Data processing method, data processing device and data processing system

Abstract

The data processing method provided by the present invention obtains, for each first compound among multiple first compounds whose true values ​​of physical property values ​​are known, a first calculated value as a physical property value by a first calculation method, obtains a second calculated value as a physical property value by a second calculation method, and generates a first correction model for correcting the first calculated value to a true value and a second correction model for correcting the second calculated value to a true value. For a second compound whose true value of the physical property value is unknown, a value obtained by correcting the calculated value obtained as a physical property value by the first calculation method in an area at least including an area where a calculation result can be obtained by the first calculation method by the first correction model is used as a true value, and a value obtained by correcting the calculated value obtained as a physical property value of the second compound by the second calculation method in an area at least including an area where a calculation result cannot be obtained by the first calculation method by the second correction model is stored in a database as a true value.

Description

Technical Field

[0001] The invention relates to a data processing method, a data processing device and a data processing system. Background Art

[0002] In recent years, materials informatics, which combines materials science and information science, has attracted attention as a method for searching for new materials. Materials informatics searches for materials by analyzing databases that record various information such as the structure and properties of materials (compounds).

[0003] In materials informatics, material searches are performed by data mining and machine learning using databases (for example, see Non-Patent Document 1).

[0004] Non-patent document 1: Joanne Hill, et al., "Materials science with large-scaledata and informatics: Unlocking new opportunities", MRS Bulletin, Volume 41, Issue 5 (Nucleation in Atomic, Molecular, and Colloidal Systems) May 2016, pp.399-409

[0005] Here, theoretically, the number of compounds that may become search targets for new compounds is enormous. However, currently, the number of compounds for which physical property values ​​are registered in the database is insufficient relative to the number of compounds that may become search targets. Summary of the invention

[0006] An object of the present invention is to provide a data processing method, a data processing device and a data processing system capable of increasing the amount of information in a database used for searching for new materials.

[0007] In order to solve the above-mentioned problems and achieve the purpose, the present invention is a data processing method, comprising: a first acquisition step, for each of a plurality of first compounds whose true value of the physical property value is known, obtaining a first calculated value as the above-mentioned physical property value by a first calculation method; a second acquisition step, for each of the plurality of first compounds, obtaining a second calculated value as the above-mentioned physical property value by a second calculation method, wherein the above-mentioned second calculation method can obtain a calculation result in an area where the calculation result cannot be obtained by the above-mentioned first calculation method; a generation step, generating a first correction model for correcting the above-mentioned first calculated value to the above-mentioned true value and a second correction model for correcting the above-mentioned second calculated value to the above-mentioned true value; a third acquisition step, for the second correction model for correcting the above-mentioned second calculated value to the above-mentioned true value of the physical property value a first storage step, correcting the third calculated value obtained in an area including at least an area where the calculation result can be obtained by the first calculation method, and storing the corrected value in a database as the true value of the physical property value of the second compound; a fourth acquisition step, obtaining a fourth calculated value as the physical property value of the second compound by the second calculation method in an area including at least an area where the calculation result cannot be obtained by the first calculation method; and a second storage step, correcting the fourth calculated value by the second correction model, and storing the corrected value in the database as the true value of the physical property value of the second compound.

[0008] In addition, the present invention is a data processing device comprising: an acquisition unit, for each of a plurality of first compounds whose true values ​​of physical property values ​​are known, acquiring a first calculated value as the above-mentioned physical property value by a first calculation method, and for each of the plurality of first compounds, acquiring a second calculated value as the above-mentioned physical property value by a second calculation method that can obtain a calculation result in an area where the calculation result cannot be obtained by the above-mentioned first calculation method; a generation unit, for generating a first correction model for correcting the above-mentioned first calculated value to the above-mentioned true value and a second correction model for correcting the above-mentioned second calculated value to the above-mentioned true value; and a storage unit for correcting the third calculated value by the above-mentioned first correction model, and storing the corrected value as the above-mentioned second compound. The true value of the above-mentioned physical property value of the object is stored in the database, the fourth calculated value is corrected by the above-mentioned second correction model, and the corrected value is stored in the above-mentioned database as the true value of the physical property value of the above-mentioned second compound, wherein the above-mentioned third calculated value is the third calculated value obtained as the physical property value by the above-mentioned acquisition unit through the above-mentioned first calculation method for the second compound whose true value of the physical property value is unknown, and the above-mentioned third calculated value is a value obtained in an area including at least an area where the calculation result can be obtained by the above-mentioned first calculation method, and the above-mentioned fourth calculated value is a value obtained by the above-mentioned acquisition unit through the above-mentioned second calculation method as the physical property value of the above-mentioned second compound in an area including at least an area where the calculation result cannot be obtained by the above-mentioned first calculation method.

[0009] In addition, the present invention is a data processing system comprising: an acquisition unit, for each of a plurality of first compounds whose true values ​​of physical property values ​​are known, acquiring a first calculated value as the above-mentioned physical property value by a first calculation method, and for each of the plurality of first compounds, acquiring a second calculated value as the above-mentioned physical property value by a second calculation method that can obtain a calculation result in an area where the calculation result cannot be obtained by the above-mentioned first calculation method; a generation unit, generating a first correction model for correcting the above-mentioned first calculated value to the above-mentioned true value and a second correction model for correcting the above-mentioned second calculated value to the above-mentioned true value; and a storage unit correcting the third calculated value by the above-mentioned first correction model, and storing the corrected value as the above-mentioned second compound. The true value of the above-mentioned physical property value of the object is stored in the database, the fourth calculated value is corrected by the above-mentioned second correction model, and the corrected value is stored in the above-mentioned database as the true value of the physical property value of the above-mentioned second compound, wherein the above-mentioned third calculated value is the third calculated value obtained as the physical property value by the above-mentioned acquisition unit through the above-mentioned first calculation method for the second compound whose true value of the physical property value is unknown, and the above-mentioned third calculated value is a value obtained in an area including at least an area where the calculation result can be obtained by the above-mentioned first calculation method, and the above-mentioned fourth calculated value is a value obtained by the above-mentioned acquisition unit through the above-mentioned second calculation method as the physical property value of the above-mentioned second compound in an area including at least an area where the calculation result cannot be obtained by the above-mentioned first calculation method.

