03.13.01 · modern-geometry / spectral-sequences

Spectral sequences — exact couples, filtered complexes, double complexes

shipped3 tiersLean: partial

Anchor (Master): Bott-Tu §14; McCleary Ch. 1–3; Massey 1952; Leray 1946

Intuition [Beginner]

A spectral sequence is a sequence of approximations to a thing you cannot compute directly, where each approximation refines the last. Imagine you want to know the cohomology of a space that is too complicated to compute in one shot, but the space comes with a structure — a fibration, a filtration, a covering — that breaks the problem into pieces. A spectral sequence is the bookkeeping device that puts the pieces back together.

The starting page, called $E_{2}$ , is what you would compute if you ignored the interactions between pieces. The next page, $E_{3}$ , fixes the simplest interactions. Each successive page fixes finer interactions. Eventually the corrections stop, and the final page $E_{\infty}$ encodes the answer.

The picture is a stack of grids, one per page. Each grid has rows and columns, and an arrow on each page tells you what cancels with what. As you turn the page the arrows get longer, the surviving entries get fewer, and after enough pages the surviving entries assemble into the cohomology you wanted.

Visual [Beginner]

Picture a grid of boxes labelled by two coordinates $p$ and $q$ . On page two, short arrows of slope $- 1$ connect adjacent boxes. On page three, longer arrows of slope $- 1/2$ . On page four, even longer arrows. On each page some boxes are erased by the arrows; the entries that survive on every page from some point onward are the answer.

The key idea: each page's surviving boxes feed into the next page; eventually the arrows leave the grid and nothing changes. What remains is the cohomology, organised by the filtration that gave rise to the spectral sequence in the first place.

Worked example [Beginner]

Consider a square divided into two strips by a horizontal line. To compute the total area you can add the area of the two strips. Now consider a more complicated decomposition: the strips overlap, and to add their areas honestly you have to subtract the overlap, then add back the over-overlap, and so on.

A spectral sequence performs the same kind of inclusion-exclusion on cohomology. Take a sphere $S^{3}$ that fibres over $S^{2}$ with fibre $S^{1}$ — the Hopf fibration. Each piece on its own is easy: the cohomology of $S^{2}$ has a $1$ in degree $0$ and a $1$ in degree $2$ ; the cohomology of $S^{1}$ has a $1$ in degree $0$ and a $1$ in degree $1$ . The naive product $H^{*} (S^{2}) \cdot H^{*} (S^{1})$ has a $1$ in degrees $0, 1, 2, 3$ — that is what $E_{2}$ records.

But $S^{3}$ has cohomology only in degrees $0$ and $3$ . So the entries in degrees $1$ and $2$ on the $E_{2}$ page must cancel. The differential on $E_{2}$ pairs them up and erases them. After this cancellation, $E_{3} = E_{\infty}$ has entries only in degrees $0$ and $3$ , matching $S^{3}$ . The spectral sequence has done its bookkeeping.

Check your understanding [Beginner]

Formal definition [Intermediate+]

The cleanest way to introduce a spectral sequence is through Massey's notion of an exact couple. The double-complex setup follows.

Exact couple. A (cohomological) exact couple is a triple of bigraded abelian groups $(A, E)$ together with three maps $$ \begin{array}{c} A \xrightarrow{,i,} A \ \nwarrow{\scriptstyle k} \quad \swarrow{\scriptstyle j} \ E \end{array} $$ such that the triangle $A i A j E k A$ is exact at each vertex. Set $d := j \circ k : E \to E$ . Then $d \circ d = j \circ k \circ j \circ k = 0$ (since $k \circ j = 0$ by exactness), so $E$ is a complex with differential $d$ .

The derived couple $(A^{'}, E^{'})$ is constructed by: $$ A' = i(A) \subset A, \qquad E' = H(E, d) = \frac{\ker d}{\operatorname{im} d}. $$ The maps $i^{'}, j^{'}, k^{'}$ are induced from $i, j, k$ , and the triangle $A^{'} \to A^{'} \to E^{'} \to A^{'}$ is again exact. Iterating produces a sequence of couples $(A^{(r)}, E^{(r)})$ , conventionally indexed so that $E^{(2)}$ is the input. The bigraded objects $E_{r} := E^{(r)}$ together with their differentials $d_{r}$ form the spectral sequence of the exact couple.

Bigrading and differential. In the cohomological convention used throughout, $E_{r} = ⨁_{p, q} E_{r}^{p, q}$ and the differential has bidegree $$ d_r : E_r^{p, q} \to E_r^{p + r, q - r + 1}, $$ i.e. of bidegree $(r, 1 - r)$ . The total degree $n = p + q$ is preserved up to a $+ 1$ shift consistent with cohomological differentials.

Convergence. A spectral sequence is said to converge to a graded object $H^{*}$ , written $$ E_2^{p, q} ;\Rightarrow; H^{p + q}, $$ when there is a filtration $\dots \subset F^{p + 1} H^{n} \subset F^{p} H^{n} \subset \dots$ on each $H^{n}$ such that $$ E_\infty^{p, q} \cong \frac{F^p H^{p+q}}{F^{p+1} H^{p+q}}. $$ The page $E_{\infty}$ is the associated graded of $H^{*}$ with respect to the filtration. Two filtrations on the same complex generally give two distinct spectral sequences with the same abutment; we denote them $_{I} E_{r}$ and $_{I I} E_{r}$ when both are in play ^{[Bott-Tu §14]}.

