Spectral Sequences — Serre and Leray-Hirsch

⚡ Key Takeaways

Serre spectral sequence: for a fibration $F \to E \to B$ with $B$ simply connected, there is a first-quadrant cohomology spectral sequence $E_2^{p,q} = H^p(B; H^q(F)) \Rightarrow H^{p+q}(E)$ . Leray–Hirsch: when the fiber cohomology is free and globally extends over $E$ , this degenerates and $H^*(E) \cong H^*(B) \otimes H^*(F)$ as an $H^*(B)$ -module. Applies across algebraic topology, geometry, and physics (gauge theory, string topology). Developed by Jean Leray (1946), refined by Jean-Pierre Serre (1951 thesis) and Guy Hirsch.

💡 The Core Insight

A fibration interpolates between base and fiber. Cohomology of the total space is "assembled" from cohomology of $B$ and $F$ , but the assembly is not a naive tensor product — there are higher-order corrections (the differentials $d_r$ ) encoding how fibers twist over the base. The spectral sequence is precisely the algebra that tracks these corrections page by page. When all differentials vanish (Leray–Hirsch), the corrections disappear and you get the clean tensor product. The art is computing differentials by exploiting multiplicative structure, naturality, and known edge homomorphisms.

DefinitionSpectral Sequence (cohomological, first quadrant)

A (cohomology) spectral sequence is a sequence of bigraded abelian groups (or $R$ -modules) $\{E_r^{p,q}\}_{r \geq r_0}$ together with differentials

1

\quad d_r : E_r^{p,q} \longrightarrow E_r^{p+r,\, q-r+1}, \qquad d_r \circ d_r = 0,

Derivation hint: the differential moves $r$ to the right and $r-1$ down, so the total degree $p+q$ increases by exactly $1$ ; squaring to zero is the defining axiom of a differential. such that the next page is the cohomology of the current one:

2

\quad E_{r+1}^{p,q} \;\cong\; \frac{\ker\left(d_r : E_r^{p,q} \to E_r^{p+r,q-r+1}\right)}{\operatorname{im}\left(d_r : E_r^{p-r,q+r-1} \to E_r^{p,q}\right)}.

Derivation hint: take cycles modulo boundaries on each page, the universal recipe for "homology of a complex." It converges to a graded module $H^n$ with a filtration $0 = F^{n+1}H^n \subseteq \cdots \subseteq F^0 H^n = H^n$ if for each $(p,q)$ the page stabilizes ( $E_r^{p,q} = E_\infty^{p,q}$ for large $r$ ) and

3

\quad E_\infty^{p,q} \;\cong\; F^p H^{p+q} / F^{p+1} H^{p+q}.

Derivation hint: $E_\infty$ records the associated graded of the filtration on the target; reconstructing $H^n$ requires solving extension problems among these quotients.

TheoremSerre Spectral Sequence

Let $\pi : E \to B$ be a fibration with fiber $F$ , where $B$ is path-connected. Suppose $\pi_1(B)$ acts trivially on $H^*(F; R)$ (automatic if $B$ is simply connected). Then there is a first-quadrant cohomology spectral sequence of $R$ -algebras with

4

\quad E_2^{p,q} \;=\; H^p\!\left(B;\, H^q(F; R)\right) \;\Longrightarrow\; H^{p+q}(E; R).

Derivation hint: filter $E$ by preimages of the skeleta of $B$ ; the associated graded is built from cellular cochains of $B$ valued in $H^*(F)$ , giving the $E_2$ identification via the universal coefficient and the cellular chain complex. The multiplicative structure is governed by

5

\quad d_r(x \cdot y) \;=\; d_r(x)\cdot y \;+\; (-1)^{p+q}\, x \cdot d_r(y), \qquad x \in E_r^{p,q},

Derivation hint: each page is a bigraded differential algebra; the Leibniz rule with Koszul sign comes from the cup product being a chain map and the total degree controlling the sign. with the product on $E_2$ being the cup product on $B$ tensored with the cup product on $F$ , twisted by sign $(-1)^{p q'}$ on $H^p(B;H^q F)\otimes H^{p'}(B;H^{q'}F)$ .

