Gauß' theorem for two submodules (Q1266193)

This paper looks at an extension of a well-known theorem, whose origins can be traced back to C. F. Gauß, which says that every finitely generated module over a principal ideal domain (PID) \(R\) is a direct sum of cyclic modules. A reformulation of this theorem gives what is now called the stacked bases theorem: If \(H\) is a submodule of a free \(R\)-module \(F\) of finite rank over the PID \(R\), then \(F\) has a basis \(a_1,\dots,a_n\) such that \(H\) has a ``stacked basis'' \(r_1a_1,\dots,r_\mu a_\mu\), where \(\mu\leq\kappa\) and the \(r_i\) form an elementary divisor chain with \(r_{i-1}\) dividing \(r_i\) for \(i=2,\dots,\mu\) [see \textit{J. M. Cohen} and \textit{H. Gluck}, J. Algebra 14, 493-505 (1970; Zbl 0191.32203)]. The goal of the paper under review is to replace the distinguished submodule \(H\) by two distinguished submodules. To clarify this, we record the following definitions due to the authors. Given any finite \(n\), the abelian category \(\text{Rep}_nR\) of \(R_n\)-modules is defined as having objects \({\mathbf F}=(F,F_0,\dots,F_{n-1})\) where \(F\) is a free \(R\)-module and each \(F_i\) is a distinguished submodule of \(F\). If \(F\) has (finite or infinite) rank \(\kappa\), then \(\mathbf F\) is also said to have rank \(\kappa\). If \(\kappa= 1\) then \(\mathbf F\) is called a cyclic \(R_n\)-module. A morphism between two \(R_n\)-modules \({\mathbf F}\) and \({\mathbf F}'\) is an \(R\)-homomorphism \(\varphi:F\to F'\) with \(F_i\varphi\subseteq F_i'\) for each \(i\). An ordinary free basis \(B\) of \(F\) is called a basis of \({\mathbf F}\) if there are elements \(r_{ib}\in R\) for \(i<n\) and \(b\in B\) such that \(\{r_{ib} b:b\in B\}\) is a basis of \(F_i\) for each \(i\). (The possibility that \(r_{ib}=0\) is not excluded.) In this language, Gauß' theorem becomes: Finite rank \(R_1\)-modules have bases. -- The question arises as to what happens if we remove the finite rank assumption. It is easily seen that, in this case, \({\mathbf F}= (F,F_0)\) has a basis only if \(F/F_0\) is a direct sum of cyclic \(R\)-modules and Cohen and Gluck (loc. cit.) showed that this was also sufficient. Their results have recently been generalized by \textit{P. Hill} and \textit{C. Megibben} [Trans. Am. Math. Soc. 312, No. 1, 377-402 (1989; Zbl 0668.20043)] and this plays a crucial rôle in the current paper. The next question that arises is if this reformulated stacked bases theorem can be extended to \(\text{Rep}_nR\) for \(n>1\). \textit{R. Göbel} and \textit{W. May} [J. Pure Appl. Algebra 65, No. 1, 29-43 (1990; Zbl 0716.16015)] have provided a negative answer for \(n\geq 4\). For \(n=2\), examples given of \(R_2\)-modules \((F,F_0,F_1)\) in section 2 of the current paper show that some purity and torsion restrictions must be imposed on the submodules. However, the tour de force of the paper is the theorem which says: if \({\mathbf F}=(F,F_0,F_1)\) is a countably generated \(R_2\)-module over a PID \(R\) such that \(F_1\) and \(F_2\) are pure in \(F\), \(F_0\cap F_1\) is a summand of \(F\), and \(F/(F_0\cap F_1)\) is the direct sum of a bounded and free \(R\)-module, then \({\mathbf F}\) is a direct sum of \(R_2\)-modules of rank at most 2. [This was recently proved in the case of finite rank \({\mathbf F}\) when \(R\subseteq \mathbb{Q}\) by \textit{D. Arnold} and \textit{M. Dugas}, Commun. Algebra 21, No. 6, 1947-1982 (1993; Zbl 0776.20017).] In the final section, applications of the main theorem are given to the theory of Butler groups.