An automated system and method for molecular dynamics simulation based on decision and execution hierarchy

Abstract

The application discloses a kind of molecular dynamics simulation automation system and method based on decision and execution hierarchy.It comprises from top to bottom user interaction layer, decision layer, deterministic checking layer, execution layer and achievement output layer, user interaction layer is used to receive the molecular dynamics simulation research demand submitted by user with natural language, decision layer is parsed research demand by large language model, carries out field literature research and simulation scheme deduction, generates structured scheme data;Deterministic checking layer carries out deterministic checking operation to structured scheme data, divides checking result, feedbacks the modification suggestion of structured scheme data that does not pass checking to user interaction layer, retains checking record and releases scheme;Execution layer is analyzed to the structured scheme data that passes checking;Achievement output layer receives the data analysis result output by execution layer, and outputs scientific result and report document with complete traceability chain.

Description

Technical Field

[0001] This invention relates to the field of computer-aided scientific computing and artificial intelligence, and more specifically to an automated system and method for molecular dynamics simulation based on decision-making and execution hierarchies. Background Technology

[0002] Molecular dynamics (MD) simulations are a crucial tool for studying microscopic mechanisms in fields such as materials science, chemistry, and energy. A complete MD study typically involves multiple steps, including literature review, determining the research protocol, selecting the force field, constructing atomic structures, generating simulation input scripts (such as LAMMPS input files), running the simulation, post-processing analysis, and generating a report. These steps heavily rely on the researcher's professional experience, involve numerous parameters, are cumbersome and error-prone, posing a high barrier to entry not only for beginners but also requiring experienced researchers to perform a large amount of repetitive, mechanical operations. Therefore, automating the entire molecular dynamics (MD) simulation process is an effective way to address these challenges.

[0003] Currently, traditional molecular dynamics research workflows and existing cloud-based automated molecular dynamics systems still have significant shortcomings in scenarios involving complex system research, multi-parameter simulation, and batch configuration analysis. (1) Traditional molecular dynamics research schemes mainly rely on the personal experience of researchers. The research ideas are easily limited by the existing knowledge and experience boundaries. It is difficult to fully form multiple candidate simulation schemes, experimental verification schemes and analysis paths in the early stage of research. At the same time, molecular dynamics simulations will generate a large number of molecular configurations. It is subjective to manually identify the molecular configurations required for the research, and it is difficult to review frame by frame to find the most needed experimental results. (2) Although existing cloud platform molecular dynamics calculation systems can lower the threshold for using molecular dynamics software through graphical workflow, automatic matching of force field parameters, job submission and trajectory analysis, they are mainly geared towards the automation of the process after the user has clearly defined the calculation simulation task. They lack a mechanism for multiple rounds of discussion, manual confirmation and approval control in key links such as the formation of research hypotheses, the formulation of modeling schemes, the screening of force fields, the submission of jobs and the formation of conclusions. It is difficult to meet the controllable design requirements in scenarios where the research scheme is uncertain or requires repeated judgment. (3) Existing automated molecular dynamics systems are mostly limited to trajectory analysis and property extraction based on preset indicators. They are difficult to accurately identify key states from a large number of configurations for specific research objectives. Furthermore, they lack a full-link logical connection mechanism that runs through research objectives, simulation parameters, experimental verification and analysis conclusions, resulting in the fragmentation of each research link.

[0004] Therefore, it is urgent to rely on large language models and intelligent agent workflows to expand research paradigms and achieve intelligent screening of key configurations while ensuring researchers' leadership, and to build a consistent logical chain from research design to theoretical conclusions, thereby significantly improving the reliability, controllability and reproducibility of molecular dynamics research. Summary of the Invention