Filtered complex. Let $K^{*}$ be a cochain complex with a decreasing filtration $\dots \supset F^{p} K^{*} \supset F^{p + 1} K^{*} \supset \dots$ by subcomplexes, exhaustive ( $⋃_{p} F^{p} K^{*} = K^{*}$ ) and Hausdorff ( $⋂_{p} F^{p} K^{*} = 0$ ). The short exact sequence $$ 0 \to F^{p+1} K^* \to F^p K^* \to F^p K^* / F^{p+1} K^* \to 0 $$ produces a long exact sequence in cohomology, and stitching these together yields an exact couple with $A^{p, q} = H^{p + q} (F^{p} K^{*})$ and $E_{1}^{p, q} = H^{p + q} (F^{p} K^{*} / F^{p + 1} K^{*})$ . The associated spectral sequence converges to $H^{*} (K)$ when the filtration is bounded (for instance, first-quadrant: $F^{0} = K$ and $F^{p} = 0$ for $p$ large).

Double complex. A double complex is a bigraded module $K^{p, q}$ with two anticommuting differentials $d^{'} : K^{p, q} \to K^{p + 1, q}$ and $d^{''} : K^{p, q} \to K^{p, q + 1}$ . The total complex is $Tot (K)^{n} = ⨁_{p + q = n} K^{p, q}$ with $d_{tot} = d^{'} + d^{''}$ . There are two natural filtrations: $$ {}I F^p \operatorname{Tot}(K) = \bigoplus{p' \geq p} K^{p', q}, \qquad {}{II} F^q \operatorname{Tot}(K) = \bigoplus{q' \geq q} K^{p, q'}, $$ giving spectral sequences $_{I} E_{r}$ and $_{I I} E_{r}$ with $$ {}I E_2^{p, q} = H^p{d'}(H^q_{d''}(K)), \qquad {}{II} E_2^{p, q} = H^p{d''}(H^q_{d'}(K)), $$ both abutting to $H^{*} (Tot (K))$ . The Čech-de Rham double complex of 03.04.11 is the prototype.

Key theorem with proof [Intermediate+]

Theorem (existence and convergence of the spectral sequence of a filtered complex). Let $K^ $b e a coc hain co m pl e x w i t hab o u n d e dd ecr e a s in g f i l t r a t i o n$ F^ $. T h er ee x i s t s a s p ec t r a l se q u e n ce$ {E_r^{p, q}, d_r}_{r \geq 0}$ with $$ E_0^{p, q} = F^p K^{p+q} / F^{p+1} K^{p+q}, \qquad d_0 = \overline{d}, \qquad E_1^{p, q} = H^{p+q}(F^p K / F^{p+1} K), $$ and $E_{r}^{p, q} = E_{\infty}^{p, q}$ for $r$ sufficiently large (depending on $p, q$ ). Moreover $$ E_\infty^{p, q} \cong F^p H^{p+q}(K) / F^{p+1} H^{p+q}(K), $$ where $F^p H^(K) := \operatorname{im}(H^(F^p K) \to H^(K))$.*

Proof. Define $$ Z_r^{p, q} = {x \in F^p K^{p+q} : dx \in F^{p+r} K^{p+q+1}}, \qquad B_r^{p, q} = F^{p+1} K^{p+q} \cap d(F^{p-r+1} K^{p+q-1}), $$ and set $$ E_r^{p, q} = \frac{Z_r^{p, q}}{Z_{r-1}^{p+1, q-1} + B_{r-1}^{p, q}}. $$ The element $Z_{r}^{p, q}$ records cocycles modulo $F^{p + r}$ , and $B_{r - 1}^{p, q}$ records elements that are coboundaries of things in $F^{p - r + 1}$ . The differential $d : K \to K$ restricts to maps $Z_{r}^{p, q} \to Z_{r}^{p + r, q - r + 1}$ and descends to $d_{r} : E_{r}^{p, q} \to E_{r}^{p + r, q - r + 1}$ .

Page advance. Direct verification shows $$ H(E_r, d_r) \cong E_{r+1}. $$ The kernel of $d_{r}$ on $E_{r}^{p, q}$ consists of classes of $x \in Z_{r}^{p, q}$ with $d x \in F^{p + r + 1}$ ; that is, $x \in Z_{r + 1}^{p, q}$ . The image of $d_{r}$ landing in $E_{r}^{p + r, q - r + 1}$ consists of classes of $d y$ for $y \in Z_{r}^{p, q}$ , i.e. $B_{r}^{p + r, q - r + 1}$ . Hence $$ \ker d_r / \operatorname{im} d_r = Z_{r+1}^{p, q} / (Z_r^{p+1, q-1} + B_r^{p, q}) = E_{r+1}^{p, q}. $$

Convergence. Boundedness of the filtration means there are integers $a (n) \leq b (n)$ with $F^{p} K^{n} = K^{n}$ for $p \leq a (n)$ and $F^{p} K^{n} = 0$ for $p > b (n)$ . For fixed $(p, q)$ and $r$ large enough that $p + r > b (p + q + 1)$ , the constraint $d x \in F^{p + r}$ in $Z_{r}^{p, q}$ becomes $d x = 0$ , so $Z_{r}^{p, q}$ stabilises to ${x \in F^{p} K^{p + q} : d x = 0}$ . Similarly $B_{r}^{p, q}$ stabilises. Thus $E_{r}^{p, q} = E_{r + 1}^{p, q} = \dots = E_{\infty}^{p, q}$ for $r$ sufficiently large. The stable value is $$ \frac{\ker(d) \cap F^p K^{p+q}}{\operatorname{im}(d) \cap F^p K^{p+q} + \ker(d) \cap F^{p+1} K^{p+q}} = \frac{F^p H^{p+q}(K)}{F^{p+1} H^{p+q}(K)}. \qquad \square $$

The same machinery, recast through the exact couple, is shorter but less transparent at first reading. Massey's exact-couple presentation isolates the algebraic structure that makes the inductive page-advance go through; the filtered-complex description gives the spectral sequence its content ^{[Massey 1952]}.