[0010] According to the present invention, the amount of information in a database used for new material search can be increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a diagram showing an example of a schematic configuration of a data processing system according to an embodiment.

[0012] Figure 2 This is a diagram showing an example of the hardware configuration of a data processing device according to an embodiment.

[0013] Figure 3 This is a diagram showing an example of the functions of the data processing device according to the embodiment.

[0014] Figure 4 It is used for Figure 3 A diagram for explaining a storage unit shown in FIG.

[0015] Figure 5 This is a flowchart showing an example of the operation of the data processing system according to the embodiment.

[0016] Figure 6 It is a diagram schematically showing the trends of DFT calculation results and VQE calculation results.

[0017] Figure 7 Yes means Figure 5 The flowchart shown is a flowchart of the details of step S1.

[0018] Figure 8 It is used for Figure 7 A diagram illustrating the processing of the acquisition unit and the generation unit in .

[0019] Fig. 9 It is used for Figure 7 A diagram illustrating the processing of the generation unit in .

[0020] Fig.10 Yes means Figure 5 1 is a flowchart showing the details of step S2 of the flowchart shown in FIG.

[0021] Fig.11 This is a diagram for explaining the database after the processing in step S2.

[0022] Fig.12 It is a diagram for explaining the processing result of step S2.

[0023] Fig.13 This is a diagram for explaining the search unit.

[0024] Fig.14 Yes means Figure 5 1 is a flowchart showing the details of step S4 of the flowchart shown in FIG. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the data processing method, data processing device, and data processing system of the present invention will be described in detail with reference to the accompanying drawings. Hereinafter, a system that executes the data processing method disclosed in this specification will be described as an embodiment.

[0026] (Implementation Method)

[0027] Figure 1 FIG. 1 is a diagram showing an example of a schematic structure of a data processing system 1 according to an embodiment. Figure 1 As shown, a data processing system 1 according to the embodiment includes a data processing device 10 and a quantum computing device 20 . Figure 1 The devices illustrated in the figure are able to communicate with each other directly or indirectly via a network such as a LAN (Local Area Network) or a WAN (Wide Area Network).

[0028] The data processing device 10 is a von Neumann type computer, which is called a so-called classical computer, compared to a quantum computer, which is a computer that realizes parallelism using superposition of quantum mechanics. Figure 1The data processing device 10 shown is, for example, a workstation capable of executing a calculation method based on density functional theory (DFT).

[0029] The quantum computing device 20 is a computing device implemented by a quantum computer or a quantum annealer. The quantum computing device 20 is not a quantum computer with a complete error correction function, but a NISQ (Noisy Intermediate-Scale Quantum Device) that outputs a computation result containing noise. Figure 1 The quantum computing device 20 shown is capable of executing, for example, a variational quantum eigensolver (VQE).

[0030] Figure 2 FIG. 1 is a diagram showing an example of the hardware configuration of the data processing device 10. Figure 2 As shown, the data processing device 10 includes a CPU (Central Processing Unit) 11 , a ROM (Read Only Memory) 12 , a RAM (Random Access Memory) 13 , an auxiliary storage device 14 , an input device 15 , a display device 16 , and an external I / F 17 .

[0031] The CPU 11 executes programs to control the operation of the data processing device 10 in a unified manner, thereby realizing various functions of the data processing device 10. The various functions of the data processing device 10 will be described later.

[0032] The ROM 12 is a nonvolatile memory that stores various data including a program for starting the data processing device 10 (information written in the manufacturing stage of the data processing device 10). The RAM 13 is a volatile memory having a work area of ​​the CPU 11. The auxiliary storage device 14 stores various data such as a program executed by the CPU 11. The auxiliary storage device 14 is composed of, for example, a HDD (Hard Disc Drive) or the like.

[0033] The input device 15 is a device for a user who uses the data processing device 10 to perform various operations. The input device 15 includes, for example, a mouse, a keyboard, a touch panel, or hardware keys.

[0034] The display device 16 displays various information. For example, the display device 16 displays the processing results of the CPU 11, a GUI (Graphical User Interface) for accepting various operations from the user, etc. The display device 16 is composed of, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or a cathode ray tube display. In addition, for example, the input device 15 and the display device 16 may be integrated in a manner such as a touch panel.

[0035] The external I / F 17 is an interface for connecting (communicating) with an external device such as the quantum computing device 20 .

[0036] Figure 3 is a diagram showing an example of the functions of the data processing device 10. Figure 3 In the example of FIG. 1 , only the functions of the embodiments are illustrated, but the functions of the data processing device 10 are not limited to these structures. Figure 3 As shown, the data processing device 10 includes a user interface unit 101 , a storage unit 102 , an acquisition unit 103 , a generation unit 104 , a storage unit 105 , and a search unit 106 .

[0037] The user interface unit 101 has a function of accepting user input and a function of displaying various information. Figure 2 The input device 15 and the display device 16 shown are implemented.