[0005] In view of this, this invention provides an automated molecular dynamics simulation system and method based on a decision-making and execution hierarchy to address the problems of traditional research's over-reliance on human experience, low efficiency in configuration screening, separation of design and analysis, and the lack of multi-round academic discussions, manual verification, and end-to-end logical consistency control in existing automated platforms. This ensures that: the natural language-driven "decision / planning" and the scientific computing "execution / output" are isolated from each other in the system architecture; all calculations, verifications, and reports affecting the final scientific results are completed by deterministic code; and deterministic verification and manual verification are set between the two layers and at key nodes in the execution layer. This method retains the advantages of large language model interaction and the ability to expand ideas, while effectively avoiding the interference of its randomness and distortion on scientific results, ensuring that the entire research process always revolves around a unified main thread, thereby improving the controllability of the scheme, the accuracy of configuration screening, the consistency of conclusions, and the reproducibility of results.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An automated molecular dynamics simulation system based on a hierarchical decision-making and execution framework includes, from top to bottom, a user interaction layer, a decision layer, a deterministic verification layer, an execution layer, and a result output layer. Each layer forms a closed-loop feedback path and a security interception path. The user interaction layer is used to receive molecular dynamics simulation research requests submitted by users in natural language and transmit the requests to the decision layer; The decision-making level analyzes research needs through large language models, conducts domain literature surveys and simulation scheme deductions, and generates structured scheme data that is strongly constrained by predefined structured schemas; The deterministic verification layer performs deterministic verification operations on the structured solution data based on physical principles and domain knowledge, divides the verification results, feeds back the modification suggestions of the structured solution data that fails the verification to the user interaction layer, retains the verification record and allows the solution to pass; The execution layer receives the validated structured scheme data and sequentially completes atomic structure construction, simulation input script generation, molecular dynamics simulation calculation, trajectory data post-processing analysis, and result visualization through deterministic code. The output layer receives the data analysis results from the execution layer and produces scientific results and reports with a complete traceability chain.

[0007] Preferably, the structured schema serves as a data exchange channel between the decision-making layer and the execution layer, used to output valid structured schema data to the execution layer.

[0008] Preferably, the deterministic verification layer, as a centrally located forced interception node, is deployed on the data path between the decision layer and the execution layer, and embedded in key nodes within the execution layer. It incorporates a rule set and quantization thresholds built based on physical principles and domain knowledge, and performs deterministic verification operations on the structured scheme objects output by the decision layer and the intermediate products generated by the execution layer.

[0009] Preferably, the verification operation includes logical consistency checks and physical rationality checks, and the verification results include fatal errors, warnings, and information.

[0010] Preferably, the execution layer provides service calls to other layers through the MCP tool unit and standardized interfaces.

[0011] Preferably, the report content is generated by assembling deterministic code through template filling. The conclusion paragraph is based solely on the numerical results obtained from deterministic calculations. Deterministic calculations refer to the use of mathematical functions and statistical algorithms to perform smoothing filtering, linear fitting, extreme value search, and feature value extraction operations on the simulated trajectory data and thermodynamic logs output by the execution layer to obtain MD simulation feature values. The MD simulation feature values ​​are then input into the system's built-in logical decision tree, and the system automatically matches and calls the objective statement sentences in the preset conclusion template library based on the data interval in which they are located. The conclusion paragraph is generated by concatenating the templates through deterministic mapping.

[0012] Preferably, an automated method for molecular dynamics simulation based on a decision-making and execution hierarchy includes: Receive molecular dynamics simulation research requests submitted by users in natural language; By analyzing research needs through large language models, conducting domain literature surveys and simulation scheme deductions, and generating structured scheme data strongly constrained by predefined structured schemas; Perform deterministic verification operations based on physical principles and domain-specific knowledge on the structured scheme data, classify the verification results, provide modification suggestions for the structured scheme data that failed the verification, retain the verification records and release the scheme; It receives verified structured scheme data and sequentially completes atomic structure construction, simulation input script generation, molecular dynamics simulation calculation, trajectory data post-processing analysis, and result visualization through deterministic code. Based on the data analysis results, scientific results and reports with a complete traceability chain are produced.

