Claude成功完成理论物理九环散射振幅计算

Anthropic的Claude AI在约一个月内成功完成了N=4超杨-米尔斯理论的九环散射振幅计算，这是物理学家Matt von Hippel向AI公司发出的挑战任务1。该计算使用成本仅约一两千美元，充分展现了AI在前沿科学计算领域的潜力1。

Anthropic采用Claude和Claude Science平台（Fable 5.1）完成了这项工作1。值得注意的是，Claude在没有外部科学监督的情况下"一次性"完成了计算，仅获得"继续工作"的指示1。随后，Lance Dixon独立验证了Claude的计算结果1。

为完成该任务，Anthropic还采用了bootstrap方法进行辅助计算，这部分的成本约为100美元，相当于96个CPU运行一周的计算量1。同期，中国科学院的物理研究者Song He的团队也完成了部分九环振幅计算（symbol部分），使用GPT-6作为辅助工具1。

Anthropic's Claude artificial intelligence has achieved a significant breakthrough in computational theoretical physics by completing a nine-loop scattering amplitude calculation in N=4 super Yang-Mills theory 1. The accomplishment represents a major milestone in demonstrating AI's capability to tackle cutting-edge scientific challenges that previously required extensive human expertise and computational resources.

The calculation was performed using Claude and Anthropic's Claude Science platform (Fable 5.1) and was completed within approximately one month at a total cost of roughly one to two thousand dollars 1. This breakthrough directly addressed a challenge issued by physicist Matt von Hippel, who had called on AI companies to achieve either N=8 supergravity to seven loops or N=4 super Yang-Mills to nine loops 1. The computational approach employed by Claude, called the bootstrap method, required resources equivalent to ninety-six CPUs running for one week, costing approximately one hundred dollars 1. Notably, Claude completed the calculation "on the first try" without external scientific oversight, receiving only instructions to continue working 1.

The results have been independently validated by Lance Dixon 1, and researchers from the Chinese Academy of Sciences, led by Song He, also completed parts of the nine-loop amplitude calculation using GPT-6 as an assistive tool around the same timeframe 1. These parallel achievements underscore the growing role of large language models in advancing fundamental physics research and suggest that AI systems may substantially accelerate progress in theoretical physics domains that have traditionally required years of specialized human computation.