[0038] The storage unit 102 is composed of, for example, Figure 2 The storage unit 102 stores a database (compound database) corresponding to the chemical characteristics of each compound as data for executing the data processing method of the embodiment. Specifically, the storage unit 102 stores a database corresponding to the numerical value (molecular descriptor) representing the characteristic of the partial structure of the compound, the true value of the physical property value representing the physical property, etc. for each compound.

[0039] Figure 4 It is used for Figure 3 For example, in the storage unit 102, as shown in FIG. Figure 4As shown, a database 102a is stored for each compound, in which a character string (molecular descriptor) representing a chemical structure, a true value of an ionization potential (IP), and a true value of an electron affinity (EA) are registered. In addition, the true value of the physical property value registered in the database 102a may be an experimental value obtained by experiment or a calculated value obtained by high-precision calculation.

[0040] In addition, as data for executing the data processing method of the embodiment, Figure 4 As shown, the storage unit 102 stores the learned model 102b. The learned model 102b is used for processing by the search unit 106 described later. The learned model 102b is generated by machine learning using teacher data. The teacher data used to generate the learned model 102b is, for example, a list of compounds whose true values ​​of ionization potential and electron affinity are known as true values ​​of physical property values ​​of physical property A. For example, the teacher data is database 102a. Alternatively, the teacher data is a database different from database 102a and is a database having the same structure as database 102a. Alternatively, the teacher data is database 102a, and a database different from database 102a and having the same structure as database 102a.

[0041] When the user inputs a property value "AX" (an example of a prescribed characteristic quantity) required for property A, such a learned model 102b infers and outputs a new compound having a property value "AX". For example, property A is ionization potential and electron affinity. For example, if the property value "AX" of ionization potential is input as the property value of property A from the search unit 106 described later, the learned model 102b outputs a character string inferred as a chemical structure having the property value "AX". In addition, if the property value "AX" of electron affinity is input as the property value of property A from the search unit 106 described later, the learned model 102b outputs a character string inferred as a chemical structure having the property value "AX". In addition, when the physical property value "AX1" of the ionization potential and the physical property value "AX2" of the electron affinity are input as the physical property value of the physical property A from the search unit 106 described later, the learned model 102b outputs a character string inferred as a chemical structure having the physical property value "AX1" and the physical property value "AX2". In addition, the learned model 102b can be generated by the data processing device 10 or by other devices.

[0042] In this way, the data processing device 10 uses the above-mentioned database 102a to perform a new material search. Here, in theory, the number of compounds that may become the search object of a new compound is very large. However, the number of compounds whose physical property values ​​are registered in the database 102a is insufficient relative to the number of compounds that may become the search object. Therefore, in order to increase the amount of information in the database 102a used for new material search, the data processing device 10 cooperates with the quantum computing device 20 to perform the data processing method described below. Figure 5 1 is a flowchart showing an example of the operation of the data processing system according to the embodiment.

[0043] (Step S1)

[0044] First, the data processing device 10 generates a correction model (step S1). Step S1 is Figure 3 The processing is performed by the acquisition unit 103 and the generation unit 104 shown.

[0045] In step S1, the acquisition unit 103 acquires a first calculated value as a physical property value for each of a plurality of first compounds whose true value of the physical property value is known by a first calculation method (first acquisition step). In an embodiment, the first calculation method is the above-mentioned DFT, which is executed by the DFT calculation unit 103a of the acquisition unit 103. In addition, in step S1, the acquisition unit 103 acquires a second calculated value as a physical property value for each of the plurality of first compounds by a second calculation method (second acquisition step). In an embodiment, the second calculation method is the above-mentioned VQE, which is executed by sending a calculation command to the quantum computing device 20 by the VQE calculation command unit 103b of the acquisition unit 103. Moreover, in step S1, the generation unit 104 generates a first correction model for correcting the first calculated value to a true value and a second correction model for correcting the second calculated value to a true value (generation step).

[0046] Here, VQE calculation, which is an example of the second calculation method, is a calculation method that can obtain a calculation result in an area where a calculation result cannot be obtained by the first calculation method (DFT). Figure 6 This point will be explained. Figure 6 The calculated value of the ionization potential is plotted on the horizontal axis, and the true value of the ionization potential is plotted on the vertical axis to show the relationship between the calculation result based on DFT and the true value, and the relationship between the calculation result based on VQE and the true value. Figure 6 The relationship shown is an example of the concept of the data processing method for explaining the embodiment, including the actual calculation results of the ionization potential, and it is not intended that the same relationship holds for the calculation results of all physical properties.

[0047] like Figure 6As shown, in the DFT calculation, there is a region 1000 where the calculation result can be obtained by DFT and a region 1100 where the calculation result cannot be obtained by DFT. Region 1000 is a region where the calculation result having correlation with the true value can be obtained by DFT calculation. In other words, region 1000 is a region where the calculation result with qualitative correctness can be obtained by DFT, and a region where the calculation result with appropriateness can be obtained by DFT.

[0048] On the other hand, region 1100 can be divided into region 1200 where invalid calculation results are obtained by DFT and region 1300 where DFT calculation ends abnormally. Region 1200 is a region where calculation results can be obtained by DFT calculation, but the calculation results obtained are less correlated with the true value than region 1000.

[0049] In contrast, in VQE calculation, in principle, there is no region that cannot be calculated, and a qualitatively correct calculation result can be obtained. Here, in VQE calculation performed by NISQ (quantum computing device 20), Figure 6 As shown, a calculated value including a certain amount of noise is obtained statistically. That is, in the VQE calculation, even for a compound included in the region 1100 where the calculation result cannot be obtained by the DFT calculation, a qualitatively correct calculation result can be obtained. However, since the VQE calculation takes a longer calculation time than the DFT calculation, the molecules that can be subjected to the DFT calculation are subjected to data collection by the DFT calculation for high-speed processing.