[0013] As can be seen from the above technical solution, compared with the prior art, this invention discloses an automated molecular dynamics simulation system and method based on decision-making and execution hierarchies. It conducts multiple rounds of academic discussions by calling large language model services and structures the discussion results into molecular dynamics task specification objects. Simultaneously, it isolates the discussion and planning processes involving large models from the deterministic scientific calculation processes. All aspects affecting the final scientific results—modeling, force field file generation, LAMMPS script generation, hierarchical verification, trajectory analysis, and report generation—are completed by deterministic code, molecular dynamics rules, and manual verification mechanisms. This invention can expand research ideas while avoiding the contamination of scientific results by the randomness or non-determinism of large language models; it can screen key configurations related to the research objectives from a large number of trajectory configurations, improving the relevance of theoretical analysis; and it can maintain a consistent research logic from research design, modeling, operation to conclusion analysis through task specification objects and simulation workflow diagrams, thereby improving the controllability, reliability, interpretability, and reproducibility of the molecular dynamics research process. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This invention provides an architectural diagram of an automated molecular dynamics simulation system based on a decision-making and execution hierarchy.

[0016] Figure 2 The planning-execution dual-mode workflow diagram provided for this invention.

[0017] Figure 3 The flowchart of the structured schema strongly constrained interface provided by this invention.

[0018] Figure 4 A flowchart for generating a deterministic report provided by this invention.

[0019] Figure 5 This is a diagram illustrating the mineral composition of clay from the Gulong region, provided for the purposes of this invention.

[0020] Figure 6 This is a diagram illustrating the initial molecular model construction of illite provided by the present invention.

[0021] Figure 7 This is a supercell model diagram of illite provided by the present invention.

[0022] Figure 8This is an image showing the automatic generation of LAMMPS simulation input files (data and in files) provided by the present invention.

[0023] Figure 9 The stress-strain curves from the simulation results of illite micromechanical properties provided by this invention.

[0024] Figure 10 A snapshot of the uniaxial tensile structure in the simulation results of the micromechanical properties of illite provided by this invention.

[0025] Figure 11 A snapshot of the uniaxial compression structure in the simulation results of illite micromechanical properties provided by this invention.

[0026] Figure 12 A snapshot of the uniaxial shear structure in the simulation results of illite micromechanical properties provided by this invention.

[0027] Figure 13 The diagram showing the uniaxial tensile mechanical properties in the molecular dynamics simulation report automatically generated by Auto MD provided for this invention.

[0028] Figure 14 The diagram showing the uniaxial compressive mechanical properties in the molecular dynamics simulation report automatically generated by Auto MD provided for this invention.

[0029] Figure 15 The diagram showing the uniaxial shear mechanical properties in the molecular dynamics simulation report automatically generated by Auto MD provided for this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that the described embodiments are only used to explain the technical concept of the present invention and are not intended to limit the scope of protection of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] This invention discloses an automated molecular dynamics simulation system based on decision-making and execution hierarchies. It adopts a three-level hierarchical architecture of "decision-verification-execution" for scientific computing tasks. From top to bottom, it includes five functional levels: user interaction layer, decision layer, deterministic verification layer, execution layer, and result output layer. Each level forms a closed-loop feedback path and a security interception path.

[0032] like Figure 1As shown, the system in this embodiment adopts a dual-mode collaborative working mechanism of "planning-execution": In planning mode, the decision-making level and users discuss and converge to form an unambiguous and executable structured solution through multiple rounds of interaction; after the solution converges, the system generates a standardized solution confirmation block, which clearly lists elements such as system composition, force field selection, key parameters and output path, and enters the confirmation signal discrimination stage—the process can only enter the execution mode when the user gives a clear affirmative instruction; if the user gives a vague response, the system will determine that it has not been effectively authorized and automatically fall back to the planning mode to continue iterating. In execution mode, the system strictly follows the confirmed solution, sequentially calls the corresponding deterministic toolchain components to perform the corresponding modeling, simulation and analysis operations, and transmits the underlying execution status and calculation results to the user end in real time; once a fatal error is detected or the system's capacity boundary is exceeded, the interruption mechanism is triggered, forcibly stopping the execution and guiding the process to fall back to the planning mode, thereby achieving safe and controllable closed-loop iteration.

