Method for determining physical properties of actinide metal oxides based on machine learning

Abstract

The invention discloses a method for determining physical properties of actinide metal oxide based on machine learning, and relates to the field of crystal simulation, and the method comprises the following steps: obtaining actinide metal elementary substances and actinide metal oxide structures; according to the actinide metal elementary substance and the actinide metal oxide structure, a first principle calculation method is combined with a dynamic machine learning method of atomic scale material simulation calculation software VASP to train a machine learning force field of a corresponding system; evaluating the trained machine learning force field; when an evaluation result meets a set requirement, determining the mechanical property, phonon dispersion and thermodynamic property of the actinide metal oxide by utilizing a trained machine learning force field; the setting requirement is that the fitting error of the force and the fitting error of the energy both meet the standard. The speed and accuracy of determining the mechanical property, phonon dispersion and thermodynamic property of the actinide metal oxide can be improved.

Description

Technical Field

[0001] The present application relates to the field of crystal simulation, and in particular to a method for determining the physical properties of actinide metal oxides based on machine learning. Background Art

[0002] Actinide metals are a general term for 15 chemical elements with atomic numbers of 89-103 in group IIIB of the periodic table, also known as the 5f transition system. The outermost electron configurations of these elements are basically the same, and only the electrons are replaced in the 5f inner layer. Actinide metals are extremely chemically active, very sensitive to the surrounding environment, and easily react with the ambient atmosphere. When exposed to air and moisture, the metal surface is easily oxidized and corroded, forming a corrosion layer on the surface. The metal corrosion layer affects the performance and storage management of metal materials. The structure and composition of the oxide layer are complex, often containing metal elements of different valence states, forming oxides of different stoichiometric ratios. These oxides are often complex in structure and low in symmetry. Studying the properties of actinide metal oxides can help understand the corrosion process of actinide metals. Using density functional theory to study such problems often requires calculating the physicochemical properties of the material, among which the physicochemical properties include mechanical properties, phonon dispersion and thermodynamic properties. The stress-strain method in the first principles is often used to calculate mechanical properties. This method applies different strains to the structure, calculates the corresponding stress, and then uses the results to fit the elastic constants. The finite displacement method is often used to calculate phonon dispersion. This method first generates several displacement structures based on the symmetry of the structure, and then calculates the force conditions of the atoms after the displacement is applied to infer the force constants of the system, and then obtains the mechanical properties or phonon dispersion of the material. Both methods will generate several different structure files, each of which requires separate computing resources. Especially for low-symmetry structures or large system structures, the required computing resources are huge, time-consuming and labor-intensive, and the computing efficiency is low.

[0003] Therefore, based on the above problems, there is an urgent need to provide a new method for calculating the physical properties of actinide metal oxides. Summary of the invention

[0004] The purpose of this application is to provide a method for determining the physical properties of actinide metal oxides based on machine learning, which can improve the speed and accuracy of calculating the mechanical properties, phonon dispersion and thermodynamic properties of actinide metal oxides.

[0005] To achieve the above objectives, this application provides the following solutions: The present application provides a method for determining the physical properties of actinide metal oxides based on machine learning, and the method for determining the physical properties of actinide metal oxides based on machine learning includes: Obtain the structures of actinide metal elements and actinide metal oxides; According to the structure of actinide metals and actinide metal oxides, the first-principles calculation method is combined with the dynamic machine learning method of the atomic-scale material simulation calculation software VASP to train the machine learning force field of the corresponding system; Evaluate the trained machine learning force field; When the evaluation results meet the set requirements, the trained machine learning force field is used to calculate the mechanical properties, phonon dispersion and thermodynamic properties of actinide metal oxides; the set requirements are that the fitting error of force and the fitting error of energy meet the standards.

