Zero-One Law Shows One Experiment Can Identify Systems

Cornell researchers proved that a single well-chosen experiment can reveal the model of a physical system in most cases.

Updated on Nov. 20, 2026 in Physics

Zero-One Law Shows One Experiment Can Identify Systems

Cornell researchers have proved a zero-one law for a broad class of machine learning models, showing that if one experiment identifies a system's model, almost any randomized experiment will succeed. The paper, "A zero-one law for one-shot system identification," was published in Proceedings of the National Academy of Sciences on Nov. 20.

Why it matters

One-shot identification matters for experiments that are costly, destructive, or difficult to repeat, where scientists cannot afford many trials. The result suggests that identifying a system is largely all-or-nothing, a finding that could guide how experiments are designed across the United States.

The proof covers linear-in-parameters machine learning models but not neural networks, and it applies only when inputs and outputs are measured perfectly. Real laboratory experiments with measurement noise are not yet covered by the result.

The players

Cornell University

Cornell University is a private Ivy League research university in Ithaca, New York, known for its engineering and applied mathematics programs.

Proceedings of the National Academy of Sciences

Proceedings of the National Academy of Sciences is a peer-reviewed multidisciplinary scientific journal published by the U.S. National Academy of Sciences.

The details

The researchers examined a hypothetical system set up as a computer simulation and proved theoretically that the set of unsuccessful inputs is either infinitesimally small or the entire input space. The study was supported by the National Science Foundation and DARPA, with collaborators from Imperial College London and the Courant Institute at New York University.

Timeline

  1. Nov. 20, 2026: The zero-one law paper was published in Proceedings of the National Academy of Sciences.

The Big Picture

The result resembles Claude Shannon's information theory, which set provable limits on channel capacity and reframed communication as a mathematical question. Here, a provable all-or-nothing boundary is placed on what one experiment can reveal about a physical system's model.

Researchers designing costly or destructive experiments may be able to plan fewer trials if a single well-chosen input is enough to identify a system. The result is still limited to noiseless, linear-in-parameters models, so practical gains will depend on the team extending it to real measurement conditions.

The takeaway

A single well-designed experiment may be enough to reveal a system's model, but only under perfect measurement, so scientists should treat the result as a strong starting point rather than a guarantee for noisy real-world work.

Further reading

Explore more research in Physics to see how theoretical proofs like this one are reshaping experimental design.

Source note: This article includes information reported by Cornell University News Service.