Exactness of direct limits in the category of firm modules (Q555569)

\(R\) is an associative ring (most interestingly, if without a unit) that has a \(k\)-algebra structure, where \(k\) is a field or \(\mathbb Z\). The ring \(R\times k=\hat R\) is defined by componentwise addition and product: \((r,z)(r',z')=(rr'+rz'+zr',zz')\). Daniel Quillen suggested examination of a full subcategory of \(\hat R\)-Mod given by the left \(\hat R\)-modules \(M\) that satisfy the canonical isomorphism \(R\otimes_R M\cong M\), \(r\otimes m\mapsto rm\). Modules that satisfy this condition are called (left) firm modules. Since this is the only category used in the paper, the authors denote it by \(R\)-Mod, through abuse of notation. The present authors already answered a question of Quillen, namely that the category of firm modules is not abelian for non-unital modules \(R\) [J. Algebra 318, No. 1, 377--392 (2007; Zbl 1137.16010)]. On the other hand, if \(R\) is an associative ring, such that \(R\)-Mod is an abelian category, then for every small filtered category \(\mathcal I\), \(\varinjlim: \text{Fun}(\mathcal I, R\text{-Mod})\to R\text{-Mod}\) preserves monomorphisms and kernels (Theorem 12). This is then used to prove that \(R\)-Mod is abelian if and only if it is a quotient category, via Gabriel-Popescu theorem. The authors also find an example where the direct limits are not exact, which is a familiar example of polynomials in two variables with no independent term: \(R=xk[x,y]+yk[x,y]\); the result is effective, since the authors prove that the category has a projective generator. An additional result proved here is that the category of firm modules is locally \(\aleph_1\)-presentable.