[0006] Optionally, the Brillouin zone in the atomic-scale material simulation calculation software VASP is 1×1×1 Gamma lattice; the cutoff energy is set to 520 eV, the NPT ensemble ISIF=3, the Langevin thermal bath MDALGO=3, the atomic friction coefficient and the lattice friction coefficient are 10 ps -1 , the starting and ending temperatures are set to 250 K and 400 K respectively, the step size is 1 fs, and the relaxation time is 20 ps.

[0007] Optionally, the evaluating the trained machine learning force field specifically includes: The force and energy errors during the training process are estimated to determine the force fitting error and the energy fitting error respectively; Determine whether the fitting error of force is distributed between 0.03 eV / Å and 0.1 eV / Å and whether the fitting error of energy is less than 1 meV / atom.

[0008] Optionally, the calculating the mechanical properties, phonon dispersion and thermodynamic properties of the actinide metal oxides using the trained machine learning force field also includes: Determine the inner core electrons of actinide metal elements and actinide metal oxide structures based on PAW pseudopotentials; The exchange-correlation functional is determined using the PBE functional within the framework of the generalized gradient approximation.

[0009] Optionally, the determining of the mechanical properties, phonon dispersion and thermodynamic properties of the actinide metal oxide using the trained machine learning force field specifically includes: Based on the trained machine learning force field, combined with Phonopy software, the finite position method is used to determine the phonon dispersion of actinide metal oxides; The mechanical properties were determined using the stress-strain method based on the trained machine learning force field combined with the Elas Tool software.

[0010] Optionally, the thermodynamic properties of the actinide metal oxide are obtained based on phonon dispersion; the thermodynamic properties include: phonon vibrational free energy, entropy and isovolumetric molar heat capacity.

[0011] Optionally, the mechanical properties include: elastic constants, bulk modulus of the structure, elastic modulus, shear modulus, Poisson's ratio and Debye temperature.

[0012] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a method for determining the physical properties of actinide metal oxides based on machine learning, wherein the machine learning force field of the corresponding system is trained by using a first-principles calculation method combined with a dynamic machine learning method of an atomic-scale material simulation calculation software (Vienna Ab-initio Simulation Package, VASP) according to actinide metal elements and actinide metal oxide structures; and when the evaluation results meet the set requirements, the trained machine learning force field is used to determine the mechanical properties, phonon dispersion and thermodynamic properties of the actinide metal oxides; the machine learning force field is used to quickly calculate the mechanical properties, phonon dispersion and thermodynamic properties of the actinide metal oxides; the machine learning force field obtained in the present application has an accuracy equivalent to that of DFT calculations, which can significantly accelerate the calculation of mechanical properties and phonon dispersion of actinide metal oxides, has the advantages of saving calculation costs, simple operation, good universality, etc., and provides a specific technical route and theoretical guidance for the calculation of material properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 This is a schematic flow chart of a method for determining the physical properties of actinide metal oxides based on machine learning in one embodiment of the present application; Figure 2 Schematic diagram of the PuO2 crystal structure in Example 1 of the present application; Figure 3 This is a schematic diagram of a force fitting error curve during the machine learning force field training process in Example 1 of the present application; Figure 4 This is a schematic diagram of the energy fitting error curve during the machine learning force field training process in Example 1 of the present application; Figure 5 This is a schematic diagram of the phonon spectrum of the PuO2 structure calculated using the machine learning force field in Example 1 of the present application; Figure 6 Schematic diagram of the phonon vibration free energy, entropy and isovolumetric molar heat capacity of PuO2 structure calculated using machine learning force field in Example 1 of the present application; Figure 7Schematic diagram of bulk modulus, elastic modulus, shear modulus and Poisson's ratio of the PuO2 structure calculated using a machine learning force field in Example 1 of the present application; Figure 8 Schematic diagram of the Pu2O3 crystal structure in Example 2 of the present application; Fig. 9 This is a schematic diagram of a force fitting error curve during the machine learning force field training process in Example 2 of the present application; Fig.10 This is a schematic diagram of the energy fitting error curve during the machine learning force field training process in Example 2 of the present application; Fig.11 A schematic diagram of the phonon spectrum of the Pu2O3 structure calculated using a machine learning force field in Example 2 of the present application; Fig.12 Schematic diagram of the phonon vibration free energy, entropy and isovolumetric molar heat capacity of the Pu2O3 structure calculated using the machine learning force field in Example 2 of the present application; Fig.13 Schematic diagram of bulk modulus, elastic modulus, shear modulus and Poisson's ratio of Pu2O3 structure calculated using machine learning force field in Example 2 of the present application; DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0016] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0017] In an exemplary embodiment, Figure 1 As shown, a method for determining the physical properties of actinide metal oxides based on machine learning is provided, and the method includes the following S101 to S104. Among them: S101, obtaining structures of actinide metal elements and actinide metal oxides; S102, based on the structure of actinide metals and actinide metal oxides, the first-principles calculation method is combined with the dynamic machine learning method of the atomic-scale material simulation calculation software VASP to train the machine learning force field of the corresponding system; VASP is used for material simulation and computational material science research; In a specific embodiment, if the object of study is a plutonium oxide structure, the corresponding system is a Pu-O system; In VASP, a 1×1×1 Gamma grid was selected for the Brillouin zone, the cutoff energy was set to 520 eV, the NPT ensemble (ISIF=3), the Langevin thermal bath (MDALGO=3), and the atomic and lattice friction coefficients were set to 10 ps. -1 (LANGEVIN_GAMMA and LANGEVIN_GAMMA_L), the start temperature (TEBEG) and end temperature (TEEND) are set to 250K and 400K respectively, the step size is 1fs (POTIM=1), and the relaxation time is 20ps (NSW=20000).

