Search method, information processing device, and program

Abstract

To improve search for a crystal structure of an inorganic material having desired characteristics.SOLUTION: A search method learns a generation model GM including a decoder DEC for outputting material information with material information on a crystal structure of an inorganic material expressed in a prescribed notation as an input and with an encoder ENC for outputting a latent variable on a latent space LS and the latent variable on the latent space LS as inputs. After that, it specifies a crystal structure of an inorganic material having scores close to prescribed target scores or exceeding the target scores by using an optimization algorithm from the learned generation model.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present disclosure relates to a method for searching for a crystal structure of an inorganic material having desired properties, an information processing apparatus, and a program used for the information processing apparatus.

Background Art

[0002] Conventionally, in the technical field of searching for a material having target properties based on information such as the structure of a material, it has been desired to reduce the number of search times required to discover a material having properties that match the specification of the target.

[0003] In contrast, for example, Patent Document 1 discloses a technique for predicting the target property value of each material point in a search space based on an algorithm of a Gaussian process regression model using the target property value and the feature vector of a known material whose target property value is known as a candidate material. The search space is a space in which points representing each candidate material are distributed in a space of the dimension of the feature vectors corresponding to a plurality of candidate materials.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In recent years, as the required performance of products has increased, it has become essential to improve the performance of the materials themselves of the members constituting the products. However, hitherto, inorganic materials having new properties have been discovered accidentally through experiments and the like, and a great deal of time has been required for their search. Therefore, the present inventors considered using the technique described in Patent Document 1 when searching for the crystal structure of an inorganic material having desired properties.

[0006] However, in the technology described in Patent Document 1, the target characteristic values of known materials and the feature vectors are used as candidate materials. As the dependence on this known information increases, it is considered difficult to search for unknown structures that are not in the known material structures. Thus, there is still room for improvement in the technology for searching the crystal structures of inorganic materials having desired characteristics.

[0007] This disclosure aims to improve the search for crystal structures of inorganic materials having desired characteristics.

Means for Solving the Problems

[0008] The invention according to claim 1 is A method for searching for a crystal structure of an inorganic material, using, as an input, material information regarding the crystal structure of an inorganic material represented by a predetermined notation, learning a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to the material information and a second function that outputs the material information using the latent variable on the latent space as an input, and identifying, from the learned model, using an optimization algorithm, a crystal structure of an inorganic material having a score close to or exceeding a predetermined target score.

[0009] The invention according to claim 6 is An information processing apparatus for searching for a crystal structure of an inorganic material, including a control unit (20) and a storage unit (30), and the control unit learns a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to material information regarding the crystal structure of an inorganic material represented by a predetermined notation and a second function that outputs the material information using the latent variable on the latent space as an input, and identifies, from the learned model, using an optimization algorithm, a crystal structure of an inorganic material having a score close to or exceeding a predetermined target score.

[0010] The invention according to claim 7 is A program used in an information processing apparatus (10) for exploring the crystal structure of inorganic materials, the information processing apparatus is functioned as a model learning unit (21) that learns a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to material information with the material information regarding the crystal structure of an inorganic material represented in a predetermined notation as an input, and a second function that outputs the material information with the latent variable on the latent space as an input, and a structure specifying unit (22) that specifies the crystal structure of an inorganic material having a score close to a predetermined target score or exceeding the target score using an optimization algorithm from the learned model.

[0011] As described above, if the crystal structure of inorganic materials is explored using a model including the first function and the second function, it becomes easier to explore the crystal structure of unknown inorganic materials compared to exploring from a data group composed of known material information.

[0012] Note that the reference numerals in parentheses attached to each component etc. show an example of the correspondence relationship between the component etc. and the specific components etc. described in the embodiments described later.

Brief Description of Drawings

[0013]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

[0014] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following embodiments, parts that are the same as or equivalent to those described in the preceding embodiments may be given the same reference numerals, and the description thereof may be omitted. Further, in the embodiments, when only a part of the components is described, the components described in the preceding embodiments can be applied to other parts of the components. The following embodiments can be partially combined with each other as long as there is no problem in the combination, even if not particularly specified.

[0015] (First Embodiment) This embodiment will be described with reference to FIGS. 1 to 6. In this embodiment, an example in which the crystal structure of an inorganic material is the object of exploration will be described. In this embodiment, the information processing apparatus 10 explores the crystal structure of an inorganic material having desired physical properties.

[0016] The information processing apparatus 10 is configured by at least one of a user terminal, a server, general-purpose or dedicated electronic equipment, or a combination thereof. As shown in FIG. 1, the information processing apparatus 10 includes a control unit 20, a storage unit 30, an input unit 40, and an output unit 50.

[0017] The control unit 20 controls each part of the information processing apparatus 10 and executes processes related to the operation of the information processing apparatus 10. The control unit 20 is constituted by a control circuit including one or more processors. The processor is constituted by a general-purpose processor such as a CPU or a GPU, or a dedicated processor specialized in the search for material structures. Note that the CPU is an abbreviation for Central Processing Unit. The GPU is an abbreviation for Graphics Processing Unit. Note that the control unit 20 may be constituted by a circuit such as an FPGA or an ASIC.

[0018] The storage unit 30 stores data used for the operation of the information processing apparatus 10 and data obtained by the operation of the information processing apparatus 10. The storage unit 30 is constituted by a non-transitory physical storage medium such as a semiconductor memory, a magnetic memory, or an optical memory. Note that the RAM is an abbreviation for Random Access Memory. The ROM is an abbreviation for Read Only Memory.

[0019] In the storage unit 30, various types of information such as experimental data ED used for searching the crystal structure of an inorganic material and a crystal structure database DB are stored as material information regarding the crystal structure of the inorganic material.

[0020] The material information includes the crystal structure itself, a powder diffraction pattern specific to the crystal structure, various physical property values, and the like. The material information regarding the crystal structure of an inorganic material is represented in a predetermined notation. Examples of this notation include a common data format for crystallography called CIF and a notation for crystal characteristics using Fourier transform called FTCP.

[0021] CIF is a data format in which a crystal structure is accommodated. CIF includes information such as atomic coordinates, cell parameters, symmetry, and atomic species of the crystal structure. Note that CIF is an abbreviation for Crystallographic Information File.

[0022] In addition, FTCP uses a notation that combines the real part and the imaginary part. The real part of FTCP consists of the atomic species, atomic occupancy, atomic coordinates, cell parameters, and atomic properties, while the imaginary part is the discrete Fourier transform of the atomic properties along various spatial frequencies from the origin of each atom in reciprocal space. Note that FTCP is an abbreviation for Fourier-Transformed Crystal Properties.

[0023] The input unit 40 is an input interface such as a keyboard or a touch panel, and accepts input operations of data used for the operation of the information processing apparatus 10. Note that the input unit 40 is not essential for the information processing apparatus 10, and may be connected to the information processing apparatus 10 as an external input device, for example. A data file including material information regarding the crystal structure of an inorganic material is input to the input unit 40, for example.

[0024] The output unit 50 is an output interface such as a display, and outputs data obtained by the operation of the information processing apparatus 10. Note that the output unit 50 is not essential for the information processing apparatus 10, and may be connected to the information processing apparatus 10 as an external output device, for example.

[0025] The various functions of the information processing apparatus 10 are realized by executing various programs stored in the storage unit 30 by the control unit 20 of the information processing apparatus 10. For example, by executing a program for searching the crystal structure of an inorganic material by the control unit 20 of the information processing apparatus 10, a search function for the crystal structure of the inorganic material is realized.

[0026] Specifically, the control unit 20 includes functional units for realizing the various functions of the information processing apparatus 10. The control unit 20 of the present embodiment includes a model learning unit 21 and a structure specifying unit 22 as functional units for realizing the various functions of the information processing apparatus 10.

[0027] As shown in FIG. 2, the model learning unit 21 learns a generation model GM including an encoder ENC and a decoder DEC using, as input, material information regarding the crystal structure of an inorganic material represented by a predetermined notation.

[0028] The encoder ENC is a function that reduces the dimensionality of data and converts the feature amounts in the data into latent variables. The encoder ENC of the present embodiment corresponds to a "first function" that outputs, as input, material information regarding the crystal structure of an inorganic material and outputs a latent variable on a latent space LS corresponding to the material information. Further, the decoder DEC is a function that restores the original data from the feature amounts. The decoder DEC corresponds to a "second function" that outputs, as input, a latent variable on the latent space LS and outputs material information. Note that the latent space LS is a space in which latent variables corresponding to the feature amounts in the data are distributed. The latent space LS may be two-dimensional or multi-dimensional with two or more dimensions.