[0033] The user interaction layer serves as the system entry point, handling natural language intelligent interaction between the user and the system. This layer also acts as a failover node; when any subsequent step detects an error, the system immediately interrupts the current process and returns control to the user interaction layer, allowing the user to intervene and correct the error, thereby achieving rapid failover and manual intervention.

[0034] The decision-making layer, driven by a large language model, is responsible for understanding user intent, retrieving relevant literature, and generating research proposals. It guides users to confirm their decisions at key points through multiple rounds of natural language interaction. This layer only outputs proposal data strongly constrained by a predefined structured schema and does not directly produce any scientific computation results. This strictly constrains the inherently non-deterministic output of the large language model, and this stage is solely used to construct research proposals and related suggestions.

[0035] Furthermore, such as Figure 3 As shown, this embodiment establishes a unique, predefined structured schema object interface, namely the structured schema, between the decision-making layer and the execution layer. This interface serves as the sole channel for data exchange between the two layers, strictly limiting the data type and value range of each field and performing whitelist verification. Simultaneously, this interface incorporates normalization and fuzzy correction mechanisms, proactively filtering out syntax and format errors that may arise from the large language model before data enters the execution layer. The execution layer only receives and processes legal objects that have passed schema verification, ensuring that its input always conforms to the parsing requirements of deterministic code. This fundamentally prevents unstructured or illegal input from interfering with subsequent scientific computation, achieving a safe, reliable, and lossless conversion from "language output" to "executable instructions."

[0036] The deterministic verification layer, acting as a central, mandatory interception node, is deployed on the data path between the decision-making layer and the execution layer, and embedded in key nodes within the execution layer. This verification layer incorporates a rule set and quantified thresholds built upon physical principles and domain knowledge, performing deterministic verification operations on the structured solution objects output by the decision-making layer and the intermediate products generated by the execution layer. Verification content includes, but is not limited to, logical consistency checks and physical rationality checks. The thresholds relied upon for each verification are derived from classical physical laws or authoritative literature. The verification results are based on hard rules and are uniformly divided into three levels: "fatal error," "warning," and "information." Firstly, fatal errors: when a defect that will inevitably lead to calculation failure, incorrect results, or damage to reproducibility is detected, the process is immediately interrupted and the specific error field and recommended value are fed back to the user. Furthermore, it is explicitly prohibited for the large language model or underlying code to automatically modify the error field, thereby ensuring the human controllability and transparency of the fault handling process. Secondly, warning: When a parameter is detected to be in a suspicious range, but logically may belong to a legitimate extreme simulation scenario, the system will record it and allow the process to continue. Thirdly, information: The system fully records the specific actions, inspection items, and pass / fail status of each verification to ensure that the entire simulation process is completely traceable.

[0037] For further details, please see the appendix. Figure 2 This embodiment embeds a deterministic physical rationality verification module at key nodes in the execution layer (such as the structure construction completion node and the simulation input generation node) to perform multi-dimensional and quantifiable scientific constraint checks on the scheme object and intermediate products. Specifically, this includes: the existence and readability of the structural files (verified through a standard atomic structure library), the total charge balance of the system, the physical consistency of the force field and pairing potential, the matching of the integration step size and the force field type (e.g., the hard constraint of the reaction force field on the upper limit of the integration step size), the order-of-magnitude rationality of parameters such as temperature / pressure / number of steps / system size, the completeness of the force field's coverage of the elements contained in the system, and the non-emptiness and validity of the literature citations of the force field parameters. All verification thresholds are set according to explicit scientific or engineering standards. Verification results are uniformly output in a three-level classification: when determined to be a "fatal error," a forced interruption is triggered, the error message is returned as is, along with specific fields to be modified and recommended values, and the system is guided back to the planning mode; when determined to be a "warning" or "information," it is recorded and the process is allowed to continue, ultimately proceeding to the next execution node. This mechanism ensures physical consistency and fault controllability throughout the entire process from planning to execution.

