Characteristic Classes

⚡ Key Takeaways

Characteristic classes are cohomology classes naturally associated to vector bundles (or principal bundles) that measure their global twisting and obstruct triviality. The central families are the Chern classes (complex bundles), Stiefel–Whitney classes (real bundles, $\mathbb{Z}/2$ coefficients), Pontryagin classes (real bundles, integral), and the Euler class (oriented bundles). They live in $H^*(X)$ , are natural under pullback, and are computed via the Chern–Weil homomorphism from curvature. Introduced by Stiefel, Whitney, Pontryagin, and Chern (1935–1948).

💡 The Core Insight

The classification of vector bundles is hard, but it factors through homotopy: isomorphism classes of rank- $n$ bundles over $X$ are in bijection with homotopy classes of maps $X \to BG$ (the classifying space). Characteristic classes are nothing more than the image of $H^*(BG)$ under this correspondence. Thus understanding all natural invariants of bundles reduces to computing the cohomology ring of a single space $BG$ — and these rings are polynomial algebras on the characteristic classes.

DefinitionCharacteristic Class

Let $G$ be a topological group and $R$ a commutative ring. A characteristic class of degree $k$ for principal $G$ -bundles with coefficients in $R$ is a natural transformation

1

\quad c: \mathrm{Bun}_G(-) \longrightarrow H^k(-; R)

Derivation hint: $\mathrm{Bun}_G(X)$ is the set of isomorphism classes of principal $G$ -bundles on $X$ ; "natural" means $c(f^*P) = f^*c(P)$ for all maps $f$ . Concretely, for each space $X$ we have an assignment $P \mapsto c(P) \in H^k(X;R)$ satisfying, for every continuous $f:Y\to X$ ,

2

\quad c(f^*P) = f^*\big(c(P)\big).

Derivation hint: naturality is the single defining axiom; everything else (e.g. Whitney sum formulas) is a theorem about specific classes.

TheoremClassification by the Classifying Space

For each topological group $G$ there exists a space $BG$ with a universal principal $G$ -bundle $EG \to BG$ (where $EG$ is contractible) such that for every paracompact space $X$ :

3

\quad \mathrm{Bun}_G(X) \;\cong\; [X, BG],

the set of homotopy classes of maps. Under this bijection the bundle $P$ corresponds to a classifying map $f_P: X \to BG$ with $P \cong f_P^* (EG)$ . Consequently

4

\quad \{\text{characteristic classes of degree }k\} \;\cong\; H^k(BG; R).

Derivation hint: $EG$ contractible forces all higher homotopy of the fibration to live in $BG$ ; naturality (eq. 2) plus the Yoneda-type argument identifies natural transformations into $H^k$ with elements of $H^k(BG)$ via $c \mapsto c(EG)$ .

DefinitionChern Classes

For a complex vector bundle $E \to X$ of rank $n$ , the total Chern class is

5

\quad c(E) = 1 + c_1(E) + c_2(E) + \cdots + c_n(E), \qquad c_i(E)\in H^{2i}(X;\mathbb{Z}),

characterized uniquely by: (i) naturality; (ii) the Whitney sum formula $c(E\oplus F)=c(E)\smile c(F)$ ; (iii) normalization $c_1(\mathcal{O}(1))$ equals the generator of $H^2(\mathbb{CP}^\infty;\mathbb{Z})$ for the tautological line bundle. Derivation hint: $BU(n)$ has $H^*(BU(n);\mathbb{Z})=\mathbb{Z}[c_1,\dots,c_n]$ ; the $c_i$ are the universal classes, and the three axioms pin them down.

DefinitionStiefel–Whitney, Pontryagin, Euler

For a real vector bundle $E\to X$ of rank $n$ :

Stiefel–Whitney: $w(E)=1+w_1+\cdots+w_n$ , with $w_i\in H^i(X;\mathbb{Z}/2)$ , satisfying Whitney sum and $w_1(\gamma^1_1)\neq 0$ on $\mathbb{RP}^\infty$ .
Pontryagin: $p_i(E)=(-1)^i c_{2i}(E\otimes\mathbb{C})\in H^{4i}(X;\mathbb{Z})$ .
Euler class (oriented $E$ , rank $n$ ): $e(E)\in H^n(X;\mathbb{Z})$ , with $e(E)^2 = (-1)^? \cdots$ related to the top Pontryagin class and $w_n \equiv e \pmod 2$ .

