Simulation device, simulation method, and program

Abstract

This simulation device is provided with: a selection unit that selects target coordinates in a structural space; an addition unit that adds a potential to the target coordinates; a calculation unit that calculates time evolution of the structural space in a first time interval by a first principle calculation; and a prediction unit that predicts time evolution in a second time interval on the basis of a prediction model in which the time is used as an explanatory variable and the target coordinates that have changed due to the application of an artificial force derived from the potential are used as objective variables.

Description

[Technical Field]

[0001] This disclosure relates to a simulation apparatus, a simulation method, and a program. [Background technology]

[0002] Techniques for obtaining free energy surfaces using molecular simulations are known. In this type of technique, reaction pathway search methods (metadynamics) are sometimes used to avoid getting stuck in local optima. Metadynamics is a method that enables global sampling by adding a Gaussian function type potential to sampled reaction coordinates, thereby avoiding resampling of already sampled reaction coordinates.

[0003] For example, Patent Document 1 discloses an invention that uses metadynamics to search for stable binding structures, with the dihedral angle between the target molecule and the drug candidate molecule as a collective variable, in order to efficiently search for multiple stable binding structures. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] International Publication No. 2019 / 130529 [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] However, conventional techniques have the problem of high computational cost to obtain the free energy surface. For example, even when using metadynamics, searching for the neighborhood of the local minimum involves repeated sampling in a narrow range, resulting in a large overall computational cost until the free energy surface is obtained.

[0006] One aspect of this disclosure aims to obtain a free energy surface with low computational cost, in view of the technical challenges described above. [Means for solving the problem]

[0007] This disclosure comprises the following configuration.

[0008] <1> A selection unit configured to select the target coordinates in the structural space, An addition unit configured to add a potential to the aforementioned target coordinates, A calculation unit configured to calculate the time evolution of the structural space in a first time interval by first-principles calculations, A prediction unit is configured to predict the time evolution in a second time interval based on a prediction model in which time is the explanatory variable and the target coordinates changed by the application of an artificial force derived from the potential are the dependent variable, A simulation device equipped with the following features.

[0009] <2> the above <1> The simulation device described above, The system further comprises a learning unit configured to generate the predictive model by learning the calculation results of the aforementioned time evolution. Simulation device.

[0010] <3> the above <1> or <2> The simulation device described above, The system further includes a decision unit configured to determine whether to execute the calculation unit or the prediction unit in the next time interval, based on the difference between the calculated result of the time evolution and the predicted result of the time evolution in the same time interval. Simulation device.

[0011] <4> the above <3> The simulation device described above, The determination unit is configured to be executed after the prediction unit has predicted the time evolution over a predetermined number of time intervals. Simulation device.

[0012] <5> The simulation device according to any one of <1> to <4> above, where the potential is a Gaussian function type potential, where the artificial force is such that b is the distance between the target coordinates in the current time interval and the target coordinates in the past time interval, a and c are fitting functions, and it is calculated by the following formula,

[0013]

Equation

[0014] <6> The simulation device according to <5> above, where the prediction model uses the target coordinates modulated based on the artificial force as the target variable, where the modulation is

[0017] using F as the artificial force, m as the mass of the particles forming the structure corresponding to the target coordinates, Δt as the length of the time interval, and it is calculated by the following formula,

[0015]

Equation

[0016] <7> A computer performs a procedure of selecting target coordinates in a structure space, a procedure of adding a potential to the target coordinates, a procedure of calculating the time evolution of the structure space in a first time interval by first principle calculation, a procedure of predicting the time evolution in a second time interval based on a prediction model that uses time as an explanatory variable and the target coordinates changed by adding an artificial force derived from the potential as the target variable, and performs a simulation method.​​​​​​​A procedure for adding a potential to the aforementioned target coordinates, A procedure for calculating the time evolution of the structural space in a first time interval by first-principles calculations, A procedure for predicting the time evolution in a second time interval based on a prediction model in which time is the explanatory variable and the target coordinates changed by the application of an artificial force derived from the potential are the dependent variable, A program to execute. [Effects of the Invention]

[0018] According to one aspect of this disclosure, a free energy surface can be obtained with low computational cost. [Brief explanation of the drawing]

[0019] [Figure 1] Figure 1 is a block diagram showing an example of the overall configuration of an information processing system. [Figure 2] Figure 2 is a block diagram showing an example of a computer hardware configuration. [Figure 3] Figure 3 is a block diagram showing an example of the functional configuration of an information processing system. [Figure 4] Figure 4 is a flowchart showing an example of a simulation method. [Modes for carrying out the invention]

[0020] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0021] [Embodiment] One embodiment of the present disclosure is an information processing system for performing molecular simulations. The information processing system in this embodiment has the function of performing molecular simulations to obtain a free energy surface.

