A bundle is a way of attaching a vector space or other structured fibre to every point of a space while allowing the attachment to twist globally. Characteristic classes measure part of that twisting, and global analysis asks what differential operators can reveal about the resulting geometry and topology.
Locally, a vector bundle looks simple: over a small open set it is just a product U×Rᵏ or U×Cᵏ. Globally, those local products can be glued together in a way that prevents one universal frame from existing. The Möbius strip is the elementary picture. Every sufficiently short segment of its base circle carries what looks like an ordinary line, yet going once around reverses the local direction.
This guide develops that local-to-global problem systematically. We begin with transition functions and sections, then build new bundles from old ones, examine clutching constructions and principal bundles, and introduce characteristic classes as cohomological invariants. Connections and curvature then produce characteristic differential forms. Finally, Hodge theory and elliptic operators show how analysis can recover global topological information.
The prerequisites are the earlier BTT differential-geometry guides: Smooth Manifolds and Maps, Tangent, Cotangent and Tensor Fields, Differential Forms and Integration, and Connections and Curvature. For the cohomological language, use Homology and Cohomology.
Level: advanced undergraduate to beginning graduate enrichment. This article is a learning route, not a claim that every theorem in bundle theory or global analysis is proved here. The BTT Mathematics Hub remains the main subject entrance.
Reading route: vector bundles → transition functions → sections → bundle operations → clutching and examples → principal bundles → characteristic classes → curvature and Chern–Weil → Hodge and elliptic theory → index ideas → practice → solutions.
1. A vector bundle is locally a product but globally may not be one
A rank-k real vector bundle over a space M consists of a total space E, a projection π:E→M and a k-dimensional real vector-space structure on each fibre Eₚ=π⁻¹(p). Around every point p there is an open neighbourhood U and a homeomorphism or diffeomorphism, depending on category, Φ:π⁻¹(U)→U×Rᵏ that is linear on each fibre.
Such a map is a local trivialisation. It says that a bundle is indistinguishable from a product when viewed through a sufficiently small window. The global question is whether all those windows can be made compatible with one product M×Rᵏ.
A bundle is trivial when it is globally isomorphic to M×Rᵏ as a vector bundle. Triviality is stronger than saying every fibre has the same dimension. Fibres always have the same dimension on each connected component of a rank-k bundle; the issue is how those fibres are glued.
For complex vector bundles, replace Rᵏ by Cᵏ and require complex-linear fibre maps. Real and complex bundle theories are related but not interchangeable. Their characteristic classes live in different degrees and coefficient systems.
Ralph Cohen’s Bundles, Manifolds, and Homotopy notes develop vector bundles, characteristic classes and their role in manifold topology. Chris Wendl’s Bundles and Connections notes provide the differential-geometric bridge from bundles to connections and curvature.
2. The tangent and cotangent bundles are the first geometric examples
For a smooth n-manifold M, the tangent bundle TM is the disjoint union of all tangent spaces TₚM, with a smooth structure built from derivatives of coordinate transitions. Its rank is n.
The cotangent bundle T*M is the dual bundle whose fibre at p is Tₚ*M. Coordinate changes act on tangent and cotangent components in complementary ways, exactly as developed in the tensor-fields guide.
The tangent bundle need not be trivial. Triviality would mean there are n smooth vector fields that form a basis of TₚM at every point. A manifold with trivial tangent bundle is called parallelisable.
The circle S¹ is parallelisable: its unit tangent vector field gives a global frame. The 2-sphere S² is not parallelisable; in fact it has no nowhere-zero continuous tangent vector field. This is the classical hairy-ball phenomenon and will later be connected to the Euler class.
Not every sphere behaves the same way. S¹, S³ and S⁷ are parallelisable, while most spheres are not. A local coordinate basis therefore should never be assumed to extend globally without a proof.
3. Transition functions store the gluing data
Choose local trivialisations over open sets Uᵢ. On an overlap Uᵢ∩Uⱼ, the two fibre-coordinate descriptions differ by an invertible linear map gᵢⱼ(p). For a real rank-k bundle, gᵢⱼ:Uᵢ∩Uⱼ→GL(k,R); for a complex bundle, the target is GL(k,C).
