Characteristic Classes — Chern and Pontryagin

⚡ Key Takeaways

Characteristic classes assign cohomology classes to vector bundles, measuring their global twisting. Chern classes $c_i(E) \in H^{2i}(M;\mathbb{Z})$ live on complex bundles; Pontryagin classes $p_i(E) \in H^{4i}(M;\mathbb{Z})$ live on real bundles. They are natural, stable under trivial summands, and computable via curvature (Chern–Weil). Key uses: obstruction theory, index theorems, characteristic numbers as cobordism invariants. Developed by Stiefel, Whitney, Pontryagin, and Chern (1930s–1940s).

💡 The Core Insight

A vector bundle's "twisting" is encoded by how transition functions glue local trivializations. The classifying space $BU(n)$ (resp. $BO(n)$ ) is a universal object: every rank- $n$ bundle is a pullback of a universal bundle by a map $f: M \to BU(n)$ . Therefore all natural invariants of bundles are pullbacks of universal cohomology classes. Since $H^*(BU(n);\mathbb{Z}) = \mathbb{Z}[c_1,\dots,c_n]$ is a polynomial ring on the Chern classes, these classes generate everything. The Chern–Weil homomorphism then makes this concrete: invariant polynomials of curvature give de Rham representatives.

DefinitionChern Classes via Classifying Space

Let $E \to M$ be a complex vector bundle of rank $n$ . There is a classifying map $f_E : M \to BU(n)$ , unique up to homotopy, with $E \cong f_E^* (EU(n))$ . The total Chern class is

1

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

where $c_i \in H^{2i}(BU(n);\mathbb{Z})$ are the universal generators with $H^*(BU(n);\mathbb{Z}) = \mathbb{Z}[c_1,\dots,c_n]$ , $\deg c_i = 2i$ .

DefinitionAxioms for Chern Classes

The Chern classes are uniquely characterized by: 1. Naturality: $c(f^*E) = f^*c(E)$ for $f: N \to M$ . 2. Whitney sum: $c(E \oplus F) = c(E)\smile c(F)$ , i.e.

2

\quad c_k(E\oplus F) = \sum_{i+j=k} c_i(E)\smile c_j(F).

3. Normalization: for the tautological line bundle $\mathcal{O}(-1) \to \mathbb{CP}^1$ , $c_1$ evaluated on the fundamental class is $-1$ (equivalently $c_1(\mathcal{O}(1)) = [\mathbb{CP}^1]^\vee$ ). 4. Rank bound: $c_i(E) = 0$ for $i > \operatorname{rank}_\mathbb{C} E$ .

DefinitionPontryagin Classes

For a real vector bundle $E \to M$ of rank $r$ , complexify $E \otimes_\mathbb{R} \mathbb{C}$ and define

3

\quad p_i(E) := (-1)^i\, c_{2i}(E\otimes_\mathbb{R}\mathbb{C}) \in H^{4i}(M;\mathbb{Z}), \qquad p(E) = 1 + p_1 + p_2 + \cdots.

The odd Chern classes $c_{2i+1}(E\otimes\mathbb{C})$ are $2$ -torsion, hence ignored over $\mathbb{Q}$ .

TheoremChern–Weil Homomorphism

Let $E \to M$ be a complex bundle with connection $\nabla$ and curvature $F = \nabla^2 \in \Omega^2(M; \operatorname{End} E)$ . For any $\operatorname{Ad}$ -invariant polynomial $P$ on $\mathfrak{gl}_n(\mathbb{C})$ , the differential form $P(F)$ is closed and its de Rham class is independent of $\nabla$ . The total Chern class is represented by

4

\quad c(E) = \left[\det\!\left(I + \frac{i}{2\pi}F\right)\right] \in H^*_{\mathrm{dR}}(M),

so $c_k(E) = \big[\tfrac{1}{k!}\,\sigma_k(\tfrac{i}{2\pi}F)\big]$ via elementary symmetric polynomials of eigenvalues.

ProofChern–Weil: closedness and independence

Step 1 (closed). With local connection $1$ -form $A$ , curvature $F = dA + A\wedge A$ . The Bianchi identity reads $dF = F\wedge A - A\wedge F = [F,A]$ . For an invariant polynomial $P$ (so $P(gXg^{-1})=P(X)$ ), differentiating invariance gives $\sum P(\dots [X_i,\xi]\dots)=0$ for all $\xi$ . Applying this to $dP(F)$ shows

5

\quad dP(F) = k\,P(\nabla F, F,\dots,F) = 0,

using $\nabla F = 0$ (Bianchi).

