WO1997046949A1 - Systeme de modelisation moleculaire et procede de modelisation moleculaire - Google Patents

Systeme de modelisation moleculaire et procede de modelisation moleculaire Download PDF

Info

Publication number
WO1997046949A1
WO1997046949A1 PCT/JP1997/001951 JP9701951W WO9746949A1 WO 1997046949 A1 WO1997046949 A1 WO 1997046949A1 JP 9701951 W JP9701951 W JP 9701951W WO 9746949 A1 WO9746949 A1 WO 9746949A1
Authority
WO
WIPO (PCT)
Prior art keywords
molecular
potential
interaction
parameter
parameters
Prior art date
Application number
PCT/JP1997/001951
Other languages
English (en)
Japanese (ja)
Inventor
Jurgen Schulte
Jiro Ushio
Yoshiaki Takemura
Takuya Maruizumi
Original Assignee
Hitachi, Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hitachi, Ltd. filed Critical Hitachi, Ltd.
Publication of WO1997046949A1 publication Critical patent/WO1997046949A1/fr

Links

Classifications

    • C—CHEMISTRY; METALLURGY
    • C07—ORGANIC CHEMISTRY
    • C07K—PEPTIDES
    • C07K1/00—General methods for the preparation of peptides, i.e. processes for the organic chemical preparation of peptides or proteins of any length
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C10/00—Computational theoretical chemistry, i.e. ICT specially adapted for theoretical aspects of quantum chemistry, molecular mechanics, molecular dynamics or the like
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/90—Programming languages; Computing architectures; Database systems; Data warehousing