The transition maps obey three consistency rules: gᵢᵢ=I, gᵢⱼ=gⱼᵢ⁻¹ and on triple overlaps gᵢⱼgⱼₖgₖᵢ=I. This last equation is the cocycle condition.
Conversely, compatible transition functions can be used to build a bundle by gluing the local products Uᵢ×Rᵏ. The bundle is the equivalence class of the resulting geometric object, not one particular collection of transition matrices.
Changing local frames replaces the transition functions by equivalent data. If hᵢ:Uᵢ→GL(k) changes the frame on Uᵢ, then the new transitions are g′ᵢⱼ=hᵢgᵢⱼhⱼ⁻¹, with convention depending on whether frame vectors are written as rows or columns. The point is that a local frame change does not change the underlying bundle.
This is the same local-versus-intrinsic pattern seen with tensor components and connection matrices. Useful coordinates change; the geometric object survives.
4. Worked example: the Möbius line bundle
View S¹ as [0,1] with endpoints identified. Start with [0,1]×R and identify (0,v) with (1,−v). The quotient is the Möbius line bundle over S¹.
Locally, away from the identification seam, this looks exactly like an ordinary strip. The twist appears only when a fibre vector is followed once around the circle: a local positive direction returns as its negative.
Suppose there were a nowhere-zero continuous section s. In the interval picture, write its fibre coordinate as f(t), with f(t)≠0. The identification requires f(1)=−f(0). By the intermediate value theorem, a continuous real function changing sign must vanish somewhere. Contradiction.
A real line bundle is trivial exactly when it has a nowhere-zero global section. Therefore the Möbius line bundle is nontrivial. This elementary argument is a prototype for obstruction theory: a global object fails to exist because the gluing imposes incompatible conditions.
5. A section is a global choice inside the fibres
A section of π:E→M is a map s:M→E satisfying π∘s=id_M. It chooses one vector s(p) from every fibre.
The zero section always exists for a vector bundle. The interesting question is whether one can choose nonzero vectors continuously or smoothly, and how many independent choices can be made.
A rank-k vector bundle is trivial if and only if it admits k global sections s₁,…,sₖ that are linearly independent in every fibre. Such a collection is a global frame.
One nowhere-zero section of a rank-k bundle only splits off one trivial line under suitable metric or complement choices; it does not by itself trivialise the remaining rank k−1 part. The line-bundle case is special because there is no remaining factor.
This distinction prevents a frequent overclaim. Showing that one tangent vector field never vanishes on a 3-manifold does not yet produce three independent vector fields. Parallelisability needs a full frame.
6. Zeros of sections carry geometric information
A section can be viewed as a geometric field whose zero set records where the chosen fibre vector disappears. If the section is transverse to the zero section, its zero set is a submanifold of the expected codimension equal to the bundle rank.
For an oriented rank-n vector bundle E over an oriented n-manifold, an appropriate signed count of isolated zeros of a transverse section evaluates the Euler class e(E) on the fundamental class. For E=TM this becomes the Poincaré–Hopf relationship with the Euler characteristic.
A nowhere-zero section of an oriented rank-n bundle forces the Euler class to vanish, because there are no zeros to represent its obstruction. The converse requires additional hypotheses and should not be asserted in arbitrary rank and base dimension.
On S², χ(S²)=2. Hence a tangent vector field cannot be nowhere zero. This turns the hairy-ball theorem from a picture about combing hair into a global statement about the tangent bundle.
7. Bundle operations build new geometric objects
If E and F are vector bundles over the same base, their direct sum E⊕F has fibre Eₚ⊕Fₚ. Tensor products, duals, exterior powers and symmetric powers are defined fibrewise and glued using the induced transition functions.
The dual bundle E* has transition functions given by the dual inverse action. The endomorphism bundle End(E)=E*⊗E has fibres of linear maps Eₚ→Eₚ. Connection curvature naturally takes values in End(E).
For a smooth map f:N→M and bundle E→M, the pullback bundle f*E→N has fibre (f*E)ₙ=E_{f(n)}. It lets geometric data on M be studied along N.
