AI首次攻克量子物理难题：DeepMind精确计算量子激发态，登上Science
伦敦，8月23日 – 谷歌旗下人工智能公司DeepMind近日宣布，其研究人员开发出一种全新的方法，能够精确计算量子激发态，并成功应用于一系列原子和分子，这一突破性成果已发表在顶级学术期刊《科学》杂志上。
量子激发态是物质在受到能量刺激后，电子跃迁至更高能量状态所处的状态。它在物理学、化学等领域至关重要，例如理解光合作用、太阳能电池板工作原理等。然而，从第一性原理出发对激发态进行准确计算一直是科学界面临的重大挑战。
DeepMind的研究人员利用神经网络，克服了这一难题。他们开发了一种名为“自然激发态变分蒙特卡罗”（NES-VMC）的方法，将寻找系统激发态的问题转化为寻找扩展系统基态的问题，从而避免了以往方法中存在的自由参数和正交化问题。
研究人员在从单个原子到苯等一系列基准系统上测试了NES-VMC方法，结果表明其准确性极高，与实验结果非常接近。这标志着深度学习首次成功解决量子物理学中一些最难的问题，为通用量子模拟开辟了新道路。
论文一作兼通讯作者David Pfau表示：“这是深度学习首次准确解决量子物理学中一些最难的问题。我们希望这将是深度学习走向通用量子模拟的重要一步。”
这项研究成果将对量子化学、材料科学、药物研发等领域产生重大影响。它将帮助科学家更深入地理解物质的性质，并设计出更先进的材料和药物。
关于DeepMind
DeepMind是一家由英国人工智能专家创立的公司，于2014年被谷歌收购。其研究领域包括机器学习、深度学习、人工智能等。DeepMind曾开发出AlphaGo、AlphaFold等知名人工智能系统，在围棋、蛋白质结构预测等领域取得了突破性成果。
英语如下：
AI Cracks Quantum Puzzle! DeepMind’s Precise Calculation of Excited States Published inScience
Keywords: AI, Quantum, Excited States
London, August23 – Google’s artificial intelligence company DeepMind announced that its researchers have developed a novel method for accurately calculating quantum excited states. This breakthrough,successfully applied to a range of atoms and molecules, has been published in the prestigious scientific journal Science.
Quantum excited states refer to the state of matter whereelectrons transition to higher energy levels after being stimulated by energy. These states are crucial in fields like physics and chemistry, playing a role in understanding phenomena such as photosynthesis and the workings of solar panels. However, accurately calculating excited states from first principleshas been a major challenge in the scientific community.
DeepMind researchers overcame this hurdle by utilizing neural networks. They developed a method called “Natural Excited State Variational Monte Carlo” (NES-VMC) that transforms the problem of findingthe system’s excited states into finding the ground state of an extended system, thus avoiding the free parameters and orthogonalization issues present in previous methods.
Researchers tested the NES-VMC method on a series of benchmark systems, ranging from single atoms to molecules like benzene. The results demonstrated remarkable accuracy, closely matching experimentalfindings. This marks the first time deep learning has successfully addressed some of the most challenging problems in quantum physics, opening new avenues for universal quantum simulation.
Lead author and corresponding author David Pfau stated, “This is the first time deep learning has accurately solved some of the hardest problems in quantum physics. We hope thiswill be a significant step towards universal quantum simulation.”
This research will have a profound impact on fields like quantum chemistry, materials science, and drug development. It will empower scientists to gain a deeper understanding of matter’s properties and design more advanced materials and pharmaceuticals.
About DeepMind
DeepMind is acompany founded by British AI experts and acquired by Google in 2014. Its research focuses on machine learning, deep learning, and artificial intelligence. DeepMind has developed renowned AI systems like AlphaGo and AlphaFold, achieving groundbreaking results in fields such as Go and protein structure prediction.
【来源】https://www.jiqizhixin.com/articles/2024-08-23-6
Views: 1