Program, information processing system, and information processing method

Abstract

This program causes a computer to execute: steps S2 and S3 for deriving, using an arbitrary molecular structure as input, two or more pieces of first state information indicating at least one of energy or a force on each of two or more potential energy surfaces from among a plurality of potential energy surfaces of the molecular structure including one or more molecules; and a step S4 for outputting, on the basis of the two or more pieces of first state information, second state information indicating at least one of energy or a force on a prescribed potential energy surface different from the two or more potential energy surfaces.

Description

Program, information processing system, and information processing method

[0001] The present disclosure relates to a technique for estimating at least one of the energy and force of a molecular structure.

[0002] Patent Document 1 discloses a technique for analyzing a chemical reaction from a plurality of reactants to a product.

[0003] Non-Patent Document 1 discloses a study on virtual screening based on intersystem crossing using machine learning potentials.

[0004] Japanese Patent Application Laid-Open No. 2022-027123

[0005] Simon Axelrod, Eugene Shakhnovich, Rafael Gomez-Bombarelli. “Thermal Half-Lives of Azobenzene Derivatives: Virtual Screening Based on Intersystem Crossing Using a Machine Learning Potential.” ACS Cent. Sci. 2023, 9, 166-176

[0006] The present disclosure provides a reaction path search program that takes as input state information indicating at least one of energy and force on a potential energy surface, and is capable of searching for reaction paths that take into account reactions involving transitions from an arbitrary potential energy surface to a different potential energy surface.

[0007] A program according to one aspect of the present disclosure causes a computer to execute the following steps: a derivation step of deriving, using an arbitrary molecular structure as input, two or more pieces of first state information indicating at least one of energy and force on each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure including one or more molecules; and an output step of outputting, based on the two or more pieces of first state information, second state information indicating at least one of energy and force on a predetermined potential energy surface different from the two or more potential energy surfaces.

[0008] According to the present disclosure, even in a reaction path search program that inputs state information indicating at least one of energy and force on one potential energy surface, it is possible to search for a reaction path that takes into account a reaction involving a transition from an arbitrary potential energy surface to a different potential energy surface.

[0009] FIG. 1 is a schematic diagram showing an example of a reaction path search method. FIG. 2 is a flowchart showing an example of the basic operation of a search program. FIG. 3 is a schematic diagram showing an example of intersystem crossing. FIG. 4A is a diagram explaining a problem when using a search program that can only input information for one potential energy surface. FIG. 4B is a diagram showing an example of an external program that outputs energy or force on a potential energy surface corresponding to a reaction path considering a reaction involving a transition from an arbitrary potential energy surface to a different potential energy surface. FIG. 5 is a block diagram showing an overall configuration including an information processing system according to an embodiment. FIG. 6 is a diagram showing an example of output of second state information. FIG. 7 is a flowchart showing an example of operation of an information processing system according to an embodiment. FIG. 8 is a diagram explaining the ratio of time required for search processing by a simulation device according to an embodiment. FIG. 9 is a diagram explaining the accuracy of search processing by a simulation device according to an embodiment. FIG. 10 is a diagram explaining a problem of the technology disclosed in Non-Patent Document 1. FIG. 11A is a diagram showing an example of an input screen according to an embodiment. FIG. 11B is a diagram showing an example of an output screen according to an embodiment.

[0010] (Findings that led to the present disclosure) In recent years, when analyzing chemical reactions of molecular structures containing one or more molecules, search programs based on, for example, NEB (Nudged Elastic Band) or the like are used. NEB is a method for searching for a minimum energy path by using reactants and products as inputs and optimizing the reaction path in a chemical reaction from the reactants to the products. Figure 1 is a schematic diagram showing an example of a reaction path search method.

[0011] FIG. 1 shows an example of a minimum-energy path connecting reactants, a transition state structure, and products on a potential energy surface. A potential energy surface can be represented by treating the reactants and products as a single system (reaction system) and plotting the potential energy of the reaction system against coordinates representing the arrangement of all atoms contained in the molecular structure. In FIG. 1 , the vertical axis represents the energy of the molecular structure, and the horizontal axis represents the reaction coordinate (a coordinate when a reaction path connecting two stable structures is displayed one-dimensionally on the potential energy surface). Also, in FIG. 1 , the transition state structure represents the molecular structure at the highest energy state passed through during the chemical reaction from reactants to products. In this way, using a search program such as NEB makes it easy for users to intuitively examine the series of molecular structure changes and energy changes of the molecular structure along the reaction coordinate path (reaction path).

[0012] Here, the basic processing executed by the search program will be described with reference to Fig. 2. Fig. 2 is a flowchart showing an example of the basic operation of the search program.

[0013] (Step S101) The search program executes a process of calculating the energy and force of a molecular structure. In step S101, the search program not only calculates the energy of the molecular structure, but also calculates the energy gradient of the molecular structure, i.e., the force acting on the molecular structure.

[0014] (Step S102) The search program executes a process to determine whether the energy and force of the molecular structure calculated in step S101 satisfy the convergence condition, based on the energy and force. If it is determined that the energy and force satisfy the convergence condition (step S102: Yes), the search program terminates the reaction path search process. On the other hand, if it is determined that the energy and force do not satisfy the convergence condition (step S102: No), the search program executes step S103.

[0015] (Step S103) The search program executes a process to update the molecular structure (step S103). In step S103, the search program updates the molecular structure by adjusting the positions of one or more atoms so as to adjust the forces acting on one or more atoms among all atoms included in the molecular structure. Thereafter, the search program repeats the above steps S101 to S103 until it is determined that the energy and forces satisfy the convergence conditions.

[0016] Meanwhile, intersystem crossing may occur along the reaction pathway. Intersystem crossing is a non-radiative transition that occurs between quantum states with different spin multiplicities. For example, when a singlet state changes non-radiatively to a triplet state, or vice versa, this process is called intersystem crossing. In other words, intersystem crossing is the reversal of the spin of excited electrons.