[0029] The generation model GM is a model including the above-described "first function" and "second function". The generation model GM is constructed as an unsupervised machine learning model using a neural network. The generation model GM of the present embodiment is configured as a learning model using a variational autoencoder called VAE. The variational autoencoder is an autoencoder AE having an encoder ENC and a decoder DEC with a probability distribution introduced into the latent variables. In the variational autoencoder, the feature amounts in the input data are used as latent variables in the probability model. In the variational autoencoder, an average vector and a dispersion vector are obtained by the encoder ENC, and a latent variable is sampled from a normal distribution based on the obtained average vector and dispersion vector. Then, in the variational autoencoder, the sampled latent variable is decoded to output the material information as data. Note that VAE is an abbreviation for Variational Autoencoder.

[0030] The structure specifying unit 22 specifies a crystal structure having a score close to or exceeding a target score corresponding to the physical property required for the crystal structure of the inorganic material from the generation model GM that has been learned by the model learning unit 21 using an optimization algorithm. In the structure specifying unit 22 of the present embodiment, a Bayesian optimization algorithm is adopted as the optimization algorithm. The Bayesian optimization algorithm is a type of optimization algorithm that explores the value to be explored next by utilizing uncertainty, and optimizes a black box function using a Gaussian process.

[0031] As shown in FIG. 3, in the structure specifying unit 22, a part of the latent variables on the latent space LS is sampled as candidate points, and the score of the crystal structure specified using the sampled candidate points is obtained. Then, the structure specifying unit 22 repeats the process of comparing the score of the crystal structure specified based on the sampled candidate points with the target score, thereby specifying a desired crystal structure having a score close to or exceeding the target score.

[0032] Next, a series of processes for searching for the crystal structure of the inorganic material using the information processing apparatus 10 will be described with reference to FIG. 4. The control routine shown in FIG. 4 is executed by the control unit 20 of the information processing apparatus 10 when, for example, a command signal instructing the start of the search is input to the information processing apparatus 10.

[0033] As shown in FIG. 4, in step S100, the information processing apparatus 10 performs a process of quantifying the experimental data ED and the material information regarding the crystal structure of the inorganic material in the database DB for learning the generation model GM. In this process, for example, notations such as CIF and FTCP are used to quantify the information regarding the crystal structure of the inorganic material. Note that this process may be a process of quantifying the material information regarding the crystal structure of the inorganic material by the information processing apparatus 10, or may be a process of acquiring the quantified material information from other devices.

[0034] Subsequently, the information processing apparatus 10 performs learning processing of the generation model GM in step S110. In this processing, for example, using the reconstruction error and the Kullback-Leibler divergence (so-called KLD) as the error function, the encoder ENC and the decoder DEC of the generation model GM are trained so that each error is minimized. Note that the reconstruction error is the difference between the data input to the generation model GM and the data reconstructed by the generation model GM. The Kullback-Leibler divergence indicates the distance between the distribution of the data input to the generation model GM and the distribution of the data output from the generation model GM.

[0035] Here, among inorganic materials, assume that materials used for a predetermined purpose are target materials, and materials other than the target materials are non-target materials. Examples of the target materials include materials for redox purposes, materials for hydrogen storage purposes, magnetic materials, materials for weight reduction purposes, electret materials, CO2 reduction materials, conductive materials, materials for heat management purposes, biosensing materials, surface treatment materials, and the like.

[0036] In the learning processing of the generation model GM, the encoder ENC and the decoder DEC are learned using target data related to the target material and non-target data related to the non-target material. The generation model GM learned based on both the target material and the non-target material may suppress a decrease in accuracy due to extrapolation more than the generation model GM learned only based on the target material used for the application of the inorganic material. Therefore, for exploring the crystal structure of unknown inorganic materials, it is preferable to learn the generation model GM based on both the target material and the non-target material.