6

\quad p_i(E) = (-1)^i\, c_{2i}(E\otimes_{\mathbb R}\mathbb{C}),\qquad e(E)\smile e(E) = p_{n/2}(E)\ (n\text{ even}).

Derivation hint: complexifying a real bundle and applying Chern classes, then noting $\overline{E\otimes\mathbb C}\cong E\otimes\mathbb C$ kills odd Chern classes mod torsion, leaving Pontryagin in degrees $4i$ .

ProofCurvature Computes Characteristic Classes

Step 1 — Connection and curvature. Let $\pi: P\to X$ be a principal $G$ -bundle with connection 1-form $\omega \in \Omega^1(P;\mathfrak g)$ . Its curvature is

7

\quad \Omega = d\omega + \tfrac{1}{2}[\omega,\omega] \in \Omega^2(P;\mathfrak g).

Derivation hint: the structure equation; $[\omega,\omega]$ combines Lie bracket on $\mathfrak g$ with wedge on forms.

Step 2 — Invariant polynomials. Let $P^k:\mathfrak g\to \mathbb C$ be an $\mathrm{Ad}$ -invariant symmetric polynomial of degree $k$ . Evaluate it on the curvature to get a $2k$ -form on $X$ (it descends because of invariance):

8

\quad P^k(\Omega) \in \Omega^{2k}(X).

Derivation hint: $\mathrm{Ad}$ -invariance under the gauge group makes $P^k(\Omega)$ basic, hence a form on the base $X$ .

Step 3 — Closedness. By the Bianchi identity $d\Omega + [\omega,\Omega]=0$ and invariance of $P^k$ ,

9

\quad d\,P^k(\Omega) = 0.

Derivation hint: differentiate $P^k(\Omega,\dots,\Omega)$ , use the derivation property and Bianchi; the bracket terms cancel by $\mathrm{Ad}$ -invariance.

Step 4 — Independence of connection. If $\omega_0,\omega_1$ are two connections with curvatures $\Omega_0,\Omega_1$ , the difference of $P^k(\Omega_i)$ is exact:

10

\quad P^k(\Omega_1) - P^k(\Omega_0) = d\,(TP^k),

where $TP^k$ is the transgression (Chern–Simons) form built from $\omega_t = \omega_0 + t(\omega_1-\omega_0)$ . Derivation hint: interpolate connections, differentiate $P^k(\Omega_t)$ in $t$ , integrate over $[0,1]$ ; the integrand is exact.

Step 5 — Chern–Weil homomorphism. Hence $[P^k(\Omega)]\in H^{2k}_{dR}(X)$ is independent of the connection, defining a ring homomorphism

11

\quad w: \big(\mathrm{Sym}^\bullet \mathfrak g^*\big)^G \longrightarrow H^*_{dR}(X), \qquad P^k \mapsto [P^k(\Omega)].

For $G=U(n)$ the elementary symmetric/Newton polynomials in the curvature eigenvalues yield exactly the (real) Chern classes:

12

\quad c(E) = \det\!\Big(I + \tfrac{i}{2\pi}\Omega\Big) = 1 + \tfrac{i}{2\pi}\mathrm{tr}\,\Omega + \cdots .

Derivation hint: expand the determinant in powers of $\Omega$ ; the $\tfrac{i}{2\pi}$ normalizes so the classes are integral, matching the topological definition over $\mathbb{Z}$ .

∎

⚠️ Real vs. de Rham coefficients

Chern–Weil produces classes in $H^*(X;\mathbb{R})$ only. It sees Chern and Pontryagin classes up to torsion but cannot detect torsion classes, and it cannot produce $\mathbb{Z}/2$ -valued Stiefel–Whitney classes at all. The integrality of Chern classes (that $\tfrac{i}{2\pi}\det(\cdots)$ is an integral class) is an extra theorem, not a consequence of Chern–Weil.

mathematician · Shiing-Shen Chern | 1911–2004

Chern, building on Whitney and Pontryagin, introduced the complex characteristic classes now bearing his name in his 1946 Annals paper, and with André Weil developed the curvature-theoretic (Chern–Weil) formulation that unified the differential-geometric and topological viewpoints. His intrinsic proof of the generalized Gauss–Bonnet theorem (1944) — expressing the Euler characteristic as the integral of a curvature Pfaffian — was the conceptual seed: a local geometric density integrating to a global topological invariant. Chern's later work (Chern–Simons invariants, 1974, with James Simons) extended these ideas to secondary characteristic classes that became central in gauge theory and 3-manifold topology.

🧠 Intuition

Think of a characteristic class as a "twist meter." A trivial bundle $X\times \mathbb{C}^n$ has total Chern class $c=1$ : no twisting. The first Chern class $c_1$ of a complex line bundle is precisely the obstruction to finding a nonvanishing section — for line bundles over a surface, $\int_\Sigma c_1$ counts the signed zeros of a generic section. The Euler class generalizes this: $e(E)$ is the primary obstruction to a nowhere-zero section, and its top-dimensional integral over a closed manifold gives the Euler characteristic (Poincaré–Hopf). The Whitney sum formula $c(E\oplus F)=c(E)c(F)$ reflects that twisting "adds up multiplicatively" when you stack bundles, just as the determinant of a block-diagonal matrix is the product of the blocks. The deepest intuition is the classifying space: every bundle is a snapshot of the one universal bundle, and characteristic classes are the universal "coordinates" $H^*(BG)$ read off through your particular snapshot.

Figure

Curvature of a line bundle integrating to its first Chern number over a sphere

ExampleChern Class of the Tautological Bundle over

\mathbb{CP}^1

Given: The tautological line bundle $\gamma \to \mathbb{CP}^1 = S^2$ , whose fiber over a line $\ell\subset\mathbb C^2$ is $\ell$ itself.

Steps: 1. $H^2(\mathbb{CP}^1;\mathbb Z)\cong\mathbb Z$ , generated by the dual of the fundamental class $[\mathbb{CP}^1]$ . 2. The hyperplane bundle $\mathcal O(1)=\gamma^*$ has $c_1(\mathcal O(1))=h$ , the positive generator, by normalization. 3. Therefore $c_1(\gamma) = c_1(\mathcal O(1)^*) = -h$ . 4. Integrate: $\int_{\mathbb{CP}^1} c_1(\gamma) = -1$ . 5. Check via Chern–Weil: with the Fubini–Study connection, $\tfrac{i}{2\pi}\int_{\mathbb{CP}^1}\Omega = -1$ .

Answer:

\boxed{\;c_1(\gamma) = -h,\qquad \int_{\mathbb{CP}^1} c_1(\gamma) = -1\;}

(dimensionless degree / Chern number).

ExampleStiefel–Whitney Obstruction to Parallelizability of

\mathbb{RP}^2

Given: Real projective plane $\mathbb{RP}^2$ ; its tangent bundle $T\mathbb{RP}^2$ . We test whether $\mathbb{RP}^2$ is parallelizable and orientable.

Steps: 1. $H^*(\mathbb{RP}^2;\mathbb Z/2)=\mathbb Z/2[a]/(a^3)$ , $|a|=1$ . 2. For $\mathbb{RP}^n$ , $w(T\mathbb{RP}^n) = (1+a)^{n+1}$ . For $n=2$ : $w = (1+a)^3 = 1+3a+3a^2 = 1+a+a^2$ (mod 2). 3. So $w_1 = a \neq 0 \Rightarrow$ non-orientable. 4. $w_2 = a^2 \neq 0 \Rightarrow$ tangent bundle is nontrivial; $\mathbb{RP}^2$ is not parallelizable. 5. $w_2 \neq 0$ also obstructs a spin structure.

Answer:

\boxed{\;w(T\mathbb{RP}^2) = 1 + a + a^2,\quad w_1\neq0,\ w_2\neq0\;}

$\mathbb{RP}^2$ is non-orientable, non-parallelizable, and admits no spin structure.

⚠️ Vanishing classes do not imply triviality

A bundle with all characteristic classes zero need not be trivial. Characteristic classes are necessary but not sufficient: they capture only the part of the classifying map $X\to BG$ seen by ordinary cohomology. For example, the tangent bundle $TS^n$ for $n$ even has $e(TS^n)=\chi(S^n)=2 \neq 0$ , but for $n$ odd $e=0$ and yet $TS^n$ is nontrivial for $n\neq 1,3,7$ (these are the only parallelizable spheres, by Adams). Higher invariants (K-theory, secondary operations) are needed.

