Characteristic Classes — Chern, Stiefel-Whitney, Pontryagin

Characteristic classes assign cohomology classes to vector bundles, measuring their global twisting. The three principal families are Chern classes

c_{i} \in H^{2 i} (B; Z)

for complex bundles, Stiefel–Whitney classes

w_{i} \in H^{i} (B; Z /2)

for real bundles, and Pontryagin classes

p_{i} \in H^{4 i} (B; Z)

for real bundles. They are natural under pullback, obstruct triviality and the existence of sections, and underpin index theory, cobordism, and gauge theory. Foundational work by Stiefel, Whitney, Pontryagin, and Chern, 1935–1948.

A characteristic class is a natural transformation from the functor of isomorphism classes of vector bundles over a space $B$ to the cohomology functor $H^{*} (B)$ . Concretely, to each rank- $n$ bundle $E \to B$ one assigns cohomology classes that are invariant under bundle isomorphism and commute with pullback along continuous maps. These classes matter because they convert the hard geometric problem of classifying bundles into computable algebraic invariants: a nonzero characteristic class certifies that a bundle is nontrivial, that a manifold is non-parallelizable, or that certain sections cannot exist. The surprising fact is that all characteristic classes are generated, via the splitting principle, by these three families — the cohomology of the classifying spaces $B U (n)$ , $B O (n)$ , and the relations among them are polynomial rings, so characteristic classes are precisely polynomials in $c_{i}$ , $w_{i}$ , $p_{i}$ .

The Core Insight

Every vector bundle is pulled back from a universal bundle over a classifying space $B G$ . Therefore characteristic classes are nothing more than cohomology classes of $B G$ pulled back to $B$ . Computing them reduces to knowing $H^{*} (B G)$ , which for $U (n)$ , $O (n)$ , $S O (n)$ is a polynomial (or exterior, mod-2 polynomial) ring on generators of fixed degree. The splitting principle lets us pretend every bundle is a sum of line bundles, reducing all identities to symmetric-function combinatorics in the "Chern roots."

Formal Statement

Characteristic Class

Let $Vect_{F}^{k} (B)$ denote isomorphism classes of rank- $k$ $F$ -vector bundles over a paracompact base $B$ ( $F = R$ or $C$ ). A characteristic class $c$ with coefficients in an abelian group $A$ is an assignment

c : Vect_{F}^{k} (B) ⟶ H^{*} (B; A), c (f^{*} E) = f^{*} (c (E))

for every continuous

f : B^{'} \to B

⟳Naturality is the defining axiom; it forces

c

to be a pullback of a fixed class on the classifying space, since every

E

f^{*} E_{univ}

Classification of Bundles

For a paracompact base $B$ , there is a natural bijection

Vect_{F}^{k} (B) ≅ [B, B G_{k}], E \mapsto [f_{E}]

where

G_{k} = O (k)

for

F = R

U (k)

for

F = C

, and

B G_{k} = Gr_{k} (F^{\infty})

is the infinite Grassmannian. Consequently characteristic classes are in bijection with

H^{*} (B G_{k}; A)

⟳

E G \to B G

is the universal principal bundle; pullback of the tautological vector bundle over

Gr_{k} (F^{\infty})

realizes every bundle, with classifying map unique up to homotopy.

Chern Classes (Axiomatic)

For each complex bundle $E \to B$ there exist classes $c_{i} (E) \in H^{2 i} (B; Z)$ , $i \geq 0$ , with $c_{0} = 1$ and $c_{i} (E) = 0$ for $i > rk_{C} E$ , uniquely determined by:

(Naturality) c_{i} (f^{*} E) = f^{*} c_{i} (E), (Whitney sum) c (E \oplus F) = c (E) ⌣ c (F), (Normalization) c_{1} (O (- 1)) = - [CP^{1}]^{\lor} \in H^{2} (CP^{\infty}; Z),

where the total Chern class is

c (E) = 1 + c_{1} (E) + c_{2} (E) + \dots

⟳Existence follows from

H^{*} (B U (n); Z) = Z [c_{1}, \dots, c_{n}]

; uniqueness from the splitting principle reducing to line bundles where

c_{1}

is the Euler class.

Stiefel–Whitney Classes (Axiomatic)

For each real bundle $E \to B$ there exist classes $w_{i} (E) \in H^{i} (B; Z /2)$ with $w_{0} = 1$ , $w_{i} = 0$ for $i > rk E$ , satisfying:

w_{i} (f^{*} E) = f^{*} w_{i} (E), w (E \oplus F) = w (E) ⌣ w (F), w_{1} (γ_{R}^{1}) \neq = 0 \in H^{1} (RP^{\infty}; Z /2),

with total class

w (E) = 1 + w_{1} (E) + \dots + w_{n} (E)

and

γ_{R}^{1}

the tautological line bundle over

RP^{\infty}

⟳

H^{*} (B O (n); Z /2) = Z /2 [w_{1}, \dots, w_{n}]

; the mod-2 reduction is forced because real line bundles have

Z /2

-valued classifying obstruction

w_{1}

Pontryagin Classes

For a real bundle $E \to B$ , complexify to $E \otimes C$ and define

p_{i} (E) = (- 1)^{i} c_{2 i} (E \otimes_{R} C) \in H^{4 i} (B; Z) .

