---
title: Spectral Sequence of a Bicomplex
url: https://www.emergentmind.com/topics/spectral-sequence-of-a-bicomplex
type: topic
---

# Spectral Sequence of a Bicomplex

The spectral sequence of a bicomplex is the spectral sequence attached to one of the two natural filtrations on the total complex of a double complex. In its standard cohomological form, one starts with a bigraded object \(V^{p,q}\) equipped with horizontal and vertical differentials \(d_h\) and \(d_v\) satisfying \(d_h^2=d_v^2=d_hd_v+d_vd_h=0\), forms the total complex with differential \(d=d_h+d_v\), and then filters by columns or by rows. The resulting first-quadrant spectral sequences compute the cohomology of the total complex through successive approximations whose first pages are vertical or horizontal cohomology and whose later pages encode the interaction of the two differentials [2508.14348].

## 1. Bicomplexes and totalization

A bicomplex is a bigraded vector space \(V^{**}\) with maps
\[
d_v:V^{p,q}\to V^{p,q+1},
\qquad
d_h:V^{p,q}\to V^{p+1,q},
\]
such that
\[
d_h^2=d_v^2=d_vd_h+d_hd_v=0.
\]
In the conventions emphasized in recent expositions, the focus is on first-quadrant bicomplexes, so \(p,q\ge 0\), and everything is cohomological: each differential raises degree by \(+1\) in its own direction [2508.14348].

The total complex is defined by
\[
\operatorname{Tot}(V^{**})^n=\bigoplus_{p+q=n}V^{p,q},
\qquad
d=d_h+d_v.
\]
With the anti-commutation relation built into the definition, one has
\[
d^2=d_h^2+d_v^2+d_hd_v+d_vd_h=0.
\]
Under this convention no extra Koszul sign is inserted in the total differential. On a homogeneous component \(x_{p,q}\), the total differential is simply \(d_hx_{p,q}+d_vx_{p,q}\) [2508.14348].

This setup isolates the central phenomenon of a bicomplex: the total cohomology is not obtained by applying \(H_h\) and \(H_v\) independently, but by organizing their interaction through filtrations of \(\operatorname{Tot}(V^{**})\). The spectral sequence is the systematic device that records this interaction page by page.

## 2. Column and row filtrations

A first-quadrant bicomplex carries two canonical decreasing filtrations on its total complex. The vertical, or column, filtration is
\[
F^s_{\mathrm{col}}\operatorname{Tot}(V^{**})^n
=
\bigoplus_{p+q=n,\ p\ge s}V^{p,q},
\]
and the horizontal, or row, filtration is
\[
F^t_{\mathrm{row}}\operatorname{Tot}(V^{**})^n
=
\bigoplus_{p+q=n,\ q\ge t}V^{p,q}.
\]
Both are canonically bounded in the first quadrant, so each gives a bounded first-quadrant spectral sequence converging to the cohomology of the total complex [2508.14348].

For a filtered cochain complex \((V^*,F)\), the associated graded object is
\[
E_0^{p,q}=\frac{F^pV^{p+q}}{F^{p+1}V^{p+q}}.
\]
Applied to the bicomplex filtrations, this yields
\[
E^{p,q}_{0,\mathrm{col}}\simeq V^{p,q},
\qquad
E^{p,q}_{0,\mathrm{row}}\simeq V^{q,p}.
\]
For the column filtration, the induced \(d_0\) is the vertical differential \(d_v\); for the row filtration, after swapping indices, \(d_0\) is the horizontal differential \(d_h\) [2508.14348].

The \(E_1\)-page is obtained by taking cohomology with respect to \(d_0\). Thus the column filtration gives
\[
E^{p,q}_{1,\mathrm{col}}\simeq H^q(V^{p,*}),
\]
the vertical cohomology in each fixed column, while the row filtration gives
\[
E^{p,q}_{1,\mathrm{row}}\simeq H^q(V^{*,p}),
\]
the horizontal cohomology in each fixed row. The first nontrivial differential after \(d_0\) is induced by the other differential: on the column spectral sequence, \(d_1\) is induced by \(d_h\) on vertical cohomology classes; on the row spectral sequence, \(d_1\) is induced by \(d_v\) on horizontal cohomology classes. Anti-commutation is exactly what makes these induced maps well defined on cohomology classes [2508.14348].

The standard \(E_2\)-identifications are
\[
E^{p,q}_{2,\mathrm{col}}\simeq H_h^p(H_v^q(V^{**})),
\qquad
E^{p,q}_{2,\mathrm{row}}\simeq H_v^p(H_h^q(V^{**})).
\]
These formulas are often the first computational target: one selects the filtration for which the \(E_1\)- or \(E_2\)-page is simplest, and then studies whether higher differentials can occur [2508.14348].