TheoremLeray–Hirsch

Let $\pi : E \to B$ be a fibration with fiber $F$ over a path-connected base, $R$ a principal ideal domain. Suppose: 1. $H^q(F; R)$ is a finitely generated free $R$ -module for each $q$ ; 2. there exist classes $c_1, \dots, c_k \in H^*(E; R)$ whose restrictions $\iota^*(c_j)$ to each fiber $F$ form a basis of $H^*(F; R)$ (where $\iota: F \hookrightarrow E$ ). Then $H^*(E; R)$ is a free $H^*(B; R)$ -module, and the map

6

\quad \Phi : H^*(B; R) \otimes_R H^*(F; R) \longrightarrow H^*(E; R), \qquad \Phi(b \otimes \iota^*(c_j)) = \pi^*(b) \smile c_j,

Derivation hint: the chosen classes provide a global section of the $E_2$ -page as a free module over $H^*(B)$ ; freeness forces all differentials to kill nothing, so $E_2 = E_\infty$ and the filtration splits as a tensor product. is an isomorphism of $H^*(B; R)$ -modules (and the Serre spectral sequence collapses at $E_2$ ).

ProofConstruction of the Serre Spectral Sequence via Skeletal Filtration

Assume $B$ is a CW complex with skeleta $B^{(0)} \subseteq B^{(1)} \subseteq \cdots$ . Pull back to filter the total space:

Step 1 — Filter the total space. Set $E_p = \pi^{-1}(B^{(p)})$ , giving an increasing filtration $E_0 \subseteq E_1 \subseteq \cdots \subseteq E$ . This induces a decreasing filtration on cochains $C^*(E)$ , hence a filtration on $H^*(E)$ .

Step 2 — Exact couple. The filtration yields long exact sequences of the pairs $(E_p, E_{p-1})$ :

7

\quad \cdots \to H^n(E_p, E_{p-1}) \to H^n(E_p) \to H^n(E_{p-1}) \to H^{n+1}(E_p, E_{p-1}) \to \cdots

Derivation hint: the long exact sequence of a pair, assembled over all $p$ into an exact couple $(D, E)$ with maps $i, j, k$ . Splicing these gives an exact couple; its derived couples are the pages of the spectral sequence.

Step 3 — Identify $E_1$ . The relative cohomology $H^{p+q}(E_p, E_{p-1})$ measures cells of dimension $p$ in $B$ with fiber cohomology. Since $B^{(p)}/B^{(p-1)}$ is a wedge of $p$ -spheres (one per $p$ -cell $e_\alpha^p$ ), and the fibration is locally trivial,

8

\quad E_1^{p,q} = H^{p+q}(E_p, E_{p-1}) \cong \bigoplus_{\alpha} H^q(F) = C^p_{\text{cell}}(B; H^q(F)),

Derivation hint: excision plus local triviality identify each relative term with cellular $p$ -cochains valued in the fiber cohomology; the local coefficient system is trivial by hypothesis.

Step 4 — Identify $d_1$ and pass to $E_2$ . The differential $d_1$ is the cellular coboundary on $B$ with coefficients in $H^q(F)$ . Therefore

9

\quad E_2^{p,q} = H^p\big(C^*_{\text{cell}}(B; H^q(F)), d_1\big) = H^p(B; H^q(F)).

Derivation hint: cohomology of the cellular cochain complex is exactly singular/CW cohomology with the given coefficient module.

Step 5 — Convergence. Because the filtration on each $H^n(E)$ is finite (first quadrant: $0 \le p, q$ , and $p+q = n$ bounds $p$ ), every $(p,q)$ stabilizes. The first-quadrant condition guarantees no differentials in or out for $r > \max(p+1, q+2)$ , so $E_\infty^{p,q} = E_r^{p,q}$ for large $r$ , and equation [3] holds. $\blacksquare$

∎

ProofLeray–Hirsch from Collapse

Given classes $c_1,\dots,c_k \in H^*(E)$ restricting to a fiber basis, define the $E_2$ -level lift. Each $c_j$ has a leading term in $E_\infty^{0, q_j}$ equal to $\iota^*(c_j) \in H^0(B; H^{q_j}(F))$ . The map

10

\quad H^*(B) \otimes H^*(F) \to E_2^{*,*}, \quad b \otimes \iota^*(c_j) \mapsto b \otimes \iota^*(c_j),