[0037] Subsequently, at step S120, the information processing apparatus 10 sets a target score corresponding to the property information required for the crystal structure of the inorganic material and a set number that is the upper limit of the number of crystal structure search times. The target score and the number of search times may be set to values pre-set in the storage unit 30, or may be set to values input by the input unit 40. The target score is set, for example, as the target property such as the stability of the crystal structure of the inorganic material such as formation energy. The target score may be set as a score to be converged, or may be set as a score to be exceeded.

[0038] Subsequently, at step S130, the information processing apparatus 10 extracts candidate points of latent variables sampled from the latent space LS using the Bayesian optimization algorithm from the learned generation model GM. Then, at step S140, the information processing apparatus 10 samples candidate points of the latent variables from the latent space LS.

[0039] Subsequently, at step S150, the information processing apparatus 10 outputs material information of the crystal structure of the inorganic material corresponding to the latent variables sampled from the latent space LS from the decoder DEC, and obtains the score of the output material information. The score of the material information is obtained, for example, by calculation processing or numerical simulation for quantifying various characteristics of the material information. Note that the score of the material information may be obtained using a machine learning model such as a surrogate model, or may be obtained from outside as a value digitized by a personal operation.

[0040] Subsequently, at step S160, the information processing apparatus 10 compares the score of the crystal structure sampled from the latent space LS with the target score, and determines the suitability of the score of the sampled crystal structure. Specifically, the information processing apparatus 10 determines whether the difference between the score of the crystal structure sampled from the latent space LS and the target score is within a preset range or the score exceeds the target score, or whether the number of search times is the set number.

[0041] If the difference between the score of the crystal structure sampled from the latent space LS and the target score is outside the preset range or the score is less than or equal to the target score, and the number of search times is less than the set number of times, the information processing apparatus 10 returns to step S130. That is, the information processing apparatus 10 extracts candidate points of latent variables to be sampled next from the latent space LS by the Bayesian optimization algorithm based on the generation model GM.

[0042] On the other hand, if the difference between the score of the sampled crystal structure and the target score is within the preset range or the score exceeds the target score, or the number of search times is equal to the set number of times, the information processing apparatus 10 proceeds to step S170. In step S170, the information processing apparatus 10 outputs the search result assuming that the search for the crystal structure is completed.

[0043] The output of the search result is performed via the output unit 50 connected to the control unit 20. The search result is, for example, as shown in FIG. 5, a two-dimensional image in which the latent variable sampled from the latent space LS and the score corresponding to the latent variable are superimposed on the latent space LS, and is displayed on the display constituting the output unit 50. In this way, it is desirable that the search result is provided in a visually graspable manner. Note that the output of the search result is not limited to a two-dimensional image, and may be a multi-dimensional image or the like.

[0044] The search technique for the crystal structure of the inorganic material according to the present embodiment described above includes a search method for the crystal structure of the inorganic material, an information processing apparatus 10 for searching for the crystal structure of the inorganic material, and a program for searching for the crystal structure of the inorganic material. The search technique of the present case has the following features. That is, the search technique of the present case learns a generation model GM including an encoder ENC that outputs latent variables on the latent space LS using material information related to the crystal structure of the inorganic material as an input, and a decoder DEC that outputs material information using latent variables on the latent space LS as an input. Then, the search technique of the present case specifies a crystal structure of an inorganic material having a score close to or exceeding a predetermined target score from the learned generation model GM using an optimization algorithm.

[0045] Thus, if a generative model GM including an encoder ENC and a decoder DEC is used to search for the crystal structure of an inorganic material, it becomes easier to search for the crystal structure of an unknown inorganic material than searching from a data group composed of known material information.

[0046] Here, in real space, when considering a ternary compound Ax, By, Cz, a space with a dimensionality of 10 6 or more needs to be optimized. Such optimization is very difficult and not realistic.

[0047] On the other hand, in the present invention, materials are represented in a latent space LS with reduced dimensionality, and multiple parameters can be optimized with a small number of dimensions, so there is an advantage that it is easy to search for the crystal structure of an unknown inorganic material.

[0048] Further, according to the present embodiment, the following effects can be obtained.

[0049] (1) When learning the generative model GM, the encoder ENC and the decoder DEC are learned using target data related to the target material and non-target data related to the non-target material. The generative model GM learned based on both the target material and the non-target material may have less degradation in accuracy due to extrapolation than the generative model GM learned only based on the target material used for the application of the inorganic material. Therefore, for searching for the crystal structure of an unknown inorganic material, it is preferable to learn the generative model GM based on the target material and the non-target material.

