An example of a fractal set of plane directions having chaotic intersections with a fixed 3-periodic surface
The problem of the asymptotic behaviour of unbounded connected components of plane sections of a 3-periodic surface in R and the structure of the associated foliations on a surface in the three-torus T = R/Z was posed by S. P. Novikov in [1]. This problem comes from the theory of normal metals and is formulated as follows. In the space R with coordinates (x, x, x) one fixes an embedded surface M that is invariant under shifts by vectors in Z ⊂ R and has null homologous projection on T = R/Z. For all covectors H = (H1, H2, H3) one considers the sections of M by planes ⟨H, x⟩ = const (we call them H-sections) and asks about the asymptotic behaviour of their unbounded regular connected components (if any). This question has been studied in a number of papers [2]–[4]. Physical consequences of the results are discussed in [5]. There are the following three possibilities for the behaviour of the H-sections, and moreover, the type of a section is the same for parallel planes. Trivial case. All connected components are bounded. Integrable case. Every regular non-closed component of an H-section is a finitely deformed straight line: γ(t) = t·v+O(1), v ∈ R\{0}. Moreover, there is a covector LM,H , defined up to sign, with relatively prime integral coordinates (L1, L2, L3) that annihilates the vector v. (If we assume our three-space is Euclidean, we can write v = LM,H ×H.) The projection of such a component on T is contained in an embedded two-torus whose homology class is Poincare dual to the covector LM,H . We denote the projective class (L1 : L2 : L3) ∈ RP of LM,H by lM,H . Chaotic case. The closure of the projection on T of any unbounded component is a surface of genus greater than two. The behaviour of such components has not been studied. In some explicitly described examples such a curve ‘wanders all around the plane’, that is, the d-neighbourhood of the curve is the whole plane for some finite d. Apparently, such behaviour is typical for the chaotic case, but this has not been proved. If H1, H2, H3 are linearly dependent over Q, then the definitions above have to be refined. Moreover, in this case the covector LM,H may be not uniquely defined. We skip these details in the present paper. For a fixed surface M and a rational point l ∈ QP ⊂ RP let DM,l denote the set DM,l = {(H1 : H2 : H3) ∈ RP; lM,H = l}. It is known that for a generic 3-periodic surface M the sets DM,l are disjoint closed domains with piecewise smooth boundary, and the set of directions H with chaotic H-sections has measure zero. (Of course, for some l ∈ QP the set DM,l may be empty.) The set of directions H with trivial H-sections is open. We call the non-empty domains DM,l stability zones. For studying the stability zones it is useful to consider a 1-parameter family Mc = {x ∈ R; f(x) = c} of level surfaces of a fixed smooth function and to introduce generalized stability zones Df,l = ⋃ cDMc,l. They are also domains with piecewise smooth boundary. If l = l′, then the zones Df,l and Df,l′ can intersect only at the boundary, and moreover, the set of their common points is at most countable (in all examples known to us, two zones have at most one common point). If there are at least two zones, then there must
Read more