[0018] The machine learning force field was trained using the dynamic machine learning method of VASP (ML_LMLFF = TRUE, ML_MODE = train) and combined with first principles molecular dynamics (IBRION = 0).

[0019] The specific process of training the machine learning force field includes: The first step is to use the actinide metal single substance structure for training. The output file ML_ABN file obtained by training contains the structure file selected from the first-principles molecular dynamics calculation results and the energy and stress information calculated based on the first-principles calculation; In the second step, ML_ABN is renamed as ML_AB, calculated simultaneously with the actinide metal oxide structure file, and the second step of training is performed; In the third step, rename the ML_ABN obtained in the second step to ML_AB, keep other input files unchanged, adjust ML_MODE=refit, perform the singular value decomposition method with Tikhonov regularization to obtain more accurate results, and disable the wrong prediction when training the force field to further optimize the force field and improve the calculation speed.

[0020] Following the above steps, we can train a machine learning force field for actinide metal oxides in the form of a binary file named ML_FFN. When using the machine learning force field, rename ML_FFN to ML_FF and adjust ML_MODE=run.

[0021] After training, VASP outputs a file named ML_LOGFILE, which contains relevant information about the training process, such as the machine learning force field parameter settings, the memory required for calculation, the status of each calculation step (force, energy prediction error, and whether to update the force field, etc.). The machine learning force field is evaluated by extracting the fitting error of the force calculated at each step of the training process.

[0022] S103, evaluating the trained machine learning force field; S103 specifically includes: Extract the calculation errors of force and energy during training (the data is stored in the calculation output file named ML_LOGFILE), determine the fitting error of force and energy respectively; calculate the accuracy of the total energy and lattice constant of the system to improve the acceleration effect of machine learning force field; Determine whether the fitting error distribution of force is between 0.03 eV / Å ~0.1eV / Å and whether the fitting error of energy is less than 1meV / atom.

[0023] S104, when the evaluation results meet the set requirements, use the trained machine learning force field to determine the mechanical properties, phonon dispersion and thermodynamic properties of the actinide metal oxides; the set requirements are that the fitting error of the force and the fitting error of the energy meet the standards.