Synthesis. The exact-couple formalism is the foundational reason every filtered complex carries a spectral sequence. This is exactly Massey's 1952 abstraction of Leray's 1946–1950 work. The exact couple generalises the long exact sequence of a pair to a recursively-iterated structure.

Bridge. The construction here builds toward 03.13.02 (leray-serre spectral sequence and the gysin sequence), where the same data is upgraded, and the symmetry side is taken up in 03.13.03 (leray-hirsch theorem and the splitting principle for vector bundles). 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 3 (medium, symbolic). Exact couple from a short exact sequence.

Let $0 \to A^{*} f B^{*} g C^{*} \to 0$ be a short exact sequence of cochain complexes. Construct the exact couple whose spectral sequence is the Bockstein-type spectral sequence associated to this short exact sequence, and compute $E_{1}$ and $d_{1}$ .

Hint

Take $A^{p, q} = H^{p + q} (B)$ for all $p$ , and $E^{p, q} = H^{p + q} (C)$ . The map $j$ is induced by $g$ ; the map $k$ is the connecting map.

Answer

The relevant exact triangle is the long exact sequence in cohomology: $$ \cdots \to H^n(A) \to H^n(B) \to H^n(C) \xrightarrow{\delta} H^{n+1}(A) \to \cdots $$ Take $A^{p, q} = H^{p + q} (A)$ if we want a Bockstein-style filtration; alternately the more common setup is $E_{1}^{p, q} = H^{p + q} (C)$ with $d_{1} =$ composition $g_{*}^{- 1} \circ δ$ . The first differential is the connecting morphism, raising total degree by one. This is the simplest non-degenerate example of an exact couple and the Bockstein construction is its avatar.

Exercise 4 (medium, proof). Derived couple is exact.

Given an exact couple $(A, E)$ with maps $(i, j, k)$ , prove that the derived couple $(A^{'}, E^{'})$ with $A^{'} = i (A)$ , $E^{'} = H (E, j k)$ , $i^{'} = i ∣_{A^{'}}$ , and $j^{'}, k^{'}$ induced from $j, k$ , is exact.

Hint

Check exactness at each of the three vertices using the original exactness of $(A, E)$ and the definition of the derived maps.

Answer

For $j^{'}$ well-defined: take $a^{'} = i (a) \in A^{'}$ . Then $j^{'} (a^{'}) := [j (a)] \in H (E)$ . We check $j k \circ j = j \circ k j = 0$ (as $k j = 0$ ). So $j (a) \in ker (j k)$ , hence $[j (a)]$ makes sense; we need $j^{'}$ independent of the lift $a$ . If $a^{'} = i (a) = i (a^{''})$ , then $a - a^{''} \in ker i = im k$ , so $a - a^{''} = k (e)$ for some $e \in E$ , and $j (a) - j (a^{''}) = j k (e) \in im (j k)$ , so $[j (a)] = [j (a^{''})]$ .

For $k^{'}$ well-defined: take $[e] \in E^{'} = H (E)$ with $j k (e) = 0$ . Then $k (e) \in ker j = im i$ , so $k (e) = i (a)$ for some $a \in A$ , and we set $k^{'} ([e]) := i (a) \in A^{'}$ . Independent of lift $e$ : if $e = e^{''} + j k (e^{'''})$ , then $k (e) = k (e^{''}) + k j k (e^{'''}) = k (e^{''})$ (since $k j = 0$ ).

Exactness at $A^{'}$ (between $i^{'}$ and $j^{'}$ ): if $a^{'} = i (a) \in ker j^{'} = ker [j]$ , then $j (a) \in im (j k)$ , i.e. $j (a) = j k (e)$ for some $e$ , so $j (a - k (e)) = 0$ , hence $a - k (e) = i (b)$ for some $b$ by exactness of the original couple. Then $a^{'} = i (a) = i^{2} (b) + ik (e) = i^{2} (b) = i^{'} (i (b))$ , so $a^{'} \in im i^{'}$ . Similar checks at $E^{'}$ and at $A^{'}$ on the other side.

The full verification is bookkeeping, but the structural punchline is that exactness at each vertex of the derived couple follows from exactness at the corresponding vertex of the original. $□$

Exercise 5 (medium, symbolic). Two filtrations on a double complex.

Let $K^{p, q}$ be a first-quadrant double complex. Compute the $E_{2}$ pages of both spectral sequences $_{I} E$ and $_{I I} E$ in terms of $H_{d^{'}}$ and $H_{d^{''}}$ .

Hint

The first filtration is by columns ( $p$ ); $E_{1}$ is column-wise cohomology of $d^{''}$ , then $E_{2}$ takes $d^{'}$ -cohomology.

Answer

For the first filtration $_{I} F^{p} Tot (K) = ⨁_{p^{'} \geq p} K^{p^{'}, q}$ : $$ {}I E_1^{p, q} = H^q{d''}(K^{p, }), \qquad {}I E_2^{p, q} = H^p{d'}(H^q_{d''}(K^{, })). $$ For the second filtration $_{I I} F^{q}$ : $$ {}{II} E_1^{p, q} = H^p{d'}(K^{, q}), \qquad {}{II} E_2^{p, q} = H^q{d''}(H^p_{d'}(K^{*, *})). $$ Both abut to $H^{p + q} (Tot (K))$ . The two iterated cohomologies $H_{d^{'}} H_{d^{''}}$ and $H_{d^{''}} H_{d^{'}}$ are not generally equal as bigraded objects — they are only required to have the same total cohomology. Comparing the two is one of the principal techniques of the subject.

