Multiphase chemical reaction transition state certainty generation method and system based on physical constraint and adaptive optimization, and storage medium

Abstract

The invention relates to the cross field of computational chemistry and artificial intelligence, in particular to a multiphase chemical reaction transition state certainty generation method and system based on physical constraint and adaptive optimization and a storage medium. The method is executed by a computer system and comprises the following steps: acquiring initial structure data of reactants and products; calling a nonlinear initial guess module to execute path search based on the coarse-grained potential energy surface so as to generate a nonlinear initial guess path; a constraint optimal transmission module is called to generate a candidate transition state TS structure; the invention provides a multiphase chemical reaction transition state certainty generation method and system based on physical constraint and adaptive optimization and a storage medium, which not only keep high efficiency of a machine learning method, but also realize accurate, robust and reliable simulation of complex and real chemical scenes.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of computational chemistry and artificial intelligence, specifically to a method, system, and storage medium for the deterministic generation of transition states in multiphase chemical reactions based on physical constraints and adaptive optimization. This invention is particularly suitable for scenarios requiring high precision and efficiency in exploring complex chemical reaction mechanisms and designing novel catalysts and functional materials. Background Technology

[0002] The transition state (TS) of a chemical reaction is the instantaneous state during which a system experiences its highest energy and least stable structure as reactants transform into products. Information on the structure and energy of the transition state is crucial for understanding the intrinsic mechanisms of chemical reactions, calculating reaction rates, predicting reaction pathways, and designing novel catalysts or functional materials that can efficiently catalyze specific reactions. Therefore, accurately and efficiently determining the structure of the transition state is one of the core challenges in modern chemical and materials science research.

[0003] Traditional transition state search methods primarily rely on high-precision quantum chemical calculations, such as those based on density functional theory (DFT). These methods, including the Climbing Image algorithm, the Nudged Elastic Band (NEB) method, and its variants, require complex point-finding and optimization calculations on multidimensional potential energy surfaces. While these methods are theoretically reliable, their computational cost is extremely high, typically requiring hours or even days of CPU computation time to process a simple reaction. This enormous computational overhead severely restricts high-throughput screening of complex reaction networks and large-scale virtual discovery of new materials, resulting in long cycles and high costs in chemical research and development.

[0004] In recent years, with the development of artificial intelligence technology, using machine learning to accelerate transition state search has become a promising research direction. Several machine learning-based generative models have been proposed, aiming to learn to directly generate transition state structures from given reactant and product structures, thereby reducing computation time from hours to seconds or even less. However, existing machine learning methods still face a series of deep-seated technical problems in practical applications: First, the ability to fit nonlinear reaction paths is insufficient. Many existing methods use simple linear interpolation of the atomic coordinates of reactants and products as input for the initial guess of the generative model. This approach is acceptable for simple reactions with near-linear reaction paths, but for a large number of real-world reactions involving significant conformational changes, multi-atom rearrangements, or complex cyclic structure changes, the true minimum energy path is highly nonlinear. The simplified linear interpolation deviates significantly from the true path, which not only greatly increases the difficulty for the neural network model to learn the optimal path but may also cause the model to converge to incorrect, high-energy, non-optimal transition states, ultimately affecting the accuracy of the generated results.

[0005] Second, the model training process lacks effective constraints on physical realism. Many generative models, during training, typically focus solely on minimizing the geometric deviation (such as root mean square error) between the predicted and actual structures using their loss functions. This purely data-driven training approach makes the model like a "black box," potentially learning atomic trajectories that are "mathematically optimal" but "physically unreasonable." For example, the model might generate structures with bond lengths or angles significantly deviating from reasonable ranges, or structures where the total energy of the system fluctuates drastically during evolution. When faced with novel reaction types not present in the training set, the generalization ability and physical realism of the generated structures of such models face severe challenges.

[0006] Third, their application scenarios are limited to idealized, isolated systems. The vast majority of practically valuable chemical reactions, such as industrial catalysis or enzymatic reactions in living organisms, occur in complex environments, such as in solution phases or on catalyst surfaces. The presence of solvent molecules or catalysts significantly alters the reaction's potential energy surface, profoundly affecting the structure and energy of the transition state. However, most existing machine learning methods model reactants and products as isolated systems in a vacuum, failing to incorporate these crucial external environmental factors (such as catalysts and solvents) into the model. This prevents their application to broader and more realistic multiphase or multicomponent chemical systems, fundamentally hindering their application in wider and more realistic industrial-scale multiphase or multicomponent chemical systems, where environmental effects often dominate the reaction outcome, severely limiting their scope of application.