Derivation hint: $E_2^{p,q} = H^p(B)\otimes H^q(F)$ by the universal coefficient theorem since $H^q(F)$ is free over a PID; the chosen classes generate the fiber factor freely. is an isomorphism of $E_2$ -pages onto the free $H^*(B)$ -module on the $\iota^*(c_j)$ . Now the differentials: $d_r(c_j) \in E_r^{r, q_j - r + 1}$ . But $c_j$ lives in $H^*(E)$ — it is a permanent cycle (it survives to $E_\infty$ because it is a genuine global class). By the Leibniz rule [5] and $H^*(B)$ -linearity of $d_r$ (elements of $E_*^{p,0}$ pulled from $B$ are permanent cycles), every element $\pi^*(b)\cdot c_j$ is a permanent cycle. Since these span $E_2$ , all differentials vanish, so $E_2 = E_\infty$ . The associated graded of $H^*(E)$ is therefore free over $H^*(B)$ ; freeness kills the extension problem, and $\Phi$ is the resulting module isomorphism. $\blacksquare$

∎

mathematician · Jean-Pierre Serre | 1926–

Serre's 1951 doctoral thesis, Homologie singulière des espaces fibrés, applied Leray's nascent theory of spectral sequences to topological fibrations, computing homotopy groups of spheres and revolutionizing algebraic topology. He introduced the now-standard use of the path-loop fibration $\Omega X \to PX \to X$ to inductively relate $H^*(\Omega X)$ and $H^*(X)$ , proving for instance that $\pi_n(S^n)$ stabilizes and that all higher homotopy groups of spheres are finite except in the diagonal and the Euler-characteristic cases. For this and related work he received the 1954 Fields Medal at age 27 — the youngest ever at the time. The earlier groundwork was laid by Jean Leray while a prisoner of war (1941–1945), who invented sheaves and spectral sequences in captivity, and by Guy Hirsch, whose collapse criterion completed the picture.

🧠 Intuition

Think of a fibration as a "thick" base: above every point of $B$ sits a copy of $F$ . Cohomology classes on $E$ come in two flavors — those that vary along $B$ (pulled back via $\pi^*$ ) and those that live "fiber-wise." The $E_2$ -page $H^p(B; H^q(F))$ is the naive guess: combine $p$ -dimensional information from the base with $q$ -dimensional information from the fiber. But this guess is usually wrong: a fiber class might not extend globally because the fibers twist. The differential $d_r$ is the obstruction to extending a fiber class over more and more of the base. If $d_r(\xi) \neq 0$ , the would-be class $\xi$ is "killed" — it doesn't survive to $H^*(E)$ . Leray–Hirsch is the statement that if you can already exhibit honest global classes restricting to a fiber basis, there are no obstructions — the twisting is cohomologically invisible, and $E$ looks like a product as far as cohomology can tell. The geometric picture of the differential: $d_2$ takes a fiber class in the corner and pushes it $2$ columns right, $1$ row down — converting "fiber data of degree $q$ " into "base data of degree $q-1$ raised by $2$ ." This is the transgression, the topological analog of integrating a fiber form to get a base form.

Figure

The first-quadrant Serre spectral sequence:

E_2

-page with the

d_2

differential

ExampleCohomology of

\mathbb{CP}^n

via the Hopf-type fibration

S^1 \to S^{2n+1} \to \mathbb{CP}^n

Given: the fibration $S^1 \hookrightarrow S^{2n+1} \xrightarrow{\pi} \mathbb{CP}^n$ . We know $H^*(S^1; \mathbb{Z}) = \mathbb{Z}[t]/(t^2)$ with $|t| = 1$ , and $H^*(S^{2n+1}; \mathbb{Z}) = \mathbb{Z}\langle 1, s\rangle$ with $|s| = 2n+1$ . We want $H^*(\mathbb{CP}^n)$ .

Step 1 — $E_2$ -page. With $b = H^*(\mathbb{CP}^n)$ unknown, $E_2^{p,q} = H^p(\mathbb{CP}^n; H^q(S^1))$ . The fiber $S^1$ contributes only $q = 0$ and $q = 1$ , each a copy of $\mathbb{Z}$ . So the page has two nonzero rows.

Step 2 — Convergence forces collapse-then-cancellation. The total cohomology $H^*(S^{2n+1})$ is $\mathbb{Z}$ in degrees $0$ and $2n+1$ and zero in between. The only differential that can move between the two rows is $d_2 : E_2^{p,1} \to E_2^{p+2,0}$ .