[0050] Therefore, in the embodiment, in step S1, the generator 104 compares the true value, the first calculated value (DFT calculation result), and the second calculated value (VQE calculation result) for each of the plurality of first compounds. Furthermore, the generator 104 generates the first correction model (DFT correction model) within a range where the first calculated value has a correlation with the true value, and generates the second correction model (VQE correction model) within a range where the second calculated value has a correlation with the true value.

[0051] use Figure 7 to Figure 9 The specific processing of step S1 will be described. Figure 7 Yes means Figure 5 The flowchart shown is a flowchart of the details of step S1, Figure 8 It is used for Figure 7 FIG. 1 is a diagram for explaining the processing of the acquisition unit 103 and the generation unit 104 in FIG. Fig. 9 It is used for Figure 7 A diagram illustrating the processing of the generation unit 104 in FIG.

[0052] like Figure 7As shown, the acquisition unit 103 acquires a list of first compounds whose true values ​​of physical property values ​​are known (step S11). For example, the acquisition unit 103 acquires data from the database 102a stored in the storage unit 102. In addition, the "list of compounds whose true values ​​of ionization potential and electron affinity are known" acquired in step S11 can also be acquired from the outside via the user interface unit 101.

[0053] The DFT calculation unit 103a creates a calculation formula for DFT calculation based on the chemical structure of the first compound, and performs DFT calculation to obtain calculation results (step S12). In addition, in parallel with step S12, the VQE calculation command unit 103b creates a calculation formula for VQE calculation based on the chemical structure of the first compound, and sends the created calculation formula to the quantum computing device 20, so that the quantum computing device 20 performs VQE calculation and obtains calculation results (step S13). In addition, step S12 can be performed before or after step S13 is performed.

[0054] Figure 8 FIG. 2 shows an example in which the acquisition unit 103 acquires the calculation result of the ionization potential. Figure 8 In the upper part of , it is shown that for "compound: CK1, true value: IP_1", the calculation result of DFT is "IPD_1" and the calculation result of VQE is "IPV_1". Figure 8 In the upper part of , it is shown that for "Compound: CK2, True Value: IP_2", the calculation result of DFT is "IPD_2" and the calculation result of VQE is "IPV_2". Figure 8 The upper part of shows that for "compound: CKn, true value: IP_n", the calculation result of VQE is "IPV_n", but the calculation result of DFT cannot be obtained. That is, "compound: CKn" is a compound corresponding to the region 1300 in the DFT calculation.

[0055] Back to Figure 7 , the generator 104 compares the calculation result with the true value (step S14). In step S14, the generator 104 determines whether the calculation result of DFT is correlated with the true value. Similarly, in step S14, the generator 104 determines whether the calculation result of VQE is correlated with the true value. Figure 8 In the lower part of , the generator 104 determines that the DFT calculation result "IPD_2" is a value that has no correlation with the true value. This determination result indicates that "Compound: CK2" is a compound whose calculation result is included in the acquisition region 1200 in the DFT calculation. In addition, Figure 6 In the example shown, all results of the VQE calculations are determined to be correlated with the true value.

[0056] Then, the generator 104 generates a DFT correction model and a VQE correction model (step S15). Fig. 9 As shown in FIG. 1 , when the DFT calculation result is included in the range of region 1000, the generator 104 generates a DFT correction model that corrects the DFT calculation result to a true value. Fig. 9 As shown, the generation unit 104 generates a VQE correction model that corrects the calculation result of VQE to a true value. In addition, in step S1, DFT calculation and VQE calculation are performed on ionization potential to generate a DFT correction model and a VQE correction model, and DFT calculation and VQE calculation are also performed on electron affinity to generate a DFT correction model and a VQE correction model.

[0057] By using the DFT correction model and the VQE correction model generated in step S1 , as described in the following step S2 , it is possible to perform a correction process applicable to a wide range of molecules.

[0058] (Step S2)

[0059] Back to Figure 5 After step S1, the data processing device 10 constructs a database (step S2). Step S2 is Figure 3 The processing is performed by the acquisition unit 103 and the storage unit 105 shown.

[0060] In step S2, the acquisition unit 103 (DFT calculation unit 103a) acquires a third calculated value as a physical property value for the second compound whose true value of the physical property value is unknown by the first calculation method (DFT) (third acquisition step). Furthermore, the storage unit 105 corrects the third calculated value obtained in the region including at least the region (region 1000) where the calculation result can be obtained by the first calculation method (DFT) by the first correction model (DFT correction model), and stores the corrected value as the true value of the physical property value of the second compound in the database 102a (first storage step).

[0061] Furthermore, the acquisition unit 13 (VQE calculation command unit 103b) acquires a fourth calculated value as the physical property value of the second compound by the second calculation method (VQE) in an area including at least an area (area 1100) where the calculation result cannot be obtained by the first calculation method (fourth acquisition step). Furthermore, the storage unit 105 corrects the fourth calculated value by the second correction model (VQE correction model), and stores the corrected value as the true value of the physical property value of the second compound in the database 102a (second storage step).

[0062] use Figure 10 to Figure 12 The specific processing of step S2 will be described. Fig.10 Yes means Figure 5The flowchart shown is a flowchart of the details of step S2, Fig.11 is a diagram for explaining the database after the processing in step S2. Fig.12 It is a diagram for explaining the processing result of step S2.