[0050] (2) In the present embodiment, a variational autoencoder is used as the generative model GM. Thus, if the generative model GM is configured with a variational autoencoder, there is an advantage that it becomes easier to identify an effective crystal structure from the latent space LS because the latent variable sampled from the normal distribution is decoded.

[0051] (3) The optimization algorithm used for searching crystal materials of inorganic materials is the Bayesian optimization algorithm. When identifying the crystal structure of inorganic materials, a part of the latent variables on the latent space LS is sampled as candidate points. Then, the score of the crystal structure of the inorganic material identified using the sampled candidate points is obtained, and the process of comparing the obtained score with the target score is repeated to identify the crystal structure of the inorganic material having a score close to or exceeding the target score.

[0052] In this way, if the latent variables are sampled according to the Bayesian optimization algorithm, the number of trials until the crystal structure having a score close to or exceeding the target score is identified can be reduced compared to the case where the latent variables are sampled randomly.

[0053] Here, FIG. 6 shows the relationship between the number of searches for the crystal structure of the inorganic material and the search results in the information processing apparatus 10. Specifically, FIG. 6 shows the search results when searching for the crystal structure of the inorganic material with a small formation energy. As shown in FIG. 6, in the information processing apparatus 10 of the present invention, a result was obtained in which the crystal structure of the inorganic material predicted to have a small formation energy could be identified with a smaller number of searches compared to the case where the latent variables were sampled randomly.

[0054] (Second Embodiment) Next, the second embodiment will be described with reference to FIGS. 7 and 8. In this embodiment, the points where ○○ is being done are different from those in the first embodiment. In this embodiment, the parts different from the first embodiment will be mainly described.

[0055] As shown in FIG. 7, the generation model GM of this embodiment is composed of a learning model using a conditional variational autoencoder called CVAE. The conditional variational autoencoder is obtained by adding a condition vector as a conditional variable to the variational autoencoder for learning. In the conditional variational autoencoder of this embodiment, in addition to the material information regarding the crystal structure of the inorganic material, a part of the characteristics of the crystal structure of the inorganic material is given as a condition vector to the variational autoencoder for learning.

[0056] Here, in a three-dimensional lattice structure, it is classified into seven crystal systems such as rhombohedral, cubic, hexagonal, monoclinic, orthorhombic, tetragonal, and triclinic by the cell parameters representing the size of the unit cell. In this embodiment, among the information included in the CIF, the cell parameters are used as the condition vector. Note that the size of the unit cell is specified by the lengths of the lattice sides (for example, x, y, z) and the angles formed by the lattice sides (for example, α, β, γ).

[0057] The control unit 20 of this embodiment learns a generation model GM including an encoder ENC and a decoder DEC by using, as inputs, the material information regarding the crystal structure of the inorganic material expressed in a predetermined notation and the condition vector.

[0058] Further, the control unit 20 uses an optimization algorithm from the generation model GM that has been learned by the model learning unit 21 to identify a crystal structure having a score close to or exceeding the target score corresponding to the physical property required for the crystal structure of the inorganic material.

[0059] As shown in FIG. 8, the control unit 20 samples a part of the latent variables on the latent space LS with the condition vector added as candidate points, and obtains the score of the crystal structure specified using the sampled candidate points. Then, the structure identification unit 22 repeats the process of comparing the score of the crystal structure identified based on the sampled candidate points with the target score to identify a desired crystal structure having a score close to or exceeding the target score.

[0060] For the rest, it is the same as the first embodiment. The exploration technique for the crystal structure of the inorganic material in this embodiment can obtain the same effects as those achieved from the common configuration or equivalent configuration as the first embodiment.

[0061] Also, the exploration technique for the crystal structure of the inorganic material in this embodiment includes the following features.

[0062] (1) In this embodiment, a conditional variational autoencoder is used as the generative model GM. In this way, by configuring the generative model GM with a conditional variational autoencoder, it is possible to impose constraints on the output from the decoder DEC, so there is an advantage that it becomes easier to identify an effective crystal structure even from a sparse latent space LS.