Pullback respects many constructions: f*(E⊕F)≈f*E⊕f*F, and characteristic classes are natural under pullback. This naturality is one of the defining reasons characteristic classes are useful.
Subbundles and quotient bundles require constant-rank behaviour. A smoothly varying family of subspaces gives a subbundle only when local triviality can be established. A kernel of a bundle map forms a vector subbundle when the bundle map has constant rank; rank jumps can produce singular behaviour instead.
8. The normal bundle records directions transverse to a submanifold
If N is an embedded submanifold of M, its tangent bundle TN sits inside TM|_N. The quotient ν(N)=TM|_N/TN is the normal bundle.
When M carries a Riemannian metric, the orthogonal complement of TN gives a concrete bundle isomorphic to this quotient. The quotient definition, however, does not require a metric.
A tubular neighbourhood theorem identifies a neighbourhood of the zero section of the normal bundle with a neighbourhood of N in M. Thus the normal bundle packages first-order information about how the submanifold sits inside the ambient manifold.
For a circle embedded in the plane, the normal line bundle is trivial. For more complicated embeddings and nonorientable situations, normal bundles can carry essential twisting information.
9. Clutching turns an equator map into a bundle over a sphere
Split Sⁿ into northern and southern hemispheres. Each hemisphere is contractible, so every vector bundle is trivial over it. The entire bundle is determined by how the two trivial bundles are identified along the equator Sⁿ⁻¹.
That identification is a clutching map g:Sⁿ⁻¹→GL(k,F), where F is R or C. Homotopic clutching maps produce isomorphic bundles under the standard classification setup.
For complex line bundles over S², the clutching map is a map S¹→C* that can be deformed to S¹→U(1). Its winding number classifies the bundle. With a fixed standard orientation and convention, the integer also evaluates the first Chern class on [S²].
The trivial bundle corresponds to winding number zero. A one-turn clutching map produces a basic nontrivial line bundle. Reversing the winding reverses the first Chern number.
This is a powerful example because a global bundle class has been converted into a homotopy class of a transition function on a lower-dimensional overlap.
10. Pulling a bundle back can simplify or reveal it
Let E→M be a bundle and f:N→M a map. Even if E is nontrivial on M, f*E may become trivial on N if the map misses the obstruction or if N has simpler topology.
For example, pull the Möbius line bundle on S¹ back along the double-cover map S¹→S¹, z↦z². Going once around the new base goes twice around the original circle, so the sign reversal occurs twice and cancels. The pullback is trivial.
Characteristic classes explain this algebraically. The first Stiefel–Whitney class of a real line bundle lives in H¹(−;Z/2). Pullback along the degree-two map multiplies the generator by two, which is zero mod 2.
Geometric reasoning and cohomological naturality agree. A useful invariant should survive changes of description while transforming predictably under maps.
11. Principal bundles store frames and symmetry directly
A principal G-bundle P→M has fibre modelled on a Lie group G acting freely and transitively on each fibre. Unlike a vector bundle, a principal fibre has no preferred zero vector; it is a torsor for G.
The frame bundle Fr(E) of a rank-k vector bundle is a principal GL(k)-bundle. Its fibre over p consists of ordered bases of Eₚ. Choosing a section of Fr(E) is exactly choosing a global frame of E.
If E has a metric, the orthonormal frame bundle reduces the structure group from GL(k,R) to O(k). If the bundle is oriented, one can reduce to SO(k). Additional structures correspond to further reductions when appropriate.
Given a principal G-bundle P and a representation ρ:G→GL(V), the associated vector bundle is P×_G V. This construction turns symmetry data into fibrewise linear geometry.
Connections can be formulated naturally on principal bundles and then induce connections on all associated bundles. This viewpoint is central in gauge theory and differential geometry.
12. Local trivialisations are not global gauge choices
In a principal or associated vector bundle, a local frame is sometimes called a local gauge. On overlaps, gauges are related by transition functions.
A connection matrix A depends on the gauge and transforms with the inhomogeneous derivative term described in the preceding connection article. The curvature F transforms homogeneously.