[0063] like Fig.10 As shown, the acquisition unit 103 acquires a list of second compounds whose true values ​​of physical property values ​​are unknown (step S21). For example, the user interface unit 101 acquires a "list of compounds whose true values ​​of ionization potential and electron affinity are unknown" input by the user and transfers it to the acquisition unit 103. The number of compounds included in the list of the second compounds is greater than the number of compounds included in the list of the first compounds. In addition, the processing after step S22 described below is repeated for each compound included in the list.

[0064] The DFT calculation unit 103a performs DFT calculation on the second compound (step S22), and the storage unit 105 determines whether the calculation result can be obtained through the DFT calculation (step S23). That is, the storage unit 105 determines whether the calculation of the DFT calculation unit 103a is abnormally terminated, or whether the calculation result obtained by the DFT calculation unit 103a is a value corresponding to the region 1200. In other words, the storage unit 105 determines whether the calculation result obtained by the DFT calculation is a value corresponding to the region 1000.

[0065] Hereinafter, a specific example of the determination method will be described. For example, the storage unit 105 obtains an approximate function by first approximation in a region where the true value known by experiment and the DFT calculation result have a correlation. In the case where the value obtained by the DFT calculation is a value that deviates from the range of the straight line extrapolated by such an approximate function, that is, in the case of a value in region 1200, the storage unit 105 determines that the calculation result cannot be obtained by the DFT calculation. In addition, the storage unit 105 uses a threshold value pre-set according to the physical property of the object to determine whether the value obtained by the DFT calculation is a valid calculation result. For example, it is known that when the IP is calculated by DFT calculation, no valid calculation result can be obtained in the region below 2eV. Therefore, when the value of the IP obtained by the DFT calculation is below 2eV, the storage unit 105 determines that the calculation result cannot be obtained by the DFT calculation. In addition, the storage unit 105 determines whether the calculation result can be obtained by the DFT calculation by performing a stability analysis of the solution. For example, the storage unit 105 investigates whether there is singlet instability in the DFT wave function based on the unrestricted method, and if so, determines that the calculation result cannot be obtained through the DFT calculation, and if not, determines that the calculation result can be obtained through the DFT calculation.

[0066] Here, when the calculation result can be obtained through DFT calculation (step S23, yes), the storage unit 105 corrects the DFT calculation result through the DFT correction model (step S24), and stores the corrected value as the true value of the physical property value of the second compound in the database 102a (step S27).

[0067] On the other hand, when the calculation result cannot be obtained by the DFT calculation (step S23, No), the VQE calculation command unit 103b causes the quantum computing device 20 to perform the VQE calculation and obtain the calculation result (step S25). Furthermore, the storage unit 105 corrects the VQE calculation result by the VQE correction model (step S26), and stores the corrected value as the true value of the physical property value of the second compound in the database 102a (step S27).

[0068] Through the processing of the above step S2 (steps S21 to S27), the true values ​​of the ionization potential and the electron affinity are obtained for the plurality of second compounds. Fig.11 As shown in FIG. 1 , the second compound can be registered in the database 102a as the first compound, which can greatly increase the amount of information in the database 102a. Fig.12 As shown in the example, for ionization potential and electron affinity, the area where the database can be constructed by DFT is added to the area where the database can be expanded by VQE, thereby expanding the database space of the molecule. In addition, NISQ can output calculated values ​​that can be correlated with true values ​​in all ranges in theory, but at present, due to hardware resource issues and cost issues, the range of compounds that can be applied by VQE calculations performed by NISQ is limited compared to the range that can be applied by DFT calculations using von Neumann-type computers, which are called classical computers. In this regard, in the above-mentioned step S2, when the calculation results cannot be obtained by DFT calculation, by performing VQE calculations performed by NISQ, the database expansion using NISQ can be efficiently performed.

[0069] (Step S3)

[0070] Back to Figure 5 After step S2, the data processing device 10 uses the database 102a whose information volume has been increased by step S2 to search for new compounds (step S3). Step S3 is performed by Figure 3 The process shown is performed by the search unit 106. Fig.13 This is a diagram for explaining the search unit 106 .

[0071] For example, if a desired property value "AX" for property A is received from the user via the input device 15, Fig.13As shown, the search unit 106 inputs the property value "AX" to the learned model 102b. The learned model 102b infers the structure of the compound X that has the possibility that the property value of property A is "AX". For example, if the property value "AX" of ionization potential is input as the property value of property A from the search unit 106, the learned model 102b outputs a character string inferred as a chemical structure having the property value "AX". In addition, if the property value "AX" of electron affinity is input as the property value of property A from the search unit 106, the learned model 102b outputs a character string inferred as a chemical structure having the property value "AX". In addition, although not shown in the figure, if the property value "AX1" of ionization potential and the property value "AX2" of electron affinity are input as the property value of property A from the search unit 106, the learned model 102b outputs a character string inferred as a chemical structure having the property values ​​"AX1" and "AX2".

[0072] In the embodiment, the database 102a whose information volume is increased in step S2 is used, thereby improving the efficiency of searching for new compounds.

[0073] (Step S4)

[0074] Back to Figure 5 After step S3, the data processing device 10 updates the database 102a (step S4) for the new compound searched in step S3. Step S4 is the same as step S2. Figure 3 The processing is performed by the acquisition unit 103 and the storage unit 105 shown.

[0075] In step S4, the acquisition unit 103 (DFT calculation unit 103a) acquires a third calculated value for the new compound by DFT. Furthermore, the storage unit 105 stores the value obtained by correcting the third calculated value obtained in the region including at least the region (region 1000) where the calculation result can be obtained by the first calculation method (DFT) in the database 102a as the true value of the physical property value of the new compound by the first correction model (DFT correction model).