Step 2 (independence of connection). Let $\nabla_0, \nabla_1$ be two connections, $\alpha = \nabla_1-\nabla_0 \in \Omega^1(\operatorname{End}E)$ . Set $\nabla_t = \nabla_0 + t\alpha$ with curvature $F_t$ . Then $\frac{d}{dt}F_t = \nabla_t \alpha$ , and

6

\quad \frac{d}{dt} P(F_t) = k\,P(\nabla_t\alpha, F_t,\dots,F_t) = d\big(k\,P(\alpha,F_t,\dots,F_t)\big),

so $P(F_1)-P(F_0) = d\!\int_0^1 k P(\alpha,F_t,\dots)\,dt$ is exact. Hence $[P(F)]$ is well-defined.

Step 3 (matches topology). Both the Chern–Weil class and the classifying-space class satisfy naturality and the Whitney formula; checking on line bundles ( $c_1 = [\frac{i}{2\pi}F]$ integrates to the degree) and invoking the splitting principle fixes normalization, so the two definitions agree over $\mathbb{R}$ . $\blacksquare$

∎

ProofSplitting Principle

For any complex bundle $E \to M$ , there exists $\pi: F(E) \to M$ (the flag/projective bundle) such that: 1. $\pi^*: H^*(M) \to H^*(F(E))$ is injective, and 2. $\pi^*E \cong L_1 \oplus \cdots \oplus L_n$ splits into line bundles.

Step 1. Form $\mathbb{P}(E) \to M$ ; the tautological line $L \subset \pi^*E$ splits off $L\oplus Q$ with $\operatorname{rank} Q = n-1$ . By Leray–Hirsch, $H^*(\mathbb{P}(E))$ is free over $H^*(M)$ , giving injectivity. Iterate on $Q$ .

Step 2. Writing $c_1(L_j) = x_j$ (the Chern roots),

7

\quad c(E) = \prod_{j=1}^n (1 + x_j), \qquad c_k(E) = \sigma_k(x_1,\dots,x_n).

Since $\pi^*$ is injective, any identity among symmetric functions of the $x_j$ that holds upstairs descends to $M$ . $\blacksquare$

∎

Figure

Chern roots and the splitting principle: a rank-2 bundle pulled back to its projectivization splits into line bundles

mathematician · Shiing-Shen Chern | 1911–2004

Chern, working at Princeton and later Chicago and Berkeley, gave in 1946 the intrinsic, curvature-based construction of the classes now bearing his name, unifying earlier work of Stiefel and Whitney on real bundles. His proof of the generalized Gauss–Bonnet theorem (1944) via the Pfaffian was a landmark, providing the local-to-global bridge that became Chern–Weil theory. He founded modern global differential geometry, mentored generations, and established the MSRI. The marriage of Élie Cartan's moving frames with topology — fiber bundles seen through curvature — is his enduring legacy.

🧠 Intuition

Think of a vector bundle as a family of vector spaces that may "twist" as you move around the base. A trivial bundle $M\times\mathbb{C}^n$ has no twisting and all characteristic classes vanish. Characteristic classes are obstructions: $c_n(E)$ is the obstruction to a nowhere-zero section, $c_1$ measures the failure of the determinant line bundle to be trivial. The Chern–Weil viewpoint says: pick any connection, look at its curvature $F$ (the local "rotation" of fibers under parallel transport), and form invariant combinations. Because $F$ measures infinitesimal holonomy, integrating its invariant polynomials over cycles counts the global twisting — and crucially the count is connection-independent, like how the Gauss–Bonnet integral of curvature equals the topological Euler characteristic regardless of the metric.

TG

The gauge-theory dictionary

In gauge theory the connection $\nabla$ is the gauge potential $A_\mu$ and the curvature $F$ is the field strength $F_{\mu\nu}$ . Chern classes are then quantized topological charges: the first Chern number $\frac{1}{2\pi}\int_\Sigma F$ counts magnetic flux (Dirac monopole quantization), and the second Chern class $\frac{1}{8\pi^2}\int \operatorname{tr}(F\wedge F)$ is the instanton number in Yang–Mills. The integrality of these numbers — that a continuous field integral lands in $\mathbb{Z}$ — is exactly the topological content of characteristic classes, explaining flux quantization and the $\theta$ -vacuum structure of QCD.

