Answers are grounded in this paper and cite their sources.
The classification of idempotent matrices up to similarity is a foundational problem in matrix theory, serving as a bridge between the algebraic properties of rings and the geometric nature of their associated projective modules. A primary objective in this field is to identify the broadest classes of rings, specifically commutative domains, where every idempotent matrix is similar to a standard diagonal idempotent [Formula: see text]. Very recently, Călugăreanu and Lee [Carpathian J. Math. 42 (2026), 613–623] posed the open question of whether this property holds for all commutative GCD domains. While their work established this diagonal reduction for [Formula: see text] idempotents, we give a definitive negative answer for the general case. Let [Formula: see text] be the real affine coordinate ring of the two-sphere. We prove, by Nagata’s criterion, that [Formula: see text] is a unique factorization domain and hence a GCD domain. We explicitly construct an idempotent matrix [Formula: see text] in [Formula: see text]. Its image is the algebraic tangent module of the sphere, which is a finitely generated projective [Formula: see text]-module of rank [Formula: see text]. This module is not free, because freeness would imply the existence of a nowhere-zero continuous tangent vector field on [Formula: see text], contradicting the Hairy Ball Theorem. Therefore, [Formula: see text] is not similar over [Formula: see text] to [Formula: see text], nor to any standard diagonal idempotent. By adjoining zero blocks, this yields counterexamples for all sizes [Formula: see text], thereby resolving the open question in the negative for every [Formula: see text] and showing that the stated [Formula: see text] diagonalization theorem cannot hold under the stated hypotheses. The argument pinpoints the exact module-theoretic obstruction: over domains, diagonal similarity for all idempotents is equivalent to projective-freeness of the coefficient ring, a condition not implied by the GCD property.
No takes yet. Share an insight, caveat, or question.
Đặng Võ Phúc (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: