The Chern-Weil Homomorphism

The Chern-Weil homomorphism is a canonical algebra map from the ring of

Ad

-invariant polynomials on a Lie algebra

g

to the de Rham cohomology of a manifold

M

carrying a principal

G

-bundle. It sends an invariant polynomial

P

to the cohomology class of

P

evaluated on the curvature of any connection — and the result is independent of the connection chosen. It applies wherever a smooth principal bundle (or vector bundle) exists, giving the characteristic classes (Chern, Pontryagin, Euler) as concrete differential forms. Introduced by Shiing-Shen Chern and André Weil around 1944–1949.

The Chern-Weil theory provides a differential-geometric construction of the characteristic classes of a principal bundle, complementing the purely topological definition via classifying spaces $B G$ . Given a connection with curvature $Ω$ , one substitutes $Ω$ into an invariant polynomial $P$ of degree $k$ to produce a closed $2 k$ -form whose cohomology class is a topological invariant of the bundle alone. The construction matters because it turns abstract topology into computable integrals of curvature, the bridge underlying the Gauss-Bonnet theorem, index theory, instanton numbers, and anomaly cancellation in physics. The surprising fact is rigidity: although the curvature form $Ω$ depends heavily on the connection (and is generally nonzero even on a trivial bundle), the cohomology class $[P (Ω)]$ does not change at all when the connection is deformed — the dependence cancels exactly through an explicit transgression form.

The Core Insight

Curvature measures the local failure of parallel transport to be path-independent; an invariant polynomial of curvature measures something global that survives all local changes. The miracle is that the difference between $P (Ω_{0})$ and $P (Ω_{1})$ for two connections is always an exact form, so the obstruction lives in cohomology, not in the choice of geometry. Invariance under the structure group $\Leftrightarrow$ closedness; choice of connection $\Leftrightarrow$ exactness.

Formal Statement

Invariant Polynomials

Let $G$ be a Lie group with Lie algebra $g$ . A symmetric $k$ -linear map $P : g^{\otimes k} \to R$ (or $C$ ) is $Ad$ -invariant if

P (Ad_{g} X_{1}, \dots, Ad_{g} X_{k}) = P (X_{1}, \dots, X_{k}) \forall g \in G, X_{i} \in g .

⟳This is the symmetric-tensor analogue of a class function; differentiate at

g = e

to obtain the infinitesimal form below.

The associated homogeneous polynomial function is $P (X) := P (X, \dots, X)$ . Write $I^{k} (G)$ for the space of such polynomials and $I^{*} (G) = ⨁_{k} I^{k} (G)$ for the graded ring of invariant polynomials. The infinitesimal invariance condition is

i = 1 \sum k P (X_{1}, \dots, [Y, X_{i}], \dots, X_{k}) = 0 \forall Y \in g .

⟳Set

g = exp (t Y)

[1]

, use

Ad_{e x p (t Y)} = e^{t ad_{Y}}

, and differentiate at

t = 0

Connection, Curvature, and the Chern-Weil Form

Let $π : P \to M$ be a principal $G$ -bundle with a connection $1$ -form $ω \in Ω^{1} (P; g)$ and curvature $2$ -form $Ω \in Ω^{2} (P; g)$ given by the structure equation

Ω = d ω + \frac{1}{2} [ω, ω] = d ω + ω \land ω .

⟳

Ω

is the

g

-valued

2

-form

D ω = (d ω)^{hor}

projected to horizontal vectors; expanding the projection gives the bracket term.

For $P \in I^{k} (G)$ , extend it to $g$ -valued forms by $P (Ω) := P (Ω, \dots, Ω)$ , inserting curvature ( $2$ -form) entries and wedging the form parts. Because $P$ is $Ad$ -invariant, $P (Ω)$ on $P$ is basic (horizontal and $G$ -invariant), so it descends to a unique $2 k$ -form on $M$ , also denoted $P (Ω) \in Ω^{2 k} (M)$ .

Chern-Weil Theorem

Let $π : P \to M$ be a principal $G$ -bundle.

For every $P \in I^{k} (G)$ and every connection $ω$ , the form $P (Ω) \in Ω^{2 k} (M)$ is closed: $d P (Ω) = 0$ .
The de Rham class $[P (Ω)] \in H_{dR}^{2 k} (M)$ is independent of the connection $ω$ .
The resulting map

c w : I^{*} (G) ⟶ H_{dR}^{*} (M), c w (P) = [P (Ω)],

is a homomorphism of graded algebras (the Chern-Weil homomorphism), and is natural with respect to bundle pullbacks:

c w_{f^{*} P} = f^{*} \circ c w_{P}

for any smooth

f : N \to M

⟳(1) and (2) follow from the Bianchi identity and a transgression argument; algebra-homomorphism from

P (Ω) Q (Ω) = (P Q) (Ω)

in cohomology; naturality because pullback connection has pullback curvature.