Definitions

  • the present invention relates to molecular modeling systems, molecular modeling, molecular dynamics, molecular mechanics, materials design, molecular design, and pharmaceutical design.
  • pharmaceutical design pharmaceutical design.
  • the present invention brings a storm to the field of methods and tools in the areas of optimization, data fitting, process optimization, and real time process control. I do. Background art
  • the Lena-Doge-Jones-type functional is only accurate if it describes the fl! Interaction of a homogeneous noble gas.
  • -Layered elaborate Buckingham type functional J. Schulte, RR Lucchese, WH Marlow, J. Chem. Phys., 99 (2), 1 178 ⁇ 1993
  • Buckingham-type functional can provide great accuracy in describing the interaction of two separate atoms, even if these atoms are not of the same type, but the function is purely binary.
  • the commonly used Stillinger-Weber type functional (F.H.Stinger, TA Weber, Phys. Rev. 31 (8), 5262, 5262, 1985) is a two-body interaction. It includes functional functions for 2-body interaction and functionals for Mikyu interaction (3-body interaction).
  • the two-body interaction term is a Lena-Doge-Yones type term, and the three-body interaction term is an angular dependent term weighted by exponential functional. Since the two-body interaction term is the principal term of the interaction, the accuracy of the Stillinger- ⁇ ⁇ bar-type potential is essentially limited, and its reliability is limited to homogeneous systems such as gayone.
  • the three-body functional does not account for the case where two atoms come together during the reaction to form a single molecule (dimer), and the three-body term (3-body term). In such a case, the size of the three-body term of the Steeringer-Eleven Weber functional increases, causing severe problems in accuracy.
  • the Stilinger-Weber functional is more expensive to compute than a pure two-body function due to the exponential and angular dependence of the three-body interaction.
  • Other known three-body functionals, the Tersoff functional (Tersoff functional) J. Tersoff, Pys. Rev., B39 (8), 5566, p.
  • each algorithm is optimized to find a local minimum in the parameter set, in which case it is necessary to determine whether this local minimum is the highest achievable or global minimum, Or because it does not check whether each found minimum is one of many local minima. Therefore, finding the global minimum or maximum parameter set for the functional to be optimized is very tedious and computationally expensive.
  • Some steepest descent methods require knowing the first and second derivatives of the functional to be optimized, increasing the computational cost and inconvenience of these methods.
  • the optimizing method is a non-polynomial method that overcomes most of the problems caused by local minima in the parameter space. It relies heavily on experience and how users intervene in the anneal process. The cost of the overnight parameter optimization process depends on the skill of the user and the sensitivity of the functional to small deviations in parameter values.
  • Pena's convergence-optimized Anil method (TJP Penna, Phys. Rev., E5 R 1, 1995) overcomes some of the high computational costs associated with the Anil S optimization, but the Pena method Parameters to be optimized—works only for small numbers of evenings and simple, well-functioning functionals. Therefore, even the Pena optimization method is not suitable for quickly optimizing complex functionals with a large number of dependent variables, requiring careful monitoring and intervention by experienced users, and The characteristics greatly depend on the user's selection of the initial value of the initial test parameter set.
  • the Pena method can overcome some of the major problems of other optimization methods, but the function to be optimized is not simple, that is, it contains two or more different types of functions.
  • Molecular modeling systems and methods use molecular dynamics (Molecular dynamics consists of molecules) when a set of parameters for potential is provided by the respective molecular modeling tool or user input. It considers the atoms that are moving as particles of Newtonian mechanics, establishes an equation that assumes that they move in the potential energy force field, and calculates them over time.
  • Molecular dynamics consists of molecules
  • the creation of accurate molecular potential functionals and molecular potential parameters is usually done away from the molecular modeling system to generate potentials, Optimization of parameters and optimization of molecular interaction potential Knowledge is a redundant task that requires a great deal of user experience, and further increases computational costs and externally generated potentials and parameters. Evening Minutes f Modeling requires input pacing for the inputs to the system.
  • the molecular modeling system and method of the present invention comprising a molecular dynamics apparatus and method, a molecular dynamics apparatus and method, and a molecular potential generator and method overcomes these problems.
  • the "fit” algorithm is used for the analytical and heterogeneous molecular interaction functionals (S, A, A and A) simulations and soft tutors.
  • the problems arising from functional functional interactions are that these functionalities are cost-efficient but very inaccurate in describing atomic interactions, or
  • the problem in the technical field of the present invention is to explain heterogeneous interatomic interactions accurately with high accuracy and low calculation cost. Can be overcome by using the general heterogeneous functional form of the molecular heterogeneous interaction potential device and method of the present invention.
  • process control and prediction can be performed simultaneously with high accuracy and reliability in a short time. It is better to do it.
  • Commonly used process control tools lack one or both of the above desirable properties. This problem can be overcome with the fast optimization device and optimization method of the present invention.
  • the molecular modeling process requires specialized knowledge of the physical and chemical properties of the constituent atoms of the modeled molecule. According to the present invention, the problem that a high degree of expert knowledge is required for the molecular modeling process is solved by automatically using a molecular modeling tool database system to automatically select the best combination of the modeling device a and the modeling method. The choice can be overcome. Disclosure of the invention
  • the number of molecular potentials is limited, and the potential is of limited accuracy, to solve the problems of molecular modeling systems and methods, and to determine the exact molecular potential and potential
  • the present invention uses the optimization tool and method of the present invention and the heterogeneous intermolecular interaction potential device E and the method of the present invention to accurately and economically generate potential parameters. Utilize the molecular potential generator of the invention.
  • the heterogeneous interaction potential apparatus and method of the present invention provides low accuracy and low reliability associated with the interaction of different types of atoms. Used to solve gender problems.
  • the molecular modeling system of the present invention can arbitrarily combine the individual choices of coordinate evolution equipment and molecular potential equipment and methods.
  • the molecular modeling system and the fractional modeling method of the present invention employ an optimization device and a molecular potential device and method. Can be arbitrarily combined.
  • the molecular potential generator of the present invention is used in the present invention. It is configured as an integral part of the molecular modeling system and the molecular modeling method.
  • a three-body level heterogeneous molecular interaction functional is included as a basis for new interaction potential devices and methods.
  • the functional consists of a small number of different kinds of functions with a small number of adjustable parameters, and the different physical and chemical properties of the two- and three-body atomic interactions are the same.
  • Function inequality- is indicated by a special form of the two-body factor and the Mikyu factor.
  • the inhomogeneous-molecular potential of the present invention is based on the first-principles three-body interaction energy surface (the exact First principle 3-body interaction energy surface).
  • the individual trial distributions of the parameters the individual annealing temperatures of the parameters ( Annealing temperature)> and individual ha 0 lame Isseki receiving probability of the parameter Isseki (parameter acceptance probability) and receiving Aninore has a temperature (acceptance annealing temperature), high-speed Aniru method of the present invention, high-speed self-regulation of the present invention It can be used as a tool for fast self-tuning optimizer and optimization method.
  • the molecular modeling expert database apparatus and method of the present invention automatically select an appropriate combination of modeling apparatus and method based on its characteristic features and molecular data base. It is then used as a basis for automatically determining the best combination of equipment and method in a molecular modeling system and method. [Brief description of the screen
  • J 1 is a diagram showing the configuration of the molecular modeling system of the present invention. "A diagram showing the configuration of the modeling system processing device 105,
  • FIG. 3 is a diagram showing a configuration of a molecular dynamics (MD) device 107 of the present invention
  • FIG. 4 is a diagram showing a configuration of a molecular mechanics (MM) device 108 of the present invention
  • FIG. 3 is a diagram showing a configuration of a molecular potential generator (MPG) device S 109 of the present invention
  • Figure 6 shows that the molecular modeling system has only a molecular dynamics device, the molecular modeling system has only a molecular dynamics device and a molecular mechanics device, or the molecular modeling system of the present invention has FIG. 4 is a diagram showing a comparison of steps required to solve the problem of Embodiment 1 when used.
  • FIG. 7 is a diagram showing a configuration of a potential device E 1073 of the molecular dynamics device 107 of the present invention.
  • FIG. 8 is a diagram showing a configuration of a tracking device 1072 of the molecular dynamics device 107 of the present invention.
  • Fig. 9 shows the relative CPU time for the calculation using the Leonard Doges potential device H as a measure of the computational efficiency of the problem of molecular modeling in Example 1, and the accuracy of the ab imtio method as a measure of accuracy.
  • the molecular modeling system of the present invention can select and use potential devices appropriately. Higher and less to achieve performance]
  • 10 is a diagram showing a configuration of a potential device 1083 of the dynamics device I108 of the present invention
  • 11 is a diagram showing a configuration of an optimization device 1082 of the molecular mechanics device 108 of the present invention
  • Figure 12 illustrates the problem of molecular modeling in Example 1, showing that the molecular modeling system of the present invention achieves the best performance in terms of accuracy and computational cost with the proper choice of combination potential device.
  • FIG. 12 illustrates the problem of molecular modeling in Example 1, showing that the molecular modeling system of the present invention achieves the best performance in terms of accuracy and computational cost with the proper choice of combination potential device.
  • FIG. 13 is a diagram comparing the performance of solving the problem of the molecular mechanics of Example 1 when each optimization device of the optimization device 1082 of the molecular mechanics device 108 is selected.
  • 1-114 is a diagram showing a configuration of a potential device 1093 of the molecular potential generator 109 of the present invention.
  • FIG. 15 is a diagram showing a configuration of the optimization device 1092 of the molecular potential generator device 109 of the present invention.
  • FIG. 16 is a diagram showing the configuration of the memory device 114, the dispersion processing device ft 117, and the input device 101 of the molecular modeling system (MMS) of the present invention,
  • Figure 17 shows the molecular dynamics of Example 1 based on the CPU time required to calculate the interatomic interaction when the devices in each of the memory device 114 and the distributed processing device fi 117 were most efficiently selected.
  • Fig. 4 is a diagram showing the performance of a problem of mathematical modeling
