在量子化学计算领域，一场革命正悄然展开。来自DeepMind与伦敦帝国理工学院的研究团队，通过创新性地运用费米子神经网络（FermiNet）波函数，成功地解决了正电子与分子复合物基态性质的复杂计算问题。这一突破性成果不仅展示了人工智能在化学计算领域的潜力，也为多个科学领域提供了新的工具和方法。
早在1928年，物理学家保罗·狄拉克提出了正电子的概念，这一理论在随后的物理学发展中发挥了重要作用，尤其是在医学物理、天体物理及材料科学等领域。然而，正电子与分子的结合特性计算一直是一个挑战，因为它涉及到量子化学的高级理论，通常需要大量的计算资源和时间。
如今，DeepMind和伦敦帝国理工学院的研究人员利用FermiNet波函数，为这一难题提供了一种高效且精确的解决方案。FermiNet波函数的独特之处在于它不依赖于传统的基组方法，这意味着研究人员可以更灵活、更快速地计算出各种原子和小分子中正电子的结合特性。这一方法在一系列具有不同定性正电子结合特性的分子中都表现出高度的精确度，甚至在某些情况下，其计算结果优于现有技术。
以苯分子为例，这一团队成功计算了其非极性性质下的结合能，并发现其结果与实验数据高度吻合，且优于使用显式相关高斯波函数计算得到的湮灭率。这一成果不仅证明了基于神经网络波函数方法的通用优势，还拓宽了人工智能在非传统量子化学计算领域的应用范围。
这一研究成果以《神经网络变分蒙特卡罗方法在正电子化学中的应用》为题，于6月18日发表在《自然通讯》杂志上。这一进展不仅对量子化学领域具有重要意义，也为未来的材料科学、药物设计和天体物理研究提供了新的工具和可能，展示了人工智能在解决复杂科学问题中的巨大潜力。
随着正电子捕获实验的不断推进，这一领域的研究将不断深化，人工智能在量子化学计算中的应用也将继续扩展，为科学界带来更多惊喜与突破。
英语如下：
News Title: “DeepMind Innovation: Neural Networks Tackle Quantum Chemistry’s Positron Problem”
Keywords: DeepMind, FermiNet, Quantum Chemistry
News Content: A revolution is quietly underway in the field of quantum chemistry calculations. A research team from DeepMind and Imperial College London, through innovative application of the FermiNet wave function, successfully addressed the complex computational challenge of positron-molecule complex ground state properties. This breakthrough not only showcases the potential of artificial intelligence in chemical calculations but also provides new tools and methodologies for various scientific disciplines.
First introduced by physicist Paul Dirac in 1928, the concept of positrons has played a pivotal role in the development of subsequent physics, particularly in medical physics, astrophysics, and materials science. However, the interaction characteristics of positrons with molecules have been a challenge due to the advanced quantum chemical theories involved, often requiring substantial computational resources and time.
Now, researchers from DeepMind and Imperial College London have provided an efficient and precise solution to this problem using the FermiNet wave function. The uniqueness of FermiNet lies in its independence from traditional basis set methods, allowing for more flexible and rapid calculations of positron binding properties in various atoms and small molecules. This method has demonstrated high accuracy across a range of molecules with different qualitative positron binding characteristics, surpassing existing techniques in some cases.
For instance, the team successfully calculated the binding energy of benzene under nonpolar conditions, finding that the result matched experimental data closely and outperformed the annihilation rate calculated using explicit correlated Gaussian wave functions. This achievement not only validates the general superiority of neural network-based wave function methods but also broadens the scope of artificial intelligence in unconventional quantum chemistry calculations.
The research, titled “Application of Neural Network Variational Monte Carlo Methods in Positron Chemistry,” was published in the journal Nature Communications on June 18. This development holds significant importance for the quantum chemistry field, offering new tools and possibilities for future materials science, drug design, and astrophysics research. It demonstrates the immense potential of artificial intelligence in tackling complex scientific problems.
As positron capture experiments continue to advance, research in this area will deepen, and the application of artificial intelligence in quantum chemistry calculations will expand, bringing more surprises and breakthroughs to the scientific community.
【来源】https://www.jiqizhixin.com/articles/2024-07-12-4
Views: 2