Why It Holds / Derivation

Closedness and Connection-Independence

Step 1 — Bianchi identity. Differentiating

[3]

and using

\nablaΩ := d Ω + [ω, Ω]

gives the second Bianchi identity

d Ω + [ω, Ω] = 0, equivalently \nablaΩ = 0.

⟳Apply

d

Ω = d ω + \frac{1}{2} [ω, ω]

, use

d^{2} = 0

and the graded Jacobi identity for

[\cdot, \cdot]

on forms.

Step 2 — Polarization and invariance. Treat

P

as symmetric multilinear. The exterior derivative of

P (Ω, \dots, Ω)

d P (Ω^{\otimes k}) = i = 1 \sum k P (Ω, \dots, d Ω, \dots, Ω) .

⟳Leibniz rule for

d

on a multilinear expression of

2

-forms; the sign

(- 1)^{2 (i - 1)} = 1

since each preceding slot is an even-degree form.

Step 3 — Replace $d Ω$ by Bianchi and use invariance. Substitute

d Ω = - [ω, Ω]

from

[5]

into

[6]

d P (Ω^{\otimes k}) = - i = 1 \sum k P (Ω, \dots, [ω, Ω], \dots, Ω) = 0.

⟳The sum is exactly the infinitesimal-invariance relation

[2]

applied with

Y = ω

and all other slots equal to

Ω

;

Ad

-invariance forces it to vanish.

This proves

P (Ω)

is closed. Step 4 — Transgression for connection-independence. Let

ω_{0}, ω_{1}

be two connections and set

ω_{t} = ω_{0} + t α

with

α = ω_{1} - ω_{0}

g

-valued (tensorial)

1

-form; let

Ω_{t}

be its curvature. Then

P (Ω_{1}) - P (Ω_{0}) = d T P (ω_{1}, ω_{0}), T P (ω_{1}, ω_{0}) = k \int_{0}^{1} P (α, Ω_{t}^{\otimes (k - 1)}) d t .

⟳Compute

\frac{d}{d t} P (Ω_{t}) = k P (\dot{Ω}_{t}, Ω_{t}^{\otimes (k - 1)})

with

\dot{Ω}_{t} = \nabla_{t} α = d α + [ω_{t}, α]

; insert into

P

, apply invariance

[2]

and Bianchi to show the integrand equals

d P (α, Ω_{t}^{\otimes (k - 1)})

, then integrate in

t

The transgression (or Chern-Simons) form

T P

exhibits the difference as exact, so

[P (Ω_{1})] = [P (Ω_{0})]

. Step 5 — Algebra and naturality. For

P \in I^{k}

Q \in I^{l}

, the product

(P Q) (Ω) = P (Ω) \land Q (Ω)

at the form level, hence

c w (P Q) = c w (P) ⌣ c w (Q)

. If

f : N \to M

and

ω

is a connection on

P \to M

, then

f^{*} ω

is a connection on

f^{*} P \to N

with curvature

f^{*} Ω

; thus

c w (P)

pulls back correctly.

■

Shiing-Shen Chern & André Weil | 1911–2004 / 1906–1998

Chern, working at the interface of differential geometry and topology, sought intrinsic curvature representatives of the topological invariants Stiefel, Whitney, and Pontryagin had defined combinatorially. Around 1944 he gave his celebrated intrinsic proof of the generalized Gauss-Bonnet theorem, constructing the Euler form directly from curvature. André Weil, in unpublished notes and correspondence (circa 1949), abstracted the algebraic mechanism — the invariant-polynomial homomorphism — giving the construction its modern, group-theoretic form. The combined framework now bears both names, and the transgression form became the seed of what Chern and James Simons developed in 1974.

Building Intuition

Picture the curvature $Ω$ as a $g$ -valued "magnetic field" living on the bundle. Two physicists using different gauges (connections) measure wildly different field configurations $Ω_{0}, Ω_{1}$ . Yet when each computes a gauge-invariant scalar density like $tr (Ω \land Ω)$ and integrates it over a cycle, they get the same number — an integer-flavored topological invariant. The invariant polynomial $P$ is the "gauge-invariant combination," and $Ad$ -invariance is precisely gauge invariance. The transgression form $T P$ is the explicit gauge-transformation correction that proves the two answers agree. So Chern-Weil says: local geometry is gauge-dependent, but the right invariant combination of curvature is a fixed topological fingerprint of the bundle.

