03.13.03 · modern-geometry / spectral-sequences

Leray-Hirsch theorem and the splitting principle for vector bundles

shipped3 tiersLean: partialpending prereqs

Anchor (Master): Bott-Tu §17, §21; Borel-Hirzebruch 1958/59; Milnor-Stasheff §16; Husemoller §16

Intuition [Beginner]

The splitting principle is the cleanest pedagogical move in the theory of characteristic classes: to compute characteristic classes of a vector bundle, pretend the bundle is a sum of line bundles. This works because every formula you can write down for sums of line bundles, that holds for all such sums, holds for arbitrary vector bundles.

The reason this is allowed is the Leray-Hirsch theorem, which says: if you have a fibre bundle whose fibre cohomology classes extend to global classes on the total space, then the total space cohomology is a free module over the base cohomology with these classes as a basis. Apply this to the bundle of complete flags inside a vector bundle, and you get a covering space (the flag bundle) on which the original vector bundle splits as a sum of line bundles, and on which the cohomology pulls back injectively. Anything you compute upstairs, you can read off downstairs.

The result is a calculus: write the total Chern class formally as a product over $i$ of factors $(1 + x_{i})$ , with $x_{i}$ the "Chern roots", expand symbolically, then re-package into actual Chern classes at the end. This is how Pontryagin classes are computed, how Whitney sum formulas are derived, how Chern character is constructed, and how almost every characteristic-class identity in twentieth-century topology was proved.

Visual [Beginner]

Picture a vector bundle as a stack of vector spaces over the base. To split it into lines, choose at every point a complete flag — a tower of subspaces $0 \subset L_{1} \subset L_{2} \subset \dots \subset L_{n} =$ the fibre, where each $L_{i}$ has dimension $i$ . The space of all such flags varies smoothly over the base; this is the flag bundle.

On the flag bundle, the original bundle splits as the direct sum of the line bundles $L_{1}, L_{2} / L_{1}, \dots, L_{n} / L_{n - 1}$ . Cohomology pulls back injectively from the base to the flag bundle, so any formula derived upstairs (where everything splits into lines) is an honest formula downstairs (where the bundle is its original twisted self).

Worked example [Beginner]

For a complex vector bundle $E$ of rank $2$ over a base $B$ , the flag bundle is the projectivisation $P (E) \to B$ — a fibre bundle with fibre $C P^{1} = S^{2}$ over each point of $B$ . On $P (E)$ there is a tautological line bundle $L$ whose fibre over $(b, ℓ) \in P (E)$ is the line $ℓ \subset E_{b}$ .

By Leray-Hirsch, the cohomology of $P (E)$ is a free $H^{*} (B)$ -module with basis $1, x$ where $x = c_{1} (L^{*}) \in H^{2} (P (E))$ is the first Chern class of the dual tautological bundle. Specifically, $H^{*} (P (E)) ≅ H^{*} (B) \cdot 1 \oplus H^{*} (B) \cdot x$ , with $x^{2}$ expressible as $- c_{1} (E) x - c_{2} (E)$ in the cohomology ring.

This relation $x^{2} + c_{1} (E) x + c_{2} (E) = 0$ is Grothendieck's definition of the Chern classes: the Chern classes are the coefficients of the unique relation that $x = c_{1} (L^{*})$ satisfies in $H^{*} (P (E))$ over $H^{*} (B)$ . The relation is the splitting principle made concrete: factoring $x^{2} + c_{1} x + c_{2} = (x + r_{1}) (x + r_{2})$ identifies the Chern roots $r_{1}, r_{2}$ as the formal "first Chern classes of the line factors" of $E$ , and the symmetric functions in $r_{1}, r_{2}$ recover $c_{1}, c_{2}$ .

Check your understanding [Beginner]

Formal definition [Intermediate+]

Theorem (Leray-Hirsch, 1948). Let $π : E \to B$ be a fibre bundle with fibre $F$ , with $B$ a connected CW complex and the base ring a principal ideal domain. Suppose:

$H^{q} (F; k)$ is a finitely generated free $k$ -module for each $q$ .
There exist classes $e_1, \ldots, e_n \in H^(E; k) $w h oser es t r i c t i o n s t oe a c h f ib r e f or ma$ k $- ba s i so f$ H^(F; k)$.

Then the map $$ H^(B; k) \otimes_k H^(F; k) \to H^(E; k), \qquad \beta \otimes \iota^(e_i) \mapsto \pi^(\beta) \cdot e_i, $$ is an isomorphism of graded $H^(B; k)$-modules.

This is a module isomorphism, not (in general) a ring isomorphism — the cup product on $H^{*} (E)$ may differ from the tensor product. The proof, given in 03.13.02 Exercise 8, is that the Leray-Serre spectral sequence collapses at $E_{2}$ .

Projective bundle. For a complex vector bundle $E \to B$ of rank $n$ , the projectivisation $P (E) \to B$ has fibre $C P^{n - 1}$ . Let $L \to P (E)$ be the tautological line bundle whose fibre at $(b, ℓ)$ is the line $ℓ \subset E_{b}$ . Set $x := c_{1} (L^{*}) \in H^{2} (P (E); Z)$ (the dual sign convention follows Bott-Tu §21).