Exercise 6 (medium, proof). Collapse at $E_{2}$ .

Show that if a first-quadrant spectral sequence has $E_{2}^{p, q}$ concentrated in a single row ( $q = 0$ ) or a single column ( $p = 0$ ), then $E_{2} = E_{\infty}$ and $H^{n} ≅ E_{2}^{n, 0}$ (respectively $E_{2}^{0, n}$ ).

Hint

$d_{r} : E_{r}^{p, q} \to E_{r}^{p + r, q - r + 1}$ . If only $q = 0$ entries are non-zero, where can the differential go?

Answer

If $E_{2}^{p, q} = 0$ for $q \neq = 0$ , then on each later page the source $E_{r}^{p, 0}$ has $d_{r}$ landing in $E_{r}^{p + r, - r + 1}$ which is zero (negative $q$ -coordinate, outside the first quadrant). And the target receiving $d_{r}$ from $E_{r}^{p - r, r - 1}$ is also outside the row $q = 0$ for $r \geq 2$ . So all later differentials vanish, and $E_{2} = E_{3} = \dots = E_{\infty}$ . The filtration on $H^{n}$ then has only one non-zero graded piece in degree $p = n$ , so $H^{n} ≅ E_{2}^{n, 0}$ . The argument for the column case is symmetric. $□$

Exercise 7 (hard, symbolic). Five-term exact sequence.

Derive the five-term exact sequence $$ 0 \to E_2^{1, 0} \to H^1 \to E_2^{0, 1} \xrightarrow{d_2} E_2^{2, 0} \to H^2 $$ for any first-quadrant cohomological spectral sequence converging to $H^{*}$ .

Hint

In low total degrees, the filtration on $H^{n}$ has few non-zero pieces; explicitly identify $E_{\infty}^{p, q}$ for $p + q \leq 2$ in terms of $E_{2}$ and $d_{2}$ .

Answer

For $p + q = 1$ , the only candidates are $(1, 0)$ and $(0, 1)$ . The page $E_{3}^{1, 0}$ is the kernel of $d_{2} : E_{2}^{1, 0} \to E_{2}^{3, - 1} = 0$ , so $E_{3}^{1, 0} = E_{2}^{1, 0}$ , and the only $d_{3}$ landing in or leaving it goes to/from outside the first quadrant — so $E_{\infty}^{1, 0} = E_{2}^{1, 0}$ . Similarly $E_{3}^{0, 1}$ is $E_{2}^{0, 1}$ modulo the image of $d_{2} : E_{2}^{- 2, 2} = 0 \to E_{2}^{0, 1}$ , so $E_{3}^{0, 1} = E_{2}^{0, 1}$ , but on later pages $d_{r}$ may cancel parts of it; specifically, $E_{\infty}^{0, 1} = ker (d_{2} : E_{2}^{0, 1} \to E_{2}^{2, 0})$ .

The filtration on $H^{1}$ : $F^{0} H^{1} = H^{1} \supset F^{1} H^{1} \supset F^{2} H^{1} = 0$ , with quotients $E_{\infty}^{0, 1}$ in $F^{0} / F^{1}$ and $E_{\infty}^{1, 0}$ in $F^{1} / F^{2}$ . So $0 \to E_{2}^{1, 0} \to H^{1} \to E_{\infty}^{0, 1} \to 0$ is exact, i.e. $0 \to E_{2}^{1, 0} \to H^{1} \to ker (d_{2}) \to 0$ . Combined with $ker (d_{2}) ↪ E_{2}^{0, 1}$ and $coker (d_{2}) ↪ E_{\infty}^{2, 0} ↪ H^{2}$ , the five-term sequence follows. $□$

Exercise 8 (hard, proof). Comparison of two double-complex spectral sequences.

Let $K^{p, q}$ be a first-quadrant double complex with anticommuting differentials $d^{'}, d^{''}$ . Suppose $_{I} E_{2}^{p, q} = 0$ for all $q \geq 1$ . Prove that $H^{n} (Tot (K)) ≅ H_{d^{'}}^{n} (K^{*, 0})$ .

Hint

If only the row $q = 0$ survives at $E_{2}$ , the spectral sequence collapses there. Use Exercise 6.

Answer

The hypothesis says $_{I} E_{2}^{p, q} = H_{d^{'}}^{p} (H_{d^{''}}^{q} (K))$ vanishes for $q \geq 1$ . By Exercise 6, $_{I} E_{2} =_{I} E_{\infty}$ and the filtration on $H^{n} (Tot)$ has only one non-zero graded piece, in $(p, q) = (n, 0)$ . Thus $$ H^n(\operatorname{Tot}) \cong {}I E_2^{n, 0} = H^n{d'}(H^0_{d''}(K^{*, *})). $$ And $H_{d^{''}}^{0} (K^{p, *}) = ker (d^{''} : K^{p, 0} \to K^{p, 1})$ . Combined with the further hypothesis that the $d^{''}$ -cohomology vanishes for $q \geq 1$ (which says $H_{d^{''}}^{0} (K^{*, *})$ controls the lot), this iterated cohomology computes $H_{d^{'}}^{n} (K^{*, 0})$ . $□$

Lean formalization [Intermediate+]

[object Promise]

Mathlib's CategoryTheory.Triangulated.SpectralObject provides one abstract framework for spectral sequences (Verdier-Deligne style), and HomologicalComplex.SpectralSequence gives the spectral sequence of a filtered complex. The two-filtrations-on-a-double-complex setup, the explicit bidegree $(r, 1 - r)$ accounting on a bigraded module, and the convergence-from-bounded-filtration theorem are stub-level in Mathlib but not stated with the explicit bigrading required for Bott-Tu's pedagogy. The Lean module above declares the structures so downstream Codex units (Leray-Serre, Leray-Hirsch) can quote them.

