What the ladder never asked
The ladder of Chapters 1–6 is a sequence of embeddings. The radial ODE sits in a plane; the plane sits in extended phase space; extended phase space sits in a contact 3-manifold; that sits in four dimensions, then in the jet space J¹, then in six. Each rung answers the question where does this live?
No rung answers the question what does it carry with it? An embedding can be bent. If a quantity changes when the ambient space changes, it was a fact about the embedding, not about the object, and it will not survive the next rung. A dimension ladder with no invariant is a stack, not a ladder.
Gauss settled this in 1827, and the settlement is the rung this arc has been missing.
Theorema Egregium
Let S be a surface in ℝ³ with unit normal field N. Its two principal curvatures κ₁, κ₂ measure how the surface bends in the two extremal directions. Both depend on the embedding. Their product does not.
Theorema Egregium (Gauss, 1827). The Gaussian curvature K = κ₁κ₂ is intrinsic: it is determined by the first fundamental form alone, and is therefore unchanged by any isometry of the surface, however the surface is bent in the ambient space.
The word Gauss chose — egregium, remarkable — is not decoration. Each factor is extrinsic; the product is not. That is the whole of what a lift preserves.
The Gauss map
The proof of the theorem needs an instrument, and the instrument is more important than the theorem. Send each point of the surface to its unit normal, regarded as a point of the unit sphere.
The Gauss map. For an oriented surface S ⊂ ℝ³, N : S → S² sends p to its unit normal N(p). Its differential dNp is the shape operator, and
— the Gaussian curvature is the local area distortion of the Gauss map, signed by orientation.
So K is not merely related to the Gauss map; K is the Gauss map's Jacobian. Curvature measures how fast the normal turns. Where the map folds — where det dN = 0 — the surface is parabolic. Hold that; §22.7 returns to it.
Gauss–Bonnet, and a constraint on the attractor
The Gauss map's degree is a topological invariant, and integrating K recovers it.
Gauss–Bonnet. For a closed oriented surface S, ∫S K dA = 2π χ(S), where χ is the Euler characteristic.
This is the first theorem in the book's lineage where a local quantity, integrated, returns something that cannot vary continuously. It is the model for everything downstream — including the degree formulas of §22.6.
It also constrains this corpus's own objects, which is worth stating because it is checkable rather than decorative. The Neimark–Sacker analysis produces invariant tori. For any such torus, χ(T²) = 0, hence
— on any invariant torus, positive and negative curvature cancel exactly. A numerical torus whose computed curvature integral drifts from zero is reporting a discretisation error, not a discovery.
The projective Gauss map
Now change what the map records. Instead of the unit normal — which needs a metric and an orientation — record the tangent hyperplane, which needs neither.
Projective Gauss map. For a projective variety X ⊂ ℙⁿ, the map γ : Xsm → (ℙⁿ)∨ sends a smooth point x to its embedded tangent hyperplane TxX, regarded as a point of the dual projective space. The closure of its image is the dual variety X∨.
This is the same construction with the metric removed. Where N needed a normal direction, γ needs only a tangent hyperplane; where N landed in S², γ lands in the dual space. Dropping the metric is what makes the map algebraic — γ is given by the partial derivatives of the defining equations, so it is a rational map of varieties, and everything after this point is algebraic geometry.
The lift
A rational map with an indeterminacy locus is awkward. The remedy is the one this book has used since Chapter 2: do not study the map, study its graph in a bigger space. That bigger space is the projectivised cotangent bundle ℙT*ℙⁿ, which carries a canonical contact structure — the same kind of structure as Chapter 3's, now holomorphic and projective.
Conormal variety. Con(X) ⊂ ℙT*ℙⁿ is the closure of { (x, H) : x ∈ Xsm, TxX ⊆ H } — that is, the closure of the graph of γ.
The conormal variety is Legendrian. For any irreducible X ⊂ ℙⁿ, Con(X) has dimension n − 1 and is an integral submanifold of the contact distribution on ℙT*ℙⁿ.
So the graph of the Gauss map is exactly the object Chapter 5's jet space was reaching for. A Legendrian submanifold of a contact manifold — the same definition, over ℂ, in a projective ambient. The ladder's fifth rung and Gauss's 1827 map are the same construction seen from two ends.
Biduality (reflexivity). Over a field of characteristic zero, Con(X) = Con(X∨) under the canonical identification ℙT*ℙⁿ ≅ ℙT*(ℙⁿ)∨; hence (X∨)∨ = X.
The characteristic-zero hypothesis is not decoration. Reflexivity fails in positive characteristic — the standard counterexample is a strange curve whose Gauss map is inseparable. A statement that holds over ℂ and fails over 𝔽p is a statement about separability, not about geometry, and it should be quoted with its hypothesis.
Where the fold was hiding
Return to §22.3. The Gauss map degenerates where det dN = 0 — the parabolic curve, K = 0. A map between surfaces degenerating along a curve is precisely the setting of Whitney's theorem: the generic singularities of a smooth map ℝ² → ℝ² are folds along a curve and isolated cusps.
So the fold is not an analogy imported into this book from catastrophe theory. It is what the Gauss map does at a parabolic point, and it was already in Chapter 3's surface before the word was used. Under projectivisation the same statement becomes algebro-geometric: the singularities of the projection π∨ in the diagram above are the singularities of the dual variety, and in low dimension these are the Ak normal forms — fold, cusp, swallowtail.
The claim being made, and the one not being made. That fold and cusp arise as singularities of the Gauss map and of Legendrian projections is a theorem. That the classification of such singularities is ADE is a theorem. What is not claimed here is that real catastrophe theory and complex Legendrian singularity theory are the same subject: the complex theory classifies, the real theory selects real forms, and the correspondence between them requires care that this chapter does not supply. Where this corpus has used fold and cusp language for real dynamical systems, the connection drawn here is that they share a normal form, not that they are the same object.
Honest inventory
| Claim | Status | What it rests on |
|---|---|---|
| K is intrinsic | Classical | Gauss 1827 |
| K = det dN | Classical | definition of the shape operator |
| ∫K dA = 2πχ | Classical | Gauss–Bonnet |
| ∫K dA = 0 on an invariant torus | Corollary | χ(T²) = 0 |
| Con(X) is Legendrian, dim n−1 | Classical | standard; over ℂ |
| (X∨)∨ = X | Classical | char 0 only — see Remark |
| fold/cusp are Gauss-map singularities | Classical | Whitney |
| dm³ contact structure algebraises | Settled in Ch 23 | contactomorphic to J¹(ℝ,ℝ); see §23.7 |
| dm³ flow algebraises | OPEN | narrowed — see Ch 23 §23.7 |
The open question this chapter creates. The dm³ contact form is α = dz − r²dθ with r² = x² + y², which is polynomial. Whether the structure it defines is the real locus of a holomorphic contact structure on a projective variety — and if so, which one — is not settled here. A positive answer places the whole framework of Books I–IV inside algebraic geometry and makes the diagram of §22.6 available to it. A negative answer is sharper still, because it would identify precisely which feature of the dm³ geometry is irreducibly real. Either outcome is worth having; neither is assumed.