  • Figure 18 shows the descent optimization device 10823 when the Lena-Doges Potential device 10821 is used to explain the interaction of the atoms in the molecular mechanics problem of Example 1 containing argon rare gas atoms. Is most efficient when used with a single instruction multi-data processing unit 1171 and a distributed shared memory unit 1142.
  • FIG 19 shows the CPU time required for an efficient choice of iS: single instruction multi-data (SIMD), multi-order multi-data (MIMO), distributed shared memory (DSM) and distributed non-shared memory.
  • SIMD single instruction multi-data
  • MIMO multi-order multi-data
  • DSM distributed shared memory
  • CPU relative to And shows the performance of the molecular potential generation problem of Example 1,
  • 1-20 is a diagram showing a configuration of a data output device 1074 of the molecular dynamics device K107 of the present invention
  • FIG. 21 is a diagram showing a configuration of a data output device 1084 of the molecular mechanics device S108 of the present invention.
  • FIG. 22 is a diagram showing a configuration of a data output device 1094 of the molecular potential generator 109 of the present invention.
  • FIG. 23 is a diagram showing the functional form of the two-rest factor f 2 and the three-body factor f 3 of the heterogeneous interaction functional of the molecular modeling method of the present invention, comprising the function terms (1) to (9).
  • , 124 is a flow chart of the self-tuning multi-parameter optimization method of the present invention
  • 25 is a diagram showing the performance of the optimization method of the present invention compared to the fast pena algorithm
  • This figure shows the optimized convergence and accuracy for 100 function values compared with the Pena's fast algorithm.
  • Figure 28 shows an optimization system for the process prediction and control of the initial unknown functional of the process dependencies of some dependent variables, as can be seen, for example, in predicting the trajectory of guided missiles and their control processes.
  • FIG. 28 shows an optimization system for the process prediction and control of the initial unknown functional of the process dependencies of some dependent variables, as can be seen, for example, in predicting the trajectory of guided missiles and their control processes.
  • FIG. 4 is a diagram showing a configuration of an optimization device 500 for predicting and controlling an initial known functional of the nature
  • FIG. 30 is a diagram showing a configuration of the expert database processing device 111 of the present invention.
  • Embodiment 1 of the present invention will be described below with reference to FIGS.
  • Figure 1 shows the molecular modeling system input device 101, molecular modeling system processing device 105, molecular modeling system output device 106, memory device 114, data mass storage device 115, network file system device 116, and distributed processing device 117.
  • 1 shows a functional block diagram of a molecular modeling system (MMS) comprising a CRT device 118 and all these devices connected by a common data / address bus 113.
  • Fig. 2 shows a molecular modeling system equipped with a molecular dynamics (MD) device 107, a molecular mechanics (MM) device 108, a molecular potential generator (MPG) device 109, and a molecular modeler Kisparte Ishiba base unit fi111.
  • MD molecular dynamics
  • MPG molecular potential generator
  • the functional block diagram of the processing device 105 is shown.
  • ⁇ Potential device 1073 and tracking device 1072 that are part of molecular dynamics (MD) device 107 Fig.
  • the data human power device 1071 performs the manual processing of the data of the molecular J J mechanical device, and the tracking device in the device 1072 and the potential device of the device E 1073 ⁇ Perform the modeling task.
  • the data output device E 1074 of the device 107 processes the modeling data of the devices E 1072 and 1073, and processes the output data by the modeling task defined by the user input.
  • the data input devices 1081 and 1091 and the data output devices 1084 and 1094 in FIGS. 4 and 5 also have the same functions as described above.
  • the expert database equipment ifr ⁇ ⁇ ⁇ (!
  • the data input device nil -Performs data input processing of the expert database device
  • the expert data base unit 1112 and the expert data base decision unit 1113 select the modeling device according to the user input.
  • Execute. The data output device 1114 of the device S111 processes the output data of the devices 1112 and 1113, processes the output data by the modeling task defined by the user input, and outputs the data of the modeling device of the device 1113. Suggest a choice.
  • the high modeling efficiency of the molecular modeling system of the present invention is illustrated in FIG. 6 by the example of 7-kinetic kinetics, ie, the problem of semiconductor processing of the interaction between argon rare gas atoms and the surface of a gay semiconductor. Efficiency results are compared to different configurations of the modeling system.
  • the interaction between the argon noble gas atoms is known, the interaction between the semiconductor atoms is known, and the interaction between the semiconductor and the noble gas atoms is not known at first.
  • the user of the system must perform a number of steps. Some of these steps include user custom program coding and difficult tasks that take a lot of time and require user knowledge.
  • the molecular dynamics system and the molecular dynamics system having the molecular dynamics system need to perform a total of 10 steps. The steps require user intervention.
  • the molecular modeling system of the present invention implements the above-described problem within only six steps, and achieves high fractional modeling efficiency because a user-specific custom step is not required.
  • users can use molecular modeling systems to design standard, complex industrial materials, chemicals, It can be used for other molecular design problems.
  • the relative CPU time is reduced for the speed-up calculation, that is, the calculation using the Lennard-Doggons device.
  • the ab initio method for obtaining accurate calculation values as a measure of accuracy [The molecular orbital method for measuring the electronic structure of a molecule uses empirical parameters to calculate each ⁇ ⁇ -Unlike the trajectory method, a method of calculating the integral by a strict method without using such empirical parameters is shown. sex We used performance-(1 + relative deviation) / (relative CPU) as a measure of performance, which means that the closer the obtained value is to 1, the higher the performance of all holidays.
  • the molecular modeling system of the present invention if the potential equipment used to calculate the individual interactions of a molecular dynamics problem can be properly selected and used, the molecular dynamics can be obtained with high accuracy, reliability and efficiency. It can be seen from Figure 9 that the best performance that solves this problem is achieved. Even higher efficiencies can be achieved by selecting the most appropriate propagation units ( Figure 8) for each potential device.
  • the molecular modeling tool kisspart database device 111 assists non-experts in providing the optimal combination of molecular modeling devices according to the intended molecular data object and the intended modeling goal. Allows you to determine your choice.
  • Such a potential device may include a three-body potentiometer in a potential device 1083 (FIG. 10), such as a Stelinger-over-potential device 10833 or a potential device 10839 of the present invention. Device.
  • the relative accuracy of the structural optimization of Si-Ga-Si molecules (the second column in Fig. 12) and the relative calculation cost when using the optimization device 10828 (Fig. 11) of the present invention (Fig. 12).
  • the best performance due to accuracy and computational cost can be achieved with the molecular modeling device of the present invention with the option of proper selection of the combination potential device.
  • Performance can be further improved by properly selecting the optimizer. Performance is shown in terms of relative deviation from the exact ab initio minimum energy structure and the required CPU time for cases where other optimizers are used for the molecular mechanics problem.
  • Another optimizer is compared with the potential optimizer K 10828 of the present invention, ie, the highest-performance potential optimizer E used for the molecular mechanics problem (FIG. 12).
  • Figure 13 shows that the performance is low when the selection of the optimizer is limited.
  • a commonly used standard optimization device such as a simplex device (Simplex unit) 10821 or a descent device fi (Gradient Descent unit) 10823, the accuracy and reliability of the optimized structure are low and Long calculation time.
  • the best performance for the molecular dynamics problem can be achieved by running the optimizer E 10828 of the present invention. Compared to other optimization equipment [S, the total time was shorter and the relative deviation from the exact value was very small.
  • the best molecular mechanics performance can be achieved by utilizing the molecular modeling system of the present invention, which provides a combination of independent choices of Charl S and £ i optimizer.
  • the molecular modeling expert database device 1 1 1 supports non-expert users to determine the optimal combination of molecular modeling devices according to their intended molecular data objects and their intended modeling goals. Allows you to determine your choice.
  • the high accuracy and reliability of the molecular potential generator system is the result of the individual intermolecular interactions used to optimize or adapt each set of potential parameters to the potential energy surface data provided at the input. This can be achieved by choosing the most appropriate potential device that describes the effect. For example, energy surface data describing the interaction between one gallium atom and two ga- gen atoms can be used to explain the interatomic interaction in order to optimize the potential parameters. The most accurate approximation can be obtained by selecting.
  • the means is a bipotential device in the device 1093 (FIG. 14), such as a Stelinger-over-potential device 10933 or the potential device ffl 10939 of the present invention.
  • Figure 13 shows the relative deviation ⁇ from the exact ab initio potential energy surface for the case of using the potential device 10939 of the present invention to generate the molecular potential parameters for the above problem using other optimization devices.
  • the performance of potential parameter overnight optimization is shown depending on the required relative CPU time.
  • FIG. 13 shows that performance is compromised when the choice of optimization equipment is limited or limited. That is, simplex device 10921 (Fig.
  • the optimization performance of the high molecular potential parameter set is based on the molecular modeling system of the present invention that independently selects the combination of the potential device fi and the optimization device that generates the potential parameters. It can be achieved by utilizing.
  • the molecular modeling system processor 105 comprises devices 107, 108, 109 and 111 connected to a common data / address bus 113 (FIG. 1) and connected by a data address bus device 110, and comprises Since the device accesses the data mass storage device 115, the network file system device 116, the distributed processing device 117, and the memory device 114, higher calculation efficiency can be achieved. This is because each of the processing units in apparatus 105 has access to the local dispersion calculation means that is most appropriate for the particular molecular modeling problem to be solved.
  • FIG. 17 shows the molecular dynamics model corresponding to the CPU time required to calculate the interatomic interaction when the device in each of the devices S 114 and 117 is most effectively selected as described above. It shows the performance of the optimization problem, and when the relative CPU time is close to 1, it shows the best performance. From Fig.
  • Molecular model Chemical kiss part database instrumentation S 111 is to assist the user of non-professionals, according to the model of the target that was intended to molecular data for the purpose (mo l ecu l ar data object ), molecular dynamics equipment, molecular modeling It is possible to determine the selection of an optimal combination of a distributed processing device, a memory device, and a data mass storage device.
  • the steepest descent optimization device 10823 is replaced by a single instruction multi-data processing device fft 1171. It is most effective to use it together with the distributed shared memory device 1142.
  • a multiple instruction multiple data unit (1172) is used for the Lena-Dogeons potential device, the relative CPU time becomes 1.3, the molecular dynamics performance decreases (Fig. 18), and the distributed non-shared With memory device 1143, the performance is further reduced. Therefore, when the molecular modeling system of the present invention is used, insufficient calculation performance due to the limited selection of the calculation means is obtained.