[0017] Figure 3 is a schematic diagram showing an example of intersystem crossing. In Figure 3, the vertical axis represents the energy of the molecular structure, and the horizontal axis represents the reaction coordinate. Also in Figure 3, the solid line A1 represents the energy of the molecular structure in the singlet state, the solid line A2 represents the energy of the molecular structure in the triplet state, and the dotted line A3 represents the reaction path to be explored. In the example shown in Figure 3, the reaction path to be explored is represented as a path that follows the lower energy of the energy in the singlet state or the energy in the triplet state, and is a path that takes into account intersystem crossing that occurs at the intersection of the solid lines A1 and A2.

[0018] Thus, when analyzing chemical reactions of molecular structures, it is important to be able to explore reactions that take intersystem crossing into account, in other words, reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface.

[0019] The search program can execute the calculation of the energy in the singlet state and the calculation of the energy in the triplet state by using an external program such as a quantum chemistry calculation program such as Gaussian (registered trademark). On the other hand, if the search program can only input information for one potential energy surface output by the external program, the following problem occurs.

[0020] 4A is an explanatory diagram of a problem that occurs when using a search program that can only input information for one potential energy surface. The search program can execute an external program, such as a quantum chemistry calculation program such as Gaussian (registered trademark). However, when using an external program, the search program can only execute a process using a molecular structure as input once each time the molecular structure is updated. Therefore, as shown in FIG. 4A, the external program can only output either the energy in the singlet state or the energy in the triplet state as the energy of the input molecular structure.

[0021] In other words, when using a search program that can only input information for one potential energy surface, it is not possible to calculate both the energy in the singlet state and the energy in the triplet state, and there is a problem in that it is not possible to search for a reaction path that takes into account a reaction involving a transition from an arbitrary potential energy surface to a different potential energy surface.

[0022] In order to solve the above-described problems, a program according to a first aspect of the present disclosure causes a computer to execute the following steps: a derivation step of deriving, using an arbitrary molecular structure as input, two or more pieces of first state information indicating at least one of energy and force on each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure including one or more molecules; and an output step of outputting, based on the two or more pieces of first state information, second state information indicating at least one of energy and force on a predetermined potential energy surface different from the two or more potential energy surfaces.

[0023] This makes it possible to utilize an external program that outputs energy or force on a potential energy surface corresponding to a reaction path that takes into account a reaction involving a transition from an arbitrary potential energy surface to a different potential energy surface, as shown in FIG. 4B, and thus enables the search for such a reaction path.

[0024] Also, for example, in the program according to the second aspect of the present disclosure, in the first aspect, the derivation step may include using the arbitrary molecular structure as input and outputting the two or more pieces of first state information using a trained model that has been trained by machine learning rather than quantum chemical calculation.

[0025] This allows the trained model to be used to search for reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface, which has the advantage of making it easier to speed up the search for reaction paths that take into account reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface.

[0026] To solve the above-mentioned problems, an information processing system according to a third aspect of the present disclosure includes an acquisition unit that acquires molecular structure information indicating a molecular structure including one or more molecules, a search unit that uses the acquired molecular structure information as an input and searches for a reaction path of the molecular structure, and an output unit that outputs, as a processing result of the search unit, second state information indicating at least one of energy and force on a predetermined potential energy surface of the molecular structure information acquired by the acquisition unit, wherein the search unit includes one or more inference units that use an arbitrary molecular structure as an input and output two or more first state information indicating at least one of energy and force on each of two or more potential energy surfaces of a plurality of potential energy surfaces of the molecular structure, the second state information indicating at least one of energy and force on the predetermined potential energy surface different from the two or more first state information output by the one or more inference units. Note that the one or more inference units may output the two or more first state information using quantum chemical calculations.

[0027] This has the advantage that it becomes possible to search for reaction paths that take into account reactions that involve transitions from an arbitrary potential energy surface to a different potential energy surface.

[0028] Also, for example, in an information processing system relating to a fourth aspect of the present disclosure, the one or more inference units are trained models trained by machine learning, and the trained models are trained to use the arbitrary molecular structure as input and output the two or more pieces of first state information.

[0029] This allows the trained model to be used to search for reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface, which has the advantage of making it easier to speed up the search for reaction paths that take into account reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface.

[0030] Furthermore, for example, in the information processing system according to the fifth aspect of the present disclosure, in the third aspect, the output unit may further output intersection position information including at least one of the energy at an intersection position where the two or more potential energy surfaces intersect, the force at the intersection position, and the molecular structure at the intersection position.

[0031] This has the advantage that information about the molecular structure at the intersection of multiple potential energy surfaces can be obtained.

[0032] Also, for example, in an information processing system according to a sixth aspect of the present disclosure, in the fifth aspect, when the second state information and the intersection position information each include a molecular structure, the molecular structure at the extreme value of energy indicated by the second state information may coincide with the molecular structure indicated by the intersection position information.

[0033] This has the advantage that the molecular structure at the intersection of a plurality of potential energy surfaces can be investigated by examining the extreme values ​​of the energy or force indicated by the second state information.

[0034] Also, for example, in the information processing system according to the seventh aspect of the present disclosure, in any one of the third to sixth aspects, the predetermined potential energy surface may be a surface that is differentiable at least once.

[0035] This has the advantage that the energy indicated by the second state information changes smoothly even at the intersection of multiple potential energy surfaces, making it possible to search for reactions involving intersections using the conventional molecular structure update method based on energy differential values.

[0036] Furthermore, for example, in the information processing system according to the eighth aspect of the present disclosure, in any one of the third to sixth aspects, the two or more potential energy surfaces may be surfaces having mutually different spin multiplicities as parameters.

[0037] This allows us to search for reactions that involve a transition from an arbitrary potential energy surface to a potential energy surface with a different spin multiplicity, which has the advantage of making it easier to search for reactions that take intersystem crossing into account.

[0038] Furthermore, for example, in the information processing system according to the ninth aspect of the present disclosure, in the third or fourth aspect, the molecular structure may be a molecular structure of a molecule that exhibits intersystem crossing, and the one or more inference units may include a first inference unit that outputs first state information indicating at least one of energy and force on a potential energy surface that represents a singlet state of the molecular structure among the plurality of potential energy surfaces, and a second inference unit that outputs first state information that indicates at least one of energy and force on a potential energy surface that represents a triplet state of the molecular structure among the plurality of potential energy surfaces.