ExampleChern classes of

\mathbb{CP}^n

via the Euler sequence

Given: The tangent bundle $T\mathbb{CP}^n$ , related to the hyperplane bundle $\mathcal{O}(1)$ by the Euler sequence

8

\quad 0 \to \mathbb{C} \to \mathcal{O}(1)^{\oplus(n+1)} \to T\mathbb{CP}^n \to 0.

Let $h = c_1(\mathcal{O}(1)) \in H^2(\mathbb{CP}^n;\mathbb{Z})$ , the generator with $h^n[\mathbb{CP}^n]=1$ .

Steps. 1. Whitney sum on the exact sequence: $c(T\mathbb{CP}^n)\cdot c(\mathbb{C}) = c(\mathcal{O}(1)^{\oplus(n+1)})$ . 2. Since $c(\mathbb{C})=1$ (trivial line) and $c(\mathcal{O}(1)) = 1+h$ ,

9

\quad c(T\mathbb{CP}^n) = (1+h)^{n+1} = \sum_{k=0}^{n}\binom{n+1}{k} h^k.

3. For $n=2$ : $c(T\mathbb{CP}^2) = (1+h)^3 = 1 + 3h + 3h^2$ , so $c_1 = 3h$ , $c_2 = 3h^2$ . 4. Euler characteristic check: $c_2[\mathbb{CP}^2] = 3 \cdot (h^2[\mathbb{CP}^2]) = 3 = \chi(\mathbb{CP}^2)$ . ✓

Answer:

\boxed{c(T\mathbb{CP}^n) = (1+h)^{n+1}, \quad c_1(T\mathbb{CP}^2)=3h,\ c_2(T\mathbb{CP}^2)=3h^2,\ \chi=3.}

ExamplePontryagin class and signature of

\mathbb{CP}^2

Given: View $\mathbb{CP}^2$ as a real $4$ -manifold. We compute $p_1(T\mathbb{CP}^2)$ and verify the Hirzebruch signature theorem $\sigma = \frac{1}{3}p_1[\mathbb{CP}^2]$ .

Steps. 1. For a complex bundle $E$ , the underlying real bundle's Pontryagin class is $p(E_\mathbb{R}) = c(E)\,c(\bar E)$ since $E_\mathbb{R}\otimes\mathbb{C}\cong E\oplus\bar E$ . With Chern roots $x_1,x_2$ ( $\bar E$ has roots $-x_1,-x_2$ ):

10

\quad p_1(E_\mathbb{R}) = -c_2(E\oplus\bar E)\big|_{4} = c_1(E)^2 - 2c_2(E).

2. From the previous example, $c_1 = 3h$ , $c_2 = 3h^2$ , so

p_1 = (3h)^2 - 2(3h^2) = 9h^2 - 6h^2 = 3h^2.

3. Evaluate: $p_1[\mathbb{CP}^2] = 3 \,(h^2[\mathbb{CP}^2]) = 3$ . 4. Signature: $H^2(\mathbb{CP}^2)$ is $1$ -dimensional with positive cup-square, so $\sigma = +1$ . Check: $\frac{1}{3}p_1[\mathbb{CP}^2] = \frac{1}{3}\cdot 3 = 1 = \sigma$ . ✓

Answer:

\boxed{p_1(T\mathbb{CP}^2) = 3h^2,\quad p_1[\mathbb{CP}^2]=3,\quad \sigma=\tfrac{1}{3}p_1[\mathbb{CP}^2]=1.}

⚠️ Common mistake

Chern classes are not additive under tensor product, and Pontryagin classes are not perfectly additive under Whitney sum. The Whitney formula $c(E\oplus F)=c(E)c(F)$ holds exactly for Chern classes, but for Pontryagin classes

11

\quad p(E\oplus F) = p(E)\,p(F) \pmod{2\text{-torsion}},

the equality only being exact in $H^*(-;\mathbb{Q})$ or $H^*(-;\mathbb{Z}[\tfrac12])$ . The discrepancy is the $2$ -torsion sitting in odd complexified Chern classes. Likewise, the integral lift of $p_i$ depends on conventions; always work rationally when in doubt.

⚠️ Common mistake

$c_1(E) = 0$ does not imply $E$ is trivial. Vanishing of all Chern classes is necessary but far from sufficient. Chern classes are stable invariants and live in rational cohomology after Chern–Weil; they miss torsion phenomena and unstable information. The full classification requires the homotopy class of $f_E: M \to BU(n)$ , of which Chern classes are merely the "primary" pieces.