[0024] Before S104, it also included: Determine the inner core electrons of actinide metal elements and actinide metal oxide structures based on PAW pseudopotentials; The exchange-correlation functional adopts the PBE functional under the framework of generalized gradient approximation (GGA).

[0025] Conventional unit cells are used in the simulation calculation process, and the structure files of actinide metal oxides are obtained from the material database or through structural optimization; S104 specifically includes: Based on the trained machine learning force field, combined with Phonopy software, the finite position method is used to determine the phonon dispersion and thermodynamic properties of actinide metal oxides; According to the trained machine learning force field combined with Elas Tool software, the stress-strain method is used to determine the mechanical properties. The actinide metal oxide structure file is placed in the folder where the Elas Tool software environment is installed. Using the stress-strain method, change the elastool.in file according to the calculation needs and run it to generate a series of VASP calculation input files. Add ML_LMLFF = TRUE and ML_MODE=run one by one in INCAR, and put the force field file ML_FF in the same path. Submit the calculation to quickly calculate the bulk modulus, elastic modulus, shear modulus, Poisson's ratio and other mechanical properties of the structure; Example 1 This example uses PuO2 to introduce a method for quickly calculating its mechanical properties, phonon dispersion, and thermodynamic properties based on a machine learning force field.

[0026] a) Obtaining the crystal structure; Get the PuO2 oxide structure at room temperature and pressure from Material Projects, such as Figure 2 (The red balls are oxygen atoms and the black balls are plutonium atoms).

[0027] b) Dynamic machine learning methods to train machine learning force fields; During the training process, the cutoff radius of the radial descriptor and the angular descriptor was adjusted to 5Å (ML_RCUT1=ML_RCUT2=5.0), the weight of the occupied kernel function was 0.5 (ML_SION1=ML_SION2=0.5), and 23 and 15 basis functions were used for expansion respectively (ML_MRB1=23=ML_MRB2=15), and the other parameters were set as above. According to the above training steps and training, the machine learning force field MLFF was obtained.

[0028] c) Evaluate the machine learning force field; Extract the fitting errors of force and energy from the training process respectively (see Figure 3 , Figure 4 ), the accuracy of calculating the total energy and lattice constant of the system (see Table 1), and the acceleration effect of the machine learning force field method were evaluated (see Table 2).

[0029] Table 1 Lattice constants and system energy of PuO2 structure calculated by different methods

[0030] Table 2 Comparison of DFT and MLFF calculation speed

[0031] d) Apply machine learning force fields to calculate the mechanical properties, phonon dispersion and thermodynamic properties of PuO2; VASP combined with Phonopy software uses the finite position method to calculate the structural phonon spectrum. The expansion cell size is 2×2×2, and two displacement structure files are generated. The calculated PuO2 phonon dispersion curve is as follows: Figure 5 As shown, the phonon vibration free energy, entropy and isovolumetric molar heat capacity are further calculated according to the phonon dispersion (see Figure 6 ).

[0032] VASP combined with Elas Tool software uses the stress-strain method to calculate the structural mechanical properties of PuO2, such as Figure 7 As shown, a total of 4 structures with applied strain were calculated, and the results are shown in Table 3.

[0033] Table 3 Mechanical properties parameters of PuO2

[0034] In Table 3, MP-1959 represents the number of the crystal configuration obtained by Material Project (MP website is a crystal database).

[0035] Example 2 This example uses Pu2O3 to introduce a method for quickly calculating its mechanical properties, phonon dispersion, and thermodynamic properties based on a machine learning force field.

[0036] a) Obtaining the crystal structure; Get the Pu2O3 oxide structure at room temperature and pressure from Material Projects, such as Figure 8 (The red balls are oxygen atoms, and the black balls are plutonium atoms); b) Dynamic machine learning methods to train machine learning force fields; c) Evaluate the machine learning force field; Extract the fitting errors of force and energy from the training process respectively (e.g. Fig. 9 , Fig.10 ), the accuracy of calculating the total energy and lattice constant of the system (see Table 4).