[0038] The execution layer is implemented entirely with deterministic code and does not call any large language models throughout the process. Based on the validated structured scheme data, it sequentially completes tasks such as atomic structure construction, simulation input script generation, molecular dynamics simulation, data post-processing and analysis, visualization, and report generation. All of the above functions are provided through the MCP tool unit (Model-Code-Pipeline) and through standardized interfaces to provide services to other levels of the system, ensuring that the same output is produced under the same input conditions, thus fundamentally guaranteeing the reproducibility of scientific results.

[0039] The output layer is responsible for producing scientific results and reports with a complete traceability chain; all report content is generated by assembling deterministic code through template filling, and the conclusion section is based solely on the numerical results obtained from deterministic calculations.

[0040] Furthermore, such as Figure 4 As shown, the report generation process in this embodiment is dominated by the deterministic report module, which strictly adopts a template-filling mechanism: this module receives structured outputs from the execution layer, including a list of documents, quantitative metrics, and images, and automatically assembles them into the final report document according to a preset template; the conclusion section only objectively states the numerical results obtained from deterministic calculations, without calling large language models throughout the process, avoiding subjective inferences or generative modifications; when a certain type of output is missing, the system executes a graceful degradation strategy, that is, automatically skips the corresponding report paragraph without triggering errors or process interruptions, thereby ensuring the continuity of the report generation process. This design ensures that the report content is strictly consistent with the underlying calculation results, is traceable and reproducible, fundamentally avoiding the distortion or exaggeration of conclusions caused by model nondeterminism, and strengthening the credibility and rigor of scientific research results. The report content is generated by assembling deterministic code using a template-filling method. The conclusion paragraph is based solely on the numerical results obtained from deterministic calculations and provides an objective statement. Deterministic calculations refer to the process of performing smoothing filtering, linear fitting, extreme value search, and feature value extraction operations on the simulated trajectory data and thermodynamic logs output by the execution layer, based on mathematical functions and statistical algorithms, to obtain MD simulation feature values. The MD simulation feature values ​​are then input into the system's built-in logic decision tree, which automatically matches and calls objective statement sentences from a pre-set conclusion template library based on the data interval in which they are located. The conclusion paragraph is generated by concatenating these objective statement sentences through deterministic mapping.

[0041] Based on the aforementioned layered architecture, this invention further adopts a "planning-execution" dual-mode collaborative working mechanism: In planning mode, the decision-making layer and the user converge to form an unambiguous structured solution through multiple rounds of interaction, and generate a standardized solution confirmation block; the system can only switch to execution mode when the user gives a clear affirmative instruction; in execution mode, the system strictly follows the confirmed solution to advance the task, and once a fatal error occurs or the system's capacity boundary is exceeded, the process is interrupted and automatically reverts to planning mode, thereby forming a safe and controllable closed-loop iterative mechanism.

[0042] In summary, this invention can automate key operational steps in molecular dynamics simulation research, including automated survey of simulation parameters, automated construction of atomic structure models, automated setting of specific simulation parameters, automated operation of LAMMPS simulations, and automated generation of research conclusion reports, thereby forming an end-to-end automated scientific research toolchain covering the understanding of research intentions to the output of conclusion reports.

[0043] Building upon this foundation, this invention provides a technical path for transforming molecular dynamics simulation research from an experience-dependent model to an automated computational model. On the one hand, by leveraging the natural language understanding and domain knowledge retrieval capabilities of large language models, decision-making processes such as force field selection, parameter configuration, and system construction, which were originally highly dependent on researchers' professional experience, are transformed into standardized processes that can be completed with system assistance, significantly reducing the technical threshold and human resource investment required for molecular dynamics simulation. On the other hand, through the mandatory interception mechanism of the deterministic verification layer and the deterministic code implementation of the execution layer, the risk of contamination of scientific calculation results by the distortion of content generated by large language models is fundamentally avoided, ensuring that the simulation results produced have scientific credibility, reproducibility, and full traceability. Therefore, this embodiment provides beginners with a low-threshold, easy-to-use introductory tool for molecular dynamics simulation, and also provides professional researchers with a high-efficiency, scalable, high-throughput computational research platform, thereby promoting the transformation and upgrading of molecular dynamics simulation from a traditional manual operation research model to an intelligent and large-scale modern scientific research paradigm.