[0007] Fourth, the system cannot simultaneously prioritize reliability and efficiency. Some advanced generative models are characterized by "single-shot generation," achieving extremely high computational efficiency, but this also means providing a "one-time" result, good or bad. When faced with particularly difficult reaction cases or those nearing the model's limits, the quality of the generated structure may be poor, or even completely erroneous. The system lacks an intrinsic mechanism to assess the confidence level of its generated results and adaptively decide whether to accept the result or initiate a more precise but costly optimization process based on that confidence level. This approach forces the system to sacrifice reliability in pursuit of speed, failing to achieve an intelligent balance between efficiency and reliability. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a deterministic generation method, system and storage medium for multiphase chemical reaction transition states based on physical constraints and adaptive optimization. The aim is not only to maintain the high efficiency of machine learning methods, but also to overcome the above-mentioned problems and achieve accurate, robust and reliable simulation of complex and real chemical scenarios.

[0009] To achieve the above objectives, this invention provides a deterministic generation method for multiphase chemical reaction transition states based on physical constraints and adaptive optimization. The method is executed by a computer system, which includes a processor and a memory. The method includes the following steps: Obtain initial structural data of reactants and products; The nonlinear initial guessing module is invoked. Based on the initial structure data of the reactants and products, the nonlinear initial guessing module performs a path search based on the coarse-grained potential energy surface to generate a nonlinear initial guessing path containing multiple intermediate mirror point structures. The constrained optimal transport module is invoked. The constrained optimal transport module takes the nonlinear initial guess path as input and performs path evolution in a pre-defined constrained optimal transport model for multi-component systems to generate candidate transition state (TS) structures. The constrained optimal transport model imposes pre-defined geometric or energy constraints on environmental factors when dealing with environmental factors including catalysts or solvents. The uncertainty quantization module is invoked, which performs a reliability assessment on the candidate transition state structure and outputs a quantified confidence score. The adaptive optimization decision-making module is invoked. Based on the confidence score, it compares the score with a preset confidence threshold and adaptively performs one of the following operations: If the confidence score is higher than the high confidence threshold, the candidate transition state (TS) structure is directly accepted as the final result. If the confidence score is lower than the low confidence threshold, the targeted optimization process is initiated, and the candidate transition state TS structure is used as the starting point to perform high-precision quantum chemical calculations to obtain the refined transition state TS structure. The core of this invention lies in proposing and implementing a highly integrated intelligent decision-making closed loop, with the constrained optimal transport module as its hub. This module combines optimal transport theory with chemical and physical constraints, solving not only the problem of how to efficiently generate transition states, but more importantly, the challenge of accurately simulating complex, realistic environments containing catalysts or solvents. It organically inherits the high-quality starting point provided by the nonlinear initial guessing module and delivers physically reasonable candidate structures to the uncertainty quantification module and the adaptive optimization decision-making module, thereby ensuring the overall technological sophistication, reliability, and industrial applicability of the solution.

[0010] Furthermore, based on the coarse-grained potential energy surface path search, specifically using the GFN2-xTB semi-empirical method or the classical force field method, the approximate minimum energy path between reactants and products is quickly calculated. The reason for using this path as the input of the constrained optimal transport model is that although the standard optimal transport can directly handle reactants and products, for highly nonlinear atomic motion paths, pre-generating a nonlinear initial path can significantly reduce the evolution difficulty and ensure physical authenticity.

[0011] Furthermore, when environmental factors act as catalysts, the geometric constraints imposed on these factors specifically involve maintaining the atomic positions of the catalyst fixed during path evolution, or confining them to vibrate slightly near predetermined active sites.

[0012] Furthermore, when the environmental factor is the solvent, the energy constraint imposed on the environmental factor is specifically as follows: at each step of the path evolution, the implicit solvent model is used to calculate and correct the energy and interatomic forces of the system in real time.

[0013] Furthermore, the uncertainty quantification module incorporates a pre-trained graph neural network (GNN) model to predict the generation reliability of candidate transition state (TS) structures.

[0014] Furthermore, the targeted optimization process specifically involves performing geometric optimization calculations on the candidate transition state (TS) structure within the density functional theory (DFT) framework.