Step 3 — $d_2$ must be an isomorphism in the middle range. For $E_\infty$ to vanish in total degrees $1, 2, \dots, 2n$ , the map

11

\quad d_2 : H^p(\mathbb{CP}^n; H^1(S^1)) \xrightarrow{\;\cong\;} H^{p+2}(\mathbb{CP}^n; H^0(S^1))

Derivation hint: anything surviving in those total degrees would appear in $H^*(S^{2n+1})$ , which is zero there; only $d_2$ can cancel both rows. must be an isomorphism for $0 \le p \le 2n-2$ , identifying $H^p$ with $H^{p+2}$ .

Step 4 — Induct. $H^0(\mathbb{CP}^n) = \mathbb{Z}$ (generator $x^0=1$ ); $d_2(t) = x$ defines $x \in H^2$ , and multiplicativity gives $d_2(t x^k) = x^{k+1}$ . Thus $H^{2k} = \mathbb{Z}\langle x^k\rangle$ for $0 \le k \le n$ and all odd degrees vanish, with $x^{n+1} = 0$ .

\boxed{H^*(\mathbb{CP}^n; \mathbb{Z}) \;\cong\; \mathbb{Z}[x]/(x^{n+1}), \quad |x| = 2.}

ExampleLeray–Hirsch: cohomology of a projective bundle

Given: a complex vector bundle $V \to B$ of rank $n$ , and its projectivization $\pi : \mathbb{P}(V) \to B$ with fiber $\mathbb{CP}^{n-1}$ . Let $\xi = \mathcal{O}(-1)$ be the tautological line bundle on $\mathbb{P}(V)$ , with first Chern class $c = c_1(\xi^\vee) \in H^2(\mathbb{P}(V))$ .

Step 1 — Verify Leray–Hirsch hypotheses. The fiber cohomology $H^*(\mathbb{CP}^{n-1}) = \mathbb{Z}[u]/(u^n)$ is free, with basis $1, u, \dots, u^{n-1}$ . The class $c$ restricts on each fiber to the generator $u$ of $H^2(\mathbb{CP}^{n-1})$ . Hence $1, c, c^2, \dots, c^{n-1}$ restrict to a fiber basis.

Step 2 — Apply the theorem. By Leray–Hirsch, $H^*(\mathbb{P}(V))$ is a free $H^*(B)$ -module:

12

\quad H^*(\mathbb{P}(V); \mathbb{Z}) \;\cong\; \bigoplus_{i=0}^{n-1} H^*(B) \cdot c^{\,i}\;=\;H^*(B)\langle 1, c, \dots, c^{n-1}\rangle.

Derivation hint: the $n$ permanent cycles $c^i$ generate freely over the base; collapse at $E_2$ gives the direct sum.

Step 3 — Ring structure (Grothendieck relation). As an algebra, the single relation among the $c^i$ is

13

\quad c^n + c_1(V)\,c^{\,n-1} + c_2(V)\,c^{\,n-2} + \cdots + c_n(V) = 0,

Derivation hint: $c^n$ must be expressible in lower powers since the module is free of rank $n$ ; matching to the Whitney/splitting formula identifies the coefficients as the Chern classes of $V$ . which is exactly the defining equation of Chern classes (Grothendieck's axiomatic approach).

\boxed{H^*(\mathbb{P}(V)) = H^*(B)[c]\big/\big(\textstyle\sum_{i=0}^n c_i(V)\,c^{\,n-i}\big).}

⚠️ Common mistake

$E_\infty \neq H^*(E)$ in general — only its associated graded. A frequent error is reading off $H^n(E) = \bigoplus_{p+q=n} E_\infty^{p,q}$ as groups. This is correct over a field, but over $\mathbb{Z}$ there are extension problems: $E_\infty$ gives only the graded pieces $F^p/F^{p+1}$ , and reassembling them may produce nontrivial extensions. For example a column $\mathbb{Z}/2$ stacked on $\mathbb{Z}/2$ could be $(\mathbb{Z}/2)^2$ or $\mathbb{Z}/4$ ; the spectral sequence alone does not decide.

Local coefficients matter. If $B$ is not simply connected and $\pi_1(B)$ acts nontrivially on $H^*(F)$ , then $E_2^{p,q} = H^p(B; \mathcal{H}^q(F))$ uses local (twisted) coefficients, not ordinary cohomology. Forgetting this gives wrong answers (e.g., the Klein bottle as an $S^1$ -bundle over $S^1$ ).

Differentials are not automatic. Knowing $E_2$ does not tell you the differentials; you must determine them by extra input — multiplicativity, naturality with respect to maps of fibrations, known target cohomology, transgression, or Steenrod operations.