[0044] Specifically, an automated method for molecular dynamics simulation based on a hierarchical decision-making and execution framework includes: Receive molecular dynamics simulation research requests submitted by users in natural language; By analyzing research needs through large language models, conducting domain literature surveys and simulation scheme deductions, and generating structured scheme data strongly constrained by predefined structured schemas; Perform deterministic verification operations based on physical principles and domain knowledge on the structured scheme data, classify the verification results, provide modification suggestions for the structured scheme data that failed the verification, retain the verification records and release the scheme; The structured scheme data that has passed verification is used to sequentially complete atomic structure construction, simulation input script generation, molecular dynamics simulation calculation, trajectory data post-processing analysis, and result visualization through deterministic code. Based on the data analysis results, scientific results and reports with a complete traceability chain are produced.

[0045] The specific embodiments provided by this invention are as follows: Taking the study of the micromechanical properties of major clay minerals in the Gulong Depression shale reservoir of the Songliao Basin as an example, this paper focuses on simulating and studying the tensile, compressive, and shear mechanical properties of the most abundant clay minerals in this region. Given the complex composition of clay minerals in the shale reservoirs of this region and the significant impact of their mechanical behavior on reservoir stimulation, this paper utilizes an automated molecular dynamics simulation system based on decision-making and execution hierarchies (hereinafter referred to as the Auto MD system) proposed in this invention to study the molecular-scale mechanical properties of clay minerals. First, the Auto MD software is used to search and summarize previous literature and core X-ray diffraction (XRD) experimental results. The types and relative contents of clay minerals in the Gulong Depression shale reservoir are meticulously counted. The analysis results show that the clay minerals in the Gulong Depression shale reservoir are mainly illite, illite / montmorillonite mixed layer, chlorite, and kaolinite, with illite content of 45%, illite / montmorillonite mixed layer content of 30%, chlorite content of 12%, and kaolinite content of 8%. The total content of clay minerals accounts for 30%-45% of the rock mineral composition. Figure 5 As shown.

[0046] Subsequently, based on the statistical results of clay mineral content obtained from Auto MD software, illite, which has a relatively high content, was selected as the research object for this molecular dynamics simulation. After determining the research object, Auto MD software provided suggestions on force field selection and model selection for key parameters in the modeling process. Combining the characteristics of illite crystal structure and experience in simulating layered silicate minerals, the ClayFF force field was selected as the simulation force field, and the standard illite unit cell was selected as the initial modeling unit. Based on the unit cell parameters of illite in the crystallography database, an initial molecular model of illite was constructed using Auto MD software, such as... Figure 6 As shown. To study the mechanical properties of illite, the initial molecular model of illite needs to be expanded. AutoMD provides three suggestions for cell expansion, but based on the research requirements, the researchers determined the cell expansion scale to be 4×4×4 (20.76×35.84×39.64Å), obtaining the illite supercell model required for this simulation. The constructed supercell model was then displayed using OVITO, as shown. Figure 7 As shown.

[0047] Subsequently, the data file of illite and the in file required for mechanical simulation were constructed using Auto MD software. The simulation included three loading modes: Z-axis uniaxial tension, Z-axis uniaxial compression, and XY shear. Stress-strain curves, dump structural evolution trajectory, and final were also set. The .data file contains deformed structural files and other output parameters, which together form the complete input file required for LAMMPS simulation. Figure 8 As shown.

[0048] The constructed data and in files were input into the supercomputing platform using AutoMD software for simulation. After simulation, stress-strain curves and OVITO visualization snapshots were obtained. The simulation results were then post-processed and analyzed using AutoMD software. The simulation results are shown below. Figures 9-12 As shown.

[0049] The AutoMD software was used to integrate the above modeling process, simulation parameters, simulation results, and post-processing analysis, automatically generating a molecular dynamics simulation study report on illite in the Gulong Depression shale reservoir of the Songliao Basin. The report systematically presents the complete research chain from clay mineral composition survey, selection of research objects, molecular model construction, mechanical simulation implementation to result analysis, providing a molecular-scale mechanical basis for optimizing the fracturing and stimulation technology of the Gulong Depression shale reservoir. Figures 13-15 As shown.