The odd Chern classes

c_{2 i + 1} (E \otimes C)

are 2-torsion, since

E \otimes C ≅ \overline{E \otimes C}

forces

c_{j} = (- 1)^{j} c_{j}

⟳

\overline{E \otimes C} ≅ E \otimes C

(real bundle is self-conjugate); conjugation sends

c_{j} \mapsto (- 1)^{j} c_{j}

, so

2 c_{2 i + 1} = 0

, and the sign

(- 1)^{i}

is a normalization convention making total class well-behaved.

Why It Holds / Derivation

Existence and Uniqueness via the Splitting Principle

Step 1 — Cohomology of the classifying space. By Leray–Hirsch and induction on the rank, using the fibration

Gr_{k - 1} ↪ Gr_{k} \to CP^{\infty}

(or directly via the cell structure of Grassmannians):

H^{*} (B U (n); Z) = Z [c_{1}, \dots, c_{n}], ∣ c_{i} ∣ = 2 i; H^{*} (B O (n); Z /2) = Z /2 [w_{1}, \dots, w_{n}] .

⟳The Gysin sequence of the sphere bundle

S^{1} \to S (γ) \to Gr

plus Leray–Hirsch gives a polynomial extension at each rank.

Step 2 — Define $c_{i}$ , $w_{i}$ as pullbacks of these generators along the classifying map

f_{E}

. Naturality is automatic. Step 3 — Splitting principle. For any bundle

E \to B

there is a map

π : Fl (E) \to B

(the flag bundle) such that

π^{*} : H^{*} (B) \to H^{*} (Fl (E))

is injective and

π^{*} E

splits as a sum of line bundles

L_{1} \oplus \dots \oplus L_{n}

. Write

x_{i} = c_{1} (L_{i})

("Chern roots"). Then

c (π^{*} E) = i = 1 \prod n (1 + x_{i}), so c_{k} (E) = e_{k} (x_{1}, \dots, x_{n}),

the

k

-th elementary symmetric polynomial.

⟳Whitney sum applied to the split bundle;

c (L_{i}) = 1 + x_{i}

since line bundles have only

c_{1}

Step 4 — Whitney formula from splitting. For

E \oplus F

, the combined roots are

{x_{i}} \cup {y_{j}}

, hence

c (E \oplus F) = \prod (1 + x_{i}) \prod (1 + y_{j}) = c (E) c (F)

. Because

π^{*}

is injective, the identity descends to

B

. Step 5 — Uniqueness. Any system satisfying [3] must agree with the Euler class on line bundles (normalization), then with elementary symmetric polynomials on split bundles, then on all bundles by injectivity of

π^{*}

■

Shiing-Shen Chern | 1911–2004

Chern's 1946 paper "Characteristic Classes of Hermitian Manifolds" gave the intrinsic, curvature-based definition of what are now Chern classes, unifying the topological invariants of Stiefel, Whitney, and Pontryagin with differential geometry through the Chern–Weil homomorphism. By representing characteristic classes as polynomials in the curvature 2-form of a connection, Chern fused algebraic topology, differential geometry, and (later) gauge theory into a single framework, work that earned him the Wolf Prize and made him the central figure of 20th-century global differential geometry.

Building Intuition

Think of a vector bundle as a family of vector spaces that may twist as you move around the base. A characteristic class is a "twist meter." For a line bundle, $c_{1}$ counts how the fiber's phase winds — for $O (1)$ over $CP^{1}$ it winds once, generating $H^{2} (CP^{1}) ≅ Z$ . The top class is the Euler class: it counts zeros of a generic section. $w_{1}$ detects orientability (it vanishes iff the bundle is orientable); $w_{2}$ detects spin structure obstruction. Pontryagin classes are the real-bundle survivors of complexification — they ignore orientation and live in degrees divisible by 4. The Chern–Weil picture makes this concrete: feed the curvature into an invariant polynomial and integrate; you get an integer, the same no matter which connection you chose, because the difference is exact (a "transgression" form).

Chern–Weil Theory | de Rham picture

On a smooth manifold with a connection $\nabla$ of curvature $Ω$ (a matrix of 2-forms), the Chern classes are represented in de Rham cohomology by

c (E) = det (I + \frac{i}{2 π} Ω), c_{k} (E) = [\frac{1}{k !} (\frac{i}{2 π})^{k} \sum (invariant degree- k poly in Ω)] .