Connection-independence via transgression: two curvature representatives differ by an exact form

Worked Examples

First Chern Class of a Line Bundle

Given: A complex line bundle

L \to M

with structure group

U (1)

g = i R

. A unitary connection has curvature

Ω = F

, an imaginary-valued

2

-form. The relevant invariant polynomial is

P (X) = \frac{i}{2 π} X

(degree

1

). Steps:

The Chern-Weil form is $c_{1} (L) = \frac{i}{2 π} F$ , a closed real $2$ -form.
For the tautological bundle over $CP^{1} = S^{2}$ with the Fubini-Study connection, in local coordinate $z$ one has $F = - \frac{d z \land d z ˉ}{( 1 + ∣ z ∣ ^{2} ) ^{2}}$ .
Integrate: $\int_{S^{2}} c_{1} = \frac{i}{2 π} \int_{C} \frac{- d z \land d z ˉ}{( 1 + ∣ z ∣ ^{2} ) ^{2}}$ . Using $d z \land d \overset{z}{ˉ} = - 2 i d x d y$ and polar coordinates $r^{2} = ∣ z ∣^{2}$ ,

\int_{S^{2}} c_{1} = \frac{1}{π} \int_{0}^{\infty} \frac{2 π r d r}{( 1 + r ^{2} ) ^{2}} = \int_{0}^{\infty} \frac{2 r d r}{( 1 + r ^{2} ) ^{2}} = [- \frac{1}{1 + r ^{2}}]_{0}^{\infty} = 1.

⟳Substitute

u = 1 + r^{2}

d u = 2 r d r

; the angular integral contributes

2 π

, cancelling the

1/ π

to leave a clean integral.

Answer:

\int_{S^{2}} c_{1} (O (- 1)) = - 1

— wait, with sign convention

O (- 1)

gives

- 1

; with the orientation/connection above the magnitude is

1

(dimensionless integer = degree of the bundle).

Second Chern Number of an SU(2) Instanton

Given: An

S U (2)

bundle over

S^{4}

with the BPST instanton connection. The relevant invariant polynomial is the second Chern polynomial

c_{2}

, with Chern-Weil form

c_{2} = \frac{1}{8 π ^{2}} tr (F \land F)

(here

tr

in the fundamental representation;

F = Ω

). Steps:

The topological charge is $Q = \int_{S^{4}} \frac{1}{8 π ^{2}} tr (F \land F)$ .
For the standard one-instanton solution of size $ρ$ , the field strength density integrates (a classic computation using $tr (F \land F) = d tr (ω d ω + \frac{2}{3} ω^{3})$ and the asymptotic gauge) to give

Q = \frac{1}{8 π ^{2}} \int_{S^{4}} tr (F \land F) = 1.

⟳

tr (F \land F)

is exact (the Chern-Simons transgression

[8]

); by Stokes it equals a boundary integral of the Chern-Simons

3

-form over

S_{\infty}^{3}

, which evaluates to the winding number

de g (g : S^{3} \to S U (2) ≅ S^{3}) = 1

Answer:

Q = 1

(the instanton number, a dimensionless integer). For a

k

-instanton configuration,

Q = k

Common Misconceptions

Pitfall 1 — " $P (Ω)$ is a topological invariant." No: the form

P (Ω)

depends on the connection and is rarely unique. Only its de Rham cohomology class is the invariant. Different gauges yield different forms differing by an exact term. Pitfall 2 — "Chern-Weil gives all characteristic classes." It gives only the real characteristic classes. Torsion phenomena — e.g. the Stiefel-Whitney classes in

H^{*} (\cdot; Z /2)

or the torsion part of integral Chern classes — are invisible to de Rham cohomology and lie outside the image. Pitfall 3 — "Any invariant polynomial gives an integral class."

c w

lands in

H^{*} (M; R)

. Integrality (that the periods are integers) is a separate theorem requiring the correct normalization constants (

\frac{i}{2 π}

etc.) and a comparison with the classifying-space definition; it is not automatic from Chern-Weil alone.

Consider a flat connection on a topologically nontrivial bundle, e.g. a flat $U (1)$ connection on a nontrivial line bundle... actually no flat connection exists if $c_{1} \neq = 0$ , which itself illustrates the point: $c_{1} = \frac{i}{2 π} F = 0$ pointwise when $F = 0$ , forcing $[c_{1}] = 0$ , so a flat bundle has vanishing real Chern class. The counterexample to "Chern-Weil sees everything": a flat bundle with nontrivial holonomy (e.g. a flat bundle on a torus with $Z / n$ monodromy) has $F \equiv 0$ , hence all Chern-Weil forms vanish, yet the bundle is topologically nontrivial in torsion cohomology. Chern-Weil cannot detect it. This shows the homomorphism $c w$ has a kernel and is not injective on torsion-sensitive information.