Counterexample

Leray–Hirsch can fail without the global extension condition. Consider the nontrivial $S^1$ -bundle $S^1 \to S^3 \xrightarrow{\text{Hopf}} S^2$ . The fiber $H^*(S^1) = \mathbb{Z}\langle 1, t\rangle$ is free, so condition (1) holds. But there is no class $c \in H^1(S^3)$ restricting to the fiber generator $t$ , because $H^1(S^3) = 0$ . Condition (2) fails. Indeed Leray–Hirsch would predict $H^*(S^3) \cong H^*(S^2) \otimes H^*(S^1)$ , giving nonzero $H^1$ and $H^2$ — contradicting $H^*(S^3) = \mathbb{Z}\langle 1, s_3\rangle$ . The truth: $d_2 : E_2^{0,1} \to E_2^{2,0}$ is an isomorphism (the Euler class is nonzero), the sequence does not collapse, and $E$ is far from a product.

connection

The Serre spectral sequence sits within a web of homological machinery. It is a special case of the Leray Spectral Sequence of a continuous map and its sheaf-theoretic pushforwards $R^q\pi_* \mathcal{F}$ , with $E_2^{p,q} = H^p(B; R^q\pi_*\mathcal{F})$ . Both descend from the abstract Spectral Sequence of a Filtered Complex via an Exact Couple. In homological algebra the analog is the Grothendieck Spectral Sequence for a composite of derived functors, and the Lyndon-Hochschild-Serre Spectral Sequence is its incarnation for group extensions $1 \to N \to G \to Q \to 1$ , computing $H^*(G)$ from $H^*(Q; H^*(N))$ . The transgression in Serre's sequence is dual to the boundary map in the Wang Sequence (fibrations over spheres) and the Gysin Sequence (sphere bundles), both of which are the two-row Serre spectral sequence written as a long exact sequence. Leray–Hirsch is the engine behind the Splitting Principle and the construction of Chern Classes and Stiefel-Whitney Classes.

🔬 Where this appears

⏳ History

- 1941–1945 — Jean Leray, prisoner of war at Oflag XVII-A in Austria, invents sheaves, sheaf cohomology, and spectral sequences (originally called "suites spectrales") to do topology without revealing his applied expertise to his captors.

- 1946 — Leray publishes his theory in the Comptes Rendus, introducing the spectral sequence of a continuous map.

- 1947–1950 — Henri Cartan's seminar at the ENS systematizes and disseminates Leray's ideas; Koszul, Cartan, and Eilenberg recast spectral sequences in algebraic language.

- 1948 — Guy Hirsch formulates the collapse/product criterion now bearing his name (with Leray).

- 1951 — Jean-Pierre Serre's thesis Homologie singulière des espaces fibrés applies the machine to fibrations of topological spaces, computing homotopy and homology of fiber spaces; the modern "Serre spectral sequence" is born.

- 1952 — William Massey introduces the exact couple formalism, giving the cleanest derivation of spectral sequences.

- 1954 — Serre awarded the Fields Medal, largely for this work.

- 1957 — Grothendieck's Tôhoku paper recasts spectral sequences in the language of derived functors; the Grothendieck spectral sequence generalizes Leray's.

- 1962 — Atiyah and Hirzebruch develop the spectral sequence for generalized cohomology theories (

K

-theory), extending the Serre paradigm.

open-problem

Determining differentials remains an art, not an algorithm. For specific spaces (sphere spectra, classifying spaces of exotic groups, iterated loop spaces) the higher differentials $d_r$ for $r \ge 3$ are genuinely hard and sometimes unknown. The Adams spectral sequence and Adams–Novikov spectral sequence — variants that compute stable homotopy groups of spheres — have differentials that are the subject of ongoing research; the determination of $\pi_*^s$ in high degrees (e.g. via the motivic Adams spectral sequence, Isaksen–Wang–Xu) is an active frontier, and many differentials are only known through heavy machine computation.

Multiplicative extension problems beyond the additive ones (reconstructing the ring structure, not just the groups, of $H^*(E)$ from $E_\infty$ ) require additional structure (Steenrod operations, Massey products, $A_\infty$ -data) and have no general solution.

Convergence subtleties for non-first-quadrant or unbounded spectral sequences (conditional vs. strong convergence, $\lim^1$ obstructions, Boardman's theory) are delicate, and naive use can give wrong answers in the half-plane or whole-plane settings arising in stable homotopy.

Spectral Sequences — Serre and Leray-Hirsch

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also