## 3. Higher differentials, convergence, and abutment

For a first-quadrant cohomological spectral sequence arising from a bicomplex, the \(r\)-th differential has bidegree \((r,1-r)\):
\[
d_r:E_r^{p,q}\to E_r^{p+r,q-r+1},
\]
and the next page is
\[
E_{r+1}^{p,q}
=
\frac{\ker(d_r:E_r^{p,q}\to E_r^{p+r,q-r+1})}
{\operatorname{im}(d_r:E_r^{p-r,q+r-1}\to E_r^{p,q})}.
\]
These bidegrees apply uniformly to the spectral sequences arising from either the column or the row filtration [2508.14348].

Boundedness of the filtration is the decisive convergence hypothesis. In the formulation used for first-quadrant bicomplexes, if the filtration on a cochain complex is bounded, then the resulting spectral sequence is bounded and converges to the cohomology of the filtered complex. Since the row and column filtrations of a first-quadrant bicomplex are canonically bounded, one obtains
\[
E^{p,q}_{r,\mathrm{col}}\Rightarrow H^{p+q}(\operatorname{Tot}(V^{**})),
\qquad
E^{p,q}_{r,\mathrm{row}}\Rightarrow H^{p+q}(\operatorname{Tot}(V^{**})).
\]
Equivalently,
\[
E_\infty^{p,q}
\simeq
\frac{F^pH^{p+q}(\operatorname{Tot}(V^{**}))}{F^{p+1}H^{p+q}(\operatorname{Tot}(V^{**}))},
\]
so the limiting page recovers the associated graded object of the induced filtration on total cohomology [2508.14348].

This identification makes clear what a spectral sequence does and does not provide. It determines the graded pieces of the filtered cohomology, not automatically the extensions between them. In bounded first-quadrant settings, there are also edge morphisms
\[
H^n(V^*)\to E_2^{0,n},
\qquad
E_2^{n,0}\to H^n(V^*),
\]
which relate the abutment to distinguished edge terms. These maps are especially useful when many rows or columns vanish [2508.14348].

The construction is functorial. A cochain map preserving the chosen filtration induces a morphism of spectral sequences, and for bicomplexes this applies in particular to morphisms commuting with \(d_h\) and \(d_v\). Functoriality underlies comparison arguments, vanishing theorems, and diagrammatic applications such as exactness proofs [2508.14348].

## 4. Indexing and sign conventions

Although the underlying mechanism is stable, conventions vary substantially across the literature. In the cohomological first-quadrant convention just described, both differentials raise degree, the total differential is \(d=d_h+d_v\), and the spectral sequence differential has bidegree \((r,1-r)\) [2508.14348].

A common alternative, used when the two structure maps commute rather than anti-commute, is to modify the total differential by a sign:
\[
d_{\mathrm{tot}}|_{C^{p,q}} = d_1 + (-1)^p d_0.
\]
In that framework the bicomplex is treated as a \(2\)-multicomplex with commuting \(d_0\) and \(d_1\), and the spectral sequence is still obtained from the row or column filtration, but the sign is transferred from the bicomplex identities into totalization [2501.13509].

Several papers use homological rather than cohomological grading. For right-half-plane bicomplexes \(X_{p,q}\) with \(d_h:X_{p,q}\to X_{p-1,q}\) and \(d_v:X_{p,q}\to X_{p,q-1}\), the filtration is increasing rather than decreasing, and the associated spectral sequence satisfies
\[
d^r:E^r_{p,q}\to E^r_{p-r,q+r-1},
\qquad
E^2_{p,q}\cong H^h_p(H^v_q(X))\Rightarrow H_{p+q}(\operatorname{Tot}(X)).
\]
This is the same formal mechanism after reindexing, but direct comparison of formulas requires care because the differential bidegrees and filtration directions are reversed [1802.07610].

There are also mixed conventions in which the bicomplex is cohomological, but the horizontal differential has bidegree \((-1,0)\) and the total differential on \(\operatorname{Tot}(A)\) is written with an explicit total-degree sign,
\[
d(a)_j=d_0(a_j)+(-1)^n d_1(a_{j+1}),
\]
so that the column filtration produces differentials
\[
\delta_r:E_r^{p,q}\to E_r^{p-r,q+1-r}.
\]
This convention is standard in model-categorical treatments of bicomplexes and filtered complexes [1805.00374].

The existence of these variants is not a discrepancy in substance. It reflects the fact that the spectral sequence of a bicomplex is determined by filtered totalization, while the displayed formulas depend on choices of grading, filtration, and sign normalization.

## 5. Computation and representative applications

A standard computation begins by choosing the filtration that simplifies the first page, writing down \(E_0\) as the associated graded object, taking \(d_0\)-cohomology to obtain \(E_1\), identifying \(d_1\) from the remaining differential, computing \(E_2\), and then using the bidegrees of \(d_r\) to determine which higher differentials can possibly be nonzero. One then tracks surviving classes to \(E_\infty\) and uses convergence and edge maps to recover information about \(H^*(\operatorname{Tot}(V^{**}))\) [2508.14348].