  • the g111 molecular database modeling kit system supports non-expert users, and provides molecular g-optimization equipment and molecular modeling according to the target molecular data (molecular data object) and the intended modeling target. It is possible to determine the selection of an optimal combination of a distributed processing device, a memory device, and a data storage device.
  • the potential energy surfaces of interacting argon noble gas sources should be adapted, for example, the distributed single instruction multi-data device 1171 and the distributed shared memory device 1142
  • the most efficient calculation is to use the Descent Optimizer 10923, which uses a single instruction multi-data processing unit 1171, together with the Rena Doges Potential Device 10931, and the Distributed Shared Memory Unit 1142.
  • FIG. 19 shows the performance of the molecular potential generation problem in terms of the relative CPU time with respect to the CPU time required by the most efficient choice of equipment.
  • Figure 19 shows that, for the S-optimizer 10923 and the potential device 10931, the performance is substantially reduced when different combinations of pre-dispersers in devices 114 and 117 are used. it can. Therefore, if only the same means of the calculation processing device and the memory device a is used for the potential device E and the entire optimization device ffi, the calculation performance becomes insufficient.
  • the molecular modeling system of the present invention overcomes insufficient computational performance due to limited choice of computational means.
  • the molecular model chemical kisspart database device 11 1 supports the non-expert user and uses the molecular potential generator device according to the target molecular data object and the intended modeling target. Appropriate combination of molecular modeling distributed processing device s, memory device s :, and data mass storage device Can be determined.
  • the data input and the operation input to the molecular modeling system are: an input device 101 including a keyboard device 1011, a graphic table device 1012, a screen human input device E 1013, and a mouse input device 1014; a memory device 114; Data storage 115; and a network file system 11G; or a combination thereof, the input of said device being provided by a common data / address bus 113 to other devices of the molecular modeling system. (See Figure 1).
  • the data output of the molecular modeling system is provided by a general output device 106 (FIG. 1) and individual output devices 1074 (FIG. 20), 1084 (FIG. 21) and 1094 (FIG. 22).
  • the molecular dynamics device 107 receives input from the common data Z address bus 113 and from the device S 108 and the device 109 (FIG. 2).
  • the molecular dynamics input is processed by the data input device 1071 ( Figure 3).
  • the parameters for molecular dynamics calculation are given by the input data of the molecular dynamics apparatus, and the calculation means is selected.
  • the potential input device 1073 and the tracking device 1072 (Fig. 3) develop the molecular input coordinates in time according to the selection of molecular interactions and the expansion of the deer mark.
  • the force acting on the molecule is calculated by the potential device 1073, and the calculated force is used by the tracking device 1072 to develop the molecular coordinates in a timely manner.
  • the expanded coordinates are used by the potential device 1073 to calculate the force acting on the molecule at the time of expansion.
  • the tracking device 1072 uses the force acting on the molecular coordinates to further develop the molecular coordinates in a timely manner, and passes the further developed molecular targets again through the g1073.
  • the above-described cycle of measuring the force acting on the splitter by the interaction and then developing the molecular coordinates allows the molecular coordinates to be developed in a timely manner.
  • the timely expansion of the molecular coordinates continues until the condition defined by the molecular dynamics parameter is satisfied.
  • the output of the molecular dynamics device is processed by the output device S 1074 according to the input specifications provided by device 1071.
  • the potential device 1073 is equipped with a means II for calculating the interatomic potential and interatomic force of molecules, and Leonard-Jones 3 ⁇ 4 10731 (stabilization term proportional to the sixth power of the reciprocal of interatomic distance and 1 2 (E.Lennai'd- Jones: Proc.Roy.Soc. (Lond.), 106A, 441 ( 1924); Proc. Roy. Soc.
  • the most appropriate type of potential device a can be used for individual molecular interactions between molecular components.
  • the problem of molecular dynamics simulation including the rare gas atom and the source of semiconductors- ⁇ is that a 15 t appropriate and accurate potential device for rare gas atoms (a Lena-Dojons potential ⁇ -10731 can be obtained) And select the most appropriate potential device for the semiconductor atoms (including a three-body potential device such as the Stillinger-Weber potential device 10733) and the noble gas and ⁇ conductor atoms.
  • the most accurate description can be made by selecting the most appropriate and accurate potential device that describes the interaction (including Buckingham-type potential device 10732).
  • the tracking device 1072 is equipped with a means to calculate the molecular coordinates that are developed in a timely manner (Fig. 8).
  • Fig. 8 L.Verlet: Phys. Rev., 159,98 (1967)
  • predictor-corrector 3 ⁇ 4 10722 moveement method 3 ⁇ 4 (M.Haile: Molecular Dynamics Simulation,?. 159-163 (John Wiley & Sons, In, 1992)
  • Taylor-Expansion type 10723 position, etc.
  • the output device 1074 processes the molecular dynamics output data according to the parameters provided by the input device 1071.
  • the data and graphics to be printed are processed by the device 10741, and the data and graphics to be displayed are processed by the device 10742.
  • the data and graphics to be printed and displayed are processed by the device 10743 to output in real time, and then processed by the device 10744 to output the set time.
  • the molecular mechanics device 108 receives inputs from the common data address bus 113 and from devices 107 and 109 (FIG. 2).
  • the molecular mechanics input is processed by the data input device fi 1081 ( Figure 4).
  • the parameters for molecular mechanics calculations are obtained from the molecular mechanics input data, and the calculation means is selected.
  • the potential device 1083 and the optimization device 1082 optimize the distribution and coordinates of the input structure according to the selection and optimization of the intermolecular interaction (Fig. 4).
  • the molecular potential is calculated by the potential device 1083 and the molecular coordinates are optimized from the molecular potential and from the forces acting on the molecules by means of optimization.
  • the optimized molecular coordinates are calculated by the optimizer-1082. From the optimized pre t coordinate, the force acting on the molecules in the case means molecules potentials, and optimization is required, it is calculated in potential device 1083. The calculated potentials and respective forces are used by the optimizer 1082 to further optimize the molecular coordinates, and then pass the coordinates through the potential device 1083 c. According to the potential With this cycle of calculating the forces acting on the core and the min-f, optimization of the molecular coordinates can be achieved continuously. Optimization of the molecular coordinates is continued until the conditions defined by the molecular mechanics input parameters are satisfied. The output of the molecular mechanics device is processed at output device 1084 according to the specifications of the input provided by device 1081.
  • the potential device 1083 provides a means to calculate the potential and force between the atoms of the molecule ( Figure 10), the Lena-Dogeons type 10831, Buckingham 3 ⁇ 4 10832, the Stillinger-Weber type 10833, the Tersov type 10834, the embedding
  • a potential device is selected using the atomic type 10835, the tightly coupled type 10836, the molecular orbital type 10837, the custom-made type 10838, the potential type 10839 of the present invention, or the potential of the combination of the above components. .
  • the problem of the low efficiency, accuracy, and reliability of the molecular mechanics of a molecular mechanics system can be attributed to the most appropriate type of potential device for the interaction of individual atoms between the components of the molecule. g can be used to overcome it.
  • the problem of semiconductor molecular mechanics of optimizing a molecular structure including one gallium atom and two gallium atoms is that of a device such as the Stillinger-Weber potential device K 10833 or the potential device S 10839 of the present invention. 1083 (Fig. 10). 1 ⁇ 2 can be explained most accurately by selecting an appropriate and accurate potential device.
  • the potential device that describes the different types of interactions is selected only once, the maximum error will increase, and as a result, the molecular structure calculated by the molecular modeling device R will be very unreliable.
  • Some optimization devices S are the Hessian matrix of the first and second derivatives of each potential used in the potential device E.
  • the S-optimizer is suitable for use with the Lena-Doges potential device 10831, but is most inappropriate for use with the two-body potential device. Therefore, by selecting an optimization device for each potential device to be used, even greater computational efficiency can be achieved.
  • Figure 12 shows the performance that solves the molecular modeling problem. As a measure of its performance, the relative accuracy of the structural optimization of Si-Ga-Si molecules (the second column in Fig. 12) is shown. This relative accuracy indicates that better accuracy can be achieved by selecting an appropriate potential device.
  • the optimizer 1082 has a means to calculate the optimized molecular coordinates.
  • Fig. 11 Simplex type 10821 (Multidimensional space is divided into two on an affine plane, and the maximum or minimum direction is searched. It has a simplex method function to narrow down the maximum / minimum value area repeatedly), steepest descent type 10822 (generates a point sequence ⁇ Xn ⁇ that reduces the value of the objective function f (X) one after another from the initial value) In the so-called descent method, the point sequence generation direction is in the differential direction of the objective function f (X), and in the opposite direction of ⁇ f (X), that is, it has the function of finding the minimum value.
  • the output device 1084 processes the molecular modeled output data according to the parameters provided by the input device 1081.
  • the data and graphics to be printed are processed in $ 10841, and the data and graphics to be displayed are processed in device 10842.
  • Data and graphics to be printed and displayed are processed by the device 10843 as real-time output, and processed by the device 10844 as fixed-time output (FIG. 21).
  • the molecular potential generator device 109 receives inputs from the common data / address bus 113 and the loading device 107 and device 108 (FIG. 2).
  • the molecular potential generator input is processed by the data input device fi 1091 (Fig. 5).
  • the molecular potential generator input data provides the parameters for the calculation of the molecular potential X and the neural network, and the calculation means is selected.
  • the molecular potential generator generates a molecular potential by optimizing the parameters of the potential interaction of the input potential, and inputs the data of the potential energy surface using the potential device 1093 and the optimizer 1092.
  • the molecular potential is calculated using the potential device 1093.
  • the optimizer 1092 the potential parameter is high and it is optimized to the given input potential energy.
  • the molecular potential energy is calculated by the potential device fi1093.
  • the calculated potential energy is used by the optimizer 1092 to further optimize the potential parameters to the given input potential energy and pass the parameters back through the potential device 1093. This cycle of calculating potential energies and optimization potential parameters overnight can continue to optimize potential parameters overnight.
  • the optimization of the potential parameters is continued until the condition defined by the input parameters of the molecular potential generator is satisfied.
  • the output of the molecular potential generator device is processed by the output device R 1094 according to the manual specifications provided by the device 1091.
