Curvature Forms and Characteristic Classes

Carry a vector around a tiny closed loop, keeping it parallel to itself the whole way. On a curved space it does not come back pointing where it started — it returns rotated, and the rotation per unit enclosed area is the curvature. Package the loop's two edge directions as the slots of a 2-form and the rotation as an element of the structure group's Lie algebra, and that single object is the curvature 2-form $\Omega = \mathrm{d}\omega + \omega\wedge\omega$ , built from the connection 1-form $\omega$ exactly as the Riemann tensor is built from Christoffel symbols. It is a local, geometric quantity: change the connection and $\Omega$ changes. The discovery of Chern and Weil is that certain polynomial combinations of curvature — those invariant under the adjoint action of the structure group — are closed forms whose de Rham cohomology class does not depend on the connection at all. Wiggle the connection and the characteristic form shifts only by an exact piece, which integrates to zero over any closed cycle. So the integral is a topological constant: an integer, computed with calculus, that counts something rigid about how the bundle is twisted. This article develops the machinery from the curvature 2-form and the Bianchi identity through the Chern–Weil homomorphism to the named classes — Chern, Pontryagin, Euler — and their integer characteristic numbers. Those integers are physics: the first Chern number quantizes magnetic charge, the second counts Yang–Mills instantons, the Euler number is Gauss–Bonnet. The prerequisites are differential forms, the exterior derivative, and the covariant derivative.

Key takeaways

A connection on a bundle has a curvature 2-form $\Omega=\mathrm{d}\omega+\omega\wedge\omega$ , valued in the Lie algebra of the structure group; it measures the failure of parallel transport to be path-independent.
The Chern–Weil homomorphism sends any $\mathrm{Ad}$ -invariant polynomial $P$ of curvature to a closed form $P(\Omega)$ whose de Rham class is independent of the chosen connection.
Those classes are the characteristic classes — Chern classes for complex bundles, Pontryagin and Euler classes for real ones — topological invariants of the bundle, not of any metric or connection.
Integrating characteristic forms over closed manifolds gives integers: Gauss–Bonnet ( $\int_{S^2}K\,\mathrm{d}A=4\pi$ , $\chi=2$ ), monopole charge, and instanton number are all instances.
In physics the curvature is the field strength — $F=\mathrm{d}A+A\wedge A$ for gauge theory, the Riemann tensor for gravity — so charge quantization, anomalies, and topological terms all flow from this one construction.
Key idea: invariant polynomials of curvature are closed forms whose de Rham class is connection-independent (Chern–Weil), so curvature — a local geometric quantity — computes global topological invariants.

Where this lives today

The abstract integers of Chern–Weil are now laboratory quantities. In condensed matter the integer quantum Hall conductance $\sigma_{xy}=\nu\,e^2/h$ is a first Chern number — the TKNN invariant — got by integrating the Berry curvature of the filled bands over the Brillouin-zone torus, and its topological protection is why the plateaus are quantized to nine figures. Topological and Chern insulators classify materials by these bundle invariants, and a Chern number of one has been measured in cold-atom and photonic lattices. In high-energy physics the second Chern number is the instanton number governing QCD's $\theta$ -vacuum, the strong-CP problem, and the axion; the same density is the chiral anomaly $\partial_\mu j^\mu_5\propto\mathrm{tr}(F\tilde F)$ that fixes the $\pi^0\to\gamma\gamma$ rate and constrains anomaly cancellation in the Standard Model. And the Atiyah–Singer index theorem, index $=\int_M\hat A(TM)\,\mathrm{ch}(E)$ , ties fermion zero modes to these integrals — live in string theory, lattice gauge theory, and the Majorana modes of quantum computing.

The physical picture

Curvature is not a number attached to a point but an instruction attached to a loop: transport a vector once around an infinitesimal parallelogram and read off how it returns rotated. Shrink the loop and the rotation shrinks with its area; the ratio, in the limit, is the curvature. The classic case is the geodesic octant of the unit sphere — a triangle with three right angles, around which a parallel-transported vector rotates by a full $90^\circ$ , precisely the area it encloses.