[0015] Furthermore, a deterministic generation system for multiphase chemical reaction transition states based on physical constraints and adaptive optimization is also provided, comprising: processor; Memory, which stores computer program instructions that can be executed by a processor; When computer program instructions are executed by the processor, the system performs the following functional modules: The nonlinear initial guessing module is used to perform a path search based on the coarse-grained potential energy surface based on the acquired initial structure data of reactants and products, so as to generate a nonlinear initial guessing path containing multiple intermediate mirror point structures. The constrained optimal transport module is used to take the nonlinear initial guess path as input and perform path evolution in a pre-defined constrained optimal transport model for multi-component systems to generate candidate transition state (TS) structures. The constrained optimal transport model imposes pre-defined geometric or energy constraints on environmental factors when dealing with environmental factors including catalysts or solvents. The uncertainty quantification module is used to evaluate the reliability of candidate transition state (TS) structures and output a quantified confidence score. The adaptive optimization decision module is used to compare the confidence score with a preset confidence threshold and adaptively perform one of the following operations: if the confidence score is higher than the high confidence threshold, the candidate transition state structure is directly accepted as the final result; and if the confidence score is lower than the low confidence threshold, the targeted optimization process is initiated, and the candidate transition state TS structure is used as the starting point to perform high-precision quantum chemical calculations to obtain the refined transition state TS structure.

[0016] Furthermore, the constrained optimal transmission module is configured as follows: When environmental factors act as catalysts, geometric constraints are achieved during the path evolution process by applying position-fixed constraints or harmonic oscillator potential constraints to the atomic coordinates of the catalyst. When the environmental factor is the solvent, energy constraint can be achieved by calling a polarized continuum model (PCM) or a COSMO-like model to correct the Hamiltonian of the system at each step of the path evolution.

[0017] Furthermore, the adaptive optimization decision module further includes a targeted optimization unit, which is configured to: when the adaptive optimization decision module is activated, call the quantum chemical calculation software in the memory, take the coordinate data of the candidate transition state TS structure as input, and perform density functional theory DFT calculation.

[0018] Furthermore, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the aforementioned deterministic generation method for multiphase chemical reaction transition states based on physical constraints and adaptive optimization.

[0019] The beneficial effects of this invention are as follows: 1. Improved Accuracy and Robustness: By employing path-aware nonlinear initial guesses, this invention provides a more reasonable starting point closer to real physical processes for subsequent accurate generation, reducing the learning difficulty of the neural network and thus improving the accuracy of generating transition state structures for complex, nonlinear reaction paths. Furthermore, by explicitly introducing physical constraints such as energy and chemical bond energies during model training, this invention enables the model to learn not only geometric transformations between atoms but also atomic evolution processes that conform to physical laws. This enhances the model's generalization ability to new reaction types not seen in the training set and ensures the physical authenticity of the generated structures, avoiding the generation of chemically unreasonable structures.

[0020] 2. Expanded Application Scope: By constructing a constrained optimal transport model for multi-component systems, this invention successfully expands its application scenarios from idealized gas-phase systems to more general and practically significant catalytic and solution-phase reactions. This invention can incorporate catalysts or solvents as constrained environmental factors into the model, accurately simulating their influence on the reaction potential surface and transition state through field effects. This expands the application scope of existing isolated system models, enabling them to handle complex chemical systems in real industrial and scientific research scenarios.

[0021] 3. Balancing Efficiency and Reliability: This invention constructs an intelligent decision-making closed loop integrating "rapid generation, confidence assessment, and adaptive optimization." For the vast majority of responses, the system can provide high-confidence, reliable results quickly, achieving extremely high computational efficiency. For a few difficult or marginal cases, thanks to the structural quality of the constraint-optimal transmission module's output being far superior to traditional methods, the system can automatically identify and allocate a small amount of additional computational resources for "targeted treatment," refining the results using high-precision methods. This avoids the predicament of over-reliance on high-cost traditional methods or accepting low-quality results. This intelligent balancing mechanism, supported by the high-quality output of the constraint-optimal transmission module, optimizes the overall performance and reliability of the system, giving it greater practical value in industrial-grade high-throughput screening.

[0022] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the hardware and functional module architecture of the transition state deterministic generation system described in Embodiment 1 of the present invention.

[0024] Figure 2 This is an overall flowchart of the deterministic generation method for transition states described in one embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram illustrating the working principle of the nonlinear initial guessing module in one embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram of a loss function containing physical constraints used in the training phase of the constrained optimal transmission model described in one embodiment of the present invention.

[0027] Figure 5 This is a schematic diagram of a constrained optimal transport module processing a multiphase reaction system containing a catalyst in one embodiment of the present invention.

