library, was used to classify the lithium diffusivity as either large or small.

[0027] Comparative Example The lithium diffusivity was classified into either large or small in the same manner as in the Examples, except that only the main component of the XRD pattern data was used as the feature amount.

[0028] "evaluation" In order to verify the prediction accuracy of the diffusivity of lithium (Li), the following simulation was carried out. First, a molecular dynamics calculation was carried out on a computer for the lithium-containing oxide of the example under the following calculation conditions: Calculation time: Increase by 1 fsec up to 100 psec Calculated temperature...723K.

[0029] This gives the diffusion coefficient of lithium (cm 2 / sec) was calculated. Furthermore, the logarithm of the lithium diffusion coefficient was calculated, and a graph showing the relationship between the number of crystal structure files and the logarithm of the calculated lithium diffusion coefficient (log(Li diffusion coefficient)) was obtained (Figure 2).

[0030] In the obtained graph, a threshold value was set to classify the magnitude of lithium diffusivity. That is, for example, as shown in Figure 2, the threshold value was set to log(Li diffusion coefficient) = -7, and when log(Li diffusion coefficient) was greater than the threshold value, the lithium diffusivity was classified as large, and when log(Li diffusion coefficient) was smaller than the threshold value, the lithium diffusivity was classified as small. Furthermore, the threshold values ​​were set to -8 and -9, and the lithium diffusivity was similarly classified.

[0031] For each threshold, the accuracy of the lithium diffusivity prediction was evaluated to see whether the magnitude of lithium diffusivity classified in the Examples and Comparative Examples coincided with the magnitude of lithium diffusivity classified based on the threshold in the graph obtained by molecular dynamics calculation. A graph showing the accuracy of the lithium diffusivity prediction for each threshold is shown in Figure 3.

[0032] As shown in FIG. 3, the accuracy rate of the lithium diffusibility prediction in the Examples was higher than the accuracy rate of the lithium diffusibility prediction in the Comparative Examples at all threshold values.

Claims

[Claim 1] 1. A method for predicting lithium diffusivity, comprising: causing a computer to execute a process comprising the steps of: Obtaining information on the crystal structure of the lithium-containing oxide from a structure database; generating X-ray diffraction pattern data of the crystal structure based on the acquired information; generating electronic density of states data for the crystal structure based on the acquired information; and Using the generated X-ray diffraction pattern data and the generated electron density of states data, the diffusibility of lithium is evaluated by machine learning.

Images