Figure

Parallel transport around a geodesic triangle on the unit sphere: a vector carried N → A → B → N returns rotated by 90° = π/2 rad, exactly the enclosed area 4π/8 = π/2 — curvature is holonomy per unit area

Curvature is local and connection-dependent; topology is global and rigid. Chern–Weil turns one into the other. Feed the curvature into a gauge-invariant polynomial — a trace, determinant, or Pfaffian — and deforming the connection shifts the answer only by an exact form, contributing nothing to any integral over a closed manifold. What survives is a de Rham class, and paired with a cycle it gives an integer — in three physical guises: the flux of a monopole, the instanton number of a Yang–Mills field, and the Euler characteristic of a surface.

Figure

The Chern–Weil bridge: a connection ω has curvature Ω = dω + ω∧ω, an invariant polynomial P(Ω) is closed, and its integral over a closed cycle is an integer — for the sphere 1 ⁄ 2π ∫ K dA = χ = 2

The core equations

Cartan's second structure equation defines the curvature of a connection $\omega$ as a matrix of 2-forms, purely from $\omega$ and the exterior derivative — no metric, no coordinates.

\Omega = \mathrm{d}\omega + \omega\wedge\omega \qquad (1)

where: $\Omega$ — curvature 2-form, valued in the Lie algebra $\mathfrak{g}$ ; $\omega$ — connection 1-form, also $\mathfrak{g}$ -valued; $\wedge$ — wedge combined with matrix multiplication, so $(\omega\wedge\omega)^a{}_b=\omega^a{}_c\wedge\omega^c{}_b$ . Reading: a matrix-valued $\omega$ has $\omega\wedge\omega\ne0$ because the fibre matrices do not commute. The object $\Omega$ wears two faces: in gravity it packages the Riemann tensor; in gauge theory it is the Yang–Mills field strength, the connection being the gauge potential.

\Omega^a{}_b = \tfrac12 R^a{}_{b\mu\nu}\,\mathrm{d}x^\mu\wedge\mathrm{d}x^\nu,\qquad \Omega = F = \mathrm{d}A + A\wedge A \qquad (2)

where: $R^a{}_{b\mu\nu}$ — Riemann tensor of the Levi-Civita connection; $F$ — non-abelian field strength; $A$ — gauge potential, with $A\wedge A=\tfrac12[A,A]$ . Reading: for $U(1)$ the bracket vanishes and $F=\mathrm{d}A$ , ordinary electromagnetism. Differentiating the structure equation forces an identity: the exterior covariant derivative of the curvature vanishes — the root of $\mathrm{d}F=0$ in electromagnetism.

D\Omega \equiv \mathrm{d}\Omega + [\omega,\Omega] = 0 \qquad (3)

where: $D$ — exterior covariant derivative; $[\omega,\Omega]=\omega\wedge\Omega-\Omega\wedge\omega$ . Reading: a differential constraint, not an equation of motion — it holds for every connection. Under a gauge transformation the connection transforms inhomogeneously, but the curvature transforms by conjugation — the fact that makes invariant polynomials of $\Omega$ globally well defined.

\omega \longmapsto g\,\omega\,g^{-1} + g\,\mathrm{d}g^{-1},\qquad \Omega \longmapsto g\,\Omega\,g^{-1} \qquad (4)

where: $g$ — local gauge transformation; $g\,\mathrm{d}g^{-1}$ — the inhomogeneous term that stops $\omega$ from being a tensor. Reading: that piece cancels in $\Omega$ , so $\Omega$ is tensorial and $\mathrm{tr}(\Omega^k)$ is gauge invariant. Now the Chern–Weil statement: any $\mathrm{Ad}$ -invariant symmetric polynomial of the curvature is closed, so it represents a de Rham cohomology class.

