Lean 4 tutorial shows formal proofs can flag subtle claims in system identification
An arXiv paper introduces the Lean 4 proof assistant to the system identification community and shows how formalizing a classic Ljung lemma reveals an almost-sure caveat.
In plain words
Lean is a tool that turns mathematical proofs into code, where a small trusted program checks every step automatically. The paper's authors wrote a classic lemma from prediction-error identification in Lean and found its claim holds only almost surely, not for every realisation as its wording may suggest.
What was announced
The paper, posted to arXiv, is an introduction to Lean 4 for readers in system identification and control who have no prior exposure to formal methods. The authors explain how a Lean proof is read through two elementary examples, one on identifiability and one on set invariance, and describe how the theorem and its proof are written as code so a small, trusted kernel checks every inference.
To show how Lean can analyse published results, the authors formalize a classic lemma on inverting the noise filter in prediction-error identification, citing Ljung's 1999 Lemma 3.1. Because Lean demands justification for every step, the formalization exposes a subtle point: the lemma's original claim holds almost surely, though not for every realisation as its statement may suggest, and a simple counterexample makes this concrete. The paper closes with guidelines and perspectives on using Lean in system identification research.
What it does not show
- The source is only the paper's abstract, so the technical details of the examples, the counterexample and the guidelines are not available for verification.
- The finding about the lemma reflects the authors' formalization; no independent confirmation of the almost-sure caveat appears in the supplied material.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.