[0028] Figure 6 This is a schematic diagram of the workflow of the adaptive optimization decision module in one embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This embodiment provides a highly integrated and intelligent deterministic generation system for multiphase chemical reaction transition states based on physical constraints and adaptive optimization, along with a corresponding implementation method. This system aims to generate physically realistic and chemically sound transition state structures with extremely high efficiency and accuracy, and can adapt to various complex chemical scenarios, including gas-phase, solution-phase, and catalytic reactions.

[0031] System hardware and software environment Reference Figure 1 The system in this embodiment can be deployed on one or more high-performance computing servers. The server's hardware includes, but is not limited to: one or more central processing units (CPUs) 101, such as Intel Xeon or AMD EPYC series processors; one or more graphics processing units (GPUs) 103, such as NVIDIA A100 or H100, for accelerating neural network computation; large-capacity random access memory (RAM) 102, such as 256GB or more, for storing program instructions, model parameters, and intermediate computation data; high-speed solid-state drives (SSDs) or network storage 104 for persistently storing the chemical structure database, trained model files, and computation results; and a network interface 105 for data interaction with external databases or user terminals.

[0032] At the software level, memory 102 stores a complete set of computer program instructions, including an operating system (such as Linux), a deep learning framework (such as PyTorch or TensorFlow), a quantum chemical computing software package (such as Gaussian, ORCA), and the core functional module program of this invention. When processor 101 and GPU 103 execute these instructions, all the functions of this invention are realized.

[0033] System functional modules like Figure 1 As shown, the system of the present invention is logically composed of four core functional modules, which are implemented by the processor 101 executing the corresponding program instructions in the memory 102: Nonlinear Initial Guessing Module 10: Responsible for generating more accurate initial reaction paths than traditional linear interpolation.

[0034] Constrained optimal transport module 20: responsible for evolving the initial path into a candidate transition state structure under physical and chemical constraints.

[0035] Uncertainty Quantification (UQ) Module 30: Responsible for evaluating the reliability of the generated results.

[0036] Adaptive optimization decision module 40: Responsible for making intelligent decisions on subsequent operations based on the reliability assessment results.

[0037] The following will combine Figure 2 The overall methodology shown provides an extremely detailed explanation of each step and its corresponding functional module.

[0038] Step S201: Obtain input data At the start of the method, the system first acquires input information for a specific chemical reaction. This information includes at least the initial three-dimensional structural data of the reactant (R) and product (P). This data is typically stored in standard chemical file formats (e.g., .xyz, .mol, .cif), which define in detail the elemental type (e.g., C, H, O, N) of each atom and its coordinates (x, y, z) in a three-dimensional Cartesian coordinate system. In some scenarios involving multiphase reactions, the input data may also include structural information about environmental factors (E), such as the atomic coordinates of the catalyst or the type of solvent. This data is read from storage device 104 into memory 102 by processor 101 for processing by subsequent modules.

[0039] Step S202: Generate a path-aware nonlinear initial guess structure This step is performed by the nonlinear initial guessing module 10. Its core objective is to overcome the limitations of existing linear interpolation methods; the necessity of this module lies in overcoming the limitations of linear interpolation (such as…). Figure 3 As shown in the figure, it provides a better starting point for constrained optimal transmission.

[0040] Linear interpolation assumes that atomic motion from reactants to products occurs along a straight line. However, real chemical reactions, especially those involving ring opening and closing, large-scale migration of groups, or complex rearrangements, involve highly curved and nonlinear atomic motion paths. (See reference...) Figure 3 The linear interpolation path (dashed line) may be far from the true minimum energy path (solid line), and may even pass through regions with extremely high energy, which brings huge and unnecessary difficulties to subsequent accurate model learning.

[0041] Therefore, this module adopts a "coarse-to-fine" strategy. It first performs a low-computational-cost path search based on a coarse-grained potential energy surface. Here, "coarse-grained potential energy surface" refers to an approximate energy description method with computational overhead far less than high-precision quantum chemical calculations. In a preferred embodiment, the coarse-grained potential energy surface-based path search specifically involves using the GFN2-xTB semi-empirical method or the classical force field method to rapidly calculate the approximate minimum energy path between reactants (R) and products (P).

[0042] GFN2-xTB is a semi-empirical tight-binding quantum chemistry method. Compared to expensive DFT calculations, it simplifies the computation process through extensive parameterization, enabling the calculation of energy and force for medium-sized molecules in seconds. Although the accuracy is lower, it is sufficient to capture the general outline of the reaction pathway.