  • the potential device 1093 provides a means to calculate the potential and force between the atoms of a molecule ( Figure 14), and this device uses the Lenard-Johnson ⁇ 10931, the Knockingham-type 10932, and the Stillinger-D Bar type 10933, Taso 10934, embedded atom type 10935, tightly coupled type 10936, molecular orbital tracking type 10937, custom-made type 10938, and the potential type of the present invention 10939 or the potential device is selected using the potential of the combination of the above components. Is done. Choosing a potential device overcomes the problem of insufficient efficiency, accuracy, and reliability in adapting the molecular modeling potential of a molecular modeling system.
  • ⁇ can also use an appropriate type of potential device to present the individual features of the molecular energy surface of the molecular interaction under consideration.
  • a potential energy surface data describing the interaction of noble gas atoms can be accurately fitted using a Lena-Dajonez potential device 10931 and a potential energy surface data describing, for example, the interaction of metal atoms. The evening is precisely fitted using the embedded atomic potential device 10935.
  • the metal atomic energy data is adapted using the Rena-Dogons potential device 10931 or the noble gas energy data is adapted using the embedded atomic potential device 10935, the accuracy becomes insufficient and the potential becomes poor. Has a very limited predictive function when modeling molecular dynamics and molecular mechanics.
  • the optimizer 1092 provides a means to calculate the optimized potential parameters (Fig. 15). —Optimum equipment is selected using 10925, high-speed pena type 10926, custom-made type 10927, optimization type 10928 of the present invention, and optimization of the combination of these components. Choosing the optimizer K overcomes the problem of insufficient efficiency, accuracy, and reliability of the molecular modeling potential of the molecular modeling system. This is because in order to adapt the potential parameters for each potential device employed, This is because the most appropriate type of optimization device can be adopted. Selecting the most appropriate molecular potential generator for each purpose of the energy surface fit will allow you to select the optimizer or other combinations of available optimizers only once. In comparison, much greater computational efficiencies can be achieved.
  • potential energy surface data describing the interaction between one gallium atom and two silicon atoms is the most appropriate and accurate description of the interatomic interaction in order to optimize potential parameters.
  • the most accurate approximation can be made by selecting the stage.
  • the potential device 10939 of the present invention is used with the optimizer 10928 of the present invention, a set of optimized potential parameters with the highest accuracy, reliability and performance can be achieved.
  • Figure 13 shows the performance of the potential parameter-evening optimization when using another optimization device for the above problem of molecular potential parameter generation and using the potential device 10939 of the present invention.
  • Potential energy table is shown as a function of the relative deviation from the relative CPU time required.
  • Figure 13 shows that, using commonly used standard optimizers, such as the Simple Rex device 10921 and the descent device 10923, when the choice of the optimizer is limited or limited, the optimized potential parameter set is It shows that the performance, ie, accuracy and reliability, is low and the calculation time is long.
  • the best performance of the molecular potential parameter set optimization problem described above can be achieved by using the optimizer 10928 of the present invention. Therefore, the optimization performance of the best set of molecular potential parameters is independent of the combination of the potential device that generates the potential parameters and the optimization device. 3 ⁇
  • the output device 1094 processes the molecular potential energy output according to the parameters provided by the input device 1091.
  • the data and graphics to be printed are processed by the device K10941, the data and graphics to be displayed are processed by the device 10942, and the data and graphics to be printed and displayed are processed and configured by the device K10943 as time output. It is processed by the device 10944 as a time output ( Figure 22).
  • the molecular modeling kisspart database processor 111 (Fig. 2) inputs data from the common data address bus 113, and the devices 107, S108, and fil09 (Fig. 2). The input processing is processed by the data input device 1111. (FIG. 30)
  • the molecular modeler's Kispert database device 111 uses the expert database processing device 1112 and the expert database determination device 1113 to input the target molecular data object and the user's input. According to the defined modeling goals, we propose the optimal combination of molecular modeling equipment.
  • the data input device 1111 provides the expert database determining device 1113 with the common data and the status information of all the devices in the molecular modeling system obtained via the Z-adder box 113.
  • the expert database determining device 1113 processes the input data from the input device 1111 and the data provided by the expert database processing device 1112, and determines an appropriate combination of the molecular modeling system devices. To give the decision result to the data output device.
  • Expert The database processing device m2 holds the characteristic data of all known chemical elements and the characteristic data of molecules treated as molecular entities as single entities.
  • the data on chemical elements and mono-molecules held in the device 1 112 are as follows: the atomic chemical element name and its symbol, the molecular entity name and its symbol, and the load (the electrical charge), the magnetic state, the ionization state and energy, the atomic ana molecular entity polarization> the original--and the orbital shape of the molecule (the atomic and molecular entity orbital configuration)
  • the expert database determination unit 1113 receives as input the target molecular data, the modeling goals defined by the user, and the modeling goals defined by the user, and the expert database in the unit 1113.
  • the expert data in the instrument 1113 is the evening is the expert information of the atomic chemical elements and molecules, the atomic information, molecular data, Molecule modeling system—has the best way to model the interaction of atoms and molecules based on evening
  • the problems that explain the heterogeneous interaction of polyatomic systems, as seen in the technical field of the present invention, by the molecular modeling method of the present invention include the term describing the heterogeneous two-body interaction and the heterogeneous three-body Min to explain the interaction It can be solved using a number form.
  • the generic analytical functional type of the present invention is convenient to use in the technical field of the present invention, and can be used with other common interaction functionals without compromising computational costs. Accuracy is higher.
  • MMS molecular modeling system
  • Molecular dynamics device 107 that tracks the evolution of molecular dynamics, molecular mechanics S108 that optimizes the structure of molecular systems by ft, and molecular potential that generates intermolecular interaction potential and intermolecular interaction potential parameters Either one of the generators 109 ", or a combination of molecular dynamics, molecular mechanics and molecular potential generators equipment, said molecular dynamics device E 107 has intermolecular interactions
  • a means for selecting a potential and a means for selecting a tracking device for time evolution of molecules or a combination of the above means are provided.
  • stage selecting means and molecular mechanics structure ⁇ apparatus for selecting intermolecular interaction potential, or a combination of said means is set vignetting, the fraction?
  • a means for selecting a molecular interaction potential device and a means for selecting a molecular potential parameter optimization device, or a combination of a U-stage, is provided in the potential generator device S 109.
  • a three-way heterogeneous atom interaction potential that explains the interaction of atoms of similar and dissimilar types of means of selecting an action potential device Selection and combination of a molecular analysis device or a molecular interaction potential device, and one of the optimization means is selection and combination of a high-speed non-polynomial optimization device or an optimization device. This can be solved by using the functional in a molecular modeling system.
  • the two-body interaction factor / 2 and the three-body interaction factor of the functional are shown in FIG. 23, and the individual terms of the functional and the usefulness of these terms in overcoming the above problems are This embodiment will be described in detail subsequently.
  • the terms (1), (2) and (3) of the interatomic interaction functional represent the two-body interaction terms because they depend only on the primordial-interval distance.
  • 3 ⁇ 4 (1) expresses the inter-atomic core repulsion by a physically significant exponential function. Is done.
  • the parameters a, and bi determine the strength of the repulsion and the decrease in the repulsion due to an increase in the interatomic distance.
  • ⁇ Attraction between atoms is governed by the electrodynamic and quantum mechanical interactions of the atoms' electron shell.
  • 3 ⁇ 4 (2) in Fig. 23 represents the main function of the two-body ⁇ ! Attractive force because 3 ⁇ 4 (2) represents the polarization of the atomic shell due to individual atomic polarization and dispersion interaction. .
  • Atomic polarization is specific to the type of individual single atom and is caused by the instantaneous relative displacement (E) of the shell charge f, and is considered to be a major factor in itself. Should be. Since the total two-body interaction functional covers the basic physics of two-body interaction more accurately than the Lena-Dojons type functional, the quadrupole ) And the Buckingham-type potential that considers higher-order two-body interactions are more economical. For different pairs of two-body interaction - a set of parameters Isseki ai, Since b, and Ci can be specified, accuracy and adaptability to account for heterogeneous phase II effects are guaranteed. Item (3) in Fig. 23 is zero when analyzed beyond a predetermined cut-off radius depending on the type of two interacting atoms 1 and j. That is guaranteed.
  • the change in the strength of its two-body interaction depends on the individual electronic properties of each of the three-dimensional atoms that participate in the action. are doing.
  • the configurational preference of the term enclosing angle for a set of three atoms is the three configuration angles where the angle (5) is closed by each three atoms.
  • the cosine of these corners is compared with the cosine value (6) of the corner where placement is preferred, described by a corner-dependent period (4).
  • Only term (4) covers the essence of the general three-body interaction, ie, the preferred angle, and by varying the angle, the three-body interaction changes in some way.
  • the fundamental physics of two-body interaction is that in the presence of a third atom, depending on the type of all atoms, the three-rest configuration of the two-rest interaction is weakened or strengthened depending on the distance and type of the third atom. Cannot be covered only by the angle-dependent terms.
  • the exponent 3 ⁇ 4 (7) which takes into account the distance to each of the two other interacting atoms, describes the original physical properties that depend on the changing distance of the three-body S action.
  • the specific function term (8) including the cut-off radius, indicates a limited range of the Mikyu interaction. As the two atoms in the three-body interaction under consideration come closer, each two-body interaction becomes predominantly contributing, as in the case of a strong bond of dimer molecules. As a result, the three-rest phase action is best described by a function that similarly limits the range and strength of the effects of short-range three-body interactions, thus explaining the different types of ::. can do. This limitation is explained in section (9) of FIG.
  • the total 3-body functional is a problem caused by the Stilinger- ⁇ ⁇ -bar functional, which means that the three-body term tends to distort the two-body term. It overcomes the problem of causing additional accuracy problems when the precision of the functional that describes the field is already limited.