CounterexampleStably trivial but nontrivial

The tangent bundle of $S^2$ satisfies $TS^2 \oplus \mathbb{R} \cong \mathbb{R}^3$ (the normal bundle of $S^2\subset\mathbb R^3$ is trivial, and $TS^2\oplus\nu = T\mathbb R^3|_{S^2}$ ). Thus $TS^2$ is stably trivial, so all its Pontryagin and stable classes vanish. Nonetheless $TS^2$ is nontrivial — its Euler class $e(TS^2)=2[S^2]^\vee \neq 0$ detects it. Stable invariants alone are blind here; the unstable Euler class is essential.

connection

Characteristic classes sit at a crossroads of geometry, topology, and analysis. Via Chern–Weil Theory they are computed from Connections on Principal Bundles and curvature. Through Classifying Spaces $BG$ they are equivalent to computing $H^*(BG)$ , linking to Cohomology of Lie Groups. They are the building blocks of Index Theory: the Atiyah–Singer index theorem expresses an analytic index as $\int_X \mathrm{ch}(E)\,\mathrm{Td}(TX)$ in characteristic classes. In K-Theory the Chern character $\mathrm{ch}: K(X)\otimes\mathbb Q \xrightarrow{\sim} H^{ev}(X;\mathbb Q)$ is a ring isomorphism. They control Cobordism Theory (Thom): Stiefel–Whitney and Pontryagin numbers are complete cobordism invariants. The Euler class connects to Poincaré–Hopf Theorem and the Gauss–Bonnet Theorem.

🔬 Where this appears

⏳ History

- 1935 — Eduard Stiefel (in his thesis under Hopf) and Hassler Whitney independently introduce the mod-2 classes of tangent bundles and manifolds, now Stiefel–Whitney classes.

- 1936 — Whitney develops the general theory of sphere bundles and their characteristic cohomology classes; states the duality/sum ideas.

- 1942 — Lev Pontryagin introduces real characteristic classes via Grassmannian cycles, the Pontryagin classes, studying real bundles.

- 1944 — S.-S. Chern gives an intrinsic proof of the generalized Gauss–Bonnet theorem, expressing

\chi(M)

as a curvature integral — the germ of Chern–Weil.

- 1946 — Chern defines the complex characteristic classes (Chern classes) in Characteristic classes of Hermitian manifolds (Annals of Math.).

- 1947–1949 — André Weil and Chern formalize the curvature/invariant-polynomial construction: the Chern–Weil homomorphism.

- 1953 — René Thom relates Stiefel–Whitney and Pontryagin numbers to cobordism, proving they are complete unoriented/oriented cobordism invariants.

- 1958 — Hirzebruch's signature theorem and Riemann–Roch express invariants as polynomials in Pontryagin/Chern classes (Todd,

\hat A

,

L

-genera).

- 1963 — Atiyah and Singer prove the index theorem, with the index expressed in characteristic classes; the canonical reference text is Milnor–Stasheff, Characteristic Classes (1974).

open-problem

Characteristic classes are incomplete invariants, and pinning down exactly what they miss remains active. Open and subtle directions include: (1) Detecting torsion and unstable phenomena — ordinary characteristic classes miss torsion in $H^*(BG)$ and the finer homotopy of classifying maps; secondary and higher operations are only partially understood. (2) Integral characteristic classes for $\mathrm{Spin}$ , $\mathrm{String}$ , and higher structures — the cohomology of $B\mathrm{String}$ and the theory of differential cohomology / characteristic classes refining $\mathbb R$ -valued data (Cheeger–Simons, differential cohomology) are still being developed for higher categorical bundles (gerbes, 2-bundles). (3) The Hopf invariant / parallelizability — although solved for spheres (Adams), the existence of various structures (almost-complex, etc.) on general manifolds is constrained but not determined by characteristic classes. (4) Equivariant and motivic refinements — characteristic classes in equivariant cohomology and in motivic/algebraic settings (Chow groups, algebraic cobordism) raise integrality and comparison questions that are not fully resolved.

Formal Statement

Why It Holds / Derivation: Chern–Weil Theory

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also