[0043] Classical force fields are a faster approximation method that treats atoms as point masses connected by springs, using pre-parameterized functions (such as Lennard-Jones potential, bond length / bond angle / dihedral angle terms) to describe the interactions between atoms. It is computationally very fast, but its applicability and accuracy are limited by the force field parameters.

[0044] Processor 101 invokes the pre-installed GFN2-xTB program or force field calculation engine in memory 102 to perform a simplified Nudged Elastic Band (NEB) calculation. The NEB method creates a series of intermediate structural "images" between reactants and products, and moves these image points using optimization algorithms (such as gradient descent) until they eventually converge to the minimum energy path. Because it uses the computationally inexpensive GFN2-xTB, the entire process is typically completed within a few seconds.

[0045] Module Output and Data Flow: The output of this module is no longer a single intermediate point obtained by linear interpolation, but a nonlinear initial guess path composed of multiple (e.g., 8 to 16) intermediate mirror point structures. This series of 3D structure data is stored in a specific area of ​​memory 102 and passed to the next module.

[0046] The initial path generated by this method, because it is optimized on an approximate but physically meaningful potential energy surface, better reflects the topological characteristics of the real reaction trajectory, such as the sequence of atomic motions and the conformation of key intermediates. This provides a high-quality starting point for the subsequent constrained optimal transport module 20, greatly reducing the difficulty of its learning and optimization, and is a key first step in improving the accuracy of the final generated structure.

[0047] Step S203: Perform constrained optimal transport to generate candidate transition states This step is performed by the constrained optimal transport module 20 and is the core computational component of this invention. The reason why this module receives nonlinear path input is to improve adaptability to complex environments, rather than the direct processing of standard OT. Although the constrained optimal transport model is based on OT theory, in chemical scenarios, pre-generated paths can inject physical prior knowledge, reduce deviations from nonlinear reactions, and ensure efficient evolution.

[0048] This module employs a pre-trained deep learning model based on the theory of Optimal Transport (OT).

[0049] Optimal Transport (OT) is a mathematical theory that aims to transform one probability distribution (or set of points) into another probability distribution (or another set of points) at the lowest cost. In this invention, it is used to find an "optimal" flow of atomic motion, "transporting" the atomic distribution on a nonlinear initial guess path to the actual, lowest-energy reaction path, and ultimately determining its highest-energy point—the transition state.

[0050] The constrained optimal transport module is crucial for applying this invention from a theoretical model to complex, real-world chemical scenarios. Its importance lies in the fact that it is not merely a simple structure generator, but rather deeply infused with physicochemical principles. Without this module, or by employing traditional generation models without physical constraints, the entire technical solution would degenerate. Firstly, it would be unable to handle real systems containing catalysts or solvents, severely limiting its application scope. Secondly, even in ideal systems, the generated structures might violate fundamental physical laws, leading to unreliable results. Therefore, this module, through its unique constraint mechanism, ensures the physical authenticity and broad applicability of the technical solution, serving as the cornerstone of this invention.

[0051] This solution differs from existing technologies; the OT model used in this module has been improved in key aspects: Physical constraints were introduced during the training phase: such as Figure 4 As shown, the loss function L of this model during the training phase total It not only includes the traditional geometric deviation term L_OT (used to minimize the geometric difference between the predicted path and the true path), but also innovatively introduces a physical constraint regularization term λ*L. phys Here, λ is a hyperparameter used to balance the weights of the two terms. phys It can consist of one or more sub-items: Energy gradient constraint: This penalty applies to atomic motions that cause drastic and unreasonable fluctuations in the total energy of the system during path evolution. It requires that the direction of atomic motion roughly follows the direction of the energy gradient (when far from the transition state), preventing the model from learning physically impossible energy abrupt changes.

[0052] • Chemical Bond Energy Constraints: This section utilizes chemical knowledge graphs or prior knowledge to constrain the range of bond energies and bond lengths of chemical bonds that break and form during a reaction. For example, the breaking energy of a C / C single bond is typically within a certain range, and its bond length may elongate in the transition state, but it should not be stretched or shortened to an unreasonable degree indefinitely. This constraint penalizes structures that generate chemically unreasonable bond lengths or bond energies, ensuring the chemical rationality of the generated structures.