[0022] There are techniques for calculating state variables such as free energy by simulating particles such as atoms or molecules. In this type of technique, the coordinates for calculating the state variables are sometimes sampled from the structural space, and the state variables of those coordinates are calculated using first-principles calculations or other methods, and this process is repeated. However, due to limitations such as computational cost relative to time or size, it may not be possible to obtain sufficient samples, potentially leading to the system becoming stuck in a local minimum.

[0023] Here, structural space refers to the space that encompasses all structures that represent the positional and bonding relationships of each particle constituting a substance. The coordinates of structural space correspond to a single structure and indicate the conformation of the particles contained within that structure. An example of structural space is the structural space of a reaction molecular system in which multiple molecules undergo a chemical reaction.

[0024] One method to prevent getting stuck in local minima is a reaction pathway search technique called metadynamics. In metadynamics, a potential based on a penalty function is added to the sampled coordinates, filling the free energy surface with the potential and avoiding returning to previously sampled coordinates. According to metadynamics, the probability of getting stuck in a local minima is reduced, and global sampling becomes possible.

[0025] The penalty function is often a Gaussian function. Hereafter, a potential based on a Gaussian function will also be called a "Gaussian potential."

[0026] In metadynamics, when filling a free energy surface with a potential, the process involves moving between similar coordinates before moving beyond the free energy surface in its vicinity. In other words, metadynamics involves repeated sampling within a narrow range during the search for the local optimum. Therefore, obtaining a free energy surface using metadynamics requires significant computational cost for sampling in the vicinity of the local optimum. On the other hand, since the conformations that a particle can take are limited in the vicinity of the local optimum, it is expected that the sampling behavior can be predicted quickly and accurately by learning these conformations.

[0027] One embodiment of this disclosure aims to obtain a free energy surface with low computational cost. In this embodiment, the time evolution of the structural space in a first time interval is calculated by first-principles calculations, and the time evolution of the structural space in a second time interval is predicted based on a predictive model that has learned the calculation results of the time evolution. In one aspect, according to this embodiment, the free energy surface of the reaction molecular system can be obtained in a short time and with high accuracy.

[0028] <Overall Structure> The overall configuration of the information processing system in this embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the information processing system in this embodiment.

[0029] As shown in Figure 1, the information processing system 1000 in this embodiment includes a simulation device 10 and a terminal device 20. The simulation device 10 and the terminal device 20 are connected via a communication network N1 such as a LAN (Local Area Network) or the Internet, enabling data communication.

[0030] The simulation device 10 is an information processing device such as a personal computer, workstation, or server that performs molecular simulations. The simulation device 10 receives simulation conditions from the terminal device 20, which indicate the conditions for performing the molecular simulation. The simulation device 10 performs the molecular simulation according to the simulation conditions and transmits the simulation results to the terminal device 20. In this embodiment, the molecular simulation is, as an example, a simulation to obtain the free energy surface in a reaction molecular system.

[0031] Terminal device 20 is an information processing terminal such as a personal computer, smartphone, or tablet terminal operated by a user of the information processing system 1000. Terminal device 20 transmits the simulation conditions entered by the user to the simulation device 10. Terminal device 20 displays the simulation results received from the simulation device 10 to the user.

[0032] The overall configuration of the information processing system 1000 shown in Figure 1 is just one example, and various system configurations are possible depending on the application and purpose. For example, one or more simulation devices 10 and terminal devices 20 may be included in the information processing system 1000. For example, the simulation device 10 may be implemented using multiple computers, or it may be implemented as a cloud computing service. The classification of devices such as the simulation device 10 and terminal device 20 shown in Figure 1 is just one example.

[0033] <Hardware Configuration> The hardware configuration of the information processing system 1000 in this embodiment will be described with reference to Figure 2.