[0076] Furthermore, the acquisition unit 103 (VQE calculation command unit 103b) acquires a fourth calculated value as a physical property value of the new compound by the second calculation method (VQE) in an area including at least an area (area 1100) where the calculation result cannot be obtained by the first calculation method. Furthermore, the storage unit 105 stores the value obtained by correcting the fourth calculated value by the second correction model (VQE correction model) as the true value of the physical property value of the new compound in the database 102a.

[0077] use Fig.14 The specific processing of step S4 will be described. Fig.14 Yes means Figure 51 is a flowchart showing the details of step S4 of the flowchart shown in FIG.

[0078] like Fig.14 As shown, if the input of a new compound is accepted (step S31), the DFT calculation unit 103a performs DFT calculation on the new compound (step S32), and the storage unit 105 determines whether the calculation result can be obtained through the DFT calculation (step S33). Here, if the calculation result can be obtained through the DFT calculation (step S33, yes), the storage unit 105 corrects the DFT calculation result through the DFT correction model (step S34), and stores the correction value as the true value of the physical property value of the new compound in the database 102a (step S37).

[0079] On the other hand, when the calculation result cannot be obtained by DFT calculation (step S33, No), the VQE calculation instruction unit 103b causes the quantum computing device 20 to perform VQE calculation and obtain the calculation result (step S35). Furthermore, the storage unit 105 corrects the VQE calculation result by the VQE correction model (step S36), and stores the corrected value as the true value of the physical property value of the new compound in the database 102a (step S37).

[0080] The processing of step S4 can also obtain the true value of the physical property value for a new compound, and the amount of information in the database 102a can be further increased. If the true value is limited to the experimental value and does not include the calculated value, the true value of the physical property value obtained in step S4 is a predicted true value.

[0081] As described above, in an embodiment, for a limited number of first compounds, calculated values ​​of ionization potential and electron affinity are obtained by DFT calculation and VQE calculation, and by comparing the calculated results with the limited number of true values, a correction model that can be applied to the calculated results of a wide range of molecules is obtained. Moreover, in an embodiment, the calculated values ​​of ionization potential and electron affinity are converted into true values ​​for a large number of second compounds using the correction model that can be applied to the calculated results of a wide range of molecules. Thus, in an embodiment, the amount of information in the database 102a used for new material search can be increased.

[0082] In addition, in the embodiment, by using the database 102a with increased information volume, the efficiency of compound search can be improved, and for new compounds, a correction model that can be applied to the calculation results of a wide range of molecules can be used to determine the true value of the physical property value, and the information volume of the database 102a can be further increased.

[0083] The above describes the embodiments of the present invention, but the present invention is not limited to the above embodiments, and the components can be deformed and embodied in the implementation stage without departing from the scope of the main purpose. In addition, various inventions can be formed by appropriate combinations of multiple components disclosed in the above embodiments. For example, some components can also be deleted from all the components shown in the embodiments.

[0084] (Variation Example)

[0085] Modifications are described below.

[0086] (1) Modification 1

[0087] In the processing of step S2 described in the above-mentioned embodiment, in the region 1000 where the (valid) calculation result can be obtained by DFT, the calculation result (third calculation value) obtained by DFT calculation is corrected by the DFT correction model and stored in the database. In addition, in the processing of step S2, in the region 1100 where the calculation result cannot be obtained by DFT, the calculation result (fourth calculation value) obtained by VQE calculation is corrected by the VQE correction model and stored in the database. Here, in the processing of step S2, in the region 1200 where the invalid calculation result is obtained by DFT, both the calculation result of DFT (third calculation value) and the calculation result of VQE (fourth calculation value) are obtained, but the calculation result (fourth calculation value) obtained by VQE calculation is corrected by the VQE correction model and stored in the database.

[0088] However, the boundary between the region 1200 and the region 1000 is the boundary between the range with correlation with the DFT calculation result and the true value and the range with no correlation with the DFT calculation result and the true value. Such a boundary is set, for example, based on a threshold value set for the correlation coefficient, etc., or a user's judgment. Therefore, near the boundary, it is unclear which value of the value obtained by correcting the third calculation value and the value obtained by correcting the fourth calculation value is appropriate as the true value.

[0089] Therefore, in Modification 1, the following processing is performed. For example, the DFT calculated value corresponding to the boundary between region 1200 and region 1000 is set to "A", and "α" is given by initial setting or user setting. In Modification 1, when the DFT calculated value "Y1" of the ionization potential of the second compound "Z" is "A-α≤Y1≤A+α", the storage unit 105 corrects "Y1" obtained by correcting "Y1" using the DFT correction model. DFT ” and the corrected value “Y2” obtained by correcting the VQE calculated value “Y2” of the second compound “Z” by the VQE correction model VQE” is displayed on the display device 16. At this time, the storage unit 105 causes the chemical structure of the second compound “Z” to be also displayed on the display device 16. The user selects the correction value “Y1” in consideration of the chemical structure. DFT " and correction value "Y2 VQE " is determined to be an appropriate correction value. Moreover, the storage unit 105 stores the correction value selected by the user as the true value of the second compound "Z" in the database 102a.

[0090] In Modification 1, even if the DFT calculation result (third calculation value) is included in region 1200, there is a case where the calculation result is corrected by the DFT correction model and stored in the database. In addition, in Modification 1, even if the DFT calculation result (third calculation value) is included in region 1000, if the third calculation value is within the above range, there is a case where the VQE calculation result (fourth calculation value) is obtained, and the calculation result is corrected by the VQE correction model and stored in the database. Through the processing of Modification 1, an appropriate value can be obtained as the true value of the physical property value in the boundary region.

