OpenAI Releases Proof Disproving the Exponential Time Hypothesis

OpenAI has released a preprint titled "The Exponential Time Hypothesis is False," accompanied by a formal proof verified in the Lean theorem prover. The paper presents a deterministic algorithm capable of counting the satisfying assignments of a canonically encoded $k$-CNF formula on $n$ variables in time $(n+m)^{O_k(1)} \cdot 2^{O_k(n \log\log n / \sqrt{\log n})}$, using polynomial space. This result formally refutes the Exponential Time Hypothesis (ETH), which posits that 3-SAT cannot be solved in subexponential time.

Understanding the Result and Its Scope

While the result is a significant theoretical breakthrough in computational complexity, it does not imply that NP-complete problems are solvable in polynomial time (P=NP). As noted by researchers in the community, the algorithm provides a subexponential-time solution for SAT, but the runtime remains "astronomically slow" for practical applications.

"The subexponential solution does carry over to any problems that can be reduced to 3SAT in a linear number of variables. They list: Counting independent sets, Counting proper q-colorings, Counting perfect matchings." — @cochne

Crucially, the subexponential advantage is relative to the number of variables ($n$). Because standard polynomial-time reductions from other NP-hard problems to 3-SAT typically inflate the number of variables, the subexponential character of the algorithm is often swallowed by the reduction overhead. Consequently, the result does not currently threaten cryptographic systems that rely on the hardness of NP-complete problems.

Theoretical vs. Empirical Implications

Discussion on Hacker News highlighted a divide between the mathematical significance of the proof and its immediate utility. Some observers noted that the algorithm's constants and the sheer size of the input required for the subexponential speedup to manifest make it purely theoretical.

"At $n = 1{,}000$ variables the exponent is already $\sim 1{,}000 \cdot 5 / 5 \approx 1{,}000$, i.e. $2^{1000}$ — far beyond any feasible computation." — @jaekwon

Despite the lack of practical impact on current hardware, the result challenges the long-standing intuition held by many complexity theorists that NP-complete problems possess an inherent exponential lower bound. The formalization in Lean provides a high degree of confidence in the correctness of the proof, moving the discourse beyond traditional pen-and-paper verification.

The Role of AI in Mathematical Discovery

This release is part of a broader series of preprints from OpenAI's math division, covering a wide array of topics from geometry to quantum complexity. The ability to generate and verify such complex proofs using automated tools has sparked debate regarding the future of mathematical research.

"I wonder if LLMs are good at math because there's very little math at this level that is wrong. With code, there's always plenty of bugs... But in math, a proof either exists or it doesn't and they can always be automatically verified using tools like Lean." — @martin-t

While some users expressed skepticism about the "black box" nature of these AI-generated proofs, the reliance on Lean verification suggests a shift toward a paradigm where the machine-checked validity of the proof is the primary standard for acceptance, rather than human intuition alone.