[0034] Computers In this embodiment, the simulation device 10 and terminal device 20 are implemented, for example, by a computer. Figure 2 is a block diagram showing an example of the hardware configuration of the computer 500 in this embodiment.

[0035] As shown in Figure 2, the computer 500 includes a CPU (Central Processing Unit) 501, ROM (Read Only Memory) 502, RAM (Random Access Memory) 503, HDD (Hard Disk Drive) 504, input device 505, display device 506, communication interface 507, and external interface 508. The CPU 501, ROM 502, and RAM 503 form what is known as a computer. Each piece of hardware in the computer 500 is interconnected via a bus line 509. The input device 505 and display device 506 may also be used by connecting them to the external interface 508.

[0036] The CPU 501 is a processing unit that controls and implements the overall functions of the computer 500 by reading programs and data from storage devices such as the ROM 502 or HDD 504 onto the RAM 503 and executing processing.

[0037] ROM502 is an example of non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. ROM502 functions as the main memory, storing various programs and data necessary for the CPU501 to execute the programs installed on HDD504. Specifically, ROM502 stores boot programs such as BIOS (Basic Input / Output System) and EFI (Extensible Firmware Interface) that are executed when the computer 500 starts up, as well as OS (Operating System) settings, network settings, and other data.

[0038] RAM503 is an example of volatile semiconductor memory (storage device) whose programs and data are erased when the power is turned off. RAM503 includes, for example, DRAM (Dynamic Random Access Memory) and SRAM (Static Random Access Memory). RAM503 provides a working area that is expanded when various programs installed on HDD504 are executed by CPU501.

[0039] HDD504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in HDD504 include the operating system (OS), which is the basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that computer 500 may use a storage device that uses flash memory as its storage medium (e.g., SSD: Solid State Drive) instead of HDD504.

[0040] The input device 505 includes a touch panel used by the user to input various signals, operation keys and buttons, a keyboard and mouse, and a microphone for inputting sound data such as voice.

[0041] The display device 506 consists of a display such as a liquid crystal or organic EL (Electro-Luminescence) that displays a screen, and a speaker that outputs sound data such as audio.

[0042] Communication I / F 507 is an interface that connects to a communication network and allows computer 500 to perform data communication.

[0043] External I / F 508 is an interface for external devices. Examples of external devices include the drive device 510.

[0044] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memory that records information electrically, such as ROMs and flash memory. This allows the computer 500 to read and / or write to the recording medium 511 via the external I / F 508.

[0045] The various programs to be installed on the HDD 504 are installed, for example, when the distributed recording medium 511 is set in a drive device 510 connected to an external I / F 508, and the various programs recorded on the recording medium 511 are read by the drive device 510. Alternatively, the various programs to be installed on the HDD 504 may be downloaded via the communication I / F 507 from a network other than the communication network and installed that way.

[0046] <Functional Configuration> The functional configuration of the information processing system in this embodiment will be described with reference to Figure 3. Figure 3 is a block diagram showing an example of the functional configuration of the information processing system in this embodiment.

[0047] <<Simulation device>> As shown in Figure 3, the simulation device 10 in this embodiment comprises a construction unit 101, a selection unit 102, an addition unit 103, a calculation unit 104, a learning unit 105, a storage unit 106, a prediction unit 107, a decision unit 108, and an output unit 109.

[0048] The construction unit 101, selection unit 102, addition unit 103, calculation unit 104, learning unit 105, prediction unit 107, decision unit 108, and output unit 109 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501. The storage unit 106 is implemented by the HDD 504 shown in Figure 2.

[0049] The construction unit 101 receives simulation conditions from the terminal device 20. The simulation conditions include information indicating multiple substances that will cause a desired chemical reaction. The construction unit 101 constructs the structural space of the reaction molecular system according to the simulation conditions.

[0050] The selection unit 102 selects the reaction coordinates to be used for calculating the state variable from the structural space constructed by the construction unit 101. In this embodiment, as an example, the state variable is free energy. Hereinafter, the reaction coordinates to be used for calculating the free energy will also be referred to as the "target coordinates".

[0051] The addition unit 103 adds a potential to the target coordinates selected by the selection unit 102. In this embodiment, as an example, a Gaussian function type potential is added. The Gaussian function type potential to be added to the target coordinates is generated based on the structure corresponding to those target coordinates.