A nontrivial bundle may have no global gauge at all. Therefore a formula written in one local frame should not be extrapolated globally unless the overlaps have been checked.
This is another version of the Atlas’s recurring principle: coordinate or frame data are local interfaces; geometric claims must survive the transition rules.
13. Characteristic classes assign cohomology to bundles
A characteristic class assigns to each bundle E→M a cohomology class c(E) in a way compatible with pullback: c(f*E)=f*c(E). It should also be invariant under bundle isomorphism.
These classes do not classify every vector bundle by themselves in every setting. They are structured invariants: powerful detectors of twisting and obstructions, but not automatically complete fingerprints.
Major families include Stiefel–Whitney classes for real vector bundles with Z/2 coefficients, Chern classes for complex vector bundles with integer coefficients, Pontryagin classes for real vector bundles in degrees divisible by four, and the Euler class for oriented real bundles of rank equal to its degree.
Cohen’s characteristic-classes chapter develops these as cohomological invariants of vector and principal bundles and places them in the broader topology of manifolds.
14. Stiefel–Whitney classes detect real-bundle phenomena
For a real rank-k bundle E, the Stiefel–Whitney classes are wᵢ(E)∈Hⁱ(M;Z/2), with w₀=1 and wᵢ=0 for i>k. The total class is w(E)=1+w₁+w₂+⋯.
The first class w₁(E) is the obstruction to orientability of the bundle. A real vector bundle is orientable exactly when w₁(E)=0.
For the tangent bundle, w₁(TM)=0 precisely when the manifold is orientable. The Möbius line bundle has nonzero w₁, recording its sign reversal around the base circle.
The Whitney sum formula says w(E⊕F)=w(E)∪w(F). This converts direct-sum geometry into cup-product algebra.
Because coefficients are mod 2, sign information disappears. This makes Stiefel–Whitney theory available without orienting the bundle, but also distinguishes it from integer-valued characteristic classes.
15. The Euler class measures an oriented top-degree obstruction
For an oriented real rank-n bundle E→M, the Euler class e(E) lies in Hⁿ(M;Z). Reversing the orientation reverses the sign of e(E).
For a transverse section, the zero set represents the Poincaré dual of the Euler class under standard compactness and orientation hypotheses. When base and rank dimensions agree, isolated zeros can be counted with signs.
For the tangent bundle of a closed oriented n-manifold, evaluating e(TM) on the fundamental class gives the Euler characteristic. This is the cohomological version of the Poincaré–Hopf theorem.
The 2-sphere has Euler characteristic 2, so e(TS²) evaluates to 2. A nowhere-zero tangent section would force that evaluation to vanish, hence no such field exists.
The torus has Euler characteristic zero and does admit a nowhere-zero tangent field; in fact its tangent bundle is trivial. The zero Euler characteristic is necessary for a nowhere-zero tangent field on a closed oriented even-dimensional manifold, but one should not turn that example into an unrestricted converse for arbitrary bundles.
16. Chern classes organise complex-bundle twisting
For a complex rank-k bundle E, Chern classes cᵢ(E) lie in H²ⁱ(M;Z), with c₀=1 and cᵢ=0 for i>k. The total Chern class is c(E)=1+c₁+⋯+cₖ.
For a complex line bundle, only c₁ can be nonzero. Over S², complex line bundles are classified by an integer, and that integer is captured by the first Chern class evaluation once conventions are fixed.
The Whitney product formula says c(E⊕F)=c(E)∪c(F). For a line bundle L, c₁(L*)=−c₁(L), and c₁(L⊗K)=c₁(L)+c₁(K).
Chern classes provide obstructions to triviality: a trivial complex bundle has all positive-degree Chern classes zero. The converse is not valid without additional hypotheses; bundles can be nontrivial while these particular classes vanish.
This illustrates the proper role of an invariant. A detected nonzero class proves nontriviality. A zero value means that detector found no obstruction, not that every possible obstruction has disappeared.
17. Pontryagin classes remember real geometry through complexification
For a real vector bundle E, Pontryagin classes pᵢ(E)∈H⁴ⁱ(M;Z) can be defined from the even Chern classes of the complexified bundle E⊗C, with a standard sign convention.
