It was proved by Hitchin that any solution of his evolution equations for a half-flat SU(3)structure on a compact six-manifold M defines an extension of M to a seven-manifold with holonomy in G 2. We give a new proof, which does not require the compactness of M . More generally, we prove that the evolution of any half-flat G-structure on a six-manifold M defines an extension of M to a Ricci-flat seven-manifold N , for any real form G of SL(3, C). If G is non-compact, then the holonomy group of N is a subgroup of the non-compact form G * 2 of G C 2 . Similar results are obtained for the extension of nearly half-flat structures by nearly parallel G 2-or G * 2 -structures, as well as for the extension of cocalibrated G2-and G * 2 -structures by parallel Spin(7)-and Spin 0 (3, 4)-structures, respectively. As an application, we obtain that any six-dimensional homogeneous manifold with an invariant half-flat structure admits a canonical extension to a seven-manifold with a parallel G 2-or G * 2 -structure. For the group H3 × H3, where H 3 is the three-dimensional Heisenberg group, we describe all left-invariant half-flat structures and develop a method to explicitly determine the resulting parallel G 2-or G * 2 -structure without integrating. In particular, we construct three eight-parameter families of metrics with holonomy equal to G 2 and G * 2 . Moreover, we obtain a strong rigidity result for the metrics induced by a half-flat structure (ω, ρ) on H 3 × H3 satisfying ω(z, z) = 0, where z denotes the centre. Finally, we describe the special geometry of the space of stable three-forms satisfying a reality condition. Considering all possible reality conditions, we find four different special Kähler manifolds and one special para-Kähler manifold.Proposition 1.1. Let V be an n-dimensional real or complex vector space. The general linear group GL(V ) has an open orbit in Λ k V * , with 0 k n/2, if and only if k 2 or if k = 3 and n = 6, 7 or 8.Proof. The representation of GL(V ) on Λ k V * is irreducible. In the complex case the result thus follows, for instance, from the classification of irreducible complex prehomogeneous vector spaces [32]. The result in the real case follows from the complex case, since the complexification of the GL(n, R)-module Λ k R n * is an irreducible GL(n, C)-module. Remark 1.2. An open orbit is unique in the complex case, since an orbit that is open in the usual topology is also Zariski-open and Zariski-dense (see [31, Proposition 2.2]). Over the reals, the number of open orbits is finite by a well-known theorem of Whitney.Proposition 1.4. Let V be an oriented real vector space of dimension n and assume that k ∈ {2, n − 2} with n even, or k ∈ {3, n − 3} with n = 6, 7 or 8. There is a GL + (V )-equivariant mappinghomogeneous of degree n/k, which assigns a volume form to a stable k-form and which vanishes on non-stable forms. Given a stable k-form ρ, the derivative of φ in ρ defines a dual (n − k)-form ρ ∈ Λ n−k V * by the property(1.1)The dual formρ is also stable and satisfiesA stable form, it...