A criterion for an abelian variety to be non-simple (Q1684742): Difference between revisions

Given a complex abelian variety \(A\) with a polarisation of type \(D=(d_1,\ldots,d_g)\) defined by an ample line bundle \(\mathcal L\), it is a matter of obvious interest to decide whether \(A\) is simple, and if not, to obtain as much information as possible about how it decomposes up to isogeny into simple pieces. This question is surprisingly difficult to answer clearly. This paper gives a very usable description of abelian subvarieties of \(A\). By examining the symmetric idempotent in the ring of Rosati-invariant rational endomorphisms associated with an abelian subvariety, the authors show that there is a bijection between abelian subvarieties of \(A\) of dimension~\(n\) and classes \(\alpha\in {\mathrm{NS}}(A)\otimes{\mathbb Q}\) such that the intersection number \(\alpha^r.{\mathcal L}^{g-r}\) vanishes for \(r>n\) and is equal to \(\chi({\mathcal L})n!\prod_{n < i\leq g}(i-r)\) for \(1\leq i\leq n\). In the case of principal polarisation, this was shown earlier by the first author in [Math. Z. 282, No. 3--4, 731--746 (2016; Zbl 1346.14112)] but the polarisation defined by restricting \({\mathcal L}\) to \(X\) will not be principal in general, so that is insufficient for recursive use. Indeed, the class \(\alpha_X\in {\mathrm{NS}}(A)\otimes {\mathbb Q}\) associated to \(X\) is the class of \(X\) divided by its exponent \(e_X\), the lcm of the elementary divisors belonging to the polarisation induced on \(X\). Using this approach, the authors are able to give a straightforward criterion that distinguishes abelian varieties that are isogenous to a product of elliptic curves. For some purposes it is more helpful to have a criterion in terms of period matrices. Translating the result above into those terms gives computable conditions for the existence of abelian subvarities of a given dimension in terms of the entries in the period matrix, but they do rapidly become unwieldy. The case of elliptic curves in abelian threefolds is worked out in full here, and the equations occupy a page. They are not unusable, though: they are able to recover a result of \textit{P. Borówka} [Ann. Mat. Pura Appl. (4) 195, No. 5, 1531--1549 (2016; Zbl 1353.14055)] that describes some period matrices giving all principally polarised abelian \(g\)-folds containing an abelian subvariety with prescribed polarisation type. The final section applies these results to refine the isotypical isogeny decomposition of abelian varieties carrying a finite group action, in two examples.