They are characteristic classes of real bundles and are insensitive to orientation reversal in the way the Euler class is not. Pontryagin classes play major roles in manifold topology, signature formulas and obstruction theory.
The precise normalisations matter when translating to curvature formulas. A conceptual article should not mix conventions for factors of 2π, i or signs. When a numerical characteristic number is needed, use one source and state its convention explicitly.
Here the important distinction is structural: Stiefel–Whitney, Euler, Chern and Pontryagin classes answer different bundle questions and live in different coefficient systems and degrees.
18. Chern–Weil theory turns curvature into closed characteristic forms
Choose a connection on a vector or principal bundle and let F be its curvature. Invariant polynomials applied to F produce differential forms on the base. The Bianchi identity implies the resulting forms are closed.
Although the differential form itself depends on the chosen connection, its de Rham cohomology class is independent of that connection. Changing the connection changes the representative by an exact form.
For a complex line bundle with unitary connection, the curvature is an imaginary-valued 2-form. Under the standard normalisation, (i/2π)F represents the first Chern class in real de Rham cohomology.
Integrating that representative over a closed oriented surface gives an integer Chern number. Local curvature has therefore produced a global, quantised topological invariant after the correct normalisation and integration.
This does not mean “curvature is topology”. The connection can be changed and its pointwise curvature redistributed. The invariant polynomial’s cohomology class is the connection-independent object.
The progression from connections and curvature to Chern classes appears explicitly in the graduate bundle/geometry route on the UC San Diego Math 250B page.
19. Worked line-bundle curvature calculation
On a local trivialisation of a unitary complex line bundle, write the connection as A=i a with a a real 1-form. Because the bundle rank is one, A∧A=0 and F=dA=i da.
If on a coordinate patch a=(k/2)(xdy−ydx), then da=k dx∧dy, so F=ik dx∧dy.
Locally this is straightforward. To obtain a global Chern number on a closed surface, the local connection forms must be related correctly on overlaps, and the curvature must extend globally. Integrating a convenient local formula over a region that does not represent the whole bundle is not enough.
Suppose the globally defined curvature on a closed oriented surface Σ satisfies ∫Σ(i/2π)F=n. Then the first Chern number is n. The integer is stable under changing the connection even though F itself may change by an exact correction in the characteristic representative.
20. Classifying spaces organise all bundles of a given structure group
For a topological group G, there is a universal principal G-bundle EG→BG. Under standard hypotheses, principal G-bundles over a suitable base M are classified up to isomorphism by homotopy classes of maps M→BG.
A bundle over M can therefore be obtained as a pullback of the universal bundle along a classifying map. Characteristic classes can be understood as universal cohomology classes on BG pulled back to M.
This viewpoint explains naturality automatically. If E is classified by f:M→BG and h:N→M, then h*E is classified by f∘h, so a universal class pulls back in the expected order.
The construction is conceptually powerful but abstract. For concrete calculations on spheres or low-dimensional spaces, clutching maps and explicit cohomology can be more efficient.
21. Stable equivalence changes the classification problem
Two vector bundles E and F are stably equivalent if adding trivial bundles makes them isomorphic: E⊕εᵐ≈F⊕εⁿ for suitable trivial summands.
K-theory organises stable bundle classes using formal differences of vector bundles. This is a different classification problem from asking whether E itself is trivial.
A bundle can be stably trivial without being trivial. For example, the tangent bundle of Sⁿ satisfies TSⁿ⊕ε¹≈εⁿ⁺¹ because the outward normal line complements the tangent spaces inside the trivial ambient bundle Sⁿ×Rⁿ⁺¹.
This relation does not imply TSⁿ is trivial. Stable information is deliberately coarser. Characteristic classes and K-theory often become easier to manipulate precisely because they respect stable operations.
22. Global analysis studies differential operators between bundles
A differential operator on a manifold often acts between spaces of sections of vector bundles. The exterior derivative d:Ωᵏ(M)→Ωᵏ⁺¹(M) is the basic example. A connection gives covariant derivatives of sections. A Riemannian metric and connection produce Laplace-type operators.