[0053] In this way, when performing gradient backpropagation and parameter updates during model training, the processor 101 not only considers geometric alignment but must also "obey" fundamental physical and chemical laws. This greatly enhances the physical insight, generalization ability, and physical realism of the trained model.

[0054] The model is extended to accommodate multi-component systems: this module can handle multiphase reaction systems that include environmental factors E (such as catalysts or solvents). It treats the entire system (R, P, E) as a multi-component system and imposes different constraints on different components; this characteristic allows this module to organically coordinate with other parts. It receives high-quality initial paths for multi-component systems from the nonlinear initial guessing module 10, generates a more physically and chemically reasonable candidate structure, provides a high-quality evaluation object for the subsequent uncertainty quantification module 30, and greatly improves the efficiency and reliability of the final decision of the adaptive optimization decision module 40.

[0055] In a preferred embodiment, when the environmental factor E is a catalyst, the geometric constraint applied to the environmental factor E specifically involves: maintaining the atomic positions of the catalyst fixed during path evolution, or confining them to vibrate slightly near predetermined active sites. (Refer to...) Figure 5When the OT model processes the "transport" of atomic coordinates, the processor 101 imposes a positional constraint on the coordinates of atoms belonging to catalyst C. This can be achieved in two ways: one is to fix their coordinates and exclude them from optimization; the other is to apply a harmonic potential, which allows them to vibrate slightly around their equilibrium positions to simulate real thermal vibrations, but any large displacement will be subject to a huge energy penalty. In this way, the model can learn how atoms of reactant R and product P move and recombine on the catalyst surface, while the catalyst surface itself remains stable.

[0056] In a preferred embodiment, when environmental factor E is a solvent, the energy constraint imposed on environmental factor E is specifically as follows: during the path evolution process, an implicit solvent model is used to calculate and correct the energy and interatomic forces of the system in real time.

[0057] Implicit solvent models, such as the polarizable continuous medium model (PCM) or COSMO-like models, are methods that approximate the solvent environment as a continuous medium with a specific dielectric constant. They avoid explicitly dealing with thousands of solvent molecules, thus significantly saving computational resources.

[0058] At each step of the path evolution, when the OT model needs to calculate the energy and interatomic forces of the current system, processor 101 invokes a PCM calculation module. This module calculates the additional energy and force contributions from solvation effects based on the current atomic configuration and the dielectric constant of the solvent, and returns these corrected energies and forces to the OT model. This means that the entire path evolution process is carried out in a simulated solution environment, thus accurately capturing the stabilizing (or destabilizing) effects of the solvent on the transition state structure and energy.

[0059] Through the creative combination of the aforementioned physical constraints and multi-component modeling, this module ensures that its output candidate transition state structures are not only geometrically accurate but also highly rational in terms of physics and chemistry, and can realistically reflect the influence of complex environments. This logic of guaranteeing high-quality output from the source enables this solution to achieve high precision, high reliability, and wide applicability.

[0060] Module Output and Data Flow: After the constrained optimal transmission module 20 completes its execution, it outputs a single, high-precision candidate transition state (TS) structure. The three-dimensional atomic coordinate data of this structure is stored in memory 102, ready for reliability assessment.

[0061] This module addresses the core issues of poor physical realism and limited application scenarios in existing technologies through physical constraints and multi-component modeling. The candidate TS structures it generates are not only geometrically accurate but also more physically and chemically plausible, and can reflect the influence of real-world reaction environments.

[0062] Step S204: Perform uncertainty quantification assessment This step is performed by the Uncertainty Quantification (UQ) module 30. Its purpose is to provide a basis for subsequent intelligent decision-making.

[0063] Although the candidate TS generated in step S203 is of high quality, the results may still contain uncertainty for some extremely difficult responses or responses that the model has not fully learned. The role of UQ module 30 is to quantify this uncertainty.

[0064] In a preferred embodiment, the uncertainty quantification module 30 incorporates a pre-trained graph neural network (GNN) model to predict the generation reliability of candidate transition state (TS) structures.

[0065] Graph Neural Networks (GNNs) are deep learning models specifically designed for processing graph-structured data. In this invention, molecular structures are naturally viewed as a graph, where atoms are nodes and chemical bonds are edges.

[0066] The GNN model was trained for a classification task: given a reaction system (including reactants R, products P, and candidate TS generated by module 20), the output is the probability that the candidate TS is "reliable". This probability value (between 0 and 1) is used as a confidence score.

