2024 Vol. 53 No. 7
2024, 53(7): 431-441.
DOI: 10.7693/wl20240701
Abstract：
With the rapid advance of artificial intelligence, its integration with physical modeling has introduced revolutionary tools for scientific research at the microscopic scale. This paper delineates the development from the approximate solutions of quantum mechanics based on the Schrödinger equation to the emergence of the large atomic model (LAM), with particular emphasis on the application of machine learning in atomic-scale simulations. The theoretical foundations underlying the synergy between artificial intelligence and physical modeling are first discussed. This is followed by a comprehensive analysis of the implementation methodologies for this integration in atomic-scale simulations, including the construction and training strategies of machine learning models. Next, the critical roles of data accumulation, software tools, and engineering infrastructure in propelling advancements in this domain are examined. The potential impact of LAM on future scientific research and industrial applications is also envisioned. Through sustained technological innovation and interdisciplinary collaboration, it is anticipated that LAM will significantly contribute to many fields, including materials science, chemical engineering, and biotechnology, thereby ushering in a new era of development in basic research and applications.
With the rapid advance of artificial intelligence, its integration with physical modeling has introduced revolutionary tools for scientific research at the microscopic scale. This paper delineates the development from the approximate solutions of quantum mechanics based on the Schrödinger equation to the emergence of the large atomic model (LAM), with particular emphasis on the application of machine learning in atomic-scale simulations. The theoretical foundations underlying the synergy between artificial intelligence and physical modeling are first discussed. This is followed by a comprehensive analysis of the implementation methodologies for this integration in atomic-scale simulations, including the construction and training strategies of machine learning models. Next, the critical roles of data accumulation, software tools, and engineering infrastructure in propelling advancements in this domain are examined. The potential impact of LAM on future scientific research and industrial applications is also envisioned. Through sustained technological innovation and interdisciplinary collaboration, it is anticipated that LAM will significantly contribute to many fields, including materials science, chemical engineering, and biotechnology, thereby ushering in a new era of development in basic research and applications.
2024, 53(7): 442-449.
DOI: 10.7693/wl20240702
Abstract：
First-principles methods have become indispensable research tools in modern physics, chemistry, materials science and other fields. Based on the fundamental principles of quantum mechanics, first-principles calculations can achieve highly accurate material property predictions by solving the complicated problem of interacting electrons. However, their widespread applications are limited by the high computational cost, posing significant challenges for large-scale materials simulations and the construction of materials big data. In recent years, the emergence of groundbreaking works, such as AlphaGo, AlphaFold, and ChatGPT, heralds the advent of a new era of artificial intelligence (AI), bringing transformative opportunities to first-principles calculations. Deep learning provides a novel research paradigm for the field, enabling us to overcome the bottlenecks of traditional methods through precise modeling and efficient prediction. In this article, we introduce a series of deep-learning based first-principles computation methods. By leveraging the deep-learning modeling of a fundamental quantity in density functional theory (DFT), the DFT Hamiltonian, we propose a neural network framework designed to satisfy the key prior knowledge of the Hamiltonian, including the principles of nearsightedness and equivariance. This method has been successfully applied to large-scale simulations of twisted van der Waals materials, as well as the construction of universal models based on materials big data. These advancements offer new opportunities for developing large materials models and advancing AI-driven materials discovery.
First-principles methods have become indispensable research tools in modern physics, chemistry, materials science and other fields. Based on the fundamental principles of quantum mechanics, first-principles calculations can achieve highly accurate material property predictions by solving the complicated problem of interacting electrons. However, their widespread applications are limited by the high computational cost, posing significant challenges for large-scale materials simulations and the construction of materials big data. In recent years, the emergence of groundbreaking works, such as AlphaGo, AlphaFold, and ChatGPT, heralds the advent of a new era of artificial intelligence (AI), bringing transformative opportunities to first-principles calculations. Deep learning provides a novel research paradigm for the field, enabling us to overcome the bottlenecks of traditional methods through precise modeling and efficient prediction. In this article, we introduce a series of deep-learning based first-principles computation methods. By leveraging the deep-learning modeling of a fundamental quantity in density functional theory (DFT), the DFT Hamiltonian, we propose a neural network framework designed to satisfy the key prior knowledge of the Hamiltonian, including the principles of nearsightedness and equivariance. This method has been successfully applied to large-scale simulations of twisted van der Waals materials, as well as the construction of universal models based on materials big data. These advancements offer new opportunities for developing large materials models and advancing AI-driven materials discovery.