CounterexampleNon-isomorphic bundles with equal Chern classes

Over $S^5$ , complex rank- $2$ bundles are classified by $\pi_4(U(2)) = \pi_4(SU(2)) = \mathbb{Z}_2$ . Since $H^{\text{even}}(S^5;\mathbb{Z})$ vanishes in positive degrees, all Chern classes are zero for both bundles, yet the nontrivial bundle is not isomorphic to the trivial one. Thus characteristic classes fail to distinguish them — torsion in homotopy groups of the classifying space carries information invisible to ordinary characteristic classes.

CounterexamplePontryagin classes do not detect exotic smooth structures directly

Milnor's exotic $7$ -spheres are total spaces of $S^3$ -bundles over $S^4$ that are homeomorphic but not diffeomorphic to $S^7$ . The invariant distinguishing them ( $\lambda(M) \in \mathbb{Z}_7$ ) is built from $p_1$ of a bounding $8$ -manifold via the signature formula, not from $p_1$ of the $7$ -manifold itself (which vanishes for dimension reasons). This shows characteristic numbers must be deployed cleverly — via cobordism — to capture smooth structure.

connection

Characteristic classes sit at the crossroads of several theories. Via Chern–Weil Theory they are computed from curvature, linking to Riemannian Geometry and Connections on Fiber Bundles. The classifying-space picture connects them to Homotopy Theory and Classifying Spaces BG. The Euler class $e(E)\in H^r(M)$ of an oriented real rank- $r$ bundle satisfies $e(E)^2 = p_{r/2}(E)$ in even rank and refines $c_n$ for complex bundles ( $e = c_n$ ). The Chern character $\operatorname{ch}(E)=\sum e^{x_j}$ and Todd class are the protagonists of the Atiyah–Singer Index Theorem and Hirzebruch–Riemann–Roch Theorem. Characteristic numbers are the values defining ring homomorphisms out of Cobordism Theory, by Thom's theorem.

🔬 Where this appears

⏳ History

- 1935 — Eduard Stiefel (thesis under Hopf) introduces obstruction classes to frame fields on real manifolds.

- 1936 — Hassler Whitney independently defines what become the Stiefel–Whitney classes via embeddings and the Whitney sum formula.

- 1942 — Lev Pontryagin defines the Pontryagin classes of real bundles, studying homotopy of Grassmannians and characteristic cycles.

- 1944 — Chern proves the intrinsic generalized Gauss–Bonnet theorem using the Pfaffian of curvature, isolating the Euler class.

- 1946 — Chern's landmark paper "Characteristic classes of Hermitian manifolds" gives the curvature definition of Chern classes; André Weil develops the invariant-polynomial formalism (Chern–Weil).

- 1953 — René Thom's cobordism theory shows Stiefel–Whitney and Pontryagin numbers are complete cobordism invariants (mod torsion).

- 1954 — Friedrich Hirzebruch proves the signature theorem

\sigma = L[M]

and Riemann–Roch via the Todd class, deploying characteristic classes massively.

- 1956 — Milnor uses Pontryagin numbers (via signature defect) to detect exotic

7

-spheres.

- 1963 — Atiyah and Singer prove the index theorem, expressing analytic indices through

\operatorname{ch}

, Todd, and

\hat A

classes — the apotheosis of the theory.

open-problem

Several deep issues persist. (1) Integral characteristic classes and torsion: the integral cohomology of $BO(n)$ and $BSpin(n)$ is intricate, and detecting torsion characteristic classes (Stiefel–Whitney, secondary classes) remains computationally hard for high-dimensional structure groups. (2) Differential refinements: $\widehat{H}$ (Cheeger–Simons / differential cohomology) refines characteristic classes to capture secondary invariants like Chern–Simons forms; the full functorial picture for general gauge groups and its relation to anomalies in QFT is an active area. (3) Beyond ordinary cohomology: characteristic classes in generalized cohomology theories ( $K$ -theory, elliptic cohomology, $tmf$ ) — the Chern character, Adams operations, the Witten genus — feed open conjectures linking geometry, physics (string theory), and homotopy theory. (4) Classification gaps: as the $S^5$ example shows, characteristic classes are incomplete invariants; the precise relation between unstable homotopy of $BG$ and "computable" invariants is unresolved in general.

Characteristic Classes — Chern and Pontryagin

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also