The classes $1, x, x^{2}, \dots, x^{n - 1}$ restrict to a basis of $H^{*} (C P^{n - 1}; Z) = Z [x] / x^{n}$ on each fibre. By Leray-Hirsch, $$ H^(\mathbb{P}(E); \mathbb{Z}) \cong H^(B; \mathbb{Z})[x] / (x^n + c_1(E) x^{n-1} + \cdots + c_n(E)), $$ where the relation defines the Chern classes $c_{1} (E), \dots, c_{n} (E) \in H^{*} (B; Z)$ . This is Grothendieck's definition of Chern classes ^{[Grothendieck 1958]}.

Splitting principle. For a complex vector bundle $E \to B$ of rank $n$ , the iterated projectivisation $$ F\ell(E) \to B $$ — the flag bundle, with fibre the variety $F ℓ_{n} (C^{n})$ of complete flags in $C^{n}$ — pulls $E$ back to a bundle that splits as a direct sum of line bundles: $$ \sigma : F\ell(E) \to B, \qquad \sigma^* E \cong L_1 \oplus L_2 \oplus \cdots \oplus L_n. $$ The pullback on cohomology $$ \sigma^* : H^(B; \mathbb{Z}) \to H^(F\ell(E); \mathbb{Z}) $$ is injective, by iterating Leray-Hirsch on each successive projectivisation. The injectivity is what powers the proof technique.

Chern roots. On $F ℓ (E)$ , write $x_{i} := c_{1} (L_{i}^{*})$ . Then $$ \sigma^* c(E) = \sigma^*\bigl(1 + c_1(E) + \cdots + c_n(E)\bigr) = \prod_{i=1}^n (1 + x_i). $$ The $x_{i}$ are the Chern roots of $E$ . Any symmetric polynomial in $x_{1}, \dots, x_{n}$ is the pullback of a polynomial in the Chern classes $c_{j} (E)$ ; specifically, $c_{j} (E) = e_{j} (x_{1}, \dots, x_{n})$ is the $j$ -th elementary symmetric function. Identities in Chern classes follow from identities in Chern roots.

Pontryagin classes (real bundles). For a real vector bundle $E_{R} \to B$ of rank $2 k$ , the Pontryagin classes are defined via complexification $E_{R} \otimes_{R} C$ : $$ p_i(E_{\mathbb{R}}) := (-1)^i c_{2i}(E_{\mathbb{R}} \otimes_{\mathbb{R}} \mathbb{C}) \in H^{4i}(B; \mathbb{Z}). $$ The splitting principle for $E_{R}$ runs through $E_{R} \otimes C$ and identifies real Chern roots $\pm ξ_{1}, \dots, \pm ξ_{k}$ in pairs, with $p_{i} (E_{R}) = e_{i} (ξ_{1}^{2}, \dots, ξ_{k}^{2})$ the elementary symmetric functions in the squared roots. The construction is the structural reason Pontryagin classes live in degree $4 i$ and have the Whitney sum formula $p (E \oplus F) = p (E) p (F)$ modulo $2$ -torsion.

Key theorem with proof [Intermediate+]

Theorem (Leray-Hirsch). In the setup above, the map $H^(B; k) \otimes H^(F; k) \to H^(E; k) $,$ \beta \otimes \iota^(e_i) \mapsto \pi^(\beta) \cdot e_i $, i s ani so m or p hi s m o f g r a d e d$ H^(B; k)$-modules.

Proof. The Leray-Serre spectral sequence of the fibration $F \to E \to B$ has $E_{2}^{p, q} = H^{p} (B; H^{q} (F; k))$ , which by hypothesis is the constant local system $H^{q} (F; k)$ . The classes $e_{i}$ supply a section of the projection $H^{*} (E; k) \to E_{2}^{0, *} = H^{*} (F; k)$ — that is, each $ι^{*} (e_{i}) \in H^{*} (F; k)$ has a global cohomology lift, namely $e_{i}$ itself.

Define a map $$ \Phi : H^(B; k) \otimes_k H^(F; k) \to H^(E; k), \qquad \beta \otimes \iota^(e_i) \mapsto \pi^*(\beta) \cdot e_i. $$ We show $Φ$ is an isomorphism via the spectral sequence.

The spectral sequence is multiplicative: each page is a bigraded ring and $d_{r}$ is a derivation. The classes $e_{i} \in H^{*} (E; k)$ are in $F^{0} H^{*} (E; k)$ (the entire cohomology) but their image in $E_{\infty}^{0, q}$ is the basis $ι^{*} (e_{i})$ of $H^{q} (F; k)$ . By multiplicativity, products $π^{*} (β) \cdot e_{i}$ for $β \in H^{*} (B; k)$ generate the entire associated graded: $$ E_\infty^{p, q} = \pi^(H^p(B; k)) \cdot e_i \quad \text{(where } \deg \iota^(e_i) = q\text{)}. $$

We claim no differential $d_{r}$ for $r \geq 2$ is non-zero. Suppose $d_{r} (e_{i}) \neq = 0$ for some $r \geq 2$ . Then $d_{r} (e_{i}) \in E_{r}^{r, q - r + 1}$ is a non-zero class in $H^{*} (B; H^{q - r + 1} (F; k))$ . But $e_{i}$ is a genuine class on $E$ (a permanent cycle), so $d_{r} (e_{i}) = 0$ on the spectral-sequence side as well. Hence each $e_{i}$ survives to $E_{\infty}^{0, *}$ .