[0067] Processor 101 inputs the structural data (atom types, coordinates, bond connections) of R, P, and candidate TS into this pre-trained GNN model. GPU 103 is responsible for performing the forward propagation computation of the GNN, which yields a confidence score in a very short time (usually milliseconds).

[0068] Module Output and Data Flow: The UQ module 30 outputs a floating-point number, namely the confidence score. This score, along with the candidate TS structure, is passed to the adaptive optimization decision module 40.

[0069] This module adds a "self-reflection" capability to the entire system. It enables the system to move beyond blindly outputting results and instead assess the reliability of its output, which is fundamental to achieving intelligent decision-making.

[0070] Step S205: Perform adaptive optimization decision This step is executed by the adaptive optimization decision module 40, which is the "command center" for achieving an intelligent balance between efficiency and reliability in this invention.

[0071] Reference Figure 6 The logic of this module is very clear. It receives a confidence score from module 30 and compares it with two preset thresholds: a high confidence threshold (e.g., 0.9) and a low confidence threshold (e.g., 0.7).

[0072] High-confidence path (e.g., score > 0.9): If the confidence score is extremely high, it indicates that the UQ module has great confidence in the quality of the candidate TS. In this case, module 40 directly accepts the candidate transition state TS structure as the final result. The system saves the structure data to storage device 104 or outputs it to the user. This path achieves the highest computational efficiency and is suitable for most "simple" reactions.

[0073] Low-confidence paths (e.g., score < 0.7): If the confidence score is very low, it indicates that the UQ module believes the quality of the candidate TS is likely unreliable. In this case, module 40 decides to initiate the targeted optimization process.

[0074] In a preferred embodiment, the targeted optimization process specifically involves performing geometric optimization calculations on the candidate transition state (TS) structure within the framework of density functional theory (DFT).

[0075] Density functional theory (DFT) is one of the most widely used high-precision quantum chemical calculation methods. It can handle more complex systems than traditional ab initio methods while maintaining high accuracy.

[0076] Module 40 uses this low-confidence candidate TS structure as the starting point for a traditional but high-precision DFT transition state optimization calculation. This is crucial because the candidate TS generated by Module 20, even with low confidence, has a structure far superior to that of random guessing or linear interpolation. Using such a high-quality starting point can significantly accelerate the convergence speed of the DFT optimization calculation, potentially completing a computation that would otherwise take hours in tens of minutes. Processor 101 invokes a quantum chemistry software package (such as Gaussian) in memory 102 to execute the calculation and obtains the refined, high-confidence transition state TS structure as the final result.

[0077] Medium confidence path (0.7 <= score <= 0.9): For structures with confidence in the middle range, the system can process them according to preset strategies, such as accepting them directly to balance efficiency, or initiating targeted optimization to ensure the highest quality.

[0078] Module output and data flow: This module ultimately outputs a defined, high-quality transition state structure.

[0079] The Adaptive Optimization Decision Module 40 perfectly combines speed and accuracy. Like an experienced expert, it quickly handles simple tasks and dedicates extra effort to difficult ones, thereby minimizing average computational costs while ensuring reliable results.

[0080] The above four modules work together to form a complete, closed-loop, and intelligent solution. The nonlinear initial guessing module 10 provides a high-quality starting point for the entire process; the constrained optimal transport module 20 utilizes physical constraints and multi-component modeling capabilities to generate candidate structures that are both physically and chemically more reasonable; the uncertainty quantification module 30 acts as a "quality inspector"; and the adaptive optimization decision module 40 intelligently selects the most efficient path based on the inspection results. These four modules are interconnected, functionally supportive of each other, and work together to solve the technical problems of existing technologies in terms of accuracy, physical realism, application scope, and reliability through innovative technical means, ultimately achieving significant technical results in the field of computational chemistry. The entire solution tightly integrates algorithmic features (OT, GNN) with technical features (chemical structure data processing, physical law constraints, and quantum chemical computational calls). Its input is the data representation of physical entities, its output is an accurate prediction of the state of physical entities, and its process represents a significant improvement in the internal performance of the computer (e.g., computational speed, efficiency, and reliability).