\mathrm{d}\,P(\Omega)=0,\qquad P(\Omega)=\mathrm{tr}(\Omega^k)\ \text{or}\ \det(\cdots) \qquad (5)

where: $P$ — $\mathrm{Ad}$ -invariant symmetric polynomial on $\mathfrak{g}$ . Reading: closedness plus connection-independence (both proved below) promotes $P(\Omega)$ from a geometric form to a topological invariant. The workhorse invariant is the total Chern class of a complex bundle, a determinant whose degree-by-degree expansion gives the individual Chern classes.

c(E)=\det\!\left(\mathbb{1}+\frac{\mathrm{i}}{2\pi}\Omega\right)=1+c_1+c_2+\cdots,\qquad c_1=\frac{\mathrm{i}}{2\pi}\,\mathrm{tr}\,\Omega \qquad (6)

where: $E$ — complex bundle; $c_k$ — $k$ -th Chern class, a $2k$ -form in $H^{2k}(M;\mathbb{Z})$ . Reading: $c_2=\frac{1}{8\pi^2}\big[\mathrm{tr}(\Omega\wedge\Omega)-\mathrm{tr}\,\Omega\wedge\mathrm{tr}\,\Omega\big]$ , which for $SU(n)$ becomes $\frac{1}{8\pi^2}\mathrm{tr}(\Omega\wedge\Omega)$ . The Chern character repackages the same data additively over sums and multiplicatively over tensor products — why it, not $c(E)$ , appears in index theorems.

\mathrm{ch}(E)=\mathrm{tr}\,\exp\!\left(\frac{\mathrm{i}}{2\pi}\Omega\right)=\mathrm{rk}\,E + c_1 + \tfrac12\!\left(c_1^2-2c_2\right)+\cdots \qquad (7)

where: $\mathrm{ch}(E)$ — Chern character; $\mathrm{rk}\,E$ — rank. Reading: $\mathrm{ch}(E\oplus F)=\mathrm{ch}(E)+\mathrm{ch}(F)$ and $\mathrm{ch}(E\otimes F)=\mathrm{ch}(E)\,\mathrm{ch}(F)$ , from $\exp$ of a block-diagonal or tensor curvature. For real oriented bundles of even rank the top invariant is the square root of the determinant, the Pfaffian; it defines the Euler class and delivers Gauss–Bonnet.

e(TM)=\frac{1}{(2\pi)^{m}}\,\mathrm{Pf}(\Omega),\qquad \dim M=2m,\qquad \mathrm{Pf}(\Omega)^2=\det\Omega \qquad (8)

where: $\mathrm{Pf}$ — Pfaffian of the antisymmetric $SO(2m)$ curvature matrix; $TM$ — tangent bundle. Reading: the Pfaffian exists only for even-size antisymmetric matrices, so the Euler class lives on even-dimensional oriented manifolds. The remaining real invariants are the Pontryagin classes — the even-degree determinant terms, the real-bundle analogue of the Chern classes.

p_k(E)=\Big[\det\!\Big(\mathbb{1}+\tfrac{1}{2\pi}\Omega\Big)\Big]_{4k},\qquad p_1=-\frac{1}{8\pi^2}\,\mathrm{tr}(\Omega\wedge\Omega) \qquad (9)

where: $p_k$ — $k$ -th Pontryagin class, a $4k$ -form. Reading: odd-degree terms vanish because a real antisymmetric $\Omega$ has $\mathrm{tr}(\Omega^{\text{odd}})=0$ , leaving only degrees $4k$ .

Derivation AThe curvature form, the wedge that survives, and the Bianchi identity

The goal here is to establish the three algebraic facts everything downstream rests on: that a matrix-valued 1-form has a nonvanishing self-wedge equal to half its graded commutator, that the curvature obeys the Bianchi identity, and that this constraint is the exterior-covariant statement $D\Omega=0$ . All follow from the graded Leibniz rule and $\mathrm{d}^2=0$ , so they hold for any connection on any bundle.

Step 1. Take $\omega=\omega^A T_A$ with $T_A$ a Lie-algebra basis. Then $\omega\wedge\omega=\omega^A\wedge\omega^B\,T_A T_B$ , and antisymmetrizing the form factors picks out the commutator: $\omega\wedge\omega=\tfrac12\,\omega^A\wedge\omega^B\,[T_A,T_B]=\tfrac12[\omega,\omega]$ . Why: the scalar wedge would vanish, but the non-commuting matrices survive antisymmetrization, so $\omega\wedge\omega\ne0$ non-abelian and $=0$ for $U(1)$ .