[0091] In the above-mentioned modification 1, the case where the correction value is selected by the user is described, but the correction value may also be selected automatically. In such a case, for example, the search unit 106 infers the value of the ionization potential of the second compound "Z" by using data mining of the database 102a. Furthermore, the storage unit 105 stores "Y1 DFT ” and “Y2 VQE " The value close to the value inferred by the search unit 106 is stored in the database 102a as the true value of the second compound "Z".

[0092] (2) Modification 2

[0093] In the processing of step S2 described in the above embodiment, when the correction value obtained by correcting the calculation result (third calculation value) obtained by the DFT calculation of the second compound by the DFT correction model is stored as a true value in the database 102a, the generation unit 104 can also perform the following modification example 2.

[0094] The generator 104 of the second modification further obtains the calculation result of VQE via the VQE calculation command unit 103b for the second compound that uses the correction value obtained by correcting the DFT calculation result by the DFT correction model as the true value. In addition, the generator 104 further generates a second correction model (VQE correction model) using the value stored as the true value and the calculated value of VQE. By performing the processing of the second modification as described above at any time, the second correction model can be updated, and as a result, the accuracy of the correction can be improved.

[0095] (3) Modification 3

[0096] In the above-mentioned embodiment, the case where the first calculation method is DFT and the second calculation method is VQE is described, but it is not limited to this. If the second calculation method can obtain the calculation result in the area where the calculation result cannot be obtained by the first calculation method, the first calculation method and the second calculation method applied by the data processing method disclosed in this specification can also be applied in any combination. For example, as the first calculation method, a calculation method based on perturbation theory and coupled clustering theory can be cited. In addition, for example, as the second calculation method, quantum phase inference can be cited.

[0097] (4) Modification 4

[0098] In the above-mentioned embodiment, the case where the ionization potential and the electron affinity are calculated as physical property values ​​is described, but any type of physical property value may be used as long as it can be calculated by both the first calculation method and the second calculation method that can be applied to the data processing method disclosed in this specification.

[0099] (5) Modification 5

[0100] In the above-mentioned embodiment, the case where a chemical structure having a desired IP, a chemical structure having a desired EA, and a chemical structure having a desired IP and EA are inferred by using the learned model 102b that has learned the correlation between the chemical structure and IP, the correlation between the chemical structure and EA, and the correlation between the chemical structure and IP and EA is described, but it is not limited to this. For example, a chemical structure having a third physical property (physical property B) different from IP and EA can also be inferred by using the learned model 102b and the second learned model described below. For example, the second learned model is generated by machine learning using a list of compounds whose true values ​​of IP, EA, and physical property B are known as teacher data. If the physical property value "BX" desired by the user for physical property B is input from the search unit 106, the second learned model outputs the value of IP (IP) that has the possibility of becoming the physical property value "BX". BX ) and the value of EA (EA BX ). The search unit 106 obtains the IP BX and EA BX The character string estimated as a chemical structure having the physical property value "BX" is input to the learned model 102b. The new compound found by such processing is also processed in step S4 to obtain the true value of the physical property value of IP and EA.

[0101] (6) Modification 6

[0102] The data processing system 1 in the above-mentioned embodiment includes a data processing device 10 and a quantum computing device 20, but the data processing system 1 in the embodiment can also be configured in a manner that multiple functions of the above-mentioned data processing device 10 and the functions of the quantum computing device 20 are dispersed into multiple devices.

[0103] For example, the data processing system 1 may include: a database construction device having an acquisition unit 103, a generation unit 104, and a storage unit 105; a quantum computing device 20; a search device having a learned model 102b and a search unit 106; and a storage device storing a database 102a. In addition, the database 102a stored in the storage device may integrate databases constructed by each of the plurality of database construction devices.

[0104] The above-described embodiment can be arbitrarily combined with the above-described modified examples, and the above-described modified examples can be arbitrarily combined with each other.

[0105] In addition, the program executed by the data processing device 10 of the above-mentioned embodiment can also be configured to be provided in a form that can be installed or executable, recorded on a computer-readable recording medium such as a CD-ROM, a floppy disk (FD), a magneto-optical disk, a CD-R, a DVD, a Blu-ray Disc (registered trademark), a USB (Universal Serial Bus), etc., or it can be configured to be provided or distributed via a network such as the Internet. In addition, it can also be configured to provide various programs by pre-setting them on a non-volatile storage medium such as a ROM.

[0106] Description of Reference Numerals

[0107] 1…data processing system; 10…data processing device; 101…user interface unit; 102…storage unit; 103…acquisition unit; 103a…DFT calculation unit; 103b…VQE calculation command unit; 104…generation unit; 105…storage unit; 106…search unit; 20…quantum computing device.