Global analysis asks how local differential formulas interact with compactness, topology, boundary conditions and spectral properties. The kernel of an operator can encode global geometric information.
The key word is global. Solving a PDE in one chart does not automatically solve it on the manifold. Sections must agree on overlaps, regularity must be controlled, and global boundary or compactness conditions influence existence and uniqueness.
This is where differential geometry meets functional analysis. Spaces of smooth sections are completed in Sobolev or Hilbert norms, differential operators become unbounded or bounded maps between appropriate spaces, and ellipticity provides estimates that control solutions.
23. The Hodge star converts metric and orientation into an operator on forms
On an oriented Riemannian n-manifold, the Hodge star maps k-forms to (n−k)-forms. It is characterised by α∧*β=⟨α,β⟩vol_g.
The star depends on both the metric and orientation. Exterior differentiation d did not need a metric; the Hodge star does. This distinction keeps the topological differential-form calculus separate from metric-dependent analysis.
Using the Hodge star, define the formal adjoint d* of d, with a sign depending on degree and dimension conventions. The Hodge Laplacian is Δ=dd*+d*d.
A form ω is harmonic when Δω=0. On a closed oriented Riemannian manifold, harmonic forms satisfy dω=0 and d*ω=0.
Metric data have entered the definition of harmonicity, but the dimension of the harmonic-form space will recover a topological quantity.
24. Hodge theory selects one harmonic representative from each de Rham class
For a closed oriented Riemannian manifold, the Hodge theorem identifies the space of harmonic k-forms with the kth de Rham cohomology group.
Every de Rham cohomology class has a unique harmonic representative. Consequently the dimension of the harmonic k-form space equals the kth Betti number.
The representative depends on the chosen metric, but the dimension and cohomology class do not. Changing the metric changes which differential form is selected as harmonic, not the underlying de Rham cohomology.
On S¹ with its standard metric, the 1-form dθ is locally meaningful and the global angular 1-form can be described invariantly; a constant multiple represents the one-dimensional H¹_dR(S¹). On S², H¹_dR=0, so there are no nonzero harmonic 1-forms on the round sphere.
This is a striking return path: a metric-dependent elliptic PDE produces a canonical representative of a metric-independent topological class.
25. Elliptic operators have controlled symbols
The principal symbol of a differential operator captures its highest-order behaviour. An operator is elliptic when its symbol is invertible for every nonzero cotangent vector.
The Laplacian is elliptic. So are many geometric operators built from Dirac-type constructions. Ellipticity is the analytic analogue of a nondegeneracy condition: it prevents the highest-order part from losing directions.
On a compact manifold, elliptic operators satisfy strong regularity properties. Distributional or weak solutions of an elliptic equation become smooth when the data are smooth, and kernels are finite-dimensional under standard closed-manifold hypotheses.
These are not consequences of compactness alone. Ellipticity supplies local estimates; compactness lets those estimates be assembled into global Fredholm behaviour.
Boundary problems require additional boundary conditions such as elliptic boundary conditions. A theorem stated for a closed manifold should not be applied unchanged to a manifold with boundary.
26. Eigenvalues turn geometry into a spectrum
On a closed Riemannian manifold, the Laplace–Beltrami operator on functions has a discrete nonnegative spectrum 0=λ₀≤λ₁≤λ₂≤⋯ tending to infinity, counted with multiplicity.
The zero eigenspace consists of functions constant on each connected component. Thus the multiplicity of λ=0 records the number of connected components.
Higher eigenvalues depend on the metric and encode geometric information such as scale and shape, but the spectrum does not always determine the manifold uniquely. Isospectral nonisometric examples show that hearing the entire spectrum need not reconstruct every geometric detail.
Scaling the metric by c² scales lengths by c and Laplace eigenvalues by 1/c². This dimensional rule is a useful check when comparing spectra under rescaling.
27. The Fredholm index subtracts two finite-dimensional failures
For a Fredholm operator D between appropriate Banach or Hilbert spaces, the kernel and cokernel are finite-dimensional and the image is closed. Its index is
ind(D)=dim ker D−dim coker D.