Connections

The Chern-Weil homomorphism sits at the crossroads of several theories. Its target classes coincide with the topological Characteristic Classes defined via the classifying space $B G$ and the universal bundle $E G \to B G$ ; indeed $c w$ factors through the map $H^{*} (B G; R) \to H^{*} (M; R)$ induced by the classifying map. The transgression form $T P$ in $[8]$ is the origin of Chern-Simons Theory, whose action is precisely a Chern-Simons $3$ -form. When $G = G L (n)$ acting on a Vector Bundle, $c w$ produces the Chern Classes and Pontryagin Classes; for $S O (2 n)$ the Pfaffian polynomial gives the Euler Class, yielding the Gauss-Bonnet-Chern Theorem. The whole apparatus feeds into the Atiyah-Singer Index Theorem, where index = integral of a characteristic-class polynomial (the $\hat{A}$ -genus or Todd class) built by Chern-Weil from curvature.

Real-World Applications

Gauge theory / Yang-Mills (physics). The instanton number $\frac{1}{8 π ^{2}} \int tr (F \land F)$ in $[10]$ labels topological sectors of $S U (N)$ gauge fields and underlies the $θ$ -vacuum structure of QCD.
Anomaly cancellation (string theory). The Green-Schwarz mechanism balances characteristic forms (built by Chern-Weil from gauge and gravitational curvatures) to cancel gauge and gravitational anomalies.
Quantum Hall effect (condensed matter). The quantized Hall conductance equals the first Chern number $\frac{i}{2 π} \int_{BZ} F$ of the Bloch bundle over the Brillouin zone — the TKNN formula.
Topological insulators and band theory. Chern numbers and Pontryagin/Euler invariants of Berry-curvature bundles classify gapped phases of matter.
General relativity. The Gauss-Bonnet-Chern theorem (Euler form integrated over a closed surface or $4$ -manifold) and gravitational instanton charges use Pontryagin densities $tr (R \land R)$ .
Index theory and elliptic PDE. Computing analytic indices of Dirac, Dolbeault, and signature operators reduces to integrating Chern-Weil characteristic forms.
Magnetic monopoles. Dirac's quantization condition $e g = \frac{1}{2} n ℏ$ is the integrality of the first Chern class of the monopole $U (1)$ bundle over $S^{2}$ — exactly example $[9]$ .

Historical Development

1935–1940: Stiefel, Whitney, and Pontryagin introduce characteristic classes combinatorially/topologically; Whitney's classes live mod 2.
1942: Pontryagin defines his classes via real bundles and relates them to curvature heuristically.
1944: Shiing-Shen Chern publishes his intrinsic proof of the generalized Gauss-Bonnet theorem, building the Euler form directly from Riemann curvature.
1946: Chern introduces the Chern classes of complex vector bundles, giving curvature representatives via $det (I + \frac{i}{2 π} Ω)$ .
1949 (circa): André Weil, in unpublished notes circulated among colleagues, abstracts the algebraic core — the homomorphism $I^{*} (G) \to H^{*} (M)$ — establishing connection-independence by the transgression argument.
1950s: Borel and Hirzebruch systematize characteristic classes and the cohomology of $B G$ ; Hirzebruch's signature theorem uses Pontryagin classes.
1958–1963: Atiyah and Singer prove the index theorem, expressing analytic indices as integrals of Chern-Weil forms (Todd, $\hat{A}$ , $L$ -genus).
1974: Chern and Simons formalize the transgression / secondary characteristic forms, founding Chern-Simons theory.
1982–1985: Witten, and TKNN (Thouless-Kohmoto-Nightingale-den Nijs), bring Chern-Weil classes into physics (topological field theory, quantum Hall effect).

Subtleties & Limits

Beyond real cohomology. Chern-Weil only computes the image of characteristic classes in

H^{*} (M; R)

; recovering integral and torsion information requires refinements such as differential characters (Cheeger-Simons) and differential cohomology / Deligne cohomology, where the secondary (Chern-Simons) data is retained. Constructing canonical Chern-Weil-type representatives in these refined theories — and doing so functorially for infinite-dimensional or higher (gerbe,

2

-bundle) settings — remains an active research direction. Non-compact and infinite-dimensional groups. For loop groups, diffeomorphism groups, and gauge groups, the invariant-polynomial ring and convergence of the transgression integral

[8]

require care; the Chern-Weil program for such infinite-dimensional bundles (e.g. determinant line bundles, the

η

-invariant, families index theory) is subtle and not fully algebraic. Equivariant and noncommutative analogues. Equivariant Chern-Weil theory (Berline-Getzler-Vergne) and the Connes-Chern character in noncommutative geometry extend the homomorphism but raise open questions about completeness and integrality of the resulting invariants.