[0039] This has the advantage that it becomes possible to search for reaction paths that take into account reactions involving a transition from a potential energy surface representing a singlet state of a molecular structure to a potential energy surface representing a triplet state of a molecular structure.Similarly, it has the advantage that it becomes possible to search for reaction paths that take into account reactions involving a transition from a potential energy surface representing a triplet state of a molecular structure to a potential energy surface representing a singlet state of a molecular structure.

[0040] Furthermore, for example, in an information processing system according to a tenth aspect of the present disclosure, in the ninth aspect, the second state information output by the output unit may indicate at least the lower of the energy on the potential energy surface representing the singlet state and the energy on the potential energy surface representing the triplet state.

[0041] This has the advantage that it is easy to derive an optimal reaction path by searching for a reaction path that takes into account a reaction involving a transition from a potential energy surface representing a singlet state of a molecular structure to a potential energy surface representing a triplet state.Similarly, it has the advantage that it is easy to derive an optimal reaction path by searching for a reaction path that takes into account a reaction involving a transition from a potential energy surface representing a triplet state of a molecular structure to a potential energy surface representing a singlet state.

[0042] Furthermore, for example, in an information processing system according to an eleventh aspect of the present disclosure, in the ninth or tenth aspect, the second state information output by the output unit may indicate at least the force on the potential energy surface representing the singlet state and the force on the potential energy surface representing the triplet state, whichever has lower energy.

[0043] This has the advantage that it is easy to derive an optimal reaction path by searching for a reaction path that takes into account a reaction involving a transition from a potential energy surface representing a singlet state of a molecular structure to a potential energy surface representing a triplet state.Similarly, it has the advantage that it is easy to derive an optimal reaction path by searching for a reaction path that takes into account a reaction involving a transition from a potential energy surface representing a triplet state of a molecular structure to a potential energy surface representing a singlet state.

[0044] Also, for example, in the information processing system according to the twelfth aspect of the present disclosure, in the fifth aspect, the search unit may further include a processing unit that performs processing at the intersection position where the two or more pieces of first state information output by the one or more inference units intersect, and generates the second state information by combining regions of the potential energy surface where at least one of the energy and the force is lower than at least one of the energy and the force at the intersection position in each of the two or more pieces of first state information.

[0045] This has the advantage that, in each of the two or more pieces of first state information, a region of the potential energy surface that is lower than at least one of the energy and force at the intersection position is combined, making it easier to derive an optimal reaction path.

[0046] Furthermore, in an information processing method according to a thirteenth aspect of the present disclosure, a computer executes a derivation step of deriving, using an arbitrary molecular structure as input, two or more pieces of first state information indicating at least one of energy and force on each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure including one or more molecules, and an output step of outputting, based on the two or more pieces of first state information, second state information indicating at least one of energy and force on a predetermined potential energy surface different from the two or more potential energy surfaces.

[0047] This makes it possible to utilize an external program that outputs energy or force on a potential energy surface corresponding to a reaction path that takes into account a reaction involving a transition from an arbitrary potential energy surface to a different potential energy surface, as shown in FIG. 4B, and thus enables the search for such a reaction path.

[0048] It goes without saying that the program of the present disclosure, that is, the computer program, can be distributed via a computer-readable non-transitory recording medium such as a CD-ROM or a communication network such as the Internet.

[0049] Hereinafter, the embodiments will be specifically described with reference to the drawings.

[0050] The embodiments described below are comprehensive or specific examples of the present disclosure. The numerical values, shapes, materials, components, component placement and connection configurations, steps, and step order shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components that are not described in the independent claims that represent the highest concepts are described as optional components. Furthermore, each figure is a schematic diagram and is not necessarily an exact illustration. Furthermore, the same components are designated by the same reference numerals in each figure.

[0051] Furthermore, the information processing system according to the embodiment of the present disclosure may be configured so that all components are included in one computer, or may be configured as a system in which multiple components are distributed across multiple computers.

[0052] (Embodiment) Hereinafter, an information processing system (information processing method or program) according to an embodiment of the present disclosure will be described with reference to the drawings.

[0053] [Configuration] First, the configuration of an information processing system used in the embodiment will be described.

[0054] 5 is a block diagram showing an overall configuration including an information processing system 100 according to an embodiment. The information processing system 100 is realized by a simulation device 1 configured as a computer such as a personal computer or a server. The information processing system 100 may be realized by edge computing or cloud computing, for example. In the embodiment, the simulation device 1 will be described as being a stationary computer.

[0055] The simulation device 1 includes a first acquisition unit 11, a search unit 12, a first output unit 13, and a storage unit 14. The simulation device 1 executes a search program that searches for a reaction path in a chemical reaction of a molecular structure using a search method such as the NEB described above. Note that the simulation device 1 can execute search programs that use, in addition to NEB, a metadynamics method, a molecular dynamics method, or the like. In this embodiment, the simulation device 1 executes a search program that uses NEB.

[0056] The simulation device 1 is connected to an input unit 2, a display control unit 30, and a display unit 3. The input unit 2, the display control unit 30, and the display unit 3 are configured by an information terminal used by a user, such as a smartphone, a tablet terminal, or a personal computer.

[0057] The input unit 2 and the display control unit 30 may both be connected to the simulation device 1 via a LAN (Local Area Network) or the like, or may be connected to the simulation device 1 via a network such as the Internet.

[0058] The input unit 2 is an input interface that accepts user input and is configured, for example, with a keyboard, a touch sensor, a touchpad, or a mouse. The input unit 2 accepts input operations by the user and outputs a signal corresponding to the input operation to the information processing system 100. In the present disclosure, the display unit 3 and the input unit 2 are configured independently of each other, but they may be configured integrally like a touch panel. In the present disclosure, the information processing system 100 does not include the display unit 3 or the input unit 2, but may include these.