Advanced results [Master]

Multiplicative structure. When $K^{*}$ is a differential graded algebra and the filtration is multiplicative ( $F^{p} \cdot F^{q} \subset F^{p + q}$ ), the spectral sequence inherits a multiplicative structure: each page $E_{r}$ is a bigraded ring, $d_{r}$ is a derivation, and the products on $E_{r + 1}$ are induced from those on $E_{r}$ . In the limit, $E_{\infty}$ is a bigraded ring whose product is the associated graded of the cup product on $H^{*} (K)$ . This is the structure that powers the spectral-sequence computation of cohomology rings of fibre bundles in 03.13.02 ^{[Bott-Tu §14, multiplicative structure]}.

Comparison theorem. A morphism of filtered complexes induces a morphism of spectral sequences. If the morphism induces an isomorphism on $E_{r}$ for some $r$ , it induces an isomorphism on all later pages and on the abutments. This is the technical lever behind the zigzag lemma for double complexes and behind proofs that two distinct spectral sequences computing the same thing must produce the same answer at some early page.

Edge homomorphisms. In a first-quadrant spectral sequence with $E_{2}^{p, q} \Rightarrow H^{p + q}$ , the edges $q = 0$ and $p = 0$ contribute edge homomorphisms: $$ H^n \twoheadrightarrow E_\infty^{n, 0} \hookrightarrow E_2^{n, 0}, \qquad E_2^{0, n} \twoheadrightarrow E_\infty^{0, n} \hookrightarrow H^n. $$ For the Leray-Serre spectral sequence of a fibration $F \to E \to B$ , the edge maps are $H^{*} (B) \to H^{*} (E)$ (the bundle pullback) and $H^{*} (E) \to H^{*} (F)$ (the fibre restriction). The five-term exact sequence of Exercise 7 splices these together in low degrees.

Transgression. A class $x \in E_{r}^{0, r - 1}$ that survives to $E_{r}$ has a differential $d_{r} (x) \in E_{r}^{r, 0}$ , called the transgression of $x$ . In the Leray-Serre setting, transgression assigns to a class on the fibre an obstruction class on the base; this is the mechanism behind the Hopf invariant calculation, the cohomology of $Ω S^{n}$ , and the entire Borel construction of $H^{*} (B G)$ from $H^{*} (G)$ . The transgression is not a homomorphism in general but a partial map, defined precisely on the elements that survive page after page.

Convergence subtleties. Boundedness of the filtration is the cleanest hypothesis. In the unbounded case, two flavours of convergence — strong (the filtration on $H^{*}$ is exhaustive and Hausdorff) and weak (only the associated graded is recovered) — can diverge. Pathologies appear in pro-finite filtrations (lim $^{1}$ obstructions) and in the Adams spectral sequence (where convergence is genuinely subtle). Bott-Tu's first-quadrant cohomological setting avoids these pathologies; McCleary's User's Guide treats them in detail.

Other spectral sequences in the wider zoo.

The Eilenberg-Moore spectral sequence computes $H^{*} (F)$ for the homotopy fibre $F$ of a map $f : X \to Y$ , with $E_{2} = Tor^{H^{*} (Y)} (H^{*} (X), Z)$ , dual to the bar construction. The Adams spectral sequence computes stable homotopy groups of spheres from $Ext$ groups in the Steenrod algebra: $E_{2} = Ext_{A}^{*, *} (H^{*} (X), F_{p}) \Rightarrow π_{*}^{s} (X)_{p}^{\land}$ . The Atiyah-Hirzebruch spectral sequence computes a generalised cohomology theory $h^{*} (X)$ from ordinary cohomology: $E_{2} = H^{p} (X; h^{q} (pt)) \Rightarrow h^{p + q} (X)$ . Each is an instance of the same algebraic machinery developed in this unit, specialised to a different filtered complex.

The Čech-de Rham prototype made explicit. The double complex $K^{p, q} = C^{p} (U; Ω^{q})$ on a good cover $U$ of a smooth manifold $M$ — developed in 03.04.11 — is the concrete prototype for which the abstract machinery of this unit is the algebraic shadow. The two filtrations on $K = Tot (C^{*} (U; Ω^{*}))$ produce the two spectral sequences $_{I} E_{r}$ and $_{I I} E_{r}$ , with $$ {}I E_2^{p, q} = H^p\delta(H^q_d(C^(\mathcal{U}; \Omega^))), \qquad {}{II} E_2^{p, q} = H^q_d(H^p\delta(C^(\mathcal{U}; \Omega^))), $$ where $δ$ is the Čech differential and $d$ is the de Rham differential. The first collapses on a good cover (each multiple intersection has vanishing positive de Rham cohomology by the Poincaré lemma, so the row spectral sequence dies), giving abutment $H_{dR}^{*} (M)$ . The second collapses to the Čech complex of the constant sheaf $\underline{R}$ (each column is exact above degree $0$ by the partition-of-unity argument), giving abutment $\overset{ˇ}{H}^{*} (U; \underline{R})$ . Both abut to $H^{*} (Tot (K))$ , hence the de Rham theorem $$ H^_{\mathrm{dR}}(M) \cong \check{H}^(\mathcal{U}; \underline{\mathbb{R}}) $$ on a good cover. The tic-tac-toe principle of 03.04.11 is the explicit zigzag that traces a class through the double complex from one abutment to the other; in spectral-sequence language, it is the explicit lift of an $E_{2}$ -class to a representative cocycle on the total complex. This is the calculation Bott and Tu invoke when they observe that the reader has already been doing spectral sequences without knowing it.