[0052] The calculation unit 104 calculates the time evolution of the structural space using first-principles calculations. In this embodiment, the first-principles calculation is, as an example, a calculation method based on Density Functional Theory (DFT). The calculation unit 104 calculates the time evolution in a predetermined time interval. The length of the time interval corresponds to the time step size in molecular simulations. The time step size may be set arbitrarily.

[0053] In this embodiment, the calculation unit 104 changes the target coordinates in the next time interval by applying an artificial force to the target coordinates in the previous time interval. The artificial force is derived from the Gaussian function type potential added to the structural space by the addition unit 103.

[0054] Specifically, artificial force is defined by F as shown in equation (1). meta That is the case. Furthermore, artificial force F meta This is applied to each particle included in the structure corresponding to the target coordinates. For simplicity, equation (1) omits the index indicating the particles, but for each particle, an artificial force F is applied. metais calculated.

[0055] [Number]

[0056] However, b is the distance between the target coordinates in the current time interval and the target coordinates in the past time interval, and a and c are fitting functions.

[0057] An artificial force F meta is applied to the target coordinates, and the target coordinates are subject to the modulation R meta defined by Equation (2). Since the artificial force F meta is applied to each particle, the modulation R meta is also calculated for each particle. In Equation (2), similar to Equation (1), the index indicating the particle is omitted.

[0058] [Number]

[0059] However, F meta is the artificial force, m is the mass of the particle, and Δt is the length of the time interval (time step width).

[0060] The learning unit 105 generates a prediction model by learning the calculation results of the time evolution calculated by the calculation unit 104. When there is a learned prediction model, the learning unit 105 updates the prediction model by learning the calculation results of the time evolution newly calculated by the calculation unit 104.

[0061] The prediction model is a machine learning model that uses time as an explanatory variable and coordinates as an objective variable. The coordinates output by the prediction model are the target coordinates sampled at the input time. The target coordinates calculated by the calculation unit 104 are subject to the artificial force F meta calculated by Equation (1), and thus are subject to the modulation R metaThese are the target coordinates that have been subjected to the artificial force F. Therefore, the coordinates output by the prediction model are the target coordinates in the previous time interval with the artificial force F. meta The addition of modulation R meta These are the predicted coordinates of the target that were received.

[0062] The structure of the prediction model can be arbitrarily selected, but for example, the MD-GAN (Molecular dynamics - Generative Adversarial Network) disclosed in Reference 1 can be used. MD-GAN is a multilayer neural network that can predict the time evolution in molecular dynamics calculations within the framework of a generative adversarial network.

[0063] [Reference 1] Katsuhiro Endo, Katsufumi Tomobe, Kenji Yasuoka, "Multi-Step Time Series Generator for Molecular Dynamics," Thirty-Second AAAI Conference on Artificial Intelligence, Vol. 32, No. 1 (2018)

[0064] The memory unit 106 stores the trained prediction model. The prediction model stored in the memory unit 106 is generated or updated by the learning unit 105.

[0065] The prediction unit 107 predicts the time evolution of the structural space based on the prediction model read from the storage unit 106. The prediction unit 107 predicts the time evolution in a predetermined time interval. The time interval in which the prediction unit 107 predicts the time evolution is different from the time interval in which the calculation unit 104 calculates the time evolution. The length of the time interval may be the same as the length of the time interval in which the calculation unit 104 calculates the time evolution. Therefore, the prediction unit 107 inputs the time when a predetermined time interval has elapsed from the current time (hereinafter also referred to as the "target time") into the prediction model. The prediction unit 107 obtains the prediction result, including the target coordinates output from the prediction model.

[0066] The decision unit 108 determines whether to execute the calculation unit 104 or the prediction unit 107 based on the calculation results of the time evolution calculated by the calculation unit 104 and the prediction results of the time evolution predicted by the prediction unit 107. The decision unit 108 determines whether to execute the calculation unit 104 or the prediction unit 107 in the next time interval by comparing the error between the calculation results and prediction results in the same time interval with a predetermined threshold.

[0067] The output unit 109 determines whether the simulation termination condition has been met. In this embodiment, the termination condition is that the desired chemical reaction has occurred. That is, the output unit 109 determines whether the desired chemical reaction has occurred in the current structural space.