Step 2. Differentiate the structure equation with the graded Leibniz rule: $\mathrm{d}\Omega=\mathrm{d}(\mathrm{d}\omega+\omega\wedge\omega)=\mathrm{d}\omega\wedge\omega-\omega\wedge\mathrm{d}\omega$ . Why: $\mathrm{d}^2\omega=0$ kills the first term, and the minus sign is the whole content of the Leibniz rule for an odd-degree form.

Step 3. Replace $\mathrm{d}\omega=\Omega-\omega\wedge\omega$ and expand: $\mathrm{d}\Omega=(\Omega-\omega\wedge\omega)\wedge\omega-\omega\wedge(\Omega-\omega\wedge\omega)=\Omega\wedge\omega-\omega\wedge\Omega$ , the triple- $\omega$ terms cancelling. Why: this exhibits $\mathrm{d}\Omega$ as a pure commutator of $\omega$ against $\Omega$ .

Step 4. Hence $\mathrm{d}\Omega+[\omega,\Omega]=0$ with $[\omega,\Omega]=\omega\wedge\Omega-\Omega\wedge\omega$ , which is $D\Omega=0$ . Why: $\mathrm{d}+[\omega,\,\cdot\,]$ is the exterior covariant derivative, so the curvature is covariantly closed — the abstract source of $\mathrm{d}F=0$ and $\nabla_{[\mu}R_{\nu\rho]}{}^{ab}=0$ .

Go deeper · Derivation B · 1 minChern–Weil — closed forms whose class forgets the connection

Derivation BChern–Weil — closed forms whose class forgets the connection

The aim is the two-part theorem that makes characteristic classes possible: any invariant polynomial of the curvature is closed, and its de Rham class is the same for every connection on the bundle. The engine of both parts is the infinitesimal form of $\mathrm{Ad}$ -invariance combined with Bianchi; connection-independence follows from an interpolation whose $t$ -derivative is exact — the transgression, or Chern–Simons, form.

Step 1. Differentiate $\mathrm{Ad}$ -invariance $P(gXg^{-1},\dots)=P(X,\dots)$ at the identity in a direction $Y$ to get $\sum_i P(X_1,\dots,[Y,X_i],\dots,X_k)=0$ , valid for all $Y$ . Why: this identity is what lets bracket terms cancel in pairs when we apply $\mathrm{d}$ .

Step 2. Apply $\mathrm{d}$ to $P(\Omega)=P(\Omega,\dots,\Omega)$ . By symmetry and the Leibniz rule (all $\Omega$ even-degree, no signs) one gets $\mathrm{d}\,P(\Omega)=k\,P(\mathrm{d}\Omega,\Omega,\dots,\Omega)$ . Why: only one factor need be differentiated, the $k$ symmetric slots contributing equally.

Step 3. Substitute Bianchi, $\mathrm{d}\Omega=-[\omega,\Omega]$ : then $\mathrm{d}\,P(\Omega)=-k\,P([\omega,\Omega],\Omega,\dots)$ , and with $Y=\omega$ , $X_i=\Omega$ the Step 1 identity makes this vanish term by term, so $\mathrm{d}\,P(\Omega)=0$ . Why: closedness is the marriage of Bianchi to the invariance of $P$ — neither alone suffices.

Step 4. For connection-independence, interpolate $\omega_t=\omega_0+t\,\eta$ with $\eta=\omega_1-\omega_0$ ; the difference of two connections is a tensorial $\mathfrak{g}$ -valued 1-form (the $g\,\mathrm{d}g^{-1}$ terms cancel), and $\Omega_t=\mathrm{d}\omega_t+\omega_t\wedge\omega_t$ has $\dot\Omega_t=D_t\eta$ . Why: a straight-line path in the affine space of connections is legitimate, and its tangent $\eta$ is tensorial.

Step 5. Differentiate in $t$ and use invariance again: $\frac{\mathrm{d}}{\mathrm{d}t}P(\Omega_t)=k\,P(D_t\eta,\Omega_t,\dots)=k\,\mathrm{d}\,P(\eta,\Omega_t,\dots,\Omega_t)$ , an exact form for each $t$ . Why: the $t$ -derivative is not merely closed but manifestly $\mathrm{d}$ of something.

