On descent in dimension two and non-split Gorenstein modules (Q913871)

This paper solves a long-standing problem on the structure of finitely generated modules of finite injective dimension over a local noetherian ring (A,\({\mathfrak m},k)\). Namely, Sharp has introduced the class of Gorenstein modules by the condition that \(Ext^ i_ A(k,M)=0\) for all but one value of i (which is then necessarily equal to depth\(_ AM\) and to \(id_ AM\); the value of \(\dim_ kExt_ A^{depth(M)}(k,M)\) is called the rank of M). It is known that if A has a Gorenstein module, then it is Cohen-Macaulay with Gorenstein formal fibres [\textit{R. Y. Sharp}, Sémin. Algèbre Paul Dubreil, Proc., Paris 1977/78, 31ème Année, Lect. Notes Math. 740, 213-229 (1979; Zbl 0421.13003)], that a Gorenstein module of rank 1 is a dualizing module, that all Gorenstein modules of minimal rank are isomorphic, and that any Gorenstein module is a direct sum of ones of minimal rank [\textit{R. Fossum}, \textit{H.-B. Foxby}, \textit{P. Griffith}, and \textit{I. Reiten}, Publ. Math., Inst. Hautes Étud. Sci. 45, 193-215 (1975; Zbl 0321.13013)].The problem referred to above then reads: does the existence of a Gorenstein A-module imply that A has a canonical module? The answer is of considerable interest, since it is known that A has a canonical module if and only if it is a homomorphic image of a Gorenstein local ring [\textit{I. Reiten}, Proc. Am. Math. Soc. 32, 417-420 (1972; Zbl 0235.13016)]. When \(\dim (A)=1\) a positive answer has been obtained by Fossum, Foxby, Griffith, and Reiten (loc. cit.). The main contribution of the present paper is the construction (for each \(d\geq 2)\) of a local ring A which is a factorial, Cohen-Macaulay, non-Gorenstein domain of dimension d, which has a Gorenstein module of rank 2, but posesses no canonical module. The construction of the ring follows the lines of \textit{M. Brodmann} and \textit{C. Rotthaus} [J. Algebra 75, 386-394 (1982; Zbl 0483.13011)] and \textit{T. Ogoma} [Mem. Fac. Sci., Kochi Univ., Ser. A 3, 65-74 (1982; Zbl 0483.13005)] but the proof of the existence of the required Gorenstein modules involves new ideas and delicate computations in the divisor class group of the completion of A.