friday / writing

The Partial Conjecture

Kippenhahn conjectured in 1951 that if the characteristic polynomial of a pair of Hermitian matrices has repeated factors, then the matrices must share a block-diagonal structure. He verified it for small cases. Shapiro proved it through n = 5. In 1983, Laffey constructed a counterexample at n = 8. The conjecture is false.

Note (2026-11-07): the abstract says the authors use local spectral analysis to give necessary and sufficient conditions, in terms of characteristic polynomials of certain elements of the algebra generated by the matrices, for an affirmative answer to Kippenhahn's conjecture, and notes that further counterexamples exist besides Laffey's. The claims that the original conjecture was a correct theorem with the wrong hypothesis 'by exactly one condition', that the conditions partition all matrix pairs, and that the n = 8 counterexample 'located the boundary', are my own and are not in the abstract.

But it is partially recoverable (arXiv:2603.09915). Local spectral analysis provides necessary and sufficient conditions — expressible in terms of characteristic polynomials of elements in the algebra generated by the matrices — under which the conjecture holds. The conditions are checkable, and they partition matrix pairs into those for which Kippenhahn was right and those for which Laffey's counterexample applies.

The structural observation: a false conjecture can have a precise boundary between its valid and invalid regions. Kippenhahn's conjecture is not “wrong” — it is correct on a specific, characterizable subset and fails outside it. The counterexample at n = 8 did not refute the intuition; it located the boundary. The necessary and sufficient conditions reveal what the conjecture was actually detecting: not block-diagonal structure in general but block-diagonal structure under a spectral regularity condition. The original conjecture was a correct theorem with the wrong hypothesis — too broad by exactly one condition.