Step 6. Integrate: $P(\Omega_1)-P(\Omega_0)=\mathrm{d}\Big(k\!\int_0^1 P(\eta,\Omega_t,\dots,\Omega_t)\,\mathrm{d}t\Big)$ , so $[P(\Omega_0)]=[P(\Omega_1)]$ in $H^{2k}_{\mathrm{dR}}(M)$ . Why: that bracketed integral is the transgression (Chern–Simons) form, and its existence is why the integral over a closed manifold is invariant.

Go deeper · Derivation C · 1 minGauss–Bonnet as the two-dimensional Euler class

Derivation CGauss–Bonnet as the two-dimensional Euler class

This specializes the Euler-class formula to a closed oriented surface and recovers the classical Gauss–Bonnet theorem, turning an abstract Pfaffian into the statement that the total Gaussian curvature equals $2\pi$ times the Euler characteristic. It is the cleanest illustration of the whole program: a pointwise-computable curvature integral equals an integer fixed by topology alone, blind to how the surface is bent.

Step 1. On a Riemannian 2-manifold the structure group is $SO(2)$ and the curvature is one antisymmetric $2\times2$ matrix of 2-forms with a single entry $\Omega^1{}_2=R^1{}_{212}\,\mathrm{d}x^1\wedge\mathrm{d}x^2=K\,\mathrm{d}A$ , $K$ the Gaussian curvature and $\mathrm{d}A$ the area form. Why: in 2D the whole Riemann tensor is carried by $K$ , so the matrix of forms degenerates to a scalar times $\mathrm{d}A$ .

Step 2. Take the Pfaffian: for $\begin{pmatrix}0&a\\-a&0\end{pmatrix}$ , $\mathrm{Pf}=a$ , so $\mathrm{Pf}(\Omega)=\Omega^1{}_2=K\,\mathrm{d}A$ . Why: the Pfaffian is the square root of the determinant that stays a 2-form, not the 4-form $\det\Omega$ would give.

Step 3. Insert into the Euler-class formula with $m=1$ , $(2\pi)^m=2\pi$ : $e(TM)=\frac{1}{2\pi}K\,\mathrm{d}A$ . Why: the $1/2\pi$ is exactly what makes the integral come out an integer.

Step 4. Integrate over the closed surface and invoke Chern–Weil plus Poincaré–Hopf: $\frac{1}{2\pi}\int_M K\,\mathrm{d}A=\int_M e(TM)=\chi(M)$ . For the unit sphere $K=1$ , area $=4\pi$ , so $\int_{S^2}K\,\mathrm{d}A=4\pi$ and $\frac{1}{2\pi}(4\pi)=2=\chi(S^2)$ ; a genus- $g$ surface has $\chi=2-2g$ . Why: the integer on the right cannot change as the metric is deformed, so a dented sphere still integrates to $4\pi$ — curvature redistributes but its total is locked. *The key numbers: $\int_{S^2}K\,\mathrm{d}A=4\pi$ and $\chi(S^2)=2$ ; a genus-2 surface gives $\chi=-2$ and $\int K\,\mathrm{d}A=-4\pi$ .*

Limiting cases

How the general machine specializes to the cases physics actually uses:

Case	Result	Meaning
Abelian / electromagnetism	$\Omega=\mathrm{i}F,\ \ F=\mathrm{d}A$	Group $U(1)$ ; connection is the potential $A$ , curvature the field strength. $c_1=[F/2\pi]$ integrates over $S^2$ to the integer monopole charge.
Flat connection	$\Omega=0\ \Rightarrow\ P(\Omega)=0,\ \deg P\ge1$	Zero curvature kills every characteristic form, so all real classes of a flat bundle vanish. Torsion classes and holonomy can still be nontrivial.
Riemannian / gravity	$\Omega^a{}_b=\tfrac12 R^a{}_{b\mu\nu}\,\mathrm{d}x^\mu\wedge\mathrm{d}x^\nu$	Group $SO(n)$ , Levi-Civita connection; $\Omega$ packages the Riemann tensor, giving Pontryagin classes and, in even dimension, the Euler class via the Pfaffian.
Self-dual (instanton)	$\Omega=\star\,\Omega$	In 4D Euclidean space self-dual Yang–Mills curvature minimizes the action at fixed $k=\frac{1}{8\pi^2}\int\mathrm{tr}(F\wedge F)\in\mathbb{Z}$ , the second Chern number.