2024, 53(7): 450-459.
DOI: 10.7693/wl20240703
Abstract：
In recent years the rapid growth of computer processing power has led to major breakthroughs in scientific computing and artificial intelligence. The deep integration of these two fields has jointly fostered a data-driven paradigm for scientific research. As a representative of artificial intelligence technology, machine learning has brought unprecedented opportunities for computational materials design, with current applications mainly focusing on property prediction, synthesis prediction, knowledge discovery, and generative inverse design. This article will briefly describe the research progress in this field, and look ahead to the future directions and challenges.
In recent years the rapid growth of computer processing power has led to major breakthroughs in scientific computing and artificial intelligence. The deep integration of these two fields has jointly fostered a data-driven paradigm for scientific research. As a representative of artificial intelligence technology, machine learning has brought unprecedented opportunities for computational materials design, with current applications mainly focusing on property prediction, synthesis prediction, knowledge discovery, and generative inverse design. This article will briefly describe the research progress in this field, and look ahead to the future directions and challenges.
2024, 53(7): 460-471.
DOI: 10.7693/wl20240704
Abstract：
The study of the response properties of electron systems, such as the magnetic susceptibility in equilibrium and various transport coefficients out of equilibrium, was one of the primal forces driving the early development of solid state physics. After the classical particle view of electrons was modified by quantum mechanics and the many-body effect and evolved into the modern semiclassical particle view, the semiclassical response theory was established with the further help of the deep insight into Hilbert-space geometry and topology, offering a complete, accurate and, most importantly, intuitive framework for discussing various response properties. In this article we introduce the semiclassical response theory using classical examples from four important aspects, namely, thermoelectric response, spin transport, nonlinear response, and extrinsic mechanisms. Its value in solid state physics will also be demonstrated.
The study of the response properties of electron systems, such as the magnetic susceptibility in equilibrium and various transport coefficients out of equilibrium, was one of the primal forces driving the early development of solid state physics. After the classical particle view of electrons was modified by quantum mechanics and the many-body effect and evolved into the modern semiclassical particle view, the semiclassical response theory was established with the further help of the deep insight into Hilbert-space geometry and topology, offering a complete, accurate and, most importantly, intuitive framework for discussing various response properties. In this article we introduce the semiclassical response theory using classical examples from four important aspects, namely, thermoelectric response, spin transport, nonlinear response, and extrinsic mechanisms. Its value in solid state physics will also be demonstrated.
2024, 53(7): 472-480.
DOI: 10.7693/wl20240705
Abstract：
Quantum mechanics has been established for roughly one hundred years, yet there remain some confusion and debate. This embarrassing situation may lie in the nonuniformity of state vector representations, and in the confusing different probabilistic interpretations of the wave function. We realized that by insisting on representing the state vector, thus the wave function, as a dimensionless abstract object, bearing in mind the differences among the interpretations of Schrödinger, Born and Dirac over the wave function, in particular noticing the fact that the probabilistic interpretations of Schrödinger and Born are related by the Dirac’s completeness relation, and that the probabilistic behavior arises beyond the dimensionless state vector/wave function and the deterministic dynamic equation, then the confusion and debate over quantum mechanics can be largely resolved.
Quantum mechanics has been established for roughly one hundred years, yet there remain some confusion and debate. This embarrassing situation may lie in the nonuniformity of state vector representations, and in the confusing different probabilistic interpretations of the wave function. We realized that by insisting on representing the state vector, thus the wave function, as a dimensionless abstract object, bearing in mind the differences among the interpretations of Schrödinger, Born and Dirac over the wave function, in particular noticing the fact that the probabilistic interpretations of Schrödinger and Born are related by the Dirac’s completeness relation, and that the probabilistic behavior arises beyond the dimensionless state vector/wave function and the deterministic dynamic equation, then the confusion and debate over quantum mechanics can be largely resolved.
