What was already known
Huneke and Lyubeznik asked whether such local cohomology support could fail to be closed. Finiteness was known through dimension three, and for local rings through dimension four.
AI involvement
ChatGPT 5.5/5.6 first suggested using elliptic-curve torsion points in a flawed answer, then produced a six-dimensional counterexample when prompted further. The authors modified and simplified it into the normal five-dimensional example. A separate ChatGPT interaction found the nonreduced arithmetic example of Theorem 1.3; this assessment concerns the normal construction.
The model supplied the essential torsion-specialization idea and a direct six-dimensional example; the human authors performed the substantive final five-dimensional repair and extension.
Importance
Gives normal examples answering the general finiteness question.
Limitations
Bhatt and Ma repaired and simplified a model-origin torsion-point construction. The four-dimensional nonlocal and five-dimensional local cases remain open. Their responsibility statement does not document a completed check of the final proof.
Technical verification
Detailed arguments are published, but neither a completed review of the final five-dimensional construction nor a machine certificate is documented. The authors' correctness responsibility alone does not establish a completed check.
Human understanding
Bhatt and Ma explain the torsion specialization of degree-zero line bundles and the resulting cohomology, and themselves simplify the model's construction into Theorems 2.4 and 3.1.
Correction and publication history
Gives normal examples answering the general finiteness question.
Read this published revision