Common trapTreating the curvature form itself, or its bare trace, as the topological invariant

The curvature $\Omega$ is not a topological invariant, and neither is $\mathrm{tr}(\Omega)$ before you have checked invariance and integrated. $\Omega$ depends on the connection and the trivialization — it transforms by conjugation, $\Omega\to g\Omega g^{-1}$ . What is invariant is only the de Rham cohomology class of an $\mathrm{Ad}$ -invariant $P(\Omega)$ : the form itself shifts by an exact (transgression) term under any change of connection, so only its integral over a closed cycle is fixed. Three traps follow. First, $\int P(\Omega)$ over a region with boundary is not topological — the exact piece contributes a boundary term, which is exactly how Chern–Simons theory lives on a boundary. Second, $\mathrm{tr}(\Omega)$ vanishes identically for an $SU(n)$ bundle, so the interesting invariant is the quadratic $\mathrm{tr}(\Omega\wedge\Omega)$ , not the linear trace. Third, a flat bundle has $\Omega=0$ yet can be topologically nontrivial through torsion and holonomy that real characteristic classes cannot see.

Worked examples

1 · The first Chern number of a Dirac monopole

A monopole of integer strength $n$ sits inside a two-sphere; its $U(1)$ field strength is $F=\frac{n}{2}\sin\theta\,\mathrm{d}\theta\wedge\mathrm{d}\varphi$ . Show the first Chern number $\frac{1}{2\pi}\int_{S^2}F$ equals $n$ , and explain why it must be an integer.

Step 1. Set up the integral with $\theta\in[0,\pi]$ , $\varphi\in[0,2\pi]$ :

\frac{1}{2\pi}\int_{S^2}F=\frac{1}{2\pi}\int_0^{2\pi}\!\!\int_0^{\pi}\frac{n}{2}\sin\theta\,\mathrm{d}\theta\,\mathrm{d}\varphi.

The factor $\frac{n}{2}$ is fixed by requiring the north and south patch potentials to differ by a single-valued gauge transformation on the equator.

Step 2. The elementary integrals are $\int_0^\pi\sin\theta\,\mathrm{d}\theta=[-\cos\theta]_0^\pi=2$ and $\int_0^{2\pi}\mathrm{d}\varphi=2\pi$ . Hence

\frac{1}{2\pi}\cdot\frac{n}{2}\cdot 2\cdot 2\pi=n.

So $\frac{1}{2\pi}\int_{S^2}F=\mathbf{n}$ : the total flux is $2\pi n$ and the first Chern number is the monopole strength.

Step 3. Why an integer? Because $c_1=[F/2\pi]\in H^2(S^2;\mathbb{Z})$ is an integral class — the pullback of the generator classifying complex line bundles. Physically, the wavefunction must be single-valued where the north and south patches overlap, and matching them around the equator forces $\frac{1}{2\pi}\int_{S^2}F\in\mathbb{Z}$ . This is Dirac's condition: with charges $e$ and $g$ , $eg=2\pi n$ , so one monopole quantizes all electric charge. No smooth deformation of $A$ can change $n$ .

2 · Euler numbers from total curvature

Use Gauss–Bonnet to find the total Gaussian curvature $\int_M K\,\mathrm{d}A$ for the unit sphere, the torus, and a genus-2 surface, checking each against the Euler characteristic.

Step 1. The tool: $\int_M K\,\mathrm{d}A=2\pi\,\chi(M)$ with $\chi=2-2g$ for a closed orientable surface of genus $g$ (Derivation 3, rearranged). The right side is a pure integer times $2\pi$ , whatever metric $M$ carries.