⟳Invariant polynomials of

Ω

are closed (Bianchi identity) and their cohomology class is connection-independent (the difference of two is

d

of a transgression form).

Worked Examples

Chern classes of $T\mathbb{CP}^2$

Given: Compute

c (T CP^{2})

and the Chern numbers

c_{1}^{2} [CP^{2}]

c_{2} [CP^{2}]

. Step 1 — Euler sequence. There is a short exact sequence

0 \to O \to O (1)^{\oplus 3} \to T CP^{2} \to 0.

⟳The tautological inclusion and the surjection from the standard frame;

O

is trivial.

Step 2 — Whitney sum. Since

c (O) = 1

, additivity gives

c (T CP^{2}) = c (O (1))^{3}

. With

h = c_{1} (O (1))

the hyperplane generator,

H^{*} (CP^{2}) = Z [h] / (h^{3})

c (T CP^{2}) = (1 + h)^{3} = 1 + 3 h + 3 h^{2} .

⟳Cube the binomial, drop

h^{3} = 0

Step 3 — Read off.

c_{1} = 3 h

c_{2} = 3 h^{2}

. Using

\int_{CP^{2}} h^{2} = 1

c_{1}^{2} [CP^{2}] = (3 h)^{2} = 9, c_{2} [CP^{2}] = 3.

Boxed answer:

c_{1}^{2} = 9, c_{2} = 3

(dimensionless integers). Check:

c_{2} [CP^{2}] = χ (CP^{2}) = 3.

✓

Stiefel–Whitney classes of $T\mathbb{RP}^n$ and non-parallelizability

Given: Determine when

RP^{n}

is parallelizable using

w (T RP^{n})

. Step 1 — Formula. With

a

the generator of

H^{1} (RP^{n}; Z /2)

H^{*} = Z /2 [a] / (a^{n + 1})

w (T RP^{n}) = (1 + a)^{n + 1} .

⟳

T RP^{n} \oplus R ≅ (γ^{1})^{\oplus (n + 1)}

where

γ^{1}

is the tautological line bundle,

w (γ^{1}) = 1 + a

; trivial summand has

w = 1

Step 2 — Take $n = 4$ .

(1 + a)^{5} = 1 + 5 a + 10 a^{2} + 10 a^{3} + 5 a^{4} \equiv 1 + a + a^{4} (mod 2)

(using

(k 5) mod 2

). So

w_{1} = a \neq = 0

. Step 3 — Conclusion.

w_{1} \neq = 0 \Rightarrow RP^{4}

is non-orientable, hence certainly non-parallelizable. Boxed answer:

w (T RP^{4}) = 1 + a + a^{4}

. Parallelizability requires

w = 1

; this holds only for

n = 1, 3, 7

(and trivially

n = 0

), matching the Hopf/Adams result on division algebras.

Common Misconceptions

The Whitney sum formula does NOT hold for tensor products. From

c (E \oplus F) = c (E) c (F)

one cannot infer anything multiplicative for

E \otimes F

; the Chern class of a tensor product is governed by the Chern roots:

c_{1} (L \otimes L^{'}) = c_{1} (L) + c_{1} (L^{'})

for line bundles, but higher tensor products require the full symmetric-function machinery. Also:

c_{i} (E) = 0

for all

i > 0

does not imply

E

is trivial — Chern classes are a complete invariant only in low dimensions or special cases (e.g. line bundles over compact surfaces). Vanishing characteristic classes ≠ trivial bundle. The complex line bundle situation is clean (

c_{1}

classifies line bundles over any

B

via

H^{2} (B; Z) ≅ Pic

), but in higher rank it fails. Take the tangent bundle

T S^{4}

: it is non-trivial (no nonvanishing vector field can be guaranteed; in fact

χ (S^{4}) = 2 \neq = 0

), yet all its rational Pontryagin classes vanish:

p_{1} (T S^{4}) \in H^{4} (S^{4}; Z) ≅ Z

equals

0

, since

S^{4}

bounds and

T S^{4} \oplus R

is trivial. The nontriviality is detected by the Euler class

e (T S^{4}) = 2

, not by Pontryagin classes — different invariants see different obstructions. Conversely, exotic phenomena: Milnor's exotic 7-spheres are distinguished by Pontryagin numbers of their bounding manifolds, showing these classes carry smooth-structural information.