[0068] The output unit 109 outputs the simulation results when the simulation termination conditions are met. The simulation results include state information that indicates the current state of the structural space. State information is, for example, information that indicates the free energy surface.

[0069] <Terminal Devices> As shown in Figure 3, the terminal device 20 in this embodiment includes an input unit 201 and a display unit 202.

[0070] The input unit 201 and the display unit 202 are realized by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0071] The input unit 201 accepts the input of simulation conditions in response to user operations. The input unit 201 transmits the simulation conditions to the simulation device 10.

[0072] The display unit 202 receives the simulation results from the simulation device 10. The display unit 202 outputs the simulation results to the display device 506 or the like.

[0073] <Processing Procedure> The simulation method performed by the information processing system 1000 in this embodiment will be described with reference to Figure 4. Figure 4 is a flowchart showing an example of the simulation method in this embodiment.

[0074] In step S1, the input unit 201 of the terminal device 20 accepts the input of simulation conditions in response to user operation. The simulation conditions include information indicating multiple substances that will cause the chemical reaction desired by the user. Next, the input unit 201 transmits the simulation conditions to the simulation device 10.

[0075] In the simulation device 10, the construction unit 101 receives simulation conditions from the terminal device 20. Next, the construction unit 101 constructs the structural space of the reaction molecular system according to the simulation conditions. Then, the construction unit 101 sends the constructed structural space to the selection unit 102.

[0076] In step S2, the selection unit 102 of the simulation device 10 receives the structural space from the construction unit 101. Next, the selection unit 102 uses metadynamics to select target coordinates from the structural space. Then, the selection unit 102 sends the structural space with the selected target coordinates to the addition unit 103.

[0077] In step S3, the addition unit 103 of the simulation device 10 receives the structural space from the selection unit 102. Next, the addition unit 103 generates a Gaussian function type potential based on the structure corresponding to the target coordinates. Subsequently, the addition unit 103 adds the Gaussian function type potential to the target coordinates. Then, the addition unit 103 sends the structural space with the Gaussian function type potential added to the target coordinates to the calculation unit 104.

[0078] In step S4, the calculation unit 104 of the simulation device 10 receives the structural space from the addition unit 103. Next, the calculation unit 104 calculates the time evolution of the structural space in the next time interval using first-principles calculations. Then, the calculation unit 104 sends the calculation result of the time evolution to the learning unit 105.

[0079] Specifically, the calculation unit 104 first calculates the artificial force F from the Gaussian function type potential added in step S3 according to equation (1). meta Next, the calculation unit 104 calculates the artificial force F according to equation (2). meta The modulation R that the target coordinates receive when this is added meta Next, the calculation unit 104 calculates the modulation R meta Based on this, the structure corresponding to the target coordinates is changed, and the target coordinates corresponding to the changed structure are selected.

[0080] The calculation unit 104 repeatedly executes step S4 until a sufficient amount of calculation results have been accumulated to generate a prediction model. The amount sufficient to generate a prediction model is predetermined according to the type of prediction model. Once a sufficient amount of calculation results has been accumulated, the calculation unit 104 sends the structure space to the prediction unit 107.

[0081] In step S5, the learning unit 105 of the simulation device 10 receives the time evolution calculation results from the calculation unit 104. Next, the learning unit 105 learns the time evolution calculation results to generate a predictive model with time as the explanatory variable and coordinates as the dependent variable. Then, the learning unit 105 stores the learned predictive model in the storage unit 106.

[0082] In step S6, the prediction unit 107 of the simulation device 10 receives the structure space from the calculation unit 104. Next, the prediction unit 107 reads out the trained prediction model from the storage unit 106.

[0083] Next, the prediction unit 107 calculates the target time by adding the length of the next time interval to the current time. Then, the prediction unit 107 inputs the target time into the trained prediction model. The prediction model predicts the target coordinates at the input target time. At this time, a Gaussian function type potential based on the structure corresponding to the target coordinates is added to the target coordinates. The prediction model outputs the predicted value of the target coordinates. The prediction unit 107 obtains the prediction result output by the prediction model. The prediction result includes the target coordinates when the next time interval has elapsed.

[0084] The prediction unit 107 repeatedly executes step S6 until it predicts the time evolution over a predetermined number of time intervals. The number of repetitions can be set arbitrarily, but for example, it may be 100 times. Once the prediction unit 107 predicts the time evolution over the predetermined number of time intervals, it sends the time evolution prediction results to the determination unit 108.