[0063] (Modification of the second embodiment) In the second embodiment, the cell parameters are exemplified as the condition vector, but the condition vector is not limited to this. The condition vector may be other information such as the atomic coordinates, symmetry, atomic species, etc. of the crystal structure instead of the cell parameters.

[0064] (Other embodiments) As described above, the representative embodiments of the present disclosure have been explained. However, the present disclosure is not limited to the above-described embodiments and can be variously modified, for example, as follows.

[0065] As in the above-described embodiment, when learning the generative model GM, it is desirable to use target data related to the target material and non-target data related to the non-target material, but it is not limited to this. The learning process of the generative model GM may be such that, for example, only the target data related to the target material is used.

[0066] As in the above-described embodiments, it is desirable that the generative model GM be composed of a variational autoencoder or a conditional variational autoencoder, but it is not limited thereto. The generative model GM may be composed of other models such as, for example, an autoencoder AE, a transformer, etc. Further, in the present embodiment, an example of searching for the crystal structure of an inorganic material using a generative model GM including an encoder ENC and a decoder DEC has been described, but the technique for searching for the crystal structure of an inorganic material is not limited thereto. The crystal structure of the inorganic material may be searched using a "model" including a "first function" and a "second function", which is different from the above-described generative model.

[0067] In the above-described embodiments, the Bayesian optimization algorithm has been exemplified as the optimization algorithm used for searching for the crystal material of the inorganic material, but it is not limited thereto. The optimization algorithm may be other algorithms such as, for example, the gradient descent method.

[0068] In the above-described embodiments, it goes without saying that the elements constituting the embodiments are not necessarily essential, except in cases where it is explicitly stated that they are particularly essential and cases where they are considered to be clearly essential in principle.

[0069] In the above-described embodiments, when numerical values such as the number, numerical value, quantity, range, etc. of the components of the embodiments are mentioned, they are not limited to that specific number, except in cases where it is explicitly stated that they are particularly essential and cases where they are clearly limited to a specific number in principle.

[0070] In the above-described embodiments, when referring to the shape, positional relationship, etc. of the components, etc., they are not limited to that shape, positional relationship, etc., except in cases where it is explicitly stated and cases where they are clearly limited to a specific shape, positional relationship, etc. in principle.

[0071] The control unit and its method of the present disclosure may be implemented by a dedicated computer provided by configuring a processor and a memory programmed to execute one or more functions embodied by a computer program. The control unit and its method of the present disclosure may be implemented by a dedicated computer provided by configuring a processor with one or more dedicated hardware logic circuits. The control unit and its method of the present disclosure may be implemented by one or more dedicated computers configured by a combination of a processor programmed to execute one or more functions and a memory and a processor configured by one or more hardware logic circuits. Further, the computer program may be stored in a computer-readable non-transitory tangible recording medium as instructions executed by a computer.

[0072] [Aspect of the present disclosure]

[0073] [First aspect] A method for searching for a crystal structure of an inorganic material, learning a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to the material information by using, as an input, material information regarding the crystal structure of the inorganic material represented by a predetermined notation, and a second function that outputs the material information by using, as an input, the latent variable on the latent space; identifying, from the learned model, a crystal structure of the inorganic material having a score close to or exceeding a predetermined target score by using an optimization algorithm.

[0074] [Second aspect] Among the inorganic materials, when a material used for a predetermined purpose is a target material and materials other than the target material are non-target materials, in learning the model, the first function and the second function are learned by using target data regarding the target material and non-target data regarding the non-target material, the search method according to the first aspect.

[0075] [Third perspective] The model is the search method described in the first or second perspective, in which a variational autoencoder is used.

[0076] [Fourth perspective] The model is the search method described in the first or second perspective, in which a conditional variational autoencoder is used.

[0077] [Fifth perspective] The optimization algorithm is a Bayesian optimization algorithm, In identifying the crystal structure of the inorganic material, a part of the latent variables on the latent space is sampled as candidate points, and the score of the crystal structure of the inorganic material identified using the sampled candidate points is obtained, and the obtained score is compared with the target score, and the process of repeating this comparison is performed to identify the crystal structure of the inorganic material having a score close to or exceeding the target score, according to any one of the first to fourth perspectives of the search method described.