Connections

Characteristic classes sit at the crossroads of several theories. The Euler class $e (E) \in H^{n} (B; Z)$ for oriented rank- $n$ bundles refines the top Stiefel–Whitney class ( $w_{n} = e mod 2$ ) and the top Chern class ( $c_{n} = e$ for complex bundles), connecting to Poincaré–Hopf Theorem via zero-counting. Through Chern–Weil Theory, they become curvature integrals, linking to Gauge Theory and instanton number $\frac{1}{8 π ^{2}} \int tr (F \land F) = c_{2}$ . The Atiyah–Singer Index Theorem expresses the analytic index of an elliptic operator as $\int_{M} ch (E) Td (T M)$ , with the Chern character $ch$ and Todd class $Td$ built from Chern classes. In Cobordism Theory, Pontryagin and Stiefel–Whitney numbers are complete invariants (Thom): two manifolds are unoriented cobordant iff all SW numbers agree. The classes also generate the cohomology of Classifying Spaces $B G$ , governing principal bundle theory.

Chern roots and the splitting principle: a bundle's total Chern class as a product over its line-bundle summands after pullback to the flag bundle

Real-World Applications

Yang–Mills / instantons (physics): The second Chern number $c_{2} = \frac{1}{8 π ^{2}} \int tr (F \land F)$ is the topological charge (instanton number) of an $S U (2)$ gauge field; it labels distinct vacua and underlies the $θ$ -vacuum structure of QCD.

Quantum Hall effect (condensed matter): The quantized Hall conductance equals $\frac{e ^{2}}{h}$ times the first Chern number of the Bloch bundle over the Brillouin torus (TKNN formula, 1982) — a directly measured integer.

Topological insulators: $Z /2$ invariants refining Stiefel–Whitney/Chern data classify time-reversal-invariant phases; band-structure topology is characterized by these classes.

Anomaly cancellation (string theory): The Green–Schwarz mechanism requires characteristic-class identities (e.g. $p_{1}$ , $\hat{A}$ -genus) to vanish, constraining gauge groups to $E_{8} \times E_{8}$ or $S O (32)$ .

Robotics / motion planning: Obstruction theory using Stiefel–Whitney classes detects when a continuous global section (e.g. a coherent assignment of orientations or frames) cannot exist on a configuration space.

Exotic smooth structures: Pontryagin numbers of bounding 8-manifolds distinguish Milnor's exotic 7-spheres, a cornerstone of differential topology.

Computer graphics / vector field design: The Euler class (Poincaré–Hopf) governs the minimal number of singularities in tangent vector fields on surfaces, used in texture synthesis and remeshing.

Historical Development

1935 — Eduard Stiefel (in his thesis under Hopf) introduces obstruction classes to constructing linearly independent vector fields on manifolds, the first Stiefel–Whitney classes.
1935–1937 — Hassler Whitney independently develops the theory of sphere bundles and their characteristic cohomology classes mod 2, proving the duality theorem; the classes are named jointly.
1942 — Lev Pontryagin introduces the integral characteristic classes of real bundles (Pontryagin classes) via his study of homotopy of manifolds and cobordism.
1946 — Shiing-Shen Chern defines Chern classes for complex (Hermitian) bundles via curvature, giving the Chern–Weil intrinsic construction; unifies the topological and differential-geometric pictures.
1948 — André Weil clarifies the curvature-to-cohomology map (Chern–Weil homomorphism) in full generality for principal bundles.
1953 — René Thom proves cobordism is determined by Stiefel–Whitney and Pontryagin numbers; computes the unoriented cobordism ring, winning the Fields Medal (1958).
1958 — Hirzebruch proves the signature theorem $sign (M) = ⟨ L (p_{1}, \dots), [M]⟩$ and the Riemann–Roch theorem via the Todd genus, both expressed in characteristic classes.
1963 — Atiyah–Singer index theorem expresses analytic indices through Chern/Pontryagin data, integrating the entire theory into global analysis.
1956–1962 — Milnor and Kervaire use Pontryagin numbers to detect exotic spheres and to compute groups of differentiable structures.

Subtleties & Limits

Integrality, torsion, and completeness remain delicate. (1) The Hodge conjecture asks which integral

(p, p)

cohomology classes are represented by algebraic cycles — equivalently, which characteristic-class-type data come from geometry; it is unresolved. (2) Torsion characteristic classes: beyond rational Pontryagin classes lie torsion phenomena (Stiefel–Whitney classes mod 2, integral lifts via Bockstein) whose full structure for

B O (n)

and

B S p in

is subtle and tied to the Steenrod algebra. (3) Topological invariance of rational Pontryagin classes was a deep theorem of Novikov (1965, Fields Medal); whether finer smooth invariants are topologically invariant connects to ongoing surgery-theoretic questions. (4) The completeness problem: no finite set of characteristic classes classifies bundles in general — secondary and higher operations (e.g. Massey products,

k

-invariants in Postnikov towers) are required, and the general classification of bundles over a fixed base is computationally open.