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for deterministic generation of transition states of multiphase chemical reactions based on physical constraints and adaptive optimization, characterized in that, The method is executed by a computer system comprising a processor (101) and a memory (102), and comprises the following steps: obtaining initial structure data of reactants (R) and products (P); calling a nonlinear initial guess module (10) to perform a coarse-grained potential energy surface-based path search based on the initial structure data of the reactants (R) and products (P) to generate a nonlinear initial guess path comprising a plurality of intermediate mirror point structures; calling a constrained optimal transmission module (20) to input the nonlinear initial guess path and perform path evolution in a preset constrained optimal transmission model for a multi-component system to generate a candidate transition state TS structure; wherein the constrained optimal transmission model imposes a preset geometric or energy constraint on an environmental factor (E) when processing an environmental factor (E) comprising a catalyst or a solvent; calling an uncertainty quantification module (30) to perform reliability evaluation on the candidate transition state (TS) structure and output a quantified confidence score; calling an adaptive optimization decision module (40) to compare the confidence score with a preset confidence threshold and adaptively perform one of the following operations: if the confidence score is higher than a high confidence threshold, directly accepting the candidate transition state (TS) structure as a final result; if the confidence score is lower than a low confidence threshold, starting a targeted optimization process, taking the candidate transition state (TS) structure as an initial point, and performing high-precision quantum chemical calculation to obtain a refined transition state (TS) structure.

2. The method of claim 1, wherein, The coarse-grained potential energy surface-based path search specifically adopts a GFN2-xTB semi-empirical method or a classical force field method to quickly calculate an approximate minimum energy path between the reactants (R) and products (P).

3. The method of claim 1, wherein the method is characterized by, When the environmental factor (E) is a catalyst, the geometric constraint imposed on the environmental factor (E) specifically includes fixing the atomic positions of the catalyst or limiting them to a preset active site for small vibrations during the path evolution process.

4. The method of claim 1, wherein, When the environmental factor (E) is a solvent, the energy constraint imposed on the environmental factor (E) specifically includes using an implicit solvent model to perform real-time calculation and correction on the energy and interatomic forces of the system at each step of the path evolution.

5. The method of claim 1, wherein, The uncertainty quantification module (30) has a pre-trained graph neural network GNN model built in for predicting the generation reliability of the candidate transition state TS structure.

6. The method of claim 1, wherein, The targeted optimization process specifically includes performing a geometry optimization calculation on the candidate transition state TS structure under a density functional theory DFT framework.

7. A physical constraint and adaptive optimization based multi-phase chemical reaction transition state deterministic generation system, characterized in that, The system comprises: a processor (101); a memory (102) having computer program instructions executable by the processor (101) stored therein; When the computer program instructions are executed by the processor (101), the system implements the following functional modules: a non-linear initial guess module (10) configured to perform a coarse-grained potential energy surface-based path search based on obtained initial structure data of reactants (R) and products (P) to generate a non-linear initial guess path comprising a plurality of intermediate mirror point structures; a constrained optimal transmission module (20) configured to input the non-linear initial guess path to evolve the path in a pre-set constrained optimal transmission model for a multi-component system to generate a candidate transition state TS structure; wherein the constrained optimal transmission model applies a pre-set geometric or energy constraint on an environmental factor (E) when processing the environmental factor (E) containing a catalyst or a solvent; an uncertainty quantification module (30) configured to perform reliability evaluation on the candidate transition state TS structure and output a quantified confidence score; an adaptive optimization decision module (40) configured to compare the confidence score with a pre-set confidence threshold and adaptively perform one of the following operations: if the confidence score is higher than a high confidence threshold, directly accepting the candidate transition state TS structure as a final result; and if the confidence score is lower than a low confidence threshold, starting a targeted optimization process to perform high-precision quantum chemical calculation on the candidate transition state TS structure as an initial point to obtain a refined transition state TS structure.

8. The physical constraint and adaptive optimization based transition state deterministic generation system for multi-phase chemical reactions of claim 7, wherein, The constrained optimal transmission module (20) is configured to: when the environmental factor (E) is a catalyst, implement the geometric constraint by applying a position fixing constraint or a harmonic oscillator potential constraint on atomic coordinates of the catalyst during path evolution; when the environmental factor (E) is a solvent, implement the energy constraint by calling a polarizable continuum model PCM or a COSMO-like model to modify the Hamiltonian of the system at each step of path evolution.

9. The physical constraint and adaptive optimization based multi-phase chemical reaction transition state deterministic generation system of claim 7, wherein, The adaptive optimization decision module (40) further comprises a targeted optimization unit configured to, when started by the adaptive optimization decision module (40), call quantum chemical calculation software in the memory (102), input coordinate data of the candidate transition state TS structure, and perform density functional theory DFT calculation.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, a physical constraint and adaptive optimization-based multi-phase chemical reaction transition state determination generation method as claimed in any one of claims 1 to 6 is implemented.

Images