Step 2. Sphere ( $g=0$ , $\chi=2$ ): $\int_{S^2}K\,\mathrm{d}A=2\pi(2)=\mathbf{4\pi}$ . Cross-check: the unit sphere has $K=1$ and area $4\pi$ , so $\int K\,\mathrm{d}A=4\pi$ directly, and $\frac{1}{2\pi}(4\pi)=2=\chi$ .

Step 3. Torus ( $g=1$ , $\chi=0$ ): $\int_{T^2}K\,\mathrm{d}A=\mathbf{0}$ — the positive curvature of the outer rim exactly cancels the negative curvature of the inner throat. Genus-2 ( $g=2$ , $\chi=-2$ ): $\int K\,\mathrm{d}A=\mathbf{-4\pi}$ , net negative because a two-holed surface is saddle-like on balance.

Step 4. The number is topological, the curvature distribution geometric. Dent the sphere and $K$ redistributes wildly, but its total stays locked at $4\pi$ ; each handle drops it by exactly $4\pi$ . Total curvature counts holes: $\int_M K\,\mathrm{d}A=2\pi(2-2g)$ .

Physics of the portrait

Characteristic classes arrived in physics fully formed — the language of monopoles, instantons, anomalies, and topological phases is Chern's, invented decades before physicists needed it.

Origins · 1 minShiing-Shen Chern and how this was discovered

Shiing-Shen Chern

1911–2004 · the geometry of fibre bundles and characteristic classes

Shiing-Shen Chern is the mathematician who taught geometry to count. Born in Jiaxing, China, he studied under Wilhelm Blaschke in Hamburg and Élie Cartan in Paris, absorbing Cartan's calculus of moving frames and forms — the language in which curvature becomes a matrix of 2-forms. At the Institute for Advanced Study from 1943 he found an intrinsic proof of the generalized Gauss–Bonnet theorem, writing the Euler characteristic as an integral of a curvature polynomial with no ambient space. From that grew the 1946 construction of the Chern classes, and — with André Weil — the homomorphism sending any invariant polynomial of curvature to a characteristic class. Decades later physics caught up: Yang–Mills instantons carry his second Chern number, the quantum Hall effect is read through his first, and the Chern–Simons form (with James Simons, 1974) became the backbone of topological field theory. He shaped modern differential geometry more than any other single figure, and died in 2004.

History

1854

— Riemann founds the intrinsic geometry of curved manifolds in his habilitation lecture

1917

— Levi-Civita defines parallel transport, exposing curvature as the holonomy of a loop

1925

— Cartan writes connection and curvature as matrix-valued differential forms

1946

— Chern constructs the characteristic classes of complex vector bundles now bearing his name

1975

— the BPST instanton realises the second Chern number as a Yang–Mills solution

Going further

Texts. Nakahara, Geometry, Topology and Physics, 2nd ed., ch. 9–11 · Stone & Goldbart, Mathematics for Physics, differential geometry and topology · Kobayashi & Nomizu, Foundations of Differential Geometry, vol. II, ch. XII (Chern–Weil) · Milnor & Stasheff, Characteristic Classes · Eguchi, Gilkey & Hanson, Gravitation, Gauge Theories and Differential Geometry, Phys. Rep. 66 (1980) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed. The key results developed above are: the structure equation and Bianchi identity $\mathrm{d}\Omega+[\omega,\Omega]=0$ ; the gauge transformation laws $\Omega\to g\Omega g^{-1}$ and the inhomogeneous $\omega$ -law; the Chern–Weil closedness-plus-transgression argument; the determinant expansion giving $c_1, c_2$ , the Chern-character series, and the Euler/Pontryagin normalizations; and the anchor numbers $\int_{S^2}K\,\mathrm{d}A=4\pi$ , $\chi(S^2)=2$ , the monopole integral $\frac{1}{2\pi}\int_{S^2}F=n$ , and $k=\frac{1}{8\pi^2}\int\mathrm{tr}(F\wedge F)$ . Conventions: $\mathrm{i}/2\pi$ for Chern classes, $1/2\pi$ for Pontryagin and Euler classes.

