At first glance, curvature and topology belong to entirely different worlds. Curvature is a differential-geometric quantity: it depends on distances, angles, and derivatives — it can vary from point to point, and you can change it by deforming a surface. Topology, by contrast, is blind to such local data: a mug and a doughnut are topologically identical, no matter how you stretch or squash them.

The Gauss–Bonnet theorem reveals a stunning bridge between these two worlds. No matter how you deform a closed orientable surface, the total integral of its Gaussian curvature — a sum that feels completely analytic — always equals \( 2\pi \) times the Euler characteristic of that surface, an integer that does not change at all under continuous deformations.

In other words: add up all the local bending of a surface, and you recover a global topological fingerprint. Deform the surface, watch the curvature redistribute across every point — yet the integral refuses to budge.

At a point \( p = \mathbf{r}(u_0, v_0) \), the partial derivatives

are tangent vectors to the coordinate curves. Together they span the tangent plane \( T_p M \subset \mathbb{R}^3 \): the best flat approximation to \(M\) at \(p\). A vector \( \mathbf{v} \in T_p M \) can be written \( \mathbf{v} = a\,\mathbf{r}_u + b\,\mathbf{r}_v \) for some scalars \(a,b\).

The unit normal to the surface is

The first fundamental form \( g \) is the restriction of the Euclidean inner product to the tangent plane. For tangent vectors \( \mathbf{u} = a_1\mathbf{r}_u + b_1\mathbf{r}_v \) and \( \mathbf{w} = a_2\mathbf{r}_u + b_2\mathbf{r}_v \), it is given by

with components (in the coordinate basis \(\{\mathbf{r}_u, \mathbf{r}_v\}\))

so the metric tensor is \( g = \begin{pmatrix} E & F \\ F & G \end{pmatrix} \). The first fundamental form encodes all intrinsic geometry: lengths of curves, areas of regions, and angles between tangent vectors.

The area element is \( dA = \sqrt{EG - F^2}\;du\,dv = \|\mathbf{r}_u \times \mathbf{r}_v\|\,du\,dv \).

While the first fundamental form measures lengths within the surface, the second fundamental form \( h \) measures how the surface curves in the ambient space \(\mathbb{R}^3\). Its components are

giving the matrix \( h = \begin{pmatrix} e & f \\ f & g_2 \end{pmatrix} \). For a tangent vector \(\mathbf{v}\), the quantity \( h(\mathbf{v},\mathbf{v}) \) is (up to sign convention) the normal curvature of the surface in the direction of \(\mathbf{v}\): how fast the surface bends away from its tangent plane in that direction.

The Gauss map is the smooth function

which sends each point of the surface to its outward unit normal, viewed as a point on the unit sphere \(S^2\). Its derivative \(d\nu_p : T_p M \to T_{\nu(p)}S^2\) measures how fast the normal direction rotates as you move across the surface. The shape operator (or Weingarten map) \(S_p = -d\nu_p\) is a self-adjoint linear map on \(T_pM\).

Intuitively: if the surface is very curved at \(p\), the normal rotates rapidly, and \(d\nu\) maps a small patch to a large patch on \(S^2\). If the surface is flat, the normal hardly moves at all.

The principal curvatures \(\kappa_1, \kappa_2\) at \(p\) are the eigenvalues of the shape operator \(S_p\). They give the maximum and minimum normal curvatures over all tangent directions.

The Gaussian curvature is their product:

Equivalently, \(K\) is the ratio of the signed area stretched by the Gauss map: if \(U\) is a small patch around \(p\),

Gauss's Theorema Egregium ("Remarkable Theorem", 1827) states that \(K\) depends only on the first fundamental form — it is an intrinsic property. You can compute \(K\) using only measurements within the surface, without reference to the ambient \(\mathbb{R}^3\)!

For a smooth curve \(\gamma(s)\) parameterised by arc length on \(M\), the geodesic curvature measures how much the curve bends within the surface (as opposed to bending caused by the ambient curvature of the surface itself):

where \(\dot{\gamma}\) is the unit tangent and \(\ddot{\gamma} = \nabla_{\dot\gamma}\dot\gamma\) is the covariant acceleration.

