Let MM be a and let π:E→M\pi:E\to M be an oriented real vector bundle of rank 2m2m. Choose a Euclidean metric on EE and a compatible ∇\nabla whose is reduced to SO(2m)SO(2m). Let F∇∈Ω2(M;so(E))F_\nabla\in\Omega^2(M;\mathfrak{so}(E)) be its .

The Pfaffian Pf\mathrm{Pf} is an AdAd-invariant polynomial of degree mm on so(2m)\mathfrak{so}(2m) characterized by Pf(A)2=det⁡(A)\mathrm{Pf}(A)^2=\det(A) for skew-symmetric matrices AA. The Euler form of ∇\nabla is the closed 2m2m-form

e(∇)  :=  Pf ⁣(12πF∇)∈Ω2m(M).e(\nabla)\;:=\;\mathrm{Pf}\!\Big(\frac{1}{2\pi}F_\nabla\Big)\in\Omega^{2m}(M).

(Here Pf\mathrm{Pf} is applied fiberwise after choosing a local oriented orthonormal frame.)

Then:

  1. d e(∇)=0d\,e(\nabla)=0, where dd is the .
  2. The de Rham class [e(∇)]∈HdR2m(M)[e(\nabla)]\in H^{2m}_{\mathrm{dR}}(M) is independent of the choice of metric-compatible oriented connection.
  3. The Euler class e(E)∈H2m(M;Z)e(E)\in H^{2m}(M;\mathbb Z) is the canonical integral class whose image under the natural map to real (or de Rham) cohomology is [e(∇)][e(\nabla)]. Its real image does not by itself determine the integral class when cohomology has torsion.

Naturality holds: for any f:N→Mf:N\to M,

e(f∗E)=f∗e(E).e(f^*E)=f^*e(E).
Examples
  1. Oriented rank-2 bundles. If rank(E)=2\mathrm{rank}(E)=2, then F∇F_\nabla is an so(2)≅R\mathfrak{so}(2)\cong\mathbb R-valued 2-form. In an oriented orthonormal frame, F∇F_\nabla is represented by a scalar 2-form Ω\Omega, and
    e(∇)=12πΩ.e(\nabla)=\frac{1}{2\pi}\Omega.
    In particular, the Euler class is represented by 12πΩ\frac{1}{2\pi}\Omega.
  1. Tangent bundle of an oriented closed surface. For a closed oriented surface Σ\Sigma, the Euler class of TΣT\Sigma satisfies
    ⟨e(TΣ),[Σ]⟩=χ(Σ),\langle e(T\Sigma),[\Sigma]\rangle=\chi(\Sigma),
    and, for a Riemannian metric and its Levi-Civita connection, e(∇)=K vol⁡g/(2π)e(\nabla)=K\,\operatorname{vol}_g/(2\pi), where KK is the and vol⁡g\operatorname{vol}_g is the oriented .
  1. Flat oriented bundles. If ∇\nabla is flat (F∇=0F_\nabla=0), then e(∇)=0e(\nabla)=0, so e(E)e(E) vanishes in real cohomology (and hence in rational cohomology).