The index is stable under compact perturbations and under continuous deformations through Fredholm operators. Individual kernel dimensions can jump while the difference remains constant.
For elliptic differential operators on closed manifolds, suitable Sobolev completions give Fredholm operators. The analytic index is therefore a robust integer attached to the operator.
This stability makes the index capable of carrying topological information even though it was defined analytically.
28. The Atiyah–Singer index theorem is a bridge, not a shortcut
The Atiyah–Singer index theorem equates the analytic index of an elliptic operator on a closed manifold with a topological expression built from the operator’s symbol and characteristic classes.
This is one of the central local-to-global results of modern geometry: a dimension difference in spaces of solutions is computed by topological data.
The exact formula depends on the operator. The de Rham operator produces the Euler characteristic; signature and Dirac operators produce other characteristic-number formulas. It would be misleading to write one universal scalar formula without specifying the operator and bundle data.
This guide therefore treats Atiyah–Singer as a boundary marker for further study. The important conceptual chain is: bundle → connection/curvature → characteristic class → elliptic symbol → Fredholm index → global topological number.
29. Gauss–Bonnet is the surface-sized prototype
For a closed oriented Riemannian surface M, Gauss–Bonnet states ∫_M K dA=2πχ(M), where K is Gaussian curvature.
The left side depends on the chosen metric point by point. The right side is topological. Changing the metric redistributes curvature while the total integral remains constrained by the Euler characteristic.
Chern–Gauss–Bonnet generalises this pattern to even-dimensional oriented Riemannian manifolds using the curvature of the tangent bundle and the Euler class.
This is the same architectural idea that recurs across the batch: flexible local geometric data can produce rigid global invariants after the correct integration or cohomological projection.
30. A reliable bundle-and-global-analysis workflow
First identify the base, fibre, structure group and rank. Decide whether the bundle is real or complex and whether it has an orientation, metric or other reduction of structure group.
Then distinguish local frames from global frames. Record transition functions and check their cocycle condition. When a section is proposed, verify that its local expressions agree on overlaps.
Choose characteristic classes whose coefficient system and degree match the bundle. A nonzero class can obstruct triviality or a desired structure; a zero class usually means only that this particular obstruction vanished.
For connection calculations, separate the gauge-dependent connection matrix from the tensorial curvature. For Chern–Weil calculations, keep the normalisation and orientation explicit.
For global analysis, state compactness and boundary assumptions, identify the bundles on which the operator acts, and check ellipticity before invoking Fredholm or Hodge conclusions. Local differential formulas do not automatically supply global solvability.
31. Independent practice: sixteen questions
- Explain why a rank-k bundle with k pointwise independent global sections is trivial.
- Use the interval model to show the Möbius line bundle has no nowhere-zero continuous section.
- Why does one nowhere-zero section trivialise a line bundle but not necessarily a rank-three bundle?
- State the cocycle condition for transition functions on a triple overlap.
- Show that TSⁿ⊕ε¹ is trivial by using the outward normal vector.
- What is the fibre of the pullback bundle f*E over n∈N?
- Explain why the double-cover pullback of the Möbius line bundle becomes trivial.
- What does w₁(E)=0 say about a real vector bundle?
- Why does nonzero Euler class obstruct a nowhere-zero section of an oriented bundle?
- For complex line bundles L and K, state c₁(L⊗K) and c₁(L*).
- If a complex bundle has c₁≠0, what can you conclude about triviality? What can you not conclude if c₁=0?
- For a unitary line-bundle connection A=ia, express F and the standard de Rham representative of c₁.
- On a closed oriented Riemannian manifold, what does Hodge theory say about a de Rham cohomology class?
- What is the multiplicity of the zero Laplace eigenvalue on a closed manifold with three connected components?
- If the metric is scaled by c², how do lengths and Laplace eigenvalues scale?
- Define the Fredholm index and explain why the Atiyah–Singer theorem needs the operator’s symbol rather than only the base manifold.
32. Worked solutions and checks
1. The sections define a fibrewise linear map M×Rᵏ→E by (p,a₁,…,aₖ)↦Σaᵢsᵢ(p). Pointwise independence makes this an isomorphism on every fibre, and smooth dependence gives a bundle isomorphism.