  • the total atomic interaction functional of the present invention includes functions of the same general type that are used for Stilinger-Dever functionals at different orders. As the specified function is reused for different ⁇ ⁇ of the functional of the present invention and the number of tunable parameters is as small as the Stilinger-Weber type functional, the heterogeneous interaction between atoms is explained. A higher accuracy is obtained for the functional, and a computational cost comparable to that of the Stillinger-Weber functional is achieved (FIG. 12).
  • the functional form of the heterogeneous interatomic interaction of the present invention is that its general form is effective without compromising the computational cost or only for a specific combination of heterogeneity of the atomic interaction. It offers high accuracy comparable to commonly used interaction functionals without compromising on performance.
  • FIG. 1 Another embodiment of the present invention will be described below with reference to FIGS. 3, 7, and 12.
  • FIG. 1 Another embodiment of the present invention will be described below with reference to FIGS. 3, 7, and 12.
  • the problem of explaining the heterogeneous interatomic interaction between one gallium atom and two silicon atoms as described in the technical field of the present invention Solved in the molecular dynamics device 107. That is, by calculating the force acting on the molecular atoms by the molecular dynamics tracking device 1072 in the molecular dynamics device 107, the timely expansion of the molecular target is calculated, and the potential processing in the molecular dynamics device 107 is performed. The cycle of calculating the force acting on the molecular origin T by the interaction jfl potential with the instrument ⁇ 1073 is repeated, and the time evolution of the molecular system is calculated by expanding the molecular coordinates in a timely manner. In this calculation, one of the choices of the means of the potential processing unit 1073 is a three-body heterogeneous interatomic interaction analyzer 10 739 force Calculating the total interaction energy between atoms of similar and dissimilar types of atoms I do.
  • the total interatomic interaction energy is calculated by two-body distance (r) -dependent interaction and three-body distance and angle-dependent interaction.
  • the two-body interaction has an exponential and reciprocal force on the six functional (r- 6 ) distance dependence between two arbitrary-type atoms, and the parameter of the two-body interaction is the atom type
  • the exponential interaction shows the interatomic repulsion caused by each atomic core
  • the parame- terized r- e term shows the polarization and dispersion caused by the electron core of each atom
  • the three-body function term is composed of an angle-dependent trigonometric function weighted by a distance-dependent exponential function weighted by distance, and the parameters of the trigonometric function and the exponential function are represented by
  • the two-body and three-body weights are weighted by a distance-dependent exponential function that controls the continuous force cutoff in the analysis of the preferred scissor angles and interaction beyond the set cutoff radius for the type and three-body structures. Functionally dependent You.
  • the total interaction energy of the similar and dissimilar types of atoms is provided at the input of the device as ⁇ JT from the atomic coordinates and parameters as the sum of two-body and three-body interactions,
  • the total phase S action energy U of similar and dissimilar 'type atoms is a function of the form:
  • ⁇ is the angle between atom j and k, calculated from the atomic coordinates at the input
  • r tj is the distance between atom i and ⁇ , calculated from the atomic coordinates at the input.
  • the parameters a, b, and Ci of the device can be the type of each atom i
  • the parameters i 7 i can be the types of the atoms i and j
  • the parameter ai is calculated at the set angle at i, which defines the configuration choice of the atom configuration.
  • Each of the above input parameters indicates the physical properties of the given atom type and the interaction of those atoms with other atoms of a similar or dissimilar type.
  • the range is shown by the following equation.
  • the input dependent variables of the three-body inhomogeneous atom interaction analyzer 10739 are the atomic distance and all three ⁇ and jk surrounded by three interacting atoms.
  • the range of the parameters described in Equations 6 and 7 is that the overall magnitude of the interatomic repulsion can be adjusted for all kinds of atoms by the range of the parameter ai ;
  • the repulsion reduction is adjustable; the magnitude of the variance falls within physically significant limits by the range of parameters ci; the magnitude of the tribody repulsion by the range of the parameter ⁇ ; Preferred bibody formation in body configurations can be accounted for without significant interference with pure binary interaction terms; depending on the range of parameters, increased forces within three-body interactions and
  • the reduction is adaptive; and the range of interaction is such that the range of parameters 3 ⁇ 4 can characterize single atom processes and internal properties that occur over long ranges of atoms.
  • This functional of the present invention has the same efficiency as the Stinger-Eber functional, is half as efficient as the pure two-body functional Renard Doge-Yons and the Buckingham quadratic, and has a first-principles calculation. It is about 1000 times more efficient than (first principle calculation).
  • the intermediate form shown in Equations 1 to 5 and the parameter set shown in Equation 8 can achieve high accuracy in explaining the interaction between gallium and silicon.
  • the accuracy of the functional of the present invention is shown with reference to the first principle calculation or the so-called ab initio calculation.
  • the accuracy of the present invention is three times higher than that of the Stillinger-Weber type functional and 5 to 10 times higher than the pure frame two-body functional.
  • the function form of the present invention overcomes the problem of providing high t and accuracy while maintaining computational efficiency.
  • the present invention uses the simulated annealing technique. Because the anneal method converges slowly or overcomes the less efficient problem than other methods, the cosiro-lenz distribution for the visiting distribution of these nomenclatures Using (Cauchy-Lorentz distribution), the convergence is improved to be proportional to the logarithm of the annealing temperature so as to be proportional to the reciprocal of the annealing temperature.
  • the acceptance of the parameters over time is controlled by a generalized metric mouth policing algorithm, in which the anneal temperature of the algorithm is decoupled from the anneal temperature of the visiting distribution of the parameters.
  • these parameters were selected from the distribution of the Lorenz type, and the initial temperature of the distribution was set at the beginning of the parameter. Because all parameters are distinct functions of the test value of, all parameters do their individual distribution.
  • the sensitivity of the functional to be optimized is different for ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ Since the depth and number of local minima covered by individual jumps over a parameter are different from the other parameters, the algorithm of the present invention uses a single parameter Follow the topological param- eters topology from the perspective view. This feature increases the reliability of optimization and the overall efficiency of optimization. This feature is realized by making the visiting distribution of the parameters proportional to the individual jump success rates. As a result, each parameter is independent of the other parameter's annealing sequence and other parameters.
  • the algorithm's strength and efficiency are further increased by adjusting the visiting distribution function by the convergence behavior of the individual parameters.
  • This feature allows the optimization algorithm to be executed in a loop that adjusts the distribution function starting from the current optimal parameters with the refreshed annealing temperature, which is a function of the current optimal parameter set, and thus the parameter This is achieved by making the local parameter space more efficiently examined so that even large and deep local minima can be overcome.
  • the distribution function is adjusted in a loop, the parameters can be scanned faster in the parameter space, so that they can converge faster than without a loop.
  • the optimization algorithm is self-regulating because the strength of the optimization algorithm increases with the performance of the loop and quickly overcomes local minima without external intervention.
  • the visiting probability distribution of Koshi bite-rents which is expressed as the fraction of the probability that the test value X of the parameter changes from to X t + 1 , is represented by the following formula.
  • Equation 10 is the individuality of each parameter overnight test jump, and q v is the co replacement sheet (Rule 26) Controls the form of the siren distribution.
  • the individual time scale t is set according to the algorithm flow chart in FIG.
  • the probability of accepting police parameters for each generalized method is expressed by the following equation.
  • the molecular modeling method of the present invention can overcome the problem that local minima occur in the parameter space with respect to the convergence speed and accuracy of annealing.
  • the parameter distribution probabilities are quenched according to the individual parameter annealing temperatures Tx according to the following equation:
  • Tx (t,) indicates that all parameters and distribution iS vectors ⁇ -X follow their individual time scale t
  • Tx "is the first parameter is an individual output 3 ⁇ 4 which parameter of Isseki a of 13 ⁇ 4] number Isseki.
  • parameter q v controls the eleven Lorenz worth particular Cauchy. q,. the parameter Isseki Tsutsushima,
  • FIG. 25 shows the strength and efficiency of the optimization method of the present invention in comparison with the high-speed Pena algorithm.
  • FIG. 25 shows the accuracy of the optimization achieved by the function on the relative deviation as a function of the required CPU time, the relative deviation being plotted on a logarithmic scale.
  • the Pena method converges fast, but has limited accuracy, even for the commonly used functionals of trigonometric functions, and the limited accuracy is maintained for a long time. This behavior of the Pena method is due to local minima and limited ability to overcome.
  • FIG. 25 shows that the optimization method of the present invention quickly converges to a three-digit high-precision parameter set. Therefore, the optimization method of the present invention can overcome the problem caused by local parameters and minimum points, and can be effectively applied to trigonometric functionals. Since the optimization method of the present invention effectively overcomes local minima, very little user intervention is required when using this method, and therefore very little user skill is required.
  • the molecular modeling method of the present invention can further overcome the problem that local minima in the parameter space cause for convergence speed.
  • the parameter to be optimized to overcome the slow convergence problem of the anneal optimization method the force of each evening, the individual acceptance probability annealing temperature, and the individual generalized parameters.
  • a tropolis acceptance probability is assigned.
  • the distribution anneal temperature and the generalized metropolis anneal temperature of each individual caustic bite lent are functions of the loop initial test parameters, and high efficiency is obtained when selecting test parameters. High efficiency convergence can be achieved.
  • the molecular modeling method of the present invention can overcome the problem of slow convergence and low accuracy when the optimization method is applied to a complex functional.
  • the optimization method of the present invention is used to overcome the problem of slow convergence and low accuracy.
  • the molecular mechanics device 108 uses the interaction potential from the molecular mechanics potential device 1083 and optimizes the molecular coordinates for molecular energy using one optimization device from the optimization device 1082. Calculating the optimized structure of the molecular system by the optimization;
  • the 1082 optimizer calculates the optimized structure of the molecular system based on the individual optimization parameter time scale and the individual parameter annealing temperatures of the Cauchy-Mouth Lenz test parameter optimization distribution.