[0059] The input unit 2 receives input of initial molecular structure information regarding an initial molecular structure containing one or more molecules desired by a user. The information received by the input unit 2 is, for example, a composition formula indicating the composition of a reactant whose chemical reaction process the user wants to investigate.

[0060] The display control unit 30 causes the display unit 3 to display images and the like based on the information output from the first output unit 13 of the simulation device 1 .

[0061] The display unit 3 displays images and the like under the control of the display control unit 30. The display unit 3 is, for example, but is not limited to, a liquid crystal display, a plasma display, or an organic EL (Electro-Luminescence) display. Note that a device including the display unit 3 and the display control unit 30 may also be referred to as the display unit.

[0062] The first acquisition unit 11 acquires the initial molecular structure information received by the input unit 2 .

[0063] The search unit 12 searches for a reaction path in a chemical reaction of the initial molecular structure indicated by the initial molecular structure information acquired by the first acquisition unit 11. In the embodiment, the search unit 12 searches for a reaction path according to the NEB. Specifically, the search unit 12 searches for a reaction path according to the basic operation example of the search program using the NEB shown in FIG. 2, which has already been described.

[0064] The first output unit 13 outputs an image or the like showing the processing result of the search unit 12 to the display control unit 30, thereby displaying the image or the like on the display unit 3. For example, the first output unit 13 displays an image showing a reaction path including reactants (corresponding to the initial molecular structure), a transition state structure, and products as shown in FIG.

[0065] The storage unit 14 is a recording medium for storing data (including programs) used in various processes that can be executed by the simulation device 1. The recording medium is, for example, a hard disk drive, a RAM (Random Access Memory), a ROM (Read Only Memory), or a semiconductor memory. Note that such a recording medium may be volatile or non-volatile.

[0066] In the embodiment, the search unit 12 includes a second acquisition unit 101, one or more inference units 102, and a second output unit 103. The second acquisition unit 101, the one or more inference units 102, and the second output unit 103 are all components of the information processing system 100. The second acquisition unit 101 corresponds to the "acquisition unit" in the information processing system 100, and the second output unit 103 corresponds to the "output unit" in the information processing system 100. Note that the second acquisition unit 101 does not have to be included in the components of the information processing system 100.

[0067] The second acquisition unit 101 acquires molecular structure information related to a molecular structure including one or more molecules. Here, when the search unit 12 starts searching for a reaction path, the second acquisition unit 101 acquires the initial molecular structure information acquired by the first acquisition unit 11 as molecular structure information. Thereafter, each time the molecular structure is updated by the search unit 12, the second acquisition unit 101 acquires molecular structure information related to the updated molecular structure.

[0068] The one or more inference units 102 take an arbitrary molecular structure as input and output two or more pieces of first state information using a trained model trained by quantum chemical calculations or machine learning, etc. Each of the two or more pieces of first state information indicates at least one of energy and force on two or more potential energy surfaces among the multiple potential energy surfaces of the molecular structure indicated by the molecular structure information acquired by the second acquisition unit 101. The one or more inference units 102 are entities that execute the inference step in the information processing method of the present disclosure.

[0069] In this embodiment, the one or more inference units 102 are made up of a first inference unit 102A and a second inference unit 102B.

[0070] The first inference unit 102A uses quantum chemical calculations or the first trained model to output first state information indicating the energy and force on a first potential energy surface corresponding to the molecular structure indicated by the molecular structure information acquired by the second acquisition unit 101. The first potential energy surface is a potential energy surface in the singlet state of the molecular structure, i.e., a potential energy surface when the spin multiplicity is set to "1." The energy and force on the first potential energy surface are values ​​obtained by substituting the molecular structure indicated by the horizontal axis of the graph in Figure 6 (described below), i.e., the reaction coordinate value corresponding to the molecular structure, into a function of the first potential energy surface. The molecular structure is the structure indicated by the molecular structure information acquired by the second acquisition unit 101.

[0071] The second inference unit 102B uses quantum chemical calculations or a second trained model to output first state information indicating the energy and force on a second potential energy surface corresponding to the molecular structure indicated by the molecular structure information acquired by the second acquisition unit 101. The second potential energy surface is a potential energy surface in the triplet state of the molecular structure, i.e., a potential energy surface when the spin multiplicity is set to "3." The energy and force on the second potential energy surface are values ​​obtained by substituting the molecular structure indicated by the horizontal axis of the graph in Figure 6 (described below), i.e., the reaction coordinate values ​​corresponding to the molecular structure, into a function of the first potential energy surface. The molecular structure is the structure indicated by the molecular structure information acquired by the second acquisition unit 101.

[0072] In this manner, in the embodiment, two or more potential energy surfaces (here, the first potential energy surface and the second potential energy surface) are surfaces with different spin multiplicities as parameters.

[0073] In the embodiment, the trained model is a machine learning potential model, and is configured, for example, by a graph neural network that receives a graph structure as input. The graph neural network is, for example, a crystal graph convolutional neural network (CGCNN) or a material graph network (MEGNet). Note that the trained model is not limited to a graph neural network and may be configured, for example, by a method using a symmetric function.

[0074] The following describes the process of machine learning a trained model. In particular, the following describes the process of machine learning a first trained model and a second trained model. Hereinafter, the first trained model before the completion of machine learning will be referred to as the "first model," and the second trained model before the completion of machine learning will be referred to as the "second model."

[0075] First, a process of machine learning a first trained model will be described. In this process, a first training dataset is used to train the first model by machine learning so that, in response to an input of an arbitrary molecular structure, the first model outputs a value indicating at least one of energy and force on a first potential energy surface of the molecular structure (hereinafter also referred to as a "first state value"). The first training dataset includes the molecular structure as input data and the first state value corresponding to the molecular structure as ground truth data.

[0076] First, a machine learning device (hereinafter referred to as a "learning device") converts the coordinates and types of each atom in a molecular structure included in a first training dataset into a graph structure. In the graph structure, nodes correspond to each atom in the molecular structure, and edges correspond to bonds between each atom in the molecular structure. Next, the learning device inputs the converted graph structure into a first model. Thereafter, the learning device compares a predicted value of the first state value output from the first model with the first state value as the correct answer data. If the predicted value of the first state value deviates from the first state value as the correct answer data, the learning device updates the weights of the first model. In this way, the learning device uses a large number of first training datasets to train a first model through supervised learning.

[0077] Next, a process of machine learning a second trained model will be described. In this process, the second model is machine-learned using a second training dataset so that, in response to an input of an arbitrary molecular structure, it outputs a value indicating at least one of the energy and force on a second potential energy surface of the molecular structure (hereinafter also referred to as a "second state value"). The second training dataset includes the molecular structure as input data and the second state value corresponding to the molecular structure as ground truth data.

[0078] First, the training device converts the coordinates and types of each atom in the molecular structure included in the second training dataset into a graph structure. Next, the training device inputs the converted graph structure into the second model. Thereafter, the training device compares the predicted value of the second state value output from the second model with the second state value as the correct answer data. If the predicted value of the second state value deviates from the second state value as the correct answer data, the training device updates the weight of the second model. In this way, the training device uses a large number of second training datasets to train the second model by supervised learning. This concludes the description of the process of machine learning a trained model.

[0079] The second output unit 103 outputs second state information based on two or more pieces of first state information output by one or more inference units 102. The second state information indicates at least one of energy and force on a predetermined potential energy surface different from the two or more potential energy surfaces. The second output unit 103 is an entity that executes the output step in the information processing method disclosed herein. The second state information output by the second output unit 103 is referenced in the process of determining whether the energy has converged in the search unit 12 of the simulation device 1 and in the process of updating the molecular structure.

[0080] In the embodiment, the second output unit 103 outputs second state information indicating both the energy and the force on the third potential energy surface, based on the first state information output by the first inference unit 102 A and the first state information output by the second inference unit 102 B. The third potential energy surface is a surface different from the first potential energy surface and the second potential energy surface, and is a surface that satisfies predetermined conditions.

[0081] Here, the predetermined condition may include, for example, a condition that the lower of the energy on the first potential energy surface and the energy on the second potential energy surface is adopted. Note that the predetermined condition is not limited to the above condition and may be other conditions.

[0082] FIG. 6 is a diagram showing an example of output of second state information. FIG. 6 shows an example of output of second state information when the above-mentioned condition is satisfied as the predetermined condition. In FIG. 6, the vertical axis represents the energy of the molecular structure, and the horizontal axis represents the reaction coordinate. Also in FIG. 6, the solid line A4 represents the energy on the first potential energy surface, i.e., the energy in the singlet state of the molecular structure. Also in FIG. 6, the dashed line A5 represents the energy on the second potential energy surface, i.e., the energy in the triplet state of the molecular structure. Also in FIG. 6, the dotted line A6 represents the energy on the third potential energy surface that satisfies the above-mentioned condition, i.e., the lower of the energy in the singlet state and the energy in the triplet state.

[0083] In the example shown in FIG. 6, the reaction coordinate path (reaction path) is a path that takes into account intersystem crossing, where the energy reaches a maximum value at the intersection of solid line A4 and dashed line A5.

[0084] As described above, it can be said that the second output unit 103 outputs intersection position information including the energy at the intersection position where two or more potential energy surfaces (here, the first potential energy surface and the second potential energy surface) intersect. Note that the intersection position information may be information including at least one of the energy at the intersection position, the force at the intersection position, and the molecular structure at the intersection position.

[0085] Furthermore, as described above, when the second state information and the intersection position information each include a molecular structure, the molecular structure at the extreme value of energy indicated by the second state information (i.e., the minimum or maximum value of energy on the third potential energy surface) coincides with the molecular structure indicated by the intersection position information.

[0086] In addition, in the embodiment, the predetermined potential energy surface (i.e., the third potential energy surface) is a surface that is differentiable at least once and is a smooth surface that does not include any sharp convex shapes. Therefore, the energy changes smoothly at the intersection position, i.e., the position where intersystem crossing occurs on the path of the reaction coordinate (reaction path). Therefore, reactions involving crossing can be searched for by a conventional molecular structure update method based on energy differential values.

[0087] [Operation] The operation of the information processing system 100 according to the embodiment (i.e., the information processing method) will be described below. FIG. 7 is a flowchart showing an example of the operation of the information processing system 100 according to the embodiment. In the example shown in FIG. 7, the one or more inference units 102 are described as being composed of a first inference unit 102A and a second inference unit 102B. Note that step S1 shown below does not necessarily have to be included in the operation (information processing method) of the information processing system 100.

[0088] (Step S1) The second acquisition unit 101 acquires molecular structure information. Here, when the search unit 12 of the simulation device 1 starts searching for a reaction path, the second acquisition unit 101 acquires, as molecular structure information, the initial molecular structure information acquired by the first acquisition unit 11 of the simulation device 1. Thereafter, every time the molecular structure is updated by the search unit 12, the second acquisition unit 101 acquires molecular structure information related to the updated molecular structure.

[0089] (Step S2) The first inference unit 102A uses the first trained model to output first state information indicating the energy and force on the first potential energy surface corresponding to the molecular structure indicated by the molecular structure information acquired by the second acquisition unit 101.

[0090] (Step S3) The second inference unit 102B uses the second trained model to output first state information indicating the energy and force on the second potential energy surface corresponding to the molecular structure indicated by the molecular structure information acquired by the second acquisition unit 101.

[0091] In the embodiment, the information processing system 100 executes steps S2 and S3 in parallel, but it may execute step S3 after executing step S2, or conversely, it may execute step S2 after executing step S3.

[0092] (Step S4) The second output unit 103 outputs second state information indicating both the energy and the force on the third potential energy surface based on the first state information output by the first inference unit 102A and the first state information output by the second inference unit 102B.

[0093] Thereafter, until the search processing by the search unit 12 of the simulation device 1 is completed, the information processing system 100 repeatedly executes the above steps S1 to S4 each time the molecular structure is updated by the search unit 12.

[0094] [Advantages] As described above, the program used in the information processing system 100 according to the embodiment causes a computer to execute the following steps: a derivation step of deriving, using an arbitrary molecular structure as input, two or more pieces of first state information indicating at least one of energy and force on each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure including one or more molecules; and an output step of outputting, based on the two or more pieces of first state information, second state information indicating at least one of energy and force on a predetermined potential energy surface different from the two or more potential energy surfaces.

[0095] The computer is, for example, the information processing system 100. The derivation step corresponds to steps S2 and S3 in Fig. 7, and the output step corresponds to step S4 in Fig. 7. Although the first trained model and the second trained model are used in the embodiment, these trained models do not necessarily have to be used.

[0096] This makes it possible to search for reaction paths that take into account reactions involving transitions from an arbitrary potential energy surface to a different potential energy surface. The information processing method using such a program can also achieve the same effects as the above-mentioned program.

[0097] The deriving step also includes using the above-mentioned arbitrary molecular structure as input and outputting two or more pieces of first state information using a trained model that has been trained by machine learning.

[0098] This allows the trained model to be used to search for reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface, which has the advantage of making it easier to speed up the search for reaction paths that take into account reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface.

[0099] The information processing system 100 according to the embodiment also includes a second acquisition unit 101, a search unit 12 including one or more inference units 102, and a second output unit 103. In the example of FIG. 5 , the search unit 12 includes the second acquisition unit 101 and the second output unit 103, but may not include these units. The second acquisition unit 101 acquires molecular structure information indicating a molecular structure including one or more molecules. The search unit 12 receives the acquired molecular structure information as input and searches for a reaction path of the molecular structure. The second output unit 103 outputs, as a processing result of the search unit 12, second state information indicating at least one of energy and force on a predetermined potential energy surface of the molecular structure information acquired by the second acquisition unit 101. Here, the search unit 12 includes one or more inference units 102 that receive an arbitrary molecular structure as input and output two or more pieces of first state information indicating at least one of energy and force on each of two or more potential energy surfaces of a plurality of potential energy surfaces of the molecular structure. The second state information indicates at least one of energy and force on a predetermined potential energy surface that is different from the two or more pieces of first state information output by the one or more inference units 102 .

[0100] This has the advantage that it becomes possible to search for reaction paths that take into account reactions that involve transitions from an arbitrary potential energy surface to a different potential energy surface.

[0101] Also, for example, in the information processing system 100 according to the embodiment, one or more inference units 102 are trained models that have been trained by machine learning, and the trained models are trained to take an arbitrary molecular structure as input and output two or more pieces of first state information.

[0102] This allows the trained model to be used to search for reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface, which has the advantage of making it easier to speed up the search for reaction paths that take into account reactions that involve a transition from an arbitrary potential energy surface to a different potential energy surface.

[0103] Specifically, the search unit 12 can be said to further include a processing unit that performs processing at an intersection position where two or more pieces of first state information output by one or more inference units 102 intersect, and generates second state information by combining areas of the potential energy surface where at least one of the energy and force is lower than at least one of the energy and force at the intersection position in each of the two or more pieces of first state information.

[0104] This has the advantage that, in each of the two or more pieces of first state information, a region of the potential energy surface that is lower than at least one of the energy and force at the intersection position is combined, making it easier to derive an optimal reaction path.

[0105] In other words, the information processing system 100 according to the embodiment can realize the processing by one or more inference units 102 and the processing by the second output unit 103 as a single processing that inputs a molecular structure and outputs at least one of energy and force. Therefore, for example, in a search program using NEB or the like, if the processing executed by the information processing system 100 according to the embodiment is adopted as an external program, it becomes possible for the external program to calculate both the energy in the singlet state and the energy in the triplet state and then output the lower energy.

[0106] Therefore, the information processing system 100 according to the embodiment can use a trained model to search for reactions involving a transition from an arbitrary potential energy surface to a different potential energy surface, which has the advantage of making it easier to speed up the search for reaction paths that take into account reactions involving a transition from an arbitrary potential energy surface to a different potential energy surface.

[0107] Hereinafter, specific examples of advantages of the information processing system 100 according to the embodiment will be described with reference to the drawings.

[0108] 8 is an explanatory diagram of the ratio of the time required for the search process by the simulation device 1 according to the embodiment. FIG. 8 shows the results of executing the search process for each of four chemical reactions, "Reaction 1," "Reaction 2," "Reaction 3," and "Reaction 4." In FIG. 8, the "reaction formula" represents the reactants and products in the chemical reaction. Also, in FIG. 8, the "magnification" represents how many times the time required for the search process using NEB when using the quantum chemistry calculation program is compared to the time required for the search process using NEB when using the machine learning potential model according to the embodiment.

[0109] 8, for each chemical reaction, the time required for the search process using the NEB when the machine learning potential model according to the embodiment is used is significantly shorter than the time required for the search process using the NEB when the quantum chemistry calculation program is used. In other words, the information processing system 100 according to the embodiment can easily speed up the search process using the search program.

[0110] FIG. 9 is an explanatory diagram of the accuracy of the search process performed by the simulation device 1 according to the embodiment. FIG. 9 shows the calculation results of the energy of a molecular structure when the search process is performed for the chemical reaction "Reaction 1." In FIG. 9, the vertical axis represents energy, and the horizontal axis represents reaction coordinates. Also in FIG. 9, the dashed line A8 represents the calculation results of the energy in the singlet state using a quantum chemistry calculation program (here, Gaussian (registered trademark)), and the solid line A9 represents the calculation results of the energy in the triplet state using a quantum chemistry calculation program. Also in FIG. 9, the circles represent the calculation results of the energy in the triplet state using the machine learning potential model according to the embodiment, and the penalties represent the calculation results of the energy in the singlet state using the machine learning potential model according to the embodiment.

[0111] 9, the energy calculation results using the machine learning potential model according to the embodiment are almost the same as the energy calculation results using the quantum chemistry calculation program. In other words, the information processing system 100 according to the embodiment can easily calculate at least one of the energy and force of a molecular structure with the same accuracy as when using the quantum chemistry calculation program.

[0112] In the above specific example, the advantages of using the information processing system 100 according to the embodiment when performing a search process in a chemical reaction of a small molecule system have been described, but the same advantages as those described above can also be enjoyed when performing a search process in a chemical reaction of a large molecule system by using the information processing system 100 according to the embodiment. In particular, when performing a search process in a chemical reaction of a large molecule system, it is necessary to calculate the energies of a large number of molecular structures compared to when performing a search process in a chemical reaction of a small molecule system, and therefore it is expected that the time required for the search process can be more effectively reduced.

[0113] The advantages of the technology disclosed in Patent Document 1 and the technology disclosed in Non-Patent Document 1 will be described below.

[0114] The technique disclosed in Patent Document 1 searches for a reaction path using NEB, but does not search for a reaction path taking intersystem crossing into consideration.

[0115] Furthermore, the technique disclosed in Non-Patent Document 1 first generates candidate structures for the transition state structure by a metadynamics method using a neural network, and then searches for intersystem crossings based on the sign inversion of the difference between the energy in the singlet state and the energy in the triplet state. However, the technique disclosed in Non-Patent Document 1 has the following problems because it searches for candidate structures for the transition state structure by taking into account only the spin state of either the singlet state or the triplet state.

[0116] FIG. 10 is an explanatory diagram of the problem of the technique disclosed in Non-Patent Document 1. In FIG. 10, the vertical axis represents the energy of the molecular structure, and the horizontal axis represents the reaction coordinate. Also, in FIG. 10, the solid line A10 represents the energy in the singlet state, and the solid line A11 represents the energy in the triplet state. In the example shown in FIG. 10, intersystem crossing occurs, but no transition state structure exists on the reaction path of either the energy in the singlet state or the energy in the triplet state. Therefore, the technique disclosed in Non-Patent Document 1 cannot generate candidate structures for the transition state structure, and therefore cannot search for intersystem crossing.

[0117] In contrast, as already described, the information processing system 100 according to the embodiment utilizes an external program that outputs energy and force on an energy surface corresponding to an intersystem crossing reaction, and therefore has the advantage of being able to search for a reaction path that takes intersystem crossing into consideration. Also, as already described, the information processing system 100 according to the embodiment calculates the energy of a molecular structure using a machine learning potential model (trained model), and therefore has the advantage of being able to easily speed up the process of calculating the energy of a molecular structure and easily shorten the time required to search for a reaction path.

[0118] [Example of Displayed Screen] FIG. 11A is a diagram showing an example of an input screen displayed on the display unit 3. As shown in FIG.

[0119] To acquire the molecular structure information, the display control unit 30 displays, for example, an input screen 31 shown in FIG. 11A on the display unit 3. The input screen 31 includes a first input field 31a, a second input field 31b, and a generate button 31c. In response to a user's input operation on the input unit 2, the display control unit 30 displays a first molecule specified by the user in the first input field 31a and a second molecule specified by the user in the second input field 31b. The first molecule and the second molecule may be displayed as a composition formula or as an image showing a molecular structure (i.e., an initial molecular structure). The first molecule and the second molecule are initial reactants. When the generate button 31c is selected in response to a user's input operation on the input unit 2, the simulation device 1 causes the simulation device 1, i.e., the information processing system 100, to perform the operation shown in FIG. 7.

[0120] FIG. 11B is a diagram showing an example of an output screen displayed on the display unit 3.

[0121] When the search process by the search unit 12 of the simulation device 1 is completed, the display control unit 30 displays, for example, an output screen 32 shown in FIG. 11B on the display unit 3. The output screen 32 shows a reaction path including reactants (corresponding to the initial molecular structure), a transition state structure, and a product. By viewing this output screen 32, the user can easily confirm a reaction path that takes into account a reaction involving a transition from the first potential energy surface to the second potential energy surface, i.e., a third potential energy surface.

[0122] (Modifications) Hereinafter, a program, an information processing system, an information processing method, etc. according to one or more aspects of the present disclosure have been described based on embodiments, but the present disclosure is not limited to the above-described embodiments. As long as the modifications do not deviate from the spirit of the present disclosure, various modifications that a person skilled in the art can conceive of to the above-described embodiments may also be included in the present disclosure.

[0123] For example, in the above embodiment, the inference unit 102 outputs the first state information indicating at least one of the energy and the force on one potential energy surface among the plurality of potential energy surfaces, but this is not limiting. For example, the inference unit 102 may be configured to output two or more pieces of first state information indicating at least one of the energy and the force on each of two or more potential energy surfaces.

[0124] Specifically, in the embodiment, the first inference unit 102A outputs first state information indicating the energy and force on the first potential energy surface, and the second inference unit 102B outputs first state information indicating the energy and force on the second potential energy surface, but one inference unit 102 may be configured to output the above two pieces of state information.

[0125] For example, in the above embodiment, the one or more inference units 102 are two inference units 102 that output two pieces of first state information for each of the two potential energy surfaces of the molecular structure, but this is not limited to this. For example, the one or more inference units 102 may be three or more inference units 102 that output three or more pieces of first state information for each of the three or more potential energy surfaces of the molecular structure.

[0126] In the above embodiment, the two or more potential energy surfaces include a potential energy surface in the singlet state of the molecular structure and a potential energy surface in the triplet state of the molecular structure, but are not limited to this. For example, the two or more potential energy surfaces may include a potential energy surface in the doublet state of the molecular structure, a potential energy surface in the quartet state, or a potential energy surface in the quintet state.

[0127] Furthermore, for example, in the above embodiment, the information processing system 100 is realized as a part of the search unit 12 of the simulation apparatus 1, but this is not limiting. For example, the information processing system 100 may be realized by one or more devices separate from the simulation apparatus 1.

[0128] In the above-described embodiments, each component may be configured with dedicated hardware, or may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU (Central Processing Unit) or a processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0129] The following cases are also included in this disclosure:

[0130] (1) The at least one device is specifically a computer system comprising a microprocessor, a ROM (Read Only Memory), a RAM (Random Access Memory), a hard disk unit, a display unit, a keyboard, a mouse, etc. A computer program is stored in the RAM or hard disk unit. The at least one device achieves its function when the microprocessor operates in accordance with the computer program. Here, the computer program is composed of a combination of multiple instruction codes that indicate commands to a computer to achieve a predetermined function.

[0131] (2) Some or all of the components constituting at least one of the above devices may be configured as a single system LSI (Large Scale Integration). A system LSI is an ultra-multifunctional LSI manufactured by integrating multiple components on a single chip, and specifically, is a computer system configured to include a microprocessor, ROM, RAM, etc. A computer program is stored in the RAM. The system LSI achieves its functions by the microprocessor operating in accordance with the computer program.

[0132] (3) Some or all of the components constituting at least one of the above devices may be configured as an IC card or a standalone module that can be attached to the device. The IC card or module is a computer system configured from a microprocessor, ROM, RAM, etc. The IC card or module may include the above-mentioned ultra-multifunctional LSI. The IC card or module achieves its functions when the microprocessor operates in accordance with a computer program. This IC card or module may be tamper-resistant.

[0133] (4) The present disclosure may be embodied as the methods described above, a computer program that implements these methods on a computer, or a digital signal that includes the computer program.

[0134] The present disclosure may also be a computer program or a digital signal recorded on a computer-readable recording medium, such as a flexible disk, a hard disk, a CD (Compact Disc)-ROM, a DVD, a DVD-ROM, a DVD-RAM, a BD (Blu-ray (registered trademark) Disc), or a semiconductor memory, or a digital signal recorded on such a recording medium.

[0135] Furthermore, the present disclosure may involve transmitting a computer program or a digital signal via a telecommunications line, a wireless or wired communication line, a network such as the Internet, or data broadcasting, etc.

[0136] Furthermore, the program or digital signal may be recorded on a recording medium and transferred, or the program or digital signal may be transferred via a network or the like, so that the program or digital signal may be implemented by another independent computer system.

[0137] The present disclosure, for example, has the effect of being able to appropriately assist a user in predicting chemical reactions, and can be used in a computer device or system for displaying information related to the prediction process.

[0138] REFERENCE SIGNS LIST 11 First acquisition unit 12 Search unit 13 First output unit 14 Storage unit 2 Input unit 3 Display unit 30 Display control unit 100 Information processing system 101 Second acquisition unit (acquisition unit) 102 Inference unit 102A First inference unit 102B Second inference unit 103 Second output unit (output unit)

Claims

1. A derivation step of taking an arbitrary molecular structure as input and deriving two or more first state information that represents at least one of the energy and force in each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure containing one or more molecules, An output step is performed on the computer to output second state information that indicates at least one of the energy and force in a potential energy surface that includes an intersection point where the two or more potential energy surfaces intersect, based on the two or more first state information. program.

2. The derivation step includes taking the arbitrary molecular structure as input and outputting the two or more first state information using a trained model that has been trained by machine learning. The program according to claim 1.

3. An acquisition unit that acquires molecular structure information showing a molecular structure containing one or more molecules, A search unit that takes the acquired molecular structure information as input and searches for the reaction pathway of the molecular structure, The system comprises an output unit that outputs a second state information indicating at least one of the energy and force on a predetermined potential energy surface of the molecular structure information acquired by the acquisition unit as a processing result of the search unit, The search unit, The system includes one or more inference units that take an arbitrary molecular structure as input and output two or more first state information units that indicate at least one of the energy and force in each of two or more potential energy surfaces among a plurality of potential energy surfaces of the molecular structure. The second state information indicates at least one of the energy and force in the predetermined potential energy surface, which is different from the two or more first state information output by the one or more inference units. Information processing system.

4. The above one or more inference units are trained models that have been trained by machine learning. The aforementioned trained model is trained to take the arbitrary molecular structure as input and output the two or more first state information items. The information processing system according to claim 3.

5. The output unit further outputs intersection position information including the energy at the intersection point where the two or more potential energy surfaces intersect, the force at the intersection point, and at least one of the molecular structure at the intersection point. The information processing system according to claim 3.

6. If the second state information and the crossing position information each include a molecular structure, the molecular structure at the energy extremum indicated by the second state information coincides with the molecular structure indicated by the crossing position information. The information processing system according to claim 5.

7. The predetermined potential energy surface is a surface that is differentiable at least once. The information processing system according to any one of claims 3 to 6.

8. The two or more potential energy surfaces mentioned above are surfaces whose parameters are different spin multiplicities. The information processing system according to any one of claims 3 to 6.

9. The aforementioned molecular structure is the molecular structure of a molecule exhibiting intersystem crossing. The one or more inference units described above are: A first inference unit outputs first state information indicating at least one of the energy and force in the potential energy surface representing the singlet state of the molecular structure among the plurality of potential energy surfaces, A second inference unit that outputs first state information indicating at least one of the energy and force in the potential energy surface representing the triplet state of the molecular structure among the plurality of potential energy surfaces, includes: The information processing system according to claim 3 or 4.

10. The second state information output by the output unit indicates at least the lower of the energy in the potential energy surface representing the singlet state and the energy in the potential energy surface representing the triplet state. The information processing system according to claim 9.

11. The second state information output by the output unit indicates at least the force in the potential energy surface with the lower energy of the two potential energy surfaces representing the singlet state and the force in the potential energy surface representing the triplet state. The information processing system according to claim 9.

12. The search unit, The processing unit further comprises a processing unit that performs processing at the intersection point where the two or more first state information outputs from the one or more inference units intersect, and generates the second state information by combining a region of a potential energy surface where at least one of the energy and force is lower than at least one of the energy and force at the intersection point in each of the two or more first state information. The information processing system according to claim 5.

13. Computers A derivation step of taking an arbitrary molecular structure as input and deriving two or more first state information that represents at least one of the energy and force in each of two or more potential energy surfaces among a plurality of potential energy surfaces of a molecular structure containing one or more molecules, The following steps are performed: an output step of outputting second state information that indicates at least one of energy and force in a potential energy surface that includes an intersection point where the two or more potential energy surfaces intersect, based on the two or more first state information, Information processing methods.