Claims

1. A data processing method, comprising: A first acquisition step, for each of a plurality of first compounds whose true values ​​of physical property values ​​are known, acquiring a first calculated value as the above-mentioned physical property value by a first calculation method; A second acquisition step of acquiring, for each of the plurality of first compounds, a second calculated value as the physical property value by a second calculation method, wherein the second calculation method is capable of acquiring a calculation result in an area where a calculation result cannot be acquired by the first calculation method; A generating step of generating a first correction model for correcting the first calculated value to the true value and a second correction model for correcting the second calculated value to the true value; A third obtaining step, for a second compound whose true value of the physical property value is unknown, obtaining a third calculated value as the physical property value by the first calculation method; A first storage step of correcting the third calculated value by the first correction model and storing the corrected value as the true value of the physical property value of the second compound in a database, wherein the third calculated value is a value obtained in a region at least including a region where a calculation result can be obtained by the first calculation method; A fourth acquisition step of acquiring a fourth calculated value as a physical property value of the second compound by the second calculation method in an area including at least an area where a calculation result cannot be obtained by the first calculation method; as well as The second storage step is to correct the fourth calculated value by using the second correction model, and store the corrected value as the true value of the physical property value of the second compound in the database. The first acquisition step and the third acquisition step are respectively performed by using a von Neumann computer to execute the first calculation method to obtain the first calculated value and the third calculated value, respectively. The second acquisition step and the fourth acquisition step respectively acquire the second calculated value and the fourth calculated value by executing the second calculation method using a quantum computing device.

2. The data processing method according to claim 1, wherein: The method further comprises a searching step of searching for a new compound having a specified characteristic quantity using the above database.

3. The data processing method according to claim 2, wherein: The third obtaining step obtains the third calculated value for the new compound. The first storage step stores a value obtained by correcting the third calculated value by the first correction model as a true value of the physical property value of the new compound in the database, wherein the third calculated value is a value obtained in a region including at least a region where a calculation result can be obtained by the first calculation method. The fourth obtaining step obtains the fourth calculated value as the physical property value of the new compound by the second calculation method in a region including at least a region where the calculation result cannot be obtained by the first calculation method, The second storage step stores a value obtained by correcting the fourth calculated value using the second correction model as a true value of the physical property value of the new compound in the database.

4. The data processing method according to claim 2 or 3, wherein: The searching step searches for the new compound by machine learning using the database.

5. The data processing method according to any one of claims 1 to 3, wherein: The above-mentioned generation step is a step of comparing the above-mentioned true value, the above-mentioned first calculated value and the above-mentioned second calculated value for each first compound among the above-mentioned multiple first compounds, generating the above-mentioned first correction model within the range where the above-mentioned first calculated value has a correlation with the above-mentioned true value, and generating the above-mentioned second correction model within the range where the above-mentioned second calculated value has a correlation with the above-mentioned true value.

6. The data processing method according to any one of claims 1 to 3, wherein: The generation step is a step of generating the second correction model for the second compound by further using the value stored in the database as the true value of the physical property value in the first storage step and the calculated value obtained as the physical property value by the second calculation method.

7. The data processing method according to any one of claims 1 to 3, wherein: The true value of the physical property value of the first compound is an experimental value obtained through experiments.

8. The data processing method according to any one of claims 1 to 3, wherein: The first calculation method mentioned above is a calculation method based on density functional theory. The second calculation method mentioned above is a variational quantum eigensolution.

9. The data processing method according to claim 1, wherein: The above-mentioned quantum computing device is NISQ (Noisy Intermediate-Scale Quantum device).

10. A data processing device comprising: An acquisition unit, for each of a plurality of first compounds whose true value of the physical property value is known, acquires a first calculated value as the physical property value by a first calculation method, and for each of the plurality of first compounds, acquires a second calculated value as the physical property value by a second calculation method, wherein the second calculation method can acquire a calculation result in an area where the calculation result cannot be acquired by the first calculation method; a generating unit for generating a first correction model for correcting the first calculated value to the true value and a second correction model for correcting the second calculated value to the true value; as well as a storage unit, which corrects the third calculated value by the first correction model and stores the corrected value in a database as the true value of the physical property value of the second compound, and corrects the fourth calculated value by the second correction model and stores the corrected value in the database as the true value of the physical property value of the second compound, wherein the third calculated value is a third calculated value obtained by the acquisition unit as a physical property value by the first calculation method for the second compound whose true value of the physical property value is unknown, and the third calculated value is a value obtained in an area at least including an area where a calculation result can be obtained by the first calculation method, and the fourth calculated value is a value obtained by the acquisition unit as a physical property value by the second calculation method in an area at least including an area where a calculation result cannot be obtained by the first calculation method, The acquisition unit acquires the first calculated value and the third calculated value by executing the first calculation method using a von Neumann computer, and acquires the second calculated value and the fourth calculated value by executing the second calculation method using a quantum computing device.

11. A data processing system comprising a data processing device and a quantum computing device, wherein the data processing device comprises: An acquisition unit, for each of a plurality of first compounds whose true value of the physical property value is known, acquires a first calculated value as the physical property value by a first calculation method, and for each of the plurality of first compounds, acquires a second calculated value as the physical property value by a second calculation method, wherein the second calculation method can acquire a calculation result in an area where the calculation result cannot be acquired by the first calculation method; a generating unit for generating a first correction model for correcting the first calculated value to the true value and a second correction model for correcting the second calculated value to the true value; as well as a storage unit, which corrects the third calculated value by the first correction model and stores the corrected value in a database as the true value of the physical property value of the second compound, and corrects the fourth calculated value by the second correction model and stores the corrected value in the database as the true value of the physical property value of the second compound, wherein the third calculated value is a third calculated value obtained by the acquisition unit as a physical property value by the first calculation method for the second compound whose true value of the physical property value is unknown, and the third calculated value is a value obtained in an area at least including an area where a calculation result can be obtained by the first calculation method, and the fourth calculated value is a value obtained by the acquisition unit as a physical property value by the second calculation method in an area at least including an area where a calculation result cannot be obtained by the first calculation method, The acquisition unit acquires the first calculated value and the third calculated value by executing the first calculation method using a von Neumann computer, and acquires the second calculated value and the fourth calculated value by executing the second calculation method using the quantum computing device.

Images