[0078] [Sixth perspective] An information processing apparatus for searching for a crystal structure of an inorganic material, a control unit (20), a storage unit (30), and includes, The control unit learns a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to the material information with the material information regarding the crystal structure of the inorganic material represented by a predetermined notation as an input, and a second function that outputs the material information with the latent variable on the latent space as an input, and uses an optimization algorithm from the learned model to identify the crystal structure of the inorganic material having a score close to or exceeding a predetermined target score, the information processing apparatus.

[0079] [Seventh perspective] A program for use in an information processing apparatus (10) for searching for a crystal structure of an inorganic material, the information processing apparatus, A program that functions as a model learning unit (21) for learning a model (GM) including a first function that outputs a latent variable on a latent space (LS) corresponding to the material information using, as input, material information regarding the crystal structure of the inorganic material represented by a predetermined notation method, and a second function that outputs the material information using, as input, the latent variable on the latent space, and a structure specifying unit (22) that specifies the crystal structure of the inorganic material having a score close to or exceeding a predetermined target score using an optimization algorithm from the learned model.

Explanation of Signs

[0080] 10 Information processing apparatus 20 Control unit 30 Storage unit ENC Encoder (first function) DEC Decoder (second function) LS Latent space GM Generative model (model)

Claims

1. A method for exploring the crystal structure of an inorganic material, comprising: training a model (GM) including a first function that receives material information on a crystal structure of the inorganic material expressed in a predetermined notation as an input and outputs latent variables on a latent space (LS) corresponding to the material information, and a second function that receives the latent variables on the latent space as an input and outputs the material information; and identifying a crystalline structure of the inorganic material having a score close to or exceeding a predetermined target score using an optimization algorithm from the trained model; Among the inorganic materials, those used for a specific purpose are defined as target materials, and those other than the target materials are defined as non-target materials. A method of searching, wherein training the model involves training the first function and the second function using target data related to the target material and non-target data related to the non-target material.

2. The searching method described in claim 1, wherein the target material is one of a material for the purpose of oxidation-reduction, a material for the purpose of hydrogen storage, a magnetic material, a material for the purpose of weight reduction, an electret material, a CO2 reduction material, a conductive material, a material for the purpose of thermal management, a biosensing material, and a surface treatment material.

3. The search method according to claim 1 or 2, wherein the model uses a variational autoencoder.

4. The search method according to claim 1 or 2, wherein the model uses a conditional variational autoencoder.

5. the optimization algorithm is a Bayesian optimization algorithm; 3. The searching method according to claim 1, wherein the step of identifying the crystalline structure of the inorganic material includes sampling some of the latent variables in the latent space as candidate points, calculating the score of the crystalline structure of the inorganic material identified using the sampled candidate points, and repeating a process of comparing the calculated score with the target score to identify the crystalline structure of the inorganic material having the score close to or exceeding the target score.

6. An information processing device for searching for a crystal structure of an inorganic material, A control unit (20); A storage unit (30), the control unit learns a model (GM) including a first function that receives material information on a crystalline structure of the inorganic material expressed in a predetermined notation as input and outputs a latent variable in a latent space (LS) corresponding to the material information, and a second function that receives the latent variable in the latent space as input and outputs the material information, and identifies a crystalline structure of the inorganic material having a score close to or exceeding a predetermined target score from the learned model using an optimization algorithm; Among the inorganic materials, those used for a specific purpose are defined as target materials, and those other than the target materials are defined as non-target materials. The control unit learns the first function and the second function using target data related to the target material and non-target data related to the non-target material.

7. A program used in an information processing device (10) for searching for a crystal structure of an inorganic material, The information processing device a model learning unit (21) that learns a model (GM) including a first function that receives material information on the crystalline structure of the inorganic material expressed in a predetermined notation and outputs latent variables in a latent space (LS) corresponding to the material information as an input, and a second function that receives the latent variables in the latent space and outputs the material information as an input, and a structure identification unit (22) that uses an optimization algorithm from the learned model to identify a crystalline structure of the inorganic material having a score close to or exceeding a predetermined target score, Among the inorganic materials, those used for a specific purpose are defined as target materials, and those other than the target materials are defined as non-target materials. The model learning unit learns the first function and the second function using target data related to the target material and non-target data related to the non-target material.

Images