[0037] Table 4 Lattice constants and system energy of Pu2O3 structure calculated by different methods

[0038] d) Apply machine learning force fields to calculate the mechanical properties, phonon dispersion and thermodynamic properties of Pu2O3; VASP combined with Phonopy software uses the finite position method to calculate the structural phonon spectrum. The expansion cell size is 2×2×2, generating 160 displacement structure files. The calculated Pu2O3 phonon dispersion curve is as follows: Fig.11 According to the phonon dispersion, the phonon vibration free energy, entropy and isovolumetric molar heat capacity are further calculated (see Fig.12 ).

[0039] VASP combined with Elas Tool software uses the stress-strain method to calculate the structural mechanical properties of Pu2O3, such as Fig.13 As shown, a total of 4 structures with applied strain were calculated, and the results are shown in Table 5.

[0040] Table 5 Mechanical properties parameters of Pu2O3

[0041] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0042] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0043] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0044] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0045] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0046] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for determining the physical properties of actinide metal oxides based on machine learning, characterized in that: The method for determining the physical properties of actinide metal oxides based on machine learning includes: Obtain the structures of actinide metal elements and actinide metal oxides; According to the structure of actinide metals and actinide metal oxides, the first-principles calculation method is combined with the dynamic machine learning method of the atomic-scale material simulation calculation software VASP to train the machine learning force field of the corresponding system; Evaluate the trained machine learning force field; When the evaluation results meet the set requirements, the trained machine learning force field is used to calculate the mechanical properties, phonon dispersion and thermodynamic properties of actinide metal oxides; the set requirements are that the fitting error of force and the fitting error of energy meet the standards.

2. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1, characterized in that: The Brillouin zone in the atomic-scale material simulation calculation software VASP is 1×1×1 Gamma lattice; the cutoff energy is set to 520 eV, the NPT ensemble ISIF=3, the Langevin thermal bath MDALGO=3, the atomic friction coefficient and the lattice friction coefficient are 10 ps -1 , the starting and ending temperatures are set to 250 K and 400 K respectively, the step size is 1 fs, and the relaxation time is 20 ps.

3. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1, characterized in that: The evaluation of the trained machine learning force field specifically includes: The force and energy errors during the training process are estimated to determine the force fitting error and the energy fitting error respectively; Determine whether the fitting error of force is distributed between 0.03 eV / Å and 0.1 eV / Å and whether the fitting error of energy is less than 1 meV / atom.

4. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1, characterized in that: The method of calculating the mechanical properties, phonon dispersion and thermodynamic properties of actinide metal oxides using the trained machine learning force field previously also includes: Determine the inner core electrons of actinide metal elements and actinide metal oxide structures based on PAW pseudopotentials; The exchange-correlation functional is determined using the PBE functional within the framework of the generalized gradient approximation.

5. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1, characterized in that: The method of using the trained machine learning force field to determine the mechanical properties, phonon dispersion and thermodynamic properties of actinide metal oxides specifically includes: Based on the trained machine learning force field, combined with Phonopy software, the finite position method is used to determine the phonon dispersion of actinide metal oxides; The mechanical properties were determined using the stress-strain method based on the trained machine learning force field combined with the Elas Tool software.

6. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1 or claim 5, characterized in that: The thermodynamic properties of actinide metal oxides are obtained based on phonon dispersion; the thermodynamic properties include: phonon vibrational free energy, entropy and isovolumetric molar heat capacity.

7. The method for determining the physical properties of actinide metal oxides based on machine learning according to claim 1 or claim 5, characterized in that: Mechanical properties include: elastic constants, bulk modulus of the structure, elastic modulus, shear modulus, Poisson's ratio and Debye temperature.

Images