2. A section has fibre coordinate f(t) with endpoint rule f(1)=−f(0). If f never vanished, it could not change sign continuously. The intermediate value theorem forces a zero.
3. In a line bundle, a nonzero vector is already a basis of its one-dimensional fibre. In rank three it spans only one line; two more independent directions are still needed for a global frame.
4. With one common convention, gᵢⱼgⱼₖgₖᵢ=I on Uᵢ∩Uⱼ∩Uₖ. Equivalent index ordering gives the same consistency requirement.
5. At x∈Sⁿ, the ambient Rⁿ⁺¹ splits as TₓSⁿ⊕span{x}. The normal vector x varies smoothly and provides a trivial normal line, so TSⁿ⊕ε¹≈Sⁿ×Rⁿ⁺¹.
6. The fibre over n is E_{f(n)}. The pullback keeps the original fibre but relabels it over the new base point.
7. One circuit of the new base corresponds to two circuits of the old base. Each old circuit reverses the fibre sign, so two reversals return the original direction. Equivalently the mod-2 obstruction pulls back to twice the generator, which is zero.
8. It says the real vector bundle is orientable. For E=TM, it says the manifold is orientable.
9. A nowhere-zero section has empty zero locus. The Euler class is represented, under the standard transverse-section framework, by the zero locus. Therefore a nonzero Euler class prevents such a section.
10. c₁(L⊗K)=c₁(L)+c₁(K), and c₁(L*)=−c₁(L).
11. If c₁≠0, the bundle cannot be trivial. If c₁=0, only that first-Chern obstruction has vanished; higher Chern classes or other invariants may still show nontriviality, and in some settings characteristic classes are not complete classifiers.
12. For A=ia on a line bundle, F=dA=i da. Under the standard unitary convention, (i/2π)F is a real closed 2-form representing c₁ in real de Rham cohomology. If a different sign convention for the connection is chosen, the corresponding normalisation must be translated consistently.
13. Every kth de Rham cohomology class has a unique harmonic k-form representative for the chosen metric. Changing the metric can change the representative while leaving the cohomology group unchanged.
14. The zero eigenspace consists of functions constant on each connected component, so its dimension is three.
15. Lengths scale by c. Laplace eigenvalues scale by 1/c².
16. ind(D)=dim kerD−dim cokerD. Different elliptic operators on the same manifold have different symbols and can have different indices. Atiyah–Singer computes the analytic index from the K-theory class of the symbol together with characteristic data; the base manifold alone is insufficient.
33. Questions that protect the conceptual boundaries
Does every locally trivial object have to be globally trivial?
No. Local triviality is part of the definition of a bundle. Global triviality is a special case. The transition functions and their homotopy or cohomological obstructions decide whether the local pieces can be assembled into one global product.
Do characteristic classes completely classify bundles?
Not in general. They are natural and powerful invariants, but different bundles can share the same characteristic classes. Classification depends on the base, rank, structure group and category.
Does Hodge theory make cohomology metric-dependent?
No. The harmonic representative depends on the metric, but the de Rham class it represents does not. Hodge theory uses metric-dependent analysis to select a canonical representative from a topological class.
Does an index count all solutions?
No. The Fredholm index is a signed difference between kernel and cokernel dimensions. It is stable precisely because those dimensions may change together while their difference remains fixed.
What does this prepare?
It prepares the final R23 cell: symplectic and contact geometry. There the cotangent bundle acquires a canonical symplectic form, Hamiltonian motion becomes a geometric flow, Lagrangian submanifolds encode constrained states, and contact geometry appears naturally on odd-dimensional hypersurfaces.
Sources and further study
Ralph L. Cohen, Bundles, Manifolds, and Homotopy develops vector bundles, classifying constructions and characteristic classes. Chris Wendl, Lecture Notes on Bundles and Connections develops differential-geometric bundle theory, connections and curvature. The UC San Diego Math 250B course route provides a graduate progression through connections, curvature and Chern classes.
For the topological background, return to Homology and Cohomology. For curvature before characteristic forms, return to Connections and Curvature. Return to the BTT Mathematics Hub.