A classical diagrammatic application is Vakil’s spectral-sequence proof of the snake lemma. Starting from a commutative diagram with exact rows, one forms the associated bicomplex. For the row filtration, the \(E_1\)-page vanishes because the rows are exact, so the spectral sequence abuts to \(0\). The column filtration therefore also converges to \(0\). Its \(E_1\)-page consists of kernels and cokernels, and the \(E_2\)- and \(E_3\)-page analysis forces exactness everywhere except at the positions where the connecting morphism must appear, yielding the snake lemma exact sequence [2508.14348].

In local BRST cohomology, bicomplex spectral sequences are used in two distinct ways. For the bicomplex with \(d_h=\gamma\) and \(d_v=\delta\), the column filtration gives \(E_1^{p,q}=H^q(C^{p,\bullet},\delta)\), and \(\delta\)-acyclicity in positive \(-\)antifield number implies that the spectral sequence degenerates at \(E_2\), yielding
\[
H^g(s)\cong H^0_\gamma(H(\delta)).
\]
For the \((s,d)\)-bicomplex, horizontal filtration by form degree gives \(E_1=H(d)\); boundedness by form degree \(n\) forces degeneration by \(E_{n+1}\), and the resulting pages organize the descent equations that classify \(H(s)\) and \(H(s\mid d)\) [1008.4984].

In holomorphic Poisson geometry, the bicomplex
\[
A^{p,q}=\Gamma\!\big(\wedge^pT^{1,0}M\otimes\wedge^qT^{*(0,1)}M\big)
\]
has differentials \(\overline{\partial}\) of bidegree \((0,1)\) and \(d_\Lambda=\operatorname{ad}_\Lambda\) of bidegree \((1,0)\). The column filtration gives
\[
E_1^{p,q}=H^q(M,\mathcal O^p),
\qquad
d_1=\operatorname{ad}_\Lambda,
\]
and the spectral sequence converges to holomorphic Poisson cohomology. On \(2\)-step nilmanifolds with abelian complex structures and \(\dim\mathfrak c^{1,0}=1\), it always degenerates at \(E_2\); under additional conditions such as centrality of \(\Lambda_2\) and solvability of the equation \(\operatorname{ad}_{\Lambda_1}(p)=\overline{\partial}V\), it degenerates already at \(E_1\) [1611.08637].

## 6. Structural results, limitations, and extensions

For bounded double complexes over a field, there is a complete indecomposable decomposition into squares and zigzags. In this description, squares contribute neither Dolbeault-type cohomology nor total cohomology, odd zigzags contribute one-dimensional total cohomology classes and already degenerate at \(E_1\), and even zigzags of length \(2r\) contribute exactly one nonzero \(d_r\). Equivalently, nontrivial higher differentials occur precisely from even zigzags, and the spectral sequence degenerates on page \(r\) exactly when all even zigzags have length \(<2r\) [1812.00865]. A related indecomposable-summand viewpoint identifies the spectral arrows of a bounded bicomplex directly with odd-length zigzags of the relevant length, so that the differential \(d_r\) can be read off from the combinatorics of the decomposition [1911.02503].

The spectral sequence of a bicomplex also appears in homotopy-theoretic formalisms. For right-half-plane bicomplexes, one model structure has weak equivalences detected by totalization, while a second Cartan–Eilenberg model structure takes weak equivalences to be the maps inducing isomorphisms on
\[
E^2_{p,q}\cong H^h_p(H^v_q(X)).
\]
Strong convergence then implies that every \(E^2\)-equivalence is a total quasi-isomorphism [1802.07610]. Related model structures on bicomplexes and filtered complexes use \(E_r\)-quasi-isomorphisms as weak equivalences, and analogous constructions extend to \(N\)-multicomplexes, where one may require quasi-isomorphisms at prescribed pages of the two bicomplex spectral sequences arising from the row and column filtrations [1805.00374, 2501.13509].

The bicomplex case is also the first instance of the more general theory of polycomplexes. For a \(k\)-complex with anti-commuting differentials \(\partial_i\), the total differential is \(\sum_i\partial_i\), and there are at least \(2^k-2\) canonically bounded filtrations obtained by restricting single indices or sums of indices. Each such filtration yields a convergent spectral sequence computing \(H^*(\operatorname{Tot}(V))\), and its \(E_1\)-page is the cohomology of the total complex of an appropriate \((k-1)\)-complex slice [2508.14348].

A common misconception is that vanishing row and column homology forces the bicomplex to carry no meaningful information. In bounded first-quadrant settings the classical spectral sequences then indeed collapse to zero, but in unbounded contexts this need not capture the invariant of interest. In Tate-theoretic situations, rows and columns can be exact while the relevant homology remains nonzero; an alternative “middle homology”
\[
H(X)=Z'(X)\cap Z''(X)\big/d'(Z''(X))
\]
recovers that information when the classical row and column spectral sequences are identically zero [1108.1100].

Taken together, these developments show that the spectral sequence of a bicomplex is not merely a computational device for iterated cohomology. It is a structured interface between bigrading, filtration, totalization, and homological algebra, with manifestations ranging from elementary diagram chasing to BRST descent, holomorphic Poisson cohomology, model structures, and higher multicomplex theories.

Source: https://www.emergentmind.com/topics/spectral-sequence-of-a-bicomplex