• A geodesic satisfies \(k_g \equiv 0\): it is the surface's notion of a "straight line." On a sphere, geodesics are great circles.
• The boundary \(\partial D\) of a planar disk traversed counterclockwise has \(k_g = 1/r\) (inverse radius), so \(\int k_g\,ds = 2\pi\).
• The geodesic curvature appears naturally in the local form of the Gauss–Bonnet theorem for regions with boundary.

Given a triangulation of a closed orientable surface \(M\) — a decomposition into triangular faces — let \(V\), \(E\), \(F\) denote the number of vertices, edges, and faces respectively. The Euler characteristic is

Remarkably, this integer is independent of the triangulation chosen. It is a topological invariant of \(M\). For an orientable surface of genus \(g\) (i.e., with \(g\) handles),

Let \(D \subset M\) be a compact region whose boundary \(\partial D\) is a piecewise smooth, closed curve with exterior angles \(\theta_1, \ldots, \theta_k\) at the corners. The local Gauss–Bonnet formula states:

This is Bonnet's generalisation of Gauss's geodesic triangle formula. Each term has a clear geometric meaning:

• \(\int_D K\,dA\): total Gaussian curvature of the interior.
• \(\oint_{\partial D} k_g\,ds\): total geodesic curvature of the boundary (how much the boundary "bends" within the surface).
• \(\sum \theta_i\): sum of the exterior angles at corner points of the boundary.

For a smooth region without corners, \(\sum \theta_i = 0\). For a convex polygon with all boundary curves being geodesics (\(k_g = 0\)), this reduces to a pure angle-excess formula.

Let \(M\) be a compact, oriented Riemannian manifold of dimension \(2m\) (note: only even dimensions occur, since \(\chi(M) = 0\) for odd-dimensional closed manifolds). Let \(\Omega\) denote the curvature 2-form of the Levi-Civita connection, viewed as a skew-symmetric matrix of 2-forms.

The Pfaffian

The key ingredient is the Pfaffian of the curvature matrix. For a \(2m \times 2m\) skew-symmetric matrix \(A\), the Pfaffian \(\text{Pf}(A)\) is the polynomial defined by

with a chosen sign convention. Explicitly,

When \(m=1\) (surfaces), \(\text{Pf}\begin{pmatrix}0 & K\, dA \\ -K\,dA & 0\end{pmatrix} = K\,dA\), recovering the usual curvature form.

The Gauss–Bonnet–Chern Formula

Recovering the 2D Case

When \(m=1\), the manifold is a surface and \(\text{Pf}(\Omega/2\pi) = K\,dA/(2\pi)\), so:

Why the Pfaffian?

The Pfaffian arises because the Euler class \(e \in H^{2m}(M;\mathbb{R})\) is a characteristic class of the tangent bundle. By Chern–Weil theory, every characteristic class can be represented by a closed differential form built from the curvature. The Euler class is precisely represented by \(\text{Pf}(\Omega/2\pi)\). The Gauss–Bonnet–Chern theorem says that integrating this representative gives the topological Euler characteristic \(\chi(M) = \langle e(TM), [M]\rangle\) — the pairing of the Euler class with the fundamental class.

Low-Dimensional Instances

• \(2m = 2\): The classical Gauss–Bonnet theorem, \(\int_M K\,dA = 2\pi\chi(M)\).
• \(2m = 4\): For a 4-dimensional Riemannian manifold, the integrand involves a quadratic expression in the Riemann curvature components — the formula becomes \(\int_M \frac{1}{8\pi^2}(|R|^2 - 4|\text{Ric}|^2 + R_{\rm sc}^2/4)\,dV = \chi(M)\), where \(R\), \(\text{Ric}\), \(R_{\rm sc}\) denote the full Riemann tensor, Ricci tensor, and scalar curvature respectively.