[0085] In step S7, the calculation unit 104 of the simulation device 10 calculates the time evolution of the structural space in the current time interval using first-principles calculations. In other words, the calculation unit 104 calculates the time evolution in the current time interval based on the prediction result of the time evolution in the previous time interval. The calculation unit 104 sends the calculation result of the time evolution to the determination unit 108.

[0086] In step S8, the decision unit 108 of the simulation device 10 receives the prediction result of the time evolution in the current time interval from the prediction unit 107. The decision unit 108 also receives the calculation result of the time evolution in the current time interval from the calculation unit 104.

[0087] Next, the decision unit 108 compares the predicted result of time evolution in the current time interval with the calculated result. Subsequently, the decision unit 108 determines whether the prediction error is below a predetermined threshold. The prediction error is the distance between the target coordinates indicated by the prediction result and the target coordinates indicated by the calculation result. In other words, the decision unit 108 determines whether the target coordinates predicted by the prediction model in the current time interval and the target coordinates calculated by the first-principles calculation are significantly different.

[0088] If the prediction error is below the threshold (YES), the determination unit 108 sends the structure space to the output unit 109 and proceeds to step S9. On the other hand, if the prediction error is greater than the threshold (NO), the determination unit 108 sends the structure space to the addition unit 103 and returns to step S3.

[0089] In step S9, the output unit 109 of the simulation device 10 receives the structural space from the determination unit 108. Next, the output unit 109 determines whether or not the desired chemical reaction has occurred based on the structural space. If the chemical reaction has occurred (YES), the output unit 109 proceeds to step S10. On the other hand, if the chemical reaction has not occurred (NO), the output unit 109 returns to step S6.

[0090] In step S10, the output unit 109 of the simulation device 10 outputs the simulation results. The simulation results include information indicating the free energy surface. The output unit 109 transmits the simulation results to the terminal device 20.

[0091] In the terminal device 20, the display unit 202 receives the simulation results from the simulation device 10. Next, the display unit 202 outputs the simulation results to the display device 506. The user can refer to the simulation results output to the display device 506 of the terminal device 20 and use the calculation results of the free energy surface.

[0092] For example, a user can use the free energy surface calculation results to search for raw materials in a reaction molecular system. Alternatively, a user may obtain actual measured values ​​of the reaction molecular system through experiments or manufacturing, update the prediction model based on the simulation results and measured values, and repeatedly perform the search and prediction of reaction pathways.

[0093] After returning from step S8 to step S3, the simulation device 10 executes the processes from step S3 to step S8 again. In the second and subsequent executions of step S5, the trained prediction model is updated based on the new calculation result calculated in the previous step S4. On the other hand, after returning from step S9 to step S6, the simulation device 10 executes the processes from step S6 to step S8 again. That is, the decision unit 108 decides whether to execute the calculation unit 104 or the prediction unit 107 in the next time interval based on the error between the calculation result and the prediction result in the same time interval.

[0094] In this way, the simulation device 10 rapidly predicts the time evolution of the structural space using the prediction model as long as the prediction error by the prediction model is below a threshold, and when the prediction error exceeds the threshold, it calculates the time evolution of the structural space with high accuracy using first-principles calculations. As a result, the simulation device 10 can calculate the time evolution of the structural space at high speed and with high accuracy, and consequently, it can obtain the free energy surface at high speed and with high accuracy.

[0095] <Effects of the Embodiment> In this embodiment, the simulation device 10 adds a potential to the target coordinates in the structural space, calculates the time evolution of the structural space in a first time interval by first-principles calculation, and predicts the time evolution of the structural space in a second time interval based on a prediction model. The prediction model uses time as the explanatory variable and the target coordinates, which have changed due to the application of artificial forces derived from the potential, as the dependent variable. By predicting the time evolution of the structural space in a portion of the time interval, the simulation device 10 can reduce the number of first-principles calculations. Therefore, according to this embodiment, the free energy surface can be obtained with low computational cost.

[0096] The simulation device 10 generates a predictive model by learning the results of time evolution calculations. The simulation device 10 updates the predictive model each time it calculates time evolution using first-principles calculations. Therefore, according to this embodiment, the time evolution of structural space can be predicted with high accuracy.