Full proof set [Master]

The page-advance theorem in detail. The key calculation is that $H (E_{r}, d_{r}) = E_{r + 1}$ , where $E_{r}^{p, q}$ is defined via the explicit nested cycle/boundary subgroups $Z_{r}^{p, q}$ and $B_{r}^{p, q}$ as in the proof of the main theorem. We expand the verification.

A class in $E_{r}^{p, q}$ is represented by $x \in F^{p} K^{p + q}$ with $d x \in F^{p + r} K^{p + q + 1}$ . The differential $$ d_r [x] = [dx] \in E_r^{p+r, q-r+1} $$ descends to a well-defined map: changing $x$ by an element of $Z_{r - 1}^{p + 1, q - 1}$ (a class with $d x^{'} \in F^{p + r}$ but lying in deeper filtration) leaves $d x$ unchanged modulo $F^{p + r + 1}$ , and changing by $B_{r - 1}^{p, q}$ (a coboundary $d y$ with $y \in F^{p - r + 1}$ ) makes $d x = d (x + d y) = d x + d^{2} y = d x$ , again unchanged.

The class $[x] \in E_{r}^{p, q}$ is in $ker d_{r}$ iff $d x \in F^{p + r + 1} K^{p + q + 1} + d (F^{p - r + 1} K^{p + q})$ , equivalently iff some representative $\tilde{x} = x - d y$ (for $y \in F^{p - r + 1}$ ) satisfies $d \tilde{x} \in F^{p + r + 1}$ . So $\tilde{x} \in Z_{r + 1}^{p, q}$ , and the class of $\tilde{x}$ in $E_{r}^{p, q}$ agrees with the class of $x$ . Hence $ker d_{r} = (Z_{r + 1}^{p, q} + Z_{r - 1}^{p + 1, q - 1} + B_{r - 1}^{p, q}) / (Z_{r - 1}^{p + 1, q - 1} + B_{r - 1}^{p, q})$ .

Similarly, the image of $d_{r}$ landing in $E_{r}^{p + r, q - r + 1}$ is $(B_{r}^{p + r, q - r + 1} + Z_{r - 1}^{p + r + 1, q - r} + B_{r - 1}^{p + r, q - r + 1}) / (Z_{r - 1}^{p + r + 1, q - r} + B_{r - 1}^{p + r, q - r + 1})$ .

Taking the quotient: $$ \frac{\ker d_r}{\operatorname{im} d_r}\bigg|{(p, q)} = \frac{Z{r+1}^{p, q}}{Z_r^{p+1, q-1} + B_r^{p, q}} = E_{r+1}^{p, q}. \qquad \square $$

Uniqueness up to canonical isomorphism. Different presentations of the same filtered complex (e.g., via the exact couple of the long exact sequence of $F^{p} / F^{p + 1}$ versus the explicit $Z, B$ formulation) yield canonically isomorphic spectral sequences. The functorial naturality of the construction in morphisms of filtered complexes is verified pagewise.

Multiplicativity. If $K^{*}$ is a differential graded algebra and $F^{p} \cdot F^{q} \subset F^{p + q}$ , then the bigraded $E_{r}$ inherits a product: represent $[x] \in E_{r}^{p, q}$ and $[y] \in E_{r}^{p^{'}, q^{'}}$ by $x \in Z_{r}^{p, q}$ and $y \in Z_{r}^{p^{'}, q^{'}}$ ; their product $x y$ lies in $F^{p + p^{'}} K^{p + p^{'} + q + q^{'}}$ with $d (x y) = (d x) y \pm x (d y) \in F^{p + p^{'} + r}$ , so $x y \in Z_{r}^{p + p^{'}, q + q^{'}}$ . The product $[x] [y] := [x y]$ is well-defined modulo $Z_{r}$ and $B_{r}$ , the differential $d_{r}$ is a derivation by Leibniz on $K$ , and the $E_{2}$ ring is a quotient of $E_{1} = ⨁ H_{d^{''}}^{q} (K^{p, *})$ .

Convergence in the bounded case. For a bounded filtration $a (n) \leq F^{p} K^{n} \leq b (n)$ , the cycles and boundaries $Z_{r}^{p, q}, B_{r}^{p, q}$ stabilise to $Z_{\infty}^{p, q} = F^{p} K^{p + q} \cap ker d$ and $B_{\infty}^{p, q} = F^{p} K^{p + q} \cap im d$ once $r > b (p + q + 1) - p$ . The limit $E_{\infty}^{p, q}$ is then the indicated quotient of cycles by the next filtration of cycles plus boundaries, which is $F^{p} H^{p + q} / F^{p + 1} H^{p + q}$ as claimed.