Builds on: Differential Forms · Exterior Calculus and the Exterior Derivative · The Covariant Derivative · Christoffel Symbols and Affine Connections · Manifolds and Tangent Spaces · Tensor Contraction and the Trace Leads to: Hodge Star and the Codifferential · Parallel Transport and Geodesics

Ask the tutor: Ask me to work any characteristic number by hand — the monopole's first Chern integral, an instanton's $k$ , or the Euler number of your favourite surface.

Why does the term

\omega\wedge\omega

survive when a scalar 1-form wedged with itself is zero?

Because $\omega$ is matrix-valued, $\omega=\omega^A T_A$ with generators $T_A$ . The wedge antisymmetrizes the form parts (which alone would vanish) while multiplying the non-commuting matrices, and the antisymmetrization projects onto their commutator: $\omega\wedge\omega=\tfrac12[\omega,\omega]=\tfrac12\omega^A\wedge\omega^B[T_A,T_B]$ , generically nonzero for a non-abelian group. Only for an abelian group like $U(1)$ , where the generators commute, does it vanish and the curvature collapse to $F=\mathrm{d}A$ . This surviving term is exactly what makes Yang–Mills theory nonlinear.

What precisely is connection-independent — the curvature, the form, or the class?

Only the de Rham cohomology class. The curvature $\Omega$ depends on the connection and transforms by conjugation. The form $P(\Omega)$ also changes when you change the connection — but by an exact form, the transgression (Chern–Simons) form. Since an exact form integrates to zero over any closed manifold, the class $[P(\Omega)]$ , and the number $\int_C P(\Omega)$ over any closed cycle, is fixed. That is the whole logic of the subject: geometry (the form) is fluid, topology (the class) is rigid, and the gap between them is always exact.

Is the curvature 2-form the same thing as the Riemann tensor?

The same information in different packaging. The Riemann tensor $R^a{}_{b\mu\nu}$ has four indices; the curvature 2-form bundles the last two antisymmetric coordinate indices into a 2-form, $\Omega^a{}_b=\tfrac12 R^a{}_{b\mu\nu}\,\mathrm{d}x^\mu\wedge\mathrm{d}x^\nu$ , leaving a matrix in the frame indices. The 2-form language is more general: it applies verbatim to gauge bundles, where there is no metric and $\Omega=F$ is the Yang–Mills field strength. The payoff is uniformity — gravity and gauge theory obey the same structure equation and Bianchi identity in this language.

Where does the factor

\mathrm{i}/2\pi

in the Chern class come from?

It is the normalization that forces the integrals to be integers. The $2\pi$ matches the period of the $U(1)$ phase — the same $2\pi$ in $\oint\mathrm{d}\theta=2\pi$ — so a flux quantum contributes exactly $1$ to the first Chern number. The factor $\mathrm{i}$ makes the class real: the curvature of a Hermitian connection is anti-Hermitian, so $\mathrm{tr}\,\Omega$ is imaginary and $\mathrm{i}\,\mathrm{tr}\,\Omega$ real. Any other constant would make the Chern "numbers" real multiples of integers, useless as labels; $\mathrm{i}/2\pi$ is fixed uniquely by integrality.

How can a smooth integral over a manifold be forced to be a whole number?

Because the integrand is secretly the pullback of an integral cohomology class, and its pairing with a fundamental cycle lands in $\mathbb{Z}$ . A complex line bundle over $S^2$ is glued from two patches, and the transition function winds an integer number of times around $U(1)$ on the equatorial overlap; that winding number is $\frac{1}{2\pi}\int_{S^2}F$ . The integral can only take the values the gluing allows, and the gluing is fixed by an integer. This is why the Hall conductance is quantized to nine digits and magnetic charge is discrete — the integral cannot be a non-integer, however you deform the fields.

What is the difference between a characteristic class and a characteristic number?

A characteristic class is a cohomology class — a differential form up to exact pieces, carrying local information at every point. A characteristic number is what you get by integrating a top-degree class over a closed manifold: a single integer. So $c_1$ is a class, a 2-form; $\frac{1}{2\pi}\int_{S^2}c_1=n$ is a number. Classes can be pulled back, multiplied, combined; numbers are the final invariants you measure — monopole charge, instanton number, Euler characteristic. Chern–Weil manufactures the classes; integration extracts the numbers.