Claude has solved two open problems in theoretical physics, both concerning black holes and the decades-old search for a theory of quantum gravity. In one case, the AI model finished a calculation that physicists set up more than 20 years ago but never completed. In the other, it pinned down the exact limit of the simplest version of a technique researchers use to rule out candidate theories.
The physics problems are hard to state without the background. General relativity describes very large things, like stars. Quantum mechanics describes very small things, like atoms. Both work remarkably well, but they cannot be fully reconciled mathematically. Black holes are one of the few places where both theories matter at once, which is why physicists study them so closely.
What is the black hole phase transition problem?
The first problem involves a model universe that physicists use because it can be analyzed in detail. At low temperatures, that universe fills with a gas of particles. Above a certain temperature, a black hole forms instead. In 1983, Stephen Hawking and Don Page showed that the switch is abrupt, a phase transition similar to water suddenly boiling. Their calculation relies on general relativity, though, and it breaks down when the universe shrinks to a tiny scale where spacetime becomes blurry. On the quantum side of the same duality, that regime can be calculated in principle.
Physicists found in 1999 that a phase transition also occurs in that regime. What they did not know was whether it was abrupt or gradual. A group of physicists laid out the calculation that would answer the question in 2003. It was so long, involving many thousands of terms, that it was never finished. Claude completed it. It showed the transition is abrupt, the same kind of sudden switch Hawking and Page found at the other extreme.
Claude did the work by effectively running a small research group made up of copies of itself. Nearly every piece of the calculation was done twice by agents working independently, and parts of it were checked against previous results in the literature. The researchers also had to tame infinities that appear when the terms are naively added up, a conceptual hurdle on top of the sheer volume of work.
The result suggests that black hole formation is not merely a feature of general relativity. It appears to be built into the underlying quantum theory. The sudden switch shows up at both the blurry and smooth ends of the range, and the most natural reading is that it holds everywhere in between.
The second problem concerns a simpler model universe with two dimensions of space instead of three. Physicists cannot yet write down a complete theory of quantum gravity, even in that setting, so they use a strategy called the bootstrap. Rather than building the correct theory directly, they try to eliminate wrong ones by showing they contradict themselves.
Since 2009, researchers have used a version of this approach, called the modular bootstrap, to set a ceiling on how heavy the lightest black hole in such a universe can be. Any theory that breaks the ceiling is ruled out. Lowering the ceiling narrows the field. Earlier work by Tom Hartman, Dalimil Mazáč and Leonardo Rastelli proved a ceiling of c/8.5, where c is a quantity that roughly counts the building blocks of the universe. Numerical calculations suggested the true limit could be as low as c/9.1, but no one could prove it.
How did Claude solve the modular bootstrap problem?
Claude noticed a pattern in those numerical results that physicists had missed. It turned the pattern into a method using a mathematical tool called a spectral curve, which is well known in other fields but had not been applied to this problem. From that, it derived an exact formula for the ceiling, with a value of about 9.1266, and proved that the ceiling cannot be pushed any lower with this version of the method. It is, in other words, optimal.
How did Claude's agents work on these problems?
The bootstrap result did not come from a single agent. In August, Anthropic researcher Hongbin Chen was testing Claude’s abilities as a side project. He asked it to list open problems in theoretical physics focused on black holes and the bootstrap method. Claude returned 22 candidates hours later, and Chen assigned separate agents to each one. The agents could read each other’s progress posts and collaborate, and they ran in the background with no special setup for about six weeks.
For the modular bootstrap problem, one agent spotted the pattern while surveying the literature. Another, working on a related question about how tightly spheres can be packed in high-dimensional space, turned that pattern into a method. That method reproduced a sphere-packing result recently proved by an OpenAI model. Claude then applied it to the bootstrap problem, which an earlier Claude agent had failed to solve in 61 parallel attempts. Two other agents independently reached the same answer. In September, the results were formally verified in Lean, a computerized proof checker. Claude wrote up the paper with minimal guidance, and a physicist checked every main result.
The black hole transition also had an independent check. Weeks after one agent reached a solution, a physicist who had used Claude on the same problem shared his result, and a colleague realized Chen’s agent had gotten the same answer earlier. The two solutions used different methods, which gave researchers a useful cross-check.
Neither result speaks directly to our own universe, and neither has practical applications. Both are contributions to theoretical physics, where progress often comes in small, exacting steps. Here, the steps were taken largely by an AI system working on its own, and the results are now public in two papers, one on the black hole phase transition and one on the modular bootstrap.
Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.