Connections [Master]

Čech-de Rham double complex 03.04.11 — the prototype double complex, with the row spectral sequence collapsing to de Rham cohomology of the manifold (Bott-Tu §8) and the column spectral sequence collapsing to Čech cohomology of the constant sheaf $R$ on a good cover (Bott-Tu §10). By conn:440.exact-couple-double-complex, exact-couple spectral sequence equivalent to the Čech-de Rham double-complex spectral sequence (equivalence). The two filtrations on the Čech-de Rham double complex give two spectral sequences; both are concrete instances of the abstract exact-couple machinery (Massey 1952). The reader has already done several spectral sequences before meeting the abstract definition.
Leray-Serre spectral sequence 03.13.02 — the principal application: given a fibration $F \to E \to B$ with $B$ simply connected, the Leray-Serre spectral sequence has $E_{2}^{p, q} = H^{p} (B; H^{q} (F))$ converging to $H^{p + q} (E)$ . By conn:441.serre-finiteness, Serre spectral sequence is the filtered-complex SS of a fibration's singular cochain filtration (specialisation). The exact-couple/double-complex machinery developed here is what Serre invokes.
Leray-Hirsch theorem 03.13.03 — when the spectral sequence collapses at $E_{2}$ (because cohomology of the fibre extends to cohomology of the total space), the answer is the Künneth product. Leray-Hirsch is the collapse-at- $E_{2}$ theorem, and the splitting principle for Pontryagin and Chern classes is its direct consequence.
De Rham cohomology 03.04.06 — the receiving end of the Čech-de Rham spectral sequence. The associated-graded structure on $H_{dR}^{*} (M)$ from a good-cover filtration gives the Čech-Mayer-Vietoris computation.
Sheaf cohomology 04.03.01 — Leray's general spectral sequence $H^{p} (B; R^{q} f_{*} F) \Rightarrow H^{p + q} (E; F)$ for a continuous map $f$ , with the Leray-Serre case as the fibre-bundle specialisation. Foundation-of relation.
Eilenberg-MacLane spaces 03.12.05 — the Serre spectral sequence of the path-loop fibration $Ω K (A, n) \to P K (A, n) \to K (A, n)$ relates $H^{*} (K (A, n))$ to $H^{*} (K (A, n - 1))$ inductively, supplying both the Cartan-Serre computation of cohomology operations and the Steenrod algebra.
Pontryagin and Chern classes 03.06.04 — the Borel-Hirzebruch derivation of characteristic classes via the splitting principle uses the Leray-Hirsch theorem on the projective bundle, which uses the Leray-Serre spectral sequence, which uses the general filtered-complex machinery developed here. The vertical line of dependence runs through this unit.

The Čech-de Rham equivalence is the cleanest example a beginner can compute. The Leray-Serre application is the structural payload. The general spectral sequence is the algebraic machinery that lets the same computation be done in twenty different settings without re-deriving the bookkeeping each time.

Throughlines and forward promises. The abstract spectral sequence is the foundational language for filtered-complex computations. We will see Leray-Serre specialise this to a fibration in 03.13.02; we will see Leray-Hirsch be the collapse-at- $E_{2}$ specialisation in 03.13.03; we will later see the Atiyah-Hirzebruch spectral sequence apply the same machinery to K-theory. This pattern recurs throughout: the Čech-de Rham SS, Mayer-Vietoris SS, Eilenberg-Moore SS, and Adams SS are all instances of this filtered-complex framework. The foundational reason the reader has already done several spectral sequences before meeting the abstract definition is exactly Bott-Tu's pedagogical inversion (§14): introduce concrete two-filtration computations on the Čech-de Rham double complex first, then abstract. Putting these together with Massey's exact-couple framework gives the canonical algebraic backbone. The bridge between geometric filtration and algebraic computation is exactly the page-by-page derived-couple iteration; this is precisely the abstraction Leray needed to organise his 1946–1950 results, and Massey 1952 made transparent.

Historical & philosophical context [Master]

The spectral sequence was invented in the most singular circumstances of any twentieth-century mathematical idea. Jean Leray, captured in 1940 during the fall of France, spent the next five years interned at Oflag XVII-A, a prisoner-of-war camp in Edelbach, Austria. Leray was a leading authority on partial differential equations and fluid mechanics — he had proved the existence of weak solutions to the Navier-Stokes equations in 1934 — and he feared that if his Nazi captors learned this, he would be coerced into wartime aerodynamic research. So he hid: he told his captors he was a topologist. To make the story credible, Leray turned himself into a topologist. In the camp, with a circle of fellow prisoner-mathematicians, he organised a université en captivité and developed in two parallel sweeps what became sheaf theory and the spectral sequence.

The first published note appeared as Leray's 1946 L'anneau d'homologie d'une représentation (C. R. Acad. Sci. Paris 222), where he announced what was at the time called the anneau spectral of a continuous map. The full development came in 1950 with two long papers in the Journal de Mathématiques Pures et Appliquées, in which Leray gave the spectral sequence of a continuous map $f : E \to B$ in terms of the cohomology of the base with coefficients in the higher direct image sheaves $R^{q} f_{*} F$ . The construction was sheaf-theoretic from the start; Leray's insight was that the cohomology of a fibration could be computed by first computing the cohomology of the fibre at each point (giving a sheaf on the base) and then computing the cohomology of the base with these coefficients. The intermediate stages were the pages of the spectral sequence.

Leray's original presentation was difficult. The pages were defined by an opaque sequence of subquotients; the differentials were given by a recursive procedure that resisted intuition. William Massey made the construction transparent in 1952 with Exact couples in algebraic topology (Ann. of Math. 56), where he isolated the algebraic structure — the exact couple — that drives the page-by-page construction. Massey's exact couple is now the standard route into the subject; Leray's original derivation, while equivalent, is rarely taught.

