Physicists Giorgio Parisi and Francesco Zamponi of Sapienza University of Rome have utilized generative artificial intelligence to solve a decade-old mathematical problem regarding “jamming”—the transition of a system from a fluid state to a rigid, disordered one. As detailed in the Journal of Statistical Mechanics: Theory and Experiment on July 1, the researchers used Anthropic’s Claude to prove why two specific parameters, a and b, always sum to 1 in the context of particle contact forces and gaps.
The Jamming Problem Explained
Jamming occurs when a disordered system, such as a container of billiard balls, becomes so congested that the individual components are locked in place by their neighbors. In their 2014 research, Parisi and Zamponi identified that the parameters a and b, which dictate how contact forces and gaps scale at the critical jamming point, consistently equaled 1. Despite years of effort, the underlying mathematical reason for this relationship remained elusive. The breakthrough arrived when the team turned to AI to re-examine the problem, eventually producing a verified analytical solution in 40 prompts.
Did you know?
The “jamming” phenomenon is not just a theoretical concept for billiard balls; it is a fundamental state of matter.
AI as a Partner in Theoretical Physics
The collaboration between the researchers and the AI model yielded a solution hidden within the existing equations. According to Francesco Zamponi, while the initial output from Claude contained errors requiring human correction, the core logic provided a path forward that the team had previously missed. Zamponi noted that the AI offered access to a broad repository of mathematical training, effectively acting as an engine for pattern recognition that identified connections the human researchers had overlooked.
This success highlights a shift in how theoretical physics may evolve. Rather than replacing human intuition, the technology serves as a tool to accelerate mundane tasks and provide fresh perspectives. Zamponi is already applying this collaborative approach to the “random sequential addition of hard hyperspheres,” though he emphasizes that human guidance remains essential for providing the necessary conceptual framework.
Comparison: Traditional Research vs. AI-Assisted Discovery
| Approach | Role of AI | Primary Driver |
|---|---|---|
| Traditional | None | Pure Human Deduction |
| AI-Assisted | Pattern matching & Data synthesis | Human-Directed Prompts |
Future Trends in Scientific Computing
Pro Tip:
When using AI for complex problem-solving, treat the output as a draft rather than a final product. As seen in the Parisi-Zamponi study, iterative prompting and human verification of the underlying logic are critical to ensuring the validity of the math.
Frequently Asked Questions
How did AI solve the jamming problem?
The researchers prompted the AI to analyze the equations from their 2014 study. The model identified a path to prove that a + b = 1 by accessing its training data on mathematical relationships, which the researchers then verified and refined.
Is human input still necessary in AI-driven physics?
Yes. According to Zamponi, while AI can accelerate coding and identify mathematical patterns, the conceptual ideas and the final verification of the physical validity of the results must come from human experts.
Will AI replace mathematicians?
The current consensus from researchers like Zamponi is that AI functions as a collaborative partner. It can provide fresh perspectives and speed up research, but it currently lacks the ability to formulate the initial conceptual framework required for new breakthroughs.
Have you experimented with using AI to solve complex technical problems in your own work? Share your experiences in the comments below or subscribe to our newsletter for more updates on the intersection of AI and scientific discovery.
Related reading