[0097] The simulation device 10 determines whether to perform a first-principles calculation or run a prediction model in the next time interval based on the difference between the calculated time evolution result and the predicted time evolution result in the same time interval. If the target coordinates move significantly due to time evolution, the accuracy of the prediction result may decrease. Therefore, according to this embodiment, the free energy surface can be obtained with high accuracy.

[0098] The simulation device 10 predicts the time evolution over a predetermined number of time intervals and then calculates the time evolution using first-principles calculations. The more times the simulation device 10 runs the prediction model, the lower the computational cost becomes. Therefore, according to this embodiment, the free energy surface can be obtained with significantly lower computational cost.

[0099] [supplement] Each of the embodiments described above can be implemented by one or more processing circuits. Hereinafter, "processing circuit" as used herein includes processors programmed to execute each function by software, such as CPUs (Central Processing Units) or GPUs (Graphics Processing Units) implemented by electronic circuits, as well as devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules designed to execute each of the functions described above.

[0100] While embodiments of the present disclosure have been described in detail above, the embodiments disclosed herein are illustrative and not restrictive in all respects. The embodiments can be modified and improved in various ways without departing from the scope and spirit of the appended claims. The features described in the above embodiments can be combined in any way that is not inconsistent with other configurations.

[0101] This application claims priority to Japanese Patent Application No. 2023-118565, filed with the Japan Patent Office on 20 July 2023, which is incorporated herein by reference to its entire contents. [Explanation of Symbols]

[0102] 10 Simulation device 20 Terminal devices 101 Construction Department 102 Selection Section 103 Addition section 104 Calculation section 105 Learning Department 106 Storage section 107 Prediction Section 108 Decision Section 109 Output section 201 Input section 202 Display section 1000 Information Processing Systems

Claims

1. A selection unit configured to select the target coordinates in the structural space, An addition unit configured to add a potential to the aforementioned target coordinates, A calculation unit configured to calculate the time evolution of the structural space in a first time interval by first-principles calculations, A prediction unit is configured to predict the time evolution in a second time interval based on a prediction model in which time is the explanatory variable and the target coordinates changed by the application of an artificial force derived from the potential are the dependent variable, A simulation device equipped with the following features.

2. A simulation apparatus according to claim 1, The system further comprises a learning unit configured to generate the predictive model by learning the calculation results of the aforementioned time evolution. Simulation device.

3. A simulation apparatus according to claim 1, The system further includes a decision unit configured to determine whether to execute the calculation unit or the prediction unit in the next time interval, based on the difference between the calculated result of the time evolution and the predicted result of the time evolution in the same time interval. Simulation device.

4. A simulation apparatus according to claim 3, The determination unit is configured to be executed after the prediction unit has predicted the time evolution over a predetermined number of time intervals. Simulation device.

5. A simulation apparatus according to any one of claims 1 to 4, The aforementioned potential is a Gaussian function type potential, The aforementioned artificial force is calculated by the following equation, where b is the distance between the target coordinates in the current time interval and the target coordinates in past time intervals, and a and c are fitting functions: [Math 1] Simulation device.

6. A simulation apparatus according to claim 5, The prediction model uses the target coordinates that have been modulated based on the artificial force as the objective variable. The aforementioned modulation is F meta Let be the artificial force, let m be the mass of the particle forming the structure corresponding to the target coordinate, and let Δt be the length of the time interval, and it is calculated by the following equation: [Math 2] Simulation device.

7. Computers Procedure for selecting target coordinates in structural space, A procedure for adding a potential to the aforementioned target coordinates, A procedure for calculating the time evolution of the structural space in a first time interval by first-principles calculations, A procedure for predicting the time evolution in a second time interval based on a prediction model in which time is the explanatory variable and the target coordinates changed by the application of an artificial force derived from the potential are the dependent variable, A simulation method for performing this.

8. On the computer, Procedure for selecting target coordinates in structural space, A procedure for adding a potential to the aforementioned target coordinates, A procedure for calculating the time evolution of the structural space in a first time interval by first-principles calculations, A procedure for predicting the time evolution in a second time interval based on a prediction model in which time is the explanatory variable and the target coordinates changed by the application of an artificial force derived from the potential are the dependent variable, A program to execute.

Images