The further development came rapidly. Jean-Pierre Serre's 1951 PhD thesis Homologie singulière des espaces fibrés. Applications (Ann. of Math. 54) recast Leray's spectral sequence for the singular cohomology of a fibre bundle, made the bidegree explicit, and used the result to compute the homotopy groups of spheres in low dimensions (in particular, $π_{4} (S^{3}) = Z /2$ , a calculation impossible without the spectral sequence). Henri Cartan and his 1950s seminar at the École Normale Supérieure made the spectral sequence the central tool of sheaf cohomology and homological algebra; the Cartan-Eilenberg textbook of 1956 codified the filtered-complex framework. Armand Borel in 1953 used Leray's spectral sequence on classifying spaces of compact Lie groups to compute $H^{*} (B G)$ as the invariants of the Weyl group on $H^{*} (B T)$ , supplying the foundation for the modern theory of characteristic classes.

The pedagogical reframing that Bott and Tu offer in their 1982 Differential Forms in Algebraic Topology §14 deserves separate mention. Bott-Tu observed that by the time a student of differential forms reaches §14, they have already done several spectral sequences without realising it: the de Rham theorem comparing de Rham and Čech cohomology is the collapse of one spectral sequence; the Mayer-Vietoris computation is the simplest non-degenerate filtered-complex argument; the Künneth formula is a collapsing spectral sequence. Bott-Tu therefore introduce the spectral sequence not as a piece of abstract homological algebra but as the natural generalisation of computations the reader has already performed. This pedagogical move makes the subject accessible in a way that Leray's 1950 papers and even Massey's 1952 paper do not. The introduction of the exact couple is delayed until after the double-complex spectral sequence has been worked through on concrete examples; the abstract algebra is then a clarification of the concrete machinery rather than a prerequisite for it.

The story of Leray, the prison camp, and the simultaneous birth of sheaves and spectral sequences is documented in Mawhin's 2000 obituary essay and in McCleary's introduction. It remains one of the most striking instances of mathematics done under duress, and one of the few cases in which an intellectual deception (Leray pretending to be a topologist) produced an enduring discipline.

Bibliography [Master]

Leray, J., "L'anneau d'homologie d'une représentation", C. R. Acad. Sci. Paris 222 (1946), 1366–1368.
Leray, J., "L'anneau spectral et l'anneau filtré d'homologie d'un espace localement compact et d'une application continue", J. Math. Pures Appl. 29 (1950), 1–80, 81–139.
Massey, W. S., "Exact couples in algebraic topology", Ann. of Math. 56 (1952), 363–396.
Serre, J.-P., "Homologie singulière des espaces fibrés. Applications", Ann. of Math. 54 (1951), 425–505.
Cartan, H. & Eilenberg, S., Homological Algebra, Princeton University Press, 1956. §XV.
Bott, R. & Tu, L. W., Differential Forms in Algebraic Topology, Graduate Texts in Mathematics 82, Springer, 1982. §14.
McCleary, J., A User's Guide to Spectral Sequences, 2nd ed., Cambridge University Press, 2001. Ch. 1–3.
Mawhin, J., "Jean Leray (1906–1998)", Gazette des Mathématiciens 84 (2000), 9–43.
Hatcher, A., Spectral Sequences, online supplement to Algebraic Topology, Cornell University, 2004.

Bott-Tu Pass 4 — Agent C — N7. Greenfield chapter 13-spectral-sequences/. Exact couples (Massey 1952), filtered complexes, double complexes, two filtrations $(_{I} E,_{I I} E)$ , convergence symbol $\Rightarrow$ , page- $r$ differential bidegree $(r, 1 - r)$ . Notation decisions #15, #29, #30 introduced here. Master Historical channels Leray 1946 directly: prison-camp origin at Oflag XVII-A, simultaneous birth of sheaves and spectral sequences, Massey's exact-couple clarification. Soft-syncs on Agent A's N3 (03.04.11); Čech-de Rham double complex referenced as the prototype.

Prerequisites

03.04.06
03.04.11

Used in

03.13.02
03.13.03

Tier anchors

beginner: Bott-Tu §14 opening intuition: a spectral sequence as a sequence of approximations
intermediate: Bott-Tu §14; McCleary *A User's Guide to Spectral Sequences* Ch. 1–2
master: Bott-Tu §14; McCleary Ch. 1–3; Massey 1952; Leray 1946

References

TODO_REF
Bott-Tu — Differential Forms in Algebraic Topology · §14, the spectral sequence of a double complex
TODO_REF
Leray — L'anneau d'homologie d'une représentation · C. R. Acad. Sci. Paris 222 (1946), 1366–1368
TODO_REF
Leray — L'anneau spectral et l'anneau filtré d'homologie d'un espace localement compact et d'une application continue · J. Math. Pures Appl. 29 (1950), 1–80, 81–139
TODO_REF
Massey — Exact couples in algebraic topology · Ann. of Math. 56 (1952), 363–396
TODO_REF
McCleary — A User's Guide to Spectral Sequences · Ch. 1–3, exact couples and filtered complexes
TODO_REF
Cartan-Eilenberg — Homological Algebra · §XV, spectral sequences of filtered complexes
TODO_REF
Hatcher — Spectral Sequences (online supplement) · §1, exact couples and Bockstein
TODO_REF pending
Sigmundur Gudmundsson — Personal communication of Leray's prison-camp story · Oflag XVII-A, 1940–1945; published correspondence in Mawhin 2000 · see docs/catalogs/NEED_TO_SOURCE.md#leray-prison-camp

Lean module

Codex.SpectralSequences.SpectralSequence

Reviewer

TBD

Estimated time

beginner: 25m
intermediate: 60m
master: 95m