[0050] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0051] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automated molecular dynamics simulation system based on a hierarchical decision-making and execution mechanism, characterized in that, include: The system is structured from top to bottom, comprising a user interaction layer, a decision-making layer, a deterministic verification layer, an execution layer, and a results output layer. These layers form a closed-loop feedback path and a security interception path. The user interaction layer is used to receive molecular dynamics simulation research requests submitted by users in natural language and transmit the requests to the decision layer; The decision-making level analyzes research needs through large language models, conducts domain literature surveys and simulation scheme deductions, and generates structured scheme data that is strongly constrained by predefined structured schemas; The deterministic verification layer performs deterministic verification operations on the structured solution data based on physical principles and domain knowledge, divides the verification results, feeds back the modification suggestions of the structured solution data that fails the verification to the user interaction layer, retains the verification record and allows the solution to pass; The execution layer receives the verified structured scheme data and sequentially completes atomic structure construction, simulation input script generation, molecular dynamics simulation calculation, trajectory data post-processing analysis, and result visualization through deterministic code. The output layer receives the data analysis results from the execution layer and produces scientific results and reports with a complete traceability chain.

2. The automated molecular dynamics simulation system based on decision-making and execution hierarchy according to claim 1, characterized in that, The structured schema serves as a data exchange channel between the decision-making layer and the execution layer, used to output valid structured schema data to the execution layer.

3. The automated molecular dynamics simulation system based on decision-making and execution hierarchy according to claim 1, characterized in that, The deterministic verification layer, acting as a central mandatory interception node, is deployed on the data path between the decision layer and the execution layer and embedded in key nodes within the execution layer. It incorporates a rule set and quantization thresholds built based on physical principles and domain knowledge, and performs deterministic verification operations on the structured scheme objects output by the decision layer and the intermediate products generated by the execution layer.

4. The automated molecular dynamics simulation system based on decision-making and execution hierarchy according to claim 3, characterized in that, The verification operation includes logical consistency checks and physical rationality checks, and the verification results include fatal errors, warnings, and messages.

5. The automated molecular dynamics simulation system based on decision-making and execution hierarchy according to claim 1, characterized in that, The execution layer provides service calls to other layers through the MCP tool unit and standardized interfaces.

6. The automated molecular dynamics simulation system based on decision-making and execution hierarchy according to claim 1, characterized in that, The report content is generated by assembling deterministic code using a template-filling method. The conclusion paragraph is based solely on the numerical results obtained from deterministic calculations and provides an objective statement. Deterministic calculations refer to the process of performing smoothing filtering, linear fitting, extreme value search, and feature value extraction operations on the simulated trajectory data and thermodynamic logs output by the execution layer, based on mathematical functions and statistical algorithms, to obtain MD simulation feature values. The MD simulation feature values ​​are then input into the system's built-in logic decision tree, which automatically matches and calls objective statement sentences from a pre-set conclusion template library based on the data interval in which they are located. The conclusion paragraph is generated by concatenating these objective statement sentences through deterministic mapping.

7. An automated method for molecular dynamics simulation based on decision-making and execution hierarchies, characterized in that, include: Receive molecular dynamics simulation research requests submitted by users in natural language; By analyzing research needs through large language models, conducting domain literature surveys and simulation scheme deductions, and generating structured scheme data strongly constrained by predefined structured schemas; Perform deterministic verification operations based on physical principles and domain-specific knowledge on the structured scheme data, classify the verification results, provide modification suggestions for the structured scheme data that failed the verification, retain the verification records and release the scheme; It receives verified structured scheme data and sequentially completes atomic structure construction, simulation input script generation, molecular dynamics simulation calculation, trajectory data post-processing analysis, and result visualization through deterministic code. Based on the data analysis results, scientific results and reports with a complete traceability chain are produced.

Images