  • the necessity of the optimization is calculated based on the generalized meteorological police acceptance probability, and the generalized acceptance probability is determined according to the order of the individual parameters for each initial input optimization parameter. Controlled within the generalized metropolis algorithm, and the anneal of the parameter-to-mouth Lenz test parameter distribution over time is restored with optimized initial test parameters according to the individual test parameter timescale and anneal order. .
  • the individual anneal temperatures of the Cauchy bite-Lent distribution are quenched in proportion to the reciprocal forces of the individual parameter time scales, and the initial anneal temperature is a function of the initial optimization test parameters.
  • the quenching of the parameter distribution occurs within the self-regulating parameter distribution loop with the acceptance of the generalized metropolis algorithm parameters, and the individual parameter temperatures of the generalized metropolis algorithm correspond to the individual parameter ties. It is proportional to the reciprocal force of the time scale.
  • the optimized initial test parameters are within the self-regulating parameter distribution loop, and that the individual caustic bite Lenz distribution annealing temperatures and the individual generalized metropolis annealing temperatures and acceptance probabilities.
  • the problem of slow convergence and low accuracy when the optimization method is applied to the complex functional can be overcome.
  • the accuracy and calculation efficiency of the optimization method of the present invention are optimized for the complex functional parameters described in the seventh embodiment to the ab initio data set having 600 data points.
  • the examples are compared with other methods.
  • the initial parameters were set off as follows.
  • Figure 27 shows the accuracy and convergence of the optimization obtained as a function of time, relative squared deviation (5 as a function of time. Note that (is shown on a logarithmic scale.
  • the method of the present invention is a simplex method, a descent method, Compared to classical Monte Carlo and Pena methods-Simplex method and steepest descent method have low accuracy and strength because optimized parameters are not easily captured and recovered by local minima, see Figure 27.
  • the metropolis optimization method overcomes the problem of local minima somewhat, but is slower to converge, while the Pena method converges much faster than the simplex method, but the steep descent method and the classical metropolis method are similar.
  • the accuracy is limited because of the occurrence of local minima after a short period of time
  • the optimization method of the present invention converges faster than all the methods listed above, and the other methods already Convergence steadily after a period of encountering local minimum problem To improve accuracy. Therefore, the optimization method of the present invention overcomes the problem of low computational efficiency and the problem of the occurrence of local minima, and is a complex functional having many different functions and many parameters to be optimized. Can be effectively applied to. Therefore, the optimization method of the present invention can be used as a general optimization and adaptation tool because ⁇ -efficiency can be applied to complex functionals with little need for user intervention and skill.
  • the problem of optimizing an initially unknown functional with some dependent variables, such as those found in methods of predicting and controlling trajectories and trajectories, is the problem of optimizing an initially unknown functional (device 400). ).
  • This problem is solved by using the parameter optimizing device (FIG. 11) for the device 400 and the parameter optimizing method of the present invention.
  • Data sensor 401 retrieves characteristic data of processes that have initially unknown functional dependencies, such as the trajectory of a guided missile.
  • the related data sensor 402 is used for fast optimization, control and prediction. Search for useful relative parameter data
  • the data collected by data sensor 401 is stored in data buffer 403, and the data collected by related data sensor 402 is stored in related data buffer 404.
  • Function analysis unit S 405 force, ', device Approximate the data set of the buffer 403 by the set of basic function devices of 405.
  • the functions used in the optimization and control processes are optimized by the optimization device »1093.
  • Optimized, and the devices 403, 405 and 1093 are connected by a data Z-address bus device 409.
  • the function form of the initial unknown function dependency of the input of the data sensor 401 is calculated using the related data. Used together with the data from the overnight buffer 404, the preset future time of the process being monitored by the sensor devices 401 and 402 Predict the data value at. Next, the previous predicted data value is used for the control device 407.
  • the control device 407 can process the measured data values so that the operation control system 408 can control the above process as desired, or can send the predicted data values to the output device 41 1. To do.
  • the operation control system unit 408 is connected by a control data / address bus unit IE 410 to deliver an output which can be used for controlling the subsequent process or directly the process.
  • a control data / address bus unit IE 410 to deliver an output which can be used for controlling the subsequent process or directly the process.
  • it is essential to provide very fast and very accurate means of process control with initial unknown function compliance. If an online process or control optimizer does not provide sufficient optimization convergence speed, it cannot properly direct or predict each control and optimization process. If the parameter predictions for the initial unknown function dependencies are not sufficiently accurate, dangerous process results can occur because the process control cannot be properly directed.
  • the optimization device of the present invention it is possible to achieve a very high speed and accuracy of the parameter prediction of the initial unknown function dependency, and to solve the problem caused by the insufficient convergence speed and accuracy of the optimization. Can be overcome.
  • the optimization device 1093 optimizes the parameters of the known function dependency provided by the function device 505.
  • the output of the optimizer 1093 provides data to the controller 507, which then invokes the control process and process predictions by delivering the respective output to the operating control system unit 508, and Device 508 connects to control data / address bus device 510 to deliver output for subsequent or direct process control.
  • a real-time control process such as a power daid control system or a conventional nuclear reactor or a fusion reactor, it is essential to have a very fast and very accurate means of process control with known function dependencies.
  • the online process or the control optimization system 5 does not provide sufficient optimization convergence speed, the respective control and optimization processes cannot be properly directed or predicted. If the parameter prediction of the known function dependencies is not sufficiently accurate, dangerous process results can occur because process control cannot be properly directed.
  • the optimization device of the present invention it is possible to achieve extremely high speed and accuracy of the overnight prediction of the parameter of the known function dependency, and to overcome the problem caused by insufficient convergence speed and accuracy of the optimization. be able to. With the optimization device of the present invention, it is possible to achieve high speed and accuracy of parameter prediction of the initial unknown function dependency, and it is possible to overcome problems caused by insufficient convergence speed and accuracy of optimization. .
  • the molecular modeling system of the present invention can be realized by selecting an appropriate intermolecular interaction potential for each interaction and adopting a time and power to select a development diagram of molecular coordinates. And modeling methods are higher than other commonly used molecular modeling systems and molecular modeling methods, achieving adaptability, efficiency and reliability.
  • the molecular interaction potential apparatus and method of the present invention through its design, further improves the adaptability, efficiency and reliability of molecular dynamics, and handles heterogeneous intermolecular interactions with high precision and efficiency.
  • the molecular modeling system and the molecule of the present invention can be obtained by selecting an appropriate molecular interaction potential that explains individual intermolecular interactions and using molecular mechanics to select an appropriate high-speed molecular coordinate optimization scheme.
  • the modeling method achieves higher efficiency, accuracy and reliability than other commonly used molecular modeling systems and methods.
  • the fast and efficient optimizer and method of the present invention further improves the accuracy, efficiency and reliability of the molecular modeling of the present invention.
  • the optimization apparatus and optimization method of the present invention can be used for process control and process prediction other than the problems of molecular dynamics and molecular modeling optimization, for example, known and unknown function dependence.
  • the molecular potential of the present invention is determined by individually selecting an appropriate intermolecular interaction potential that describes the interaction energy surface and using a molecular potential generator that selects an appropriate fast potential parameter overnight optimization scheme. Are modeling systems and molecular modeling methods different from other commonly used molecular modeling systems? Achieve higher efficiency and adaptability than the modeling method.
  • the fast, efficient optimizer and optimization method of the present invention further improves the accuracy, efficiency and reliability of the molecular potential generator, which reduces the potential parameters of reliability and accuracy with less effort. Evening can be provided.
  • Individual molecular interphase code writing of ft action potential and molecular model via pre-potential and infinity I-source Since the connection to the Dellization system is completely eliminated by the molecular modeling system of the present invention and the molecular potential generator which is part of the molecular modeling method, we try to solve the problem of molecular modeling The burden on existing users is significantly reduced.
  • the molecular modeling system and the molecular modeling method of the present invention further Achieve great efficiency and convenience for the user. This is because any part of the intermolecular interaction can be designed appropriately in a molecular modeling system using appropriate knowledge of the modeling problem and the core of the user's problem.
  • the problems that arise in the modeling of semiconductors for the interaction of gallium and gallium can be calculated accurately and at a much higher cost-efficiency than currently used axioms or the so-called ab initio method.
  • the adaptability and accuracy of the molecular modeling system and the molecular modeling method of the present invention are enhanced by the heterogeneous separation-potential device of the present invention based on the heterogeneous molecular interaction method of the present invention. Can solve various problems that could not be solved before.
  • the fast self-tuning optimization method of the present invention for analytic arbitrary functionals allows for adaptation tools that require little user intervention and little user skill, such as software tools such as Kaleida Graph® , Cricket Graph (registered trademark), Dalta Graph (registered trademark), and other graphical user interface (GUI) tools, and wide spread non-GUI mainframe optimization tools It can be implemented by inserting it into graphic data display software instead of the unreliable simplex method and the descent method (seen on personal computers, ⁇ —stations).
  • software tools such as Kaleida Graph® , Cricket Graph (registered trademark), Dalta Graph (registered trademark), and other graphical user interface (GUI) tools
  • GUI graphical user interface
  • the current adaptive optimization apparatus and method are inefficient and the complex multi-parameter functionals, which could not be optimized in the past, are replaced by the fii It can be optimized by equipment and optimization method.
  • the high-speed, self-adjusting optimization device and optimization method of the present invention can be found in the topics of control and prediction of nuclear reactors, aircraft, missile tracking and concealment, 3 ⁇ 4 grids, satellites, etc. It can be used for real-time process control and process prediction of known and unknown function substitutability.

