Large annihilators in Cayley-Dickson algebras. II (Q1016741)

From the introduction: Cayley-Dickson algebras are nonassociative finite-dimensional \(R\)-division algebras that generalize the real numbers, the complex numbers, the quaternions, and the octonions. This paper is a sequel to Part I [Commun. Algebra 36, No. 2, 632--664 (2008; Zbl 1137.17007)], which explores some detailed algebraic properties of these algebras. Classically, the first four Cayley-Dickson algebras, i.e., \(\mathbb R\), \(\mathbb C\), \(\mathbb H\), and \(\mathbb O\), are viewed as at least somewhat well-behaved, while the larger Cayley-Dickson algebras are considered to be pathological. There are several different ways of making this distinction. One difference is that the first four algebras do not possess zero-divisors, while the higher algebras do have zero-divisors. Our primary long-term goal is to understand the zero-divisors in as much detail as possible. The specific purpose of this paper is to build directly on the ideas of Part I (loc. cit.) about zero-divisors with large annihilators. \dots Let \(A_n\) be the Cayley-Dickson algebra of dimension \(2n\). The central idea of the paper is to use a certain additive splitting of \(A_n\) to simplify multiplication formulas. Multiplication does not quite respect the splitting, but it almost does (see Proposition 4.1). Theorem 4.5 is the technical heart of the paper; it supplies expressions for multiplication of elements of a codimension 4 subspace of \(A_n\) that are simpler than one might expect. These simple multiplication formulas lead to detailed information about zero divisors and their annihilators. Section 5 takes a straightforward approach: just write out equations and solve them as explicitly as possible. Our simple multiplication formulas make this feasible. This leads to Theorem 5.10, which almost completely computes the dimension of the annihilator of any element. There are two ways in which the theorem fails to be complete. First, it only treats annihilators of elements in a codimension 4 subspace of \(A_n\). Second, rather than determining the dimension of an annihilator precisely, it gives two options, which differ by 4. We currently have no solution to the first problem. However, in this regard, it was already known that one codimension 2 slice is easy to deal with, so the restriction is really only codimension 2. We intend to address this question in future work. The second problem has a partial solution in Theorems 6.7 and 6.12, which distinguish between the two possible cases. We find that the answer for \(A_{n+1}\) depends inductively not just on an understanding of zero-divisors in \(A_n\) but also on a detailed understanding of annihilators in \(A_n\). Therefore, the description in these theorems is not as explicit as we might like. Fortunately, we have a complete understanding of zero-divisors and their annihilators in \(A_4\). This allows us to make calculations about zero-divisors in \(A_5\) that are not yet possible for \(A_n\) with \(n\geq 6\). Section 7 contains the details of these calculations in \(A_5\). Consequently, even though we have not made this result explicit in this article, it is possible to completely understand in geometric terms the zero-divisors in a codimension 4 subspace of \(A_5\). This goes a long way towards completely describing the zero-divisors of \(A_5\). In addition to the concrete results in Section 7 about \(A_5\), Section 8 gives a number of results about spaces of zero-divisors in \(A_n\) for arbitrary \(n\). Consider for a moment only the zero-divisors whose annihilators have dimension differing from the maximum possible dimension by a fixed constant. We show in Theorem 8.12 that, in a certain sense, the space of such zero-divisors does not depend on \(n\). This is a kind of stability result for zero-divisors with large annihilators; it was alluded to in Part I (Remark 15.8). The basic approach is to use the previous calculations of dimensions of annihilators, together with bounds on the dimensions of annihilators from Part~I (see Theorem 2.11).