Multiplicativity then forces all $d_{r}$ on $E_{2}^{p, q}$ for any $(p, q)$ to vanish: write a class $α \in E_{2}^{p, q}$ as $\sum β_{j} \cdot ι^{*} (e_{i_{j}})$ with $β_{j} \in H^{p} (B)$ and $ι^{*} (e_{i_{j}})$ a basis of $H^{*} (F; k)$ . Then $d_{r} (α) = \sum d_{r} (β_{j}) ι^{*} (e_{i_{j}}) + \sum (- 1)^{∣ β_{j} ∣} β_{j} d_{r} (ι^{*} (e_{i_{j}})) = 0$ since both terms vanish (the first because $β_{j}$ is in the bottom row and $d_{r}$ from row $0$ goes to negative row $1 - r$ , outside support; the second because each $ι^{*} (e_{i_{j}})$ has a permanent-cycle lift $e_{i_{j}}$ ).

Hence $E_{2} = E_{\infty}$ , and the filtration on $H^{n} (E)$ has associated graded $E_{\infty}^{p, q}$ with $p + q = n$ . The map $Φ$ on associated gradeds is an isomorphism by inspection. Since both sides are filtered by $π^{*}$ -degree and $Φ$ respects filtration, $Φ$ itself is an isomorphism. $□$

Theorem (Splitting principle). Let $E \to B$ be a complex vector bundle of rank $n$ over a paracompact base. There exists a space $F ℓ (E) \to B$ (the complete flag bundle) such that:

The pullback $\sigma^ E $o n$ F\ell(E) $s pl i t s a s a d i r ec t s u m o f l in e b u n d l es :$ \sigma^* E \cong L_1 \oplus \cdots \oplus L_n$.*
The pullback on cohomology $\sigma^ : H^(B; R) \to H^(F\ell(E); R) $i s inj ec t i v e f or an y coe f f i c i e n t r in g$ R$.*

Proof. Construct $F ℓ (E)$ inductively. Let $π_{1} : P (E) \to B$ be the projectivisation. The pullback $π_{1}^{*} E$ on $P (E)$ contains the tautological line subbundle $L_{1}$ , with quotient $E_{1} := π_{1}^{*} E / L_{1}$ a rank- $(n - 1)$ bundle. By induction (replace $B \to E$ by $P (E) \to E_{1}$ and iterate), we obtain a tower of projectivisations $$ F\ell(E) = F_n \to F_{n-1} \to \cdots \to F_1 = \mathbb{P}(E) \to B $$ with each step a projectivisation. On $F ℓ (E)$ , the iterated quotient produces line bundles $L_{1}, L_{2}, \dots, L_{n}$ with $σ^{*} E ≅ L_{1} \oplus \dots \oplus L_{n}$ .

Injectivity of $σ^{*}$ : each step $π_{i} : F_{i} \to F_{i - 1}$ is a projective bundle, and the pullback $π_{i}^{*}$ is injective by Leray-Hirsch (the cohomology of $F_{i}$ is a free $H^{*} (F_{i - 1})$ -module of rank $n - i + 1$ , with basis $1, x, x^{2}, \dots, x^{n - i}$ for $x = c_{1}$ of the tautological line on $F_{i}$ ). Composing $n$ injections gives the injectivity of $σ^{*}$ . $□$

Corollary (Whitney sum formula via splitting). For complex vector bundles $E, F \to B$ , $$ c(E \oplus F) = c(E) \cdot c(F). $$

Proof. Pull back to $F ℓ (E \oplus F)$ , where $E \oplus F$ splits as $L_{1} \oplus \dots \oplus L_{n + m}$ with the first $n$ lines coming from $E$ and the last $m$ from $F$ . Then $$ \sigma^* c(E \oplus F) = \prod_{i=1}^{n+m}(1 + x_i) = \prod_{i=1}^n (1 + x_i) \cdot \prod_{j=n+1}^{n+m}(1 + x_j) = \sigma^* c(E) \cdot \sigma^* c(F). $$ Since $σ^{*}$ is injective, $c (E \oplus F) = c (E) \cdot c (F)$ on $B$ . $□$

Synthesis. Leray-Hirsch is exactly the collapse-at- $E_{2}$ specialisation of Leray-Serre. The splitting principle is the foundational technique that reduces characteristic-class identities on arbitrary bundles to symmetric-function identities on Chern roots. The flag bundle is the bridge between the simple case and the general case.

Bridge. The construction here builds toward later units of the strand, where the same pattern is taken up at higher structure. The defining pattern appears again in those units in a sharpened form, where the local data is glued or quotiented. Putting these together, the foundational insight is that the data of this unit gives the structural signature that the rest of the strand reads off.

Exercises [Intermediate+]

Exercise 6 (medium, symbolic). Chern character of a sum.

Define the Chern character via Chern roots: $ch (E) = \sum_{i} e^{x_{i}}$ . Show that $ch (E \oplus F) = ch (E) + ch (F)$ and $ch (E \otimes F) = ch (E) \cdot ch (F)$ .

Hint

Chern roots of a sum are the union; Chern roots of a tensor product are pairwise sums.

Answer

Sum: the Chern roots of $E \oplus F$ are $x_{1}, \dots, x_{n}, y_{1}, \dots, y_{m}$ (the union), so $ch (E \oplus F) = \sum_{i} e^{x_{i}} + \sum_{j} e^{y_{j}} = ch (E) + ch (F)$ .

Tensor: the Chern roots of $E \otimes F$ are $x_{i} + y_{j}$ for $1 \leq i \leq n$ , $1 \leq j \leq m$ (since the line bundle $L_{i} \otimes L_{j}$ has $c_{1} = c_{1} (L_{i}) + c_{1} (L_{j}) = x_{i} + y_{j}$ ). Then $$ \mathrm{ch}(E \otimes F) = \sum_{i, j} e^{x_i + y_j} = \sum_{i, j} e^{x_i} e^{y_j} = \Bigl(\sum_i e^{x_i}\Bigr) \Bigl(\sum_j e^{y_j}\Bigr) = \mathrm{ch}(E) \cdot \mathrm{ch}(F). $$ The splitting principle is what makes these "formal" computations on Chern roots into honest identities on Chern classes. $□$

Exercise 7 (hard, symbolic). $H^(\mathrm{Gr}(2, 4))$ via Borel presentation.*

Compute the cohomology ring of the Grassmannian $Gr (2, 4)$ of $2$ -planes in $C^{4}$ , using the splitting principle on the tautological subbundle $S \to Gr (2, 4)$ and the universal quotient $Q$ .

Hint

$S \oplus Q = C^{4}$ (the standard frame bundle), so $c (S) c (Q) = 1$ .

Answer

Let $a = c_{1} (S^{*}), b = c_{2} (S^{*})$ be Chern classes of the dual tautological subbundle ( $S$ has rank $2$ ), and let $a^{'}, b^{'}$ denote $c_{1} (Q^{*}), c_{2} (Q^{*})$ ( $Q$ also rank $2$ ). The Whitney relation $S \oplus Q = C^{4}$ gives $c (S) c (Q) = 1$ , equivalently $c (S^{*}) c (Q^{*}) = 1$ in coefficients (passing to duals matches signs): $$ (1 + a + b)(1 + a' + b') = 1 \pmod{\text{degree} > 4}. $$ Expanding: $a + a^{'} = 0$ , $b + a a^{'} + b^{'} = 0$ , $a b^{'} + a^{'} b = 0$ , $b b^{'} = 0$ . Solving: $a^{'} = - a$ , $b^{'} = - b - a a^{'} = - b + a^{2}$ , then $b b^{'} = b (- b + a^{2}) = - b^{2} + a^{2} b = 0$ , and $a b^{'} + a^{'} b = - ab + a^{3} - ab = a^{3} - 2 ab = 0$ .

The cohomology ring is generated by $a, b$ subject to the relations $b^{2} = a^{2} b$ and $a^{3} = 2 ab$ . Equivalently, $H^{*} (Gr (2, 4); Z) = Z [a, b] / (b^{2} - a^{2} b, a^{3} - 2 ab)$ . The Poincaré series is $1 + t^{2} + 2 t^{4} + t^{6} + t^{8}$ , total dimension $6$ , matching the count of Schubert cells.

Exercise 8 (hard, proof). Whitney via Mayer-Vietoris alternative.

The standard proof of the Whitney sum formula uses the splitting principle (Exercise 5 corollary). Sketch the alternative proof via Mayer-Vietoris induction on the base, following Bott-Tu §6 or Milnor-Stasheff §9. Identify where the two proofs diverge.

Hint

The MV-induction proof works on a good cover, where the bundle is locally a product and the formula holds tautologically. Glue using MV and the naturality of Chern classes.

Answer

MV-induction proof (sketch). Take a good cover of $B$ by sets $U_{α}$ on which $E$ and $F$ are both products. On each $U_{α}$ , $E \oplus F$ is also a product, so $c (E \oplus F) ∣_{U_{α}} = 1 = c (E) ∣_{U_{α}} \cdot c (F) ∣_{U_{α}}$ . The naturality of Chern classes (proved separately) lets us glue local computations via the Mayer-Vietoris exact sequence. Inductively, the formula holds on every finite intersection, and a partition-of-unity argument concludes that it holds on $B$ .

Splitting-principle proof (sketch). Pull back to $F ℓ (E \oplus F)$ where everything splits into lines. Compute the formula for sums of lines (a $\prod (1 + x_{i})$ identity, immediate). Use injectivity of $σ^{*}$ to conclude on $B$ .

Where they diverge. The MV proof requires (i) naturality of Chern classes, which is itself a substantial theorem, and (ii) good-cover compactness arguments. The splitting-principle proof requires (i) construction of $F ℓ$ (a substantial geometric input), and (ii) Leray-Hirsch (a substantial spectral-sequence input). The two proofs invest in different infrastructure: MV-induction trades local-to-global gluing for the upfront cost of naturality and good covers; splitting trades the upfront cost of Leray-Hirsch and flag bundles for an instant computation. Bott-Tu emphasises this "dual proof discipline" as a structural feature of differential topology — the same theorem proved by complementary techniques. $□$

Lean formalization [Intermediate+]

[object Promise]

The flag bundle, Leray-Hirsch, and the splitting principle are all absent from Mathlib at this level of generality. Mathlib has projective spaces and projective bundles in its algebraic-geometry layer but not the topological projective bundle theorem; the cohomology-ring computation for $P (E) \to B$ would be a substantial Mathlib contribution.

Advanced results [Master]

Borel presentation of flag-manifold cohomology. For a compact Lie group $G$ with maximal torus $T$ and Weyl group $W$ , the flag manifold $G / T$ has cohomology $$ H^(G/T; \mathbb{Q}) \cong H^(BT; \mathbb{Q}) / I_W, $$ where $I_{W}$ is the ideal generated by positive-degree $W$ -invariants. For $G = U (n)$ , this is the Schubert presentation: $H^{*} (F ℓ_{n}; Q) = Q [x_{1}, \dots, x_{n}] / (e_{1}, \dots, e_{n})$ , the polynomial ring modulo the ideal of elementary symmetric functions. The proof uses the Serre spectral sequence of $G / T \to B T \to B G$ together with Borel's identification $H^{*} (B G; Q) = H^{*} (B T; Q)^{W}$ ^{[Borel-Hirzebruch 1958]}. The flag-bundle splitting principle is the geometric counterpart.

Schubert calculus. The cohomology of the Grassmannian $Gr (k, n)$ has an additive basis given by Schubert classes — Poincaré duals of Schubert varieties. The cup-product structure constants are the Littlewood-Richardson coefficients, governing how products of Schubert classes decompose. The splitting principle gives an alternative presentation of $H^{*} (Gr (k, n); Q)$ as the symmetric polynomials in the Chern roots of the tautological subbundle modulo the Whitney relation; the bridge between the Schubert and Borel presentations is one of the central themes of geometric representation theory.

Generalised splitting principles. The technique generalises beyond complex vector bundles. For an oriented real bundle, the oriented flag bundle parametrises oriented complete flags; its cohomology is computed by Leray-Hirsch with $Z /2$ coefficients (or with $Z$ coefficients on the orientable case using $E_{R} \otimes C$ ). For a principal $G$ -bundle, the Borel construction $E \times_{G} E G$ supplies an analogous reduction; classifying maps to $B G$ recover characteristic classes through $H^{*} (B G)$ . The Borel-Weil theorem identifies cohomology of $G / B$ as Lie-algebra representations.

Splitting principle for $K$ -theory. A direct $K$ -theoretic analogue: for a complex vector bundle $E \to B$ , the total space of the complete flag bundle has $K^{*} (F ℓ (E)) ≅ K^{*} (B) [L_{1}^{\pm 1}, \dots, L_{n}^{\pm 1}] / I$ for a Weyl-relation ideal $I$ . This identifies $K^{*} (B)$ -classes of $E$ as symmetric functions in the line factors $L_{i} \in K^{*} (F ℓ (E))$ . The K-theoretic Chern character $ch : K^{*} (B) \otimes Q \to H^{*} (B; Q)$ is the universal natural transformation that this picture makes computable.

Grothendieck's perspective. Alexander Grothendieck reformulated Chern classes in his 1958 Bull. Soc. Math. France paper using exactly the projective bundle theorem of Leray-Hirsch. For $Pic (P (E)) = π^{*} Pic (B) \oplus Z \cdot [O_{E} (1)]$ , the relation $\sum c_{i} (E) ξ^{n - i} = 0$ for $ξ = c_{1} (O_{E} (1))$ in $H^{*} (P (E))$ defines $c_{i} (E)$ uniquely. This axiomatic definition is the route used in algebraic geometry, where the spectral-sequence machinery is replaced by sheaf-cohomology and the projective bundle formula becomes a statement about the structure of the Chow ring (or the K-theory ring) of $P (E)$ . Grothendieck-Riemann-Roch is the global generalisation.

Splitting principle for orientifolds and equivariant bundles. In string theory and equivariant K-theory, the splitting principle generalises to settings with finite-group symmetries. For a $Z /2$ -equivariant bundle, the real flag bundle parametrises $Z /2$ -equivariant flags; its cohomology computes equivariant characteristic classes (Stiefel-Whitney for the real case, Pontryagin for the complex involution case). The framework of Atiyah-Segal completion and equivariant K-theory makes the splitting principle work in arbitrary equivariant settings.

Full proof set [Master]

Leray-Hirsch — alternative proof via the bar construction. An alternative proof of Leray-Hirsch, following the modern operadic perspective: model the fibration $F \to E \to B$ by the bar construction $B (F, F, *) \to B (F, F, B)$ . The cohomology of the total space is computed by the dual cobar construction, and the existence of global lifts $e_{i}$ is exactly the hypothesis that the cobar differential vanishes on the basis. The collapse-at- $E_{2}$ statement becomes a statement about the cobar complex being a free module. This is Eilenberg-Moore in reverse.

Projective bundle theorem — full computation. For a complex vector bundle $E \to B$ of rank $n$ , write $P (E) = (E ∖ 0_{E}) / C^{*}$ . The tautological line bundle $L \to P (E)$ has fibre at $(b, ℓ)$ equal to the line $ℓ$ . Set $x := c_{1} (L^{*}) \in H^{2} (P (E); Z)$ .

Step 1: each fibre $C P^{n - 1}$ of $π : P (E) \to B$ has cohomology $Z [x] / x^{n}$ , with the same generator $x$ pulled back from the projection. The classes $1, x, x^{2}, \dots, x^{n - 1}$ on $P (E)$ restrict to a $Z$ -basis of fibre cohomology in each fibre.

Step 2: by Leray-Hirsch, $H^{*} (P (E); Z) = ⨁_{i = 0}^{n - 1} π^{*} H^{*} (B; Z) \cdot x^{i}$ as a graded $H^{*} (B)$ -module.

Step 3: the relation. The class $x^{n} \in H^{2 n} (P (E))$ must lie in the module spanned by $1, x, \dots, x^{n - 1}$ , so there exist unique classes $c_{1}, c_{2}, \dots, c_{n} \in H^{*} (B)$ (with $de g c_{i} = 2 i$ ) such that $$ x^n + c_1 \cdot x^{n-1} + c_2 \cdot x^{n-2} + \cdots + c_n = 0. $$ The classes $c_{i} = c_{i} (E)$ are the Chern classes; this is Grothendieck's definition.

Step 4: matching with the curvature definition. For a smooth bundle with a connection of curvature $F$ , the Chern-Weil definition gives $c_{i} (E) =$ coefficients of $det (I + \frac{1}{2 π i} F)$ , and one verifies on the projective bundle that the Grothendieck and Chern-Weil definitions coincide ^{[Bott-Tu §21 + §23]}.

Splitting principle — naturality and uniqueness. The flag bundle construction is functorial: a bundle map $E \to E^{'}$ over $B \to B^{'}$ induces a map $F ℓ (E) \to F ℓ (E^{'})$ over $B \to B^{'}$ . The pullback of Chern roots respects this functoriality. Hence any natural transformation of cohomology that vanishes on sums of line bundles vanishes universally, by the splitting principle.

Pontryagin classes of complex bundles. For a complex vector bundle $E$ of complex rank $n$ , view it as a real bundle $E_{R}$ of real rank $2 n$ . The complexification $E_{R} \otimes_{R} C$ splits as $E \oplus \overline{E}$ , where $\overline{E}$ is the complex conjugate bundle. The Chern roots of $\overline{E}$ are $- x_{i}$ (negatives of the Chern roots of $E$ ). Hence $$ c(E_{\mathbb{R}} \otimes \mathbb{C}) = c(E) c(\overline{E}) = \prod_i (1 + x_i)(1 - x_i) = \prod_i (1 - x_i^2). $$ The Pontryagin classes follow: $$ p_i(E_{\mathbb{R}}) = (-1)^i c_{2i}(E_{\mathbb{R}} \otimes \mathbb{C}) = e_i(x_1^2, \ldots, x_n^2). $$ A complex bundle has only "even" Pontryagin classes; this is the structural reason for the degree- $4 i$ shift.

Connections [Master]

Leray-Serre spectral sequence 03.13.02 — the Leray-Hirsch theorem is the collapse-at- $E_{2}$ specialisation of Leray-Serre. The hypothesis that fibre classes lift to total-space classes forces the spectral sequence to degenerate. Foundation-of relation; anchor phrase: Leray-Hirsch is the collapse-at- $E_{2}$ case of Leray-Serre.
Künneth formula 03.04.12 — by conn:436.kunneth-leray-hirsch, Leray-Hirsch built on Künneth on each fiber (foundation-of). When fibre cohomology is finite-dimensional and a global lift exists, Leray-Hirsch reduces to fibrewise Künneth times base cohomology — fibre-Künneth is exactly the Leray-Hirsch input.
Pontryagin and Chern classes 03.06.04 — by conn:444.leray-hirsch-splitting, splitting principle built on Leray-Hirsch theorem applied iteratively to flag-bundle projections (foundation-of). The splitting principle is the cleanest route to the Whitney sum formula, the multiplicativity of Chern character, and the symmetric-function identities that govern characteristic classes.
Classifying spaces 03.08.04 — the splitting principle factors through the classifying-space picture: $B U (n) \to B U (1)^{n} / S_{n}$ , where the right-hand side has the Borel-Weil cohomology presentation $H^{*} (B T)^{W}$ . By conn:445.splitting-flag-borel, splitting principle equivalent to Borel presentation H(BG) = H*(BT)^W* (equivalence). For $G = U (n)$ , $W = S_{n}$ acts by permuting Chern roots — the universal splitting principle is exactly this cohomology fact.
K-theory 03.08.01 — K-theoretic splitting principle: $K^{*} (B)$ -classes of vector bundles are symmetric functions in line-bundle classes on the flag bundle. The Chern character is the universal natural transformation realising this picture rationally.
Vector bundle 03.05.02 — Chern classes are characteristic invariants of vector bundles, computed by the splitting principle.
Spectral sequence (general) 03.13.01 — the Leray-Hirsch theorem rests on the multiplicative structure of the Leray-Serre spectral sequence, which itself rests on the multiplicative structure of a filtered DGA (developed in the apex unit).
Stiefel-Whitney classes 03.06.03 — the real splitting principle (with $Z /2$ coefficients) gives Stiefel-Whitney classes of a real bundle as symmetric functions of "Stiefel-Whitney roots" in the real flag bundle.

The splitting principle is the bridge between the pedagogically simple case (sums of line bundles, where everything is computable) and the general case (arbitrary vector bundles, where bundle isomorphism is hard). It is one of the core techniques of twentieth-century algebraic topology.

Throughlines and forward promises. The splitting principle is the foundational technique for characteristic-class identities. We will see the Whitney sum formula, the multiplicativity of the Chern character, and every symmetric-function calculation in characteristic-class theory route through this machine. The pattern recurs in K-theory (Chern character), in equivariant cohomology (Borel construction), and in Schubert calculus (divided-difference operators). The foundational reason characteristic-class identities reduce to symmetric-function identities is exactly the Borel presentation $H^{*} (B G; Q) = H^{*} (B T; Q)^{W}$ — this is precisely the equivalence the splitting principle realises geometrically on flag bundles. Putting these together: Leray-Hirsch is a specialisation of Leray-Serre at $E_{2}$ -collapse; the splitting principle is an instance of fibrewise Künneth; the Borel presentation is the cohomology fact behind the universal splitting. The bridge between the simple case (sums of line bundles) and the general case (arbitrary bundles) is exactly the cohomologically-injective pullback to the flag bundle.

Historical & philosophical context [Master]

The Leray-Hirsch theorem was announced by Guy Hirsch in his 1948 C. R. Acad. Sci. Paris note Un isomorphisme attaché aux structures fibrées (227), where he established that for a fibre bundle whose fibre cohomology is finitely generated and admits global lifts to the total space, the cohomology of the total space is a free module over the base cohomology. Hirsch was a Belgian topologist working in the Brussels group, and his note appeared two years after Leray's original spectral-sequence announcement. The theorem was developed independently of Leray's and Serre's work but turned out, with hindsight, to be the cleanest collapse criterion for what would become the Leray-Serre spectral sequence — the hypothesis on global lifts is exactly what guarantees the spectral sequence degenerates at $E_{2}$ .

The splitting principle as a systematic technique for characteristic-class computations was made into a discipline by Armand Borel and Friedrich Hirzebruch in their 1958 and 1959 papers Characteristic classes and homogeneous spaces I, II (Amer. J. Math. 80, 81). Borel and Hirzebruch worked in the framework of compact Lie groups: for a compact connected Lie group $G$ with maximal torus $T$ , they observed that the natural map $B T \to B G$ is rationally a finite covering with deck group $W$ (the Weyl group), so $H^{*} (B G; Q) = H^{*} (B T; Q)^{W}$ . Specialising to $G = U (n)$ and $T = U (1)^{n}$ : the map $H^{*} (B U (n); Q) \to H^{*} (B U (1)^{n}; Q)^{S_{n}}$ is an isomorphism, identifying Chern classes with symmetric polynomials in the universal Chern roots $x_{1}, \dots, x_{n} \in H^{*} (B U (1)^{n})$ . Pulling back along a classifying map $f : B \to B U (n)$ recovers the splitting principle for any rank- $n$ complex bundle.

The key insight of Borel-Hirzebruch was that this picture gives a universal splitting principle: the universal vector bundle $E U (n) \to B U (n)$ pulls back to the universal sum of line bundles $E U (1)^{n} \to B U (1)^{n}$ along the inclusion $B T \to B G$ , and any natural identity in characteristic classes that holds on $B U (1)^{n}$ must hold on $B U (n)$ by pullback injectivity. This is the form of the principle most useful in calculations: rather than constructing a flag bundle for each individual bundle, one works with the universal flag bundle $B U (1)^{n} \to B U (n)$ once and forever.

The geometric form via flag bundles — the form Bott and Tu emphasise in §21 — was developed in parallel by Steven Halperin and others in the 1960s and 1970s. This form has the advantage of being immediately constructive: given a specific bundle $E \to B$ , one builds the specific flag bundle $F ℓ (E) \to B$ and verifies that Chern roots make sense pointwise. The Borel-Hirzebruch universal form is the abstract shadow.

Alexander Grothendieck's 1958 Bull. Soc. Math. France paper La théorie des classes de Chern gave a different — and ultimately deeper — perspective. Grothendieck observed that the projective bundle theorem (the Leray-Hirsch identification of $H^{*} (P (E))$ ) could itself be taken as the definition of Chern classes: the unique sequence of base classes $c_{1}, \dots, c_{n}$ that make the relation $\sum c_{i} x^{n - i} = 0$ hold in the cohomology ring of the projective bundle. This axiomatic definition lifts directly to algebraic geometry, where $P (E)$ is a projective bundle in the algebraic sense and $H^{*}$ is replaced by the Chow ring (or the K-theory ring, or any cohomological functor satisfying naturality and projective-bundle decomposition). The splitting principle in algebraic geometry — for a coherent sheaf on a Noetherian scheme — is a direct consequence of Grothendieck's projective-bundle reformulation.

The pedagogical reframing of Bott-Tu §17 + §21 + §22 deserves separate mention. Bott and Tu treat Leray-Hirsch and the splitting principle as the practical computational engine that powers the entire theory of characteristic classes. Their §21 — titled simply "The splitting principle" — is barely four pages: state Leray-Hirsch, observe that the projective bundle satisfies the hypothesis with $1, x, x^{2}, \dots, x^{n - 1}$ as fibre basis, iterate to get the flag bundle, conclude with the formal computation of Whitney sum and Chern character. The brevity is the point: once Leray-Hirsch is in hand from §17, the splitting principle is a one-page corollary, and the entire characteristic-class calculus follows from this single observation. This is the cleanest derivation in algebraic topology of a piece of machinery that had taken thirty years to develop.

Bibliography [Master]

Hirsch, G., "Un isomorphisme attaché aux structures fibrées", C. R. Acad. Sci. Paris 227 (1948), 1328–1330.
Borel, A. & Hirzebruch, F., "Characteristic classes and homogeneous spaces I", Amer. J. Math. 80 (1958), 458–538.
Borel, A. & Hirzebruch, F., "Characteristic classes and homogeneous spaces II", Amer. J. Math. 81 (1959), 315–382.
Grothendieck, A., "La théorie des classes de Chern", Bull. Soc. Math. France 86 (1958), 137–154.
Bott, R. & Tu, L. W., Differential Forms in Algebraic Topology, Graduate Texts in Mathematics 82, Springer, 1982. §17, §21, §22.
Milnor, J. & Stasheff, J., Characteristic Classes, Annals of Mathematics Studies 76, Princeton University Press, 1974. §16.
Husemoller, D., Fibre Bundles, 3rd ed., Graduate Texts in Mathematics 20, Springer, 1994. §16.
Atiyah, M. F., K-Theory, Benjamin, 1967. §2 — splitting principle for K-theory.
Hirzebruch, F., Topological Methods in Algebraic Geometry, 3rd ed., Springer, 1966.

Bott-Tu Pass 4 — Agent C — N9. Leray-Hirsch theorem (collapse-at- $E_{2}$ of Leray-Serre); projective bundle theorem with Grothendieck's definition of Chern classes; splitting principle via iterated projectivisation; Whitney sum, Pontryagin, Chern character via Chern roots. Master Historical channels Borel-Hirzebruch 1958/1959 directly: universal form $H^(BG; \mathbb{Q}) = H^(BT; \mathbb{Q})^W$ gives a universal splitting principle.

Prerequisites

03.13.02 pending
03.06.04 pending
03.05.02 pending

Tier anchors

beginner: Bott-Tu §17 + §21 opening: 'pretend the bundle is a sum of line bundles'
intermediate: Bott-Tu §17 (Leray-Hirsch); Bott-Tu §21 (splitting principle); Milnor-Stasheff §16
master: Bott-Tu §17, §21; Borel-Hirzebruch 1958/59; Milnor-Stasheff §16; Husemoller §16

References

TODO_REF
Bott-Tu — Differential Forms in Algebraic Topology · §17 (Leray-Hirsch theorem); §21 (splitting principle for complex vector bundles); §22 (real splitting and Pontryagin classes)
TODO_REF
Hirsch — Un isomorphisme attaché aux structures fibrées · C. R. Acad. Sci. Paris 227 (1948), 1328–1330
TODO_REF
Borel-Hirzebruch — Characteristic classes and homogeneous spaces I · Amer. J. Math. 80 (1958), 458–538
TODO_REF
Borel-Hirzebruch — Characteristic classes and homogeneous spaces II · Amer. J. Math. 81 (1959), 315–382
TODO_REF
Milnor-Stasheff — Characteristic Classes · §16, splitting principle and unique-axiomatisation of Chern classes
TODO_REF
Husemoller — Fibre Bundles · §16, Leray-Hirsch theorem and applications
TODO_REF
Grothendieck — La théorie des classes de Chern · Bull. Soc. Math. France 86 (1958), 137–154

Lean module

Codex.SpectralSequences.LerayHirschSplitting

Reviewer

TBD

Estimated time

beginner: 20m
intermediate: 50m
master: 80m