Abstract

Cette invention se rapporte à un système de modélisation moléculaire et à un procédé de modélisation moléculaire qui utilisent un outil de dynamique moléculaire (MD) se servant d'un nouveau potentiel correspondant à une interaction intermoléculaire hétérogène, un outil de modélisation moléculaire (MM) effectuant une nouvelle optimisation d'autorégulation rapide, un nouvel outil générateur de potentiel moléculaire (MPG), et un outil base de données expertes de modélisation moléculaire, afin de résoudre les problèmes concernant les interactions intermoléculaires spécifiées lors de la modélisation des interactions intermoléculaires homogènes et hétérogènes, ces problèmes étant tels que la précision de la fonctionnalité du potentiel est faible et que soit la fonction du potentiel soit le paramètre du potentiel est inconnu. Ce système et ce procédé servent à résoudre le problème dû au fait qu'un paramètre concernant la fonctionnalité d'une interaction spécifiée est cinconnu, le problème dû au fait que le coût du calcul pour l'optimisation d'une nouvelle fonctionnalité de potentiel est élevé et le problème de la sélection d'une valeur initiale des procédures d'optimisation.
PCT/JP1997/001951 1996-06-07 1997-06-09 Systeme de modelisation moleculaire et procede de modelisation moleculaire WO1997046949A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP14543096 1996-06-07
JP8/145430 1996-06-07

Publications (1)

Publication Number Publication Date
WO1997046949A1 true WO1997046949A1 (fr) 1997-12-11

Family

ID=15385071

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP1997/001951 WO1997046949A1 (fr) 1996-06-07 1997-06-09 Systeme de modelisation moleculaire et procede de modelisation moleculaire

Country Status (1)

Country Link
WO (1) WO1997046949A1 (fr)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005034843A (ja) * 1998-06-26 2005-02-10 Toshiba Corp 反応メカニズム算出法、および反応メカニズム算出方法を記録した記録媒体
WO2006102228A1 (fr) * 2005-03-18 2006-09-28 Eve Zoebisch Procede et systeme de modelisation moleculaire
JPWO2008041304A1 (ja) * 2006-09-29 2010-01-28 富士通株式会社 分子力場割り当て方法、分子力場割り当て装置、および分子力場割り当てプログラム
CN110634537A (zh) * 2019-07-24 2019-12-31 深圳晶泰科技有限公司 用于有机分子晶体结构高精度能量计算的双层神经网算法
CN110858505A (zh) * 2018-08-23 2020-03-03 塔塔咨询服务有限公司 用于预测原子元素及其合金材料的结构和性质的系统和方法
CN111863141A (zh) * 2020-07-08 2020-10-30 深圳晶泰科技有限公司 分子力场多目标拟合算法库系统及工作流程方法
JP7314097B2 (ja) 2020-06-11 2023-07-25 住友重機械工業株式会社 シミュレーション方法、シミュレーション装置、及びプログラム

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08161290A (ja) * 1994-12-06 1996-06-21 Fujitsu Ltd ポテンシャルライブラリ装置

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08161290A (ja) * 1994-12-06 1996-06-21 Fujitsu Ltd ポテンシャルライブラリ装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
PHYSICAL REVIEW E, Vol. 53, No. 2, February 1996, JURGEN SCHULTE, "Nonpolynomial Fitting of Multiparameter Functions", p. R1348-R1350. *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005034843A (ja) * 1998-06-26 2005-02-10 Toshiba Corp 反応メカニズム算出法、および反応メカニズム算出方法を記録した記録媒体
JP4713856B2 (ja) * 1998-06-26 2011-06-29 株式会社東芝 反応メカニズム算出装置及び反応メカニズム算出方法
WO2006102228A1 (fr) * 2005-03-18 2006-09-28 Eve Zoebisch Procede et systeme de modelisation moleculaire
US7797144B2 (en) 2005-03-18 2010-09-14 Eve Zoebisch Molecular modeling method and system
JPWO2008041304A1 (ja) * 2006-09-29 2010-01-28 富士通株式会社 分子力場割り当て方法、分子力場割り当て装置、および分子力場割り当てプログラム
CN110858505A (zh) * 2018-08-23 2020-03-03 塔塔咨询服务有限公司 用于预测原子元素及其合金材料的结构和性质的系统和方法
CN110858505B (zh) * 2018-08-23 2023-06-20 塔塔咨询服务有限公司 用于预测原子元素及其合金材料的结构和性质的系统和方法
CN110634537A (zh) * 2019-07-24 2019-12-31 深圳晶泰科技有限公司 用于有机分子晶体结构高精度能量计算的双层神经网算法
CN110634537B (zh) * 2019-07-24 2022-03-18 深圳晶泰科技有限公司 用于有机分子晶体结构高精度能量计算的双层神经网方法
JP7314097B2 (ja) 2020-06-11 2023-07-25 住友重機械工業株式会社 シミュレーション方法、シミュレーション装置、及びプログラム
CN111863141A (zh) * 2020-07-08 2020-10-30 深圳晶泰科技有限公司 分子力场多目标拟合算法库系统及工作流程方法

Similar Documents

Publication Publication Date Title
Kluger et al. Quantum Vlasov equation and its Markov limit
Jasper et al. Introductory lecture: Nonadiabatic effects in chemical dynamics
Yuan et al. An extended quadrature method of moments for population balance equations
Rosso et al. On the use of the adiabatic molecular dynamics technique in the calculation of free energy profiles
Saakian et al. Dynamics of the Eigen and the Crow-Kimura models for molecular evolution
KR100582969B1 (ko) 플라스마-처리의 표면 프로파일을 예측하는 방법 및 장치
Nakayama et al. Forward–backward semiclassical dynamics for quantum fluids using pair propagators: Application to liquid para-hydrogen
Rico-Martınez et al. Coarse projective kMC integration: forward/reverse initial and boundary value problems
Berendsen et al. Molecular dynamics simulations: Techniques and approaches
Wang et al. Implicit-solvent coarse-grained modeling for polymer solutions via Mori-Zwanzig formalism
Fang et al. Adaptive Euler–Maruyama method for SDEs with non-globally Lipschitz drift
WO1997046949A1 (fr) Systeme de modelisation moleculaire et procede de modelisation moleculaire
Gamboa et al. Influence of thermostats on the calculations of heat capacities from Born‐Oppenheimer molecular dynamics simulations
Mittal et al. Path integral Monte Carlo method for the quantum anharmonic oscillator
Alfonso et al. A flexible nudged elastic band program for optimization of minimum energy pathways using ab initio electronic structure methods
Ohtsuki et al. Monotonically convergent algorithms for solving quantum optimal control problems described by an integrodifferential equation of motion
Tzorakoleftherakis et al. Model-based reactive control for hybrid and high-dimensional robotic systems
Mitrakos et al. Sectional modeling of aerosol dynamics in multi-dimensional flows
Zhang et al. Kinetic energy conserving integrators for Gaussian thermostatted SLLOD
Mandyam et al. Statistical simulations of diffusional coarsening in finite clusters
Bindal et al. Equation-free, coarse-grained computational optimization using timesteppers
Biring et al. A new adaptive mutation simulated annealing algorithm: application to the study of pure and mixed Pt–Pd clusters
Pesce et al. Variational wave packet method for dissipative photodesorption problems
Ni et al. Construction of stochastic PDEs for feedback control of surface roughness in thin film deposition
Nguyen et al. Model order reduction for reacting flows: Laminar Gaussian flame applications

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): CN JP KR US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FI FR GB GR IE IT LU MC NL PT SE

DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase