Try
A compact boundaryless surface immersed in R³ cannot be minimal everywhere. A catenoid avoids this obstruction by being noncompact. A finite catenoid patch has a boundary, so it also lies outside the theorem’s hypotheses.
The proof uses coordinate functions and integration by parts, not a visual argument about enclosed volume. This is one way local curvature identities produce global restrictions when compactness and boundary conditions are added.
Predict
Predict which part of the argument fails on a surface with boundary. Integration by parts produces a boundary term that can supply a nonzero contribution.
Predict whether a globally nonvanishing tangent field can exist on the sphere. The Poincaré–Hopf theorem rules it out through Euler characteristic, while a torus admits such fields. The field obstruction is topological, not a consequence of one particular embedding.
Experiment
Compare three global questions.
| Question | Local data sufficient? | Additional issue |
|---|
| Compute K at one regular point | Yes, metric derivatives | Coordinate consistency |
| Reconstruct a local immersion | g and b with compatibility | Integrability theorem |
| Obtain a global embedding | No | Topology, injectivity and continuation |
| Find a nowhere-zero tangent field | No | Euler characteristic and index |
A successful local construction can self-intersect or fail to extend globally. Do not silently upgrade a patch theorem into a classification of complete surfaces.
Understand
For an immersed surface position vector X, the vector-valued Laplace–Beltrami identity under S=−dN is Δ_MX=2HN. In local orthonormal directions, the tangential connection correction cancels the tangential second derivatives and the remaining normal trace is (tr S)N. This derives the identity from the Gauss decomposition.
If H=0 on a connected compact boundaryless surface, each coordinate X_i is harmonic. Integration by parts gives
0=∫X_iΔX_i dA=−∫||∇X_i||²dA.
Hence every coordinate is constant, contradicting rank two of the immersion. Compactness and absence of boundary are essential to that conclusion.
For a continuous tangent vector field with an isolated zero, choose an oriented local frame and normalize the field on a small positively oriented loop around the zero. Its winding number in S¹ is the index. Poincaré–Hopf states that on a compact oriented surface without boundary, a continuous tangent field with only isolated zeros has sum of indices equal to χ. The boundary restriction matters: a disk has Euler characteristic one but admits a constant nonzero tangent field. Boundary versions require additional hypotheses or terms. The local degree definition is given here; the smooth oriented-surface version is proved below using the already derived connection and curvature identities.
Prove the oriented surface index theorem
We now prove the smooth version rather than leave it as an unexplained global dependency. Let M be a compact oriented smooth Riemannian surface without boundary, and V a C² tangent field whose zeros are isolated. The zero set is closed; if it were infinite, compactness would give a zero that is an accumulation point of zeros, contradicting isolation. Thus there are finitely many zeros p₁,…,pᵣ. Use disjoint small oriented coordinate disks around them. The proof also works with no zeros, when no disks are removed.
On their complement Mε put E₁=V/||V|| and let E₂ be its positive quarter-turn. This is a globally defined orthonormal frame on Mε. Define Ω by ∇E₁=ΩE₂. The structure equation already derived in local Gauss–Bonnet gives dΩ=−K dA. Stokes therefore gives ∫Mε K dA=−∮∂Mε Ω.
Near one zero choose a smooth oriented orthonormal frame e₁,e₂ extending over the entire coordinate disk, with connection form ω. Along its positively oriented circle Cε write E₁=cosθ e₁+sinθ e₂. A real angle lift exists along a parametrized interval; it need not have the same endpoint value. Differentiation gives Ω=dθ+ω on that circle. Consequently
∮Cε Ω = 2π index(V,p)+∮Cε ω.
The index is the integer change of θ divided by 2π. It is independent of sufficiently small radius: the nonvanishing field on an annulus gives a homotopy of normalized boundary maps, whose integer degree cannot change continuously. It is independent of the extending frame too: the frame-change rotation extends across the disk and so has boundary degree zero. The latter fact follows by contracting the disk radially to its center; the restricted rotations are a homotopy to a constant map.
The boundary circles of Mε run clockwise relative to the disks, so substituting them in Stokes reverses their signs. We obtain ∫Mε K dA=Σ[2π index(V,pⱼ)+∮Cⱼ,ε ωⱼ]. The ωⱼ are smooth near their centers, their integrals tend to zero with the circle lengths, and the removed curvature integrals also tend to zero. Taking the limit yields ∫M K dA=2πΣ index(V,pⱼ). Combining with the independently established global Gauss–Bonnet formula gives
Σ index(V,pⱼ)=χ(M).
This is a deduction from the structure equation, Stokes and Gauss–Bonnet; the surface triangulation and planar turning inputs to Gauss–Bonnet remain identified in those lessons. It is not a proof of the all-dimensional index theorem. For a merely continuous field, a separate approximation argument preserving nonvanishing on the chosen boundary loops is needed; the derivation here explicitly assumes C² and does not silently differentiate a continuous field.
Compute indices, then use the global constraint
In an oriented local chart, V=(x,y) winds once counterclockwise around a small circle, so its index is+1. The sink V=(−x,−y) also has index+1: multiplying a unit vector by−1 adds a fixed π to its angle, not a reversal of winding. The saddle V=(x,−y) has index−1. More generally an invertible linear field V(x)=Ax has index sign(det A): continuously deform its positive-definite factor to the identity in a polar decomposition; the orthogonal factor is a rotation of degree+1 or a reflected rotation of degree−1. No zero is introduced on the loop during that deformation.
Changed case. A smooth field on an oriented genus-two surface has four isolated saddle zeros and no other zeros claimed. Is the report possible? What if two additional source zeros are reported? Can the theorem alone certify that the second report actually occurs?
Feedback: necessary index balance is not existence
The Euler characteristic is 2−2(2)=−2. Four saddles contribute−4, so the first report is impossible under the compact/no-boundary/isolated-zero hypotheses. Adding two sources contributes+2 and yields−2, removing this index-sum obstruction. It does not construct such a field or exclude other unverified features: a necessary sum condition is not an automatic existence certificate. A torus with one source and one saddle similarly has permitted sum zero, whereas a sphere with only a saddle has sum−1 rather than 2.Primary comparison: UC San Diego, Indices of Vector Fields and the Gauß–Bonnet Theorem, inspected 30 October 2026. The sign convention and clockwise deleted-disk boundaries are worked out here so the curvature/index relation can be checked rather than quoted.
Apply
A nowhere-zero field on the sphere would have empty index sum zero, contradicting χ=2. On a torus χ=0, and a coordinate-direction field in its quotient description gives an example without zeros. The theorem predicts an obstruction when the sum cannot vanish; it does not locate zeros or determine their individual indices.
Repair “Matching Gaussian curvature makes two surfaces rigidly congruent.” A plane and a cylinder both have K=0 but different shape operators and global topology. Local reconstruction uses both fundamental forms with Gauss–Codazzi compatibility, and global rigidity requires additional hypotheses and theorems.
For optional depth, Hopf–Rinow, all-dimensional index theory, general surface reconstruction and higher-dimensional comparison theorems extend this sequence. They have been identified where used; the directly proved core includes Frenet reconstruction, curvature operators, Gauss, first variation, Gauss’s lemma and the local-to-global Gauss–Bonnet calculation.
Continue
Prepare a synthesis with one exact surface calculation, one invariant-based obstruction and one complete proof whose dependencies are listed. The Leeds 2026/27 differential geometry outline and Illinois Math 423 syllabus informed breadth; saved exercises and software checks do not establish equivalent assessment or independently reviewed proof proficiency.
