5
$\begingroup$

I have to deal with unbounded filtrations and want to use the conditional convergence of spectral sequences and the results from

[1]: J. Michael Boardman, Conditionally Convergent Spectral Sequences, March 1999 (http://hopf.math.purdue.edu/Boardman/ccspseq.pdf)

The article uses cohomological spectral sequences derived from the exact couple coming from a cochain complex $C$ and a decreasing filtration $F$ of $C$. The system of inclusions is $$A^s := H(F_s C) \leftarrow A^{s+1}$$ and the pages are denoted by $E^s_r$ for $s\in \mathbb{Z}$ and $r\in \mathbb{N}$ ($r$ is the page number and $s$ the ``filtration degree''). The symbol $A^\infty$ denotes the limit and the symbol $A^{-\infty}$ the colimit. The symbol $RA^\infty$ denotes the right derived module of the limit. I basically work over $\mathbb{R}$.

The following are the two theorems (or their parts) from [1] which I am interested in:

Theorem 6.1 (p.19): Let $C$ be a filtered cochain complex. Suppose that \begin{equation}\label{Eq:Exit}\tag{C1} E^s = 0\quad\text{for all } s>0.\end{equation} If $A^\infty = 0$, then the spectral sequence converges strongly to $A^{-\infty}$.

Theorem 7.2 (p.21): Let $f: C \rightarrow \bar{C}$ be a morphism of filtered cochain complexes and suppose that $E^s$, resp. $\bar{E}^s$ converge conditionally to $A^{-\infty}$, resp. $\bar{A}^{-\infty}$. Suppose, moreover, that \begin{equation}\tag{C2} E^s = \bar{E}^s = 0\quad\text{for all }s<0. \end{equation} If $f$ induces the isomorphisms $E^\infty\simeq \bar{E}^\infty$ and $RE^\infty\simeq R\bar{E}^\infty$, then it induces the isomorphism $H(C)\simeq H(\bar{C})$.

Let me introduce the standard (degree shifted) bigrading on $E_r$ and visualize $E_r^{s,d}$ as sitting at the coordinate $(s,d)$ in plane. The differentials are then $$ d_r: E_r^{s,d}\rightarrow E_r^{s+r,d-r+1}. $$ My questions are the following:

  1. How does Theorem 6.1 generalize if (C1) is replaced by the following condition of exiting differentials? $$ E_r \text{ sit in a half-plane and if we fix any coordinate }(s,d), \text{ then all but finitely many }d_r\text{ starting at }(s,d)\text{ leave the half-plane.}$$

  2. How does Theorem 7.2 generalize if (C2) is replaced by the following condition of entering differentials? $$ E_r \text{ sit in a half-plane and if we fix any coordinate }(s,d), \text{ then all but finitely many }d_r\text{ ending at }(s,d)\text{ start outside of the half-plane.}$$

The author of [1] addresses the questions as follows:

  1. On p.19, Chapter 6 in brackets right before Theorem 6.1:

    ...The results generalize appropriately, as all arguments can be carried out degreewise; the main difficulty is to find notation that would help rather than hinder the exposition

  2. On p.20, Chapter 7 in brackets a couple of paragraphs before Theorem 7.2:

    ...The results remain valid when appropriately modified, as all arguments can be carried out degreewise; the difficulty is to find notation that helps rather than hinders.

How do these theorems generalize precisely? Has it been done anywhere? Thanks!

P.S. I come from differential geometry and am not familiar with the proof methods for spectral sequences at all. I use it merely as a black box.

$\endgroup$
5
$\begingroup$

After a long life in preprint form, Boardman's paper was published in the conference proceedings celebrating his 60th birthday:

  \bib{MR1718076}{article}{
       author={Boardman, J. Michael},
       title={Conditionally convergent spectral sequences},
       conference={
          title={Homotopy invariant algebraic structures},
          address={Baltimore, MD},
          date={1998},
       },
       book={
          series={Contemp. Math.},
          volume={239},
          publisher={Amer. Math. Soc., Providence, RI},
       },
       date={1999},
       pages={49--84},
       review={\MR{1718076}},
       doi={10.1090/conm/239/03597},
    }

In the $(s,d)$-bigraded case, you can replace Boardman's (left half-plane) condition that $E_1^{s,d} = 0$ for $s > 0$ (or $s > s_0$ for some fixed integer $s_0$) by the (upper half-plane) condition that $E_1^{s,d} = 0$ for all $d < 0$ (or $d < d_0$ for some fixed integer $d_0$).

Similarly, you can replace his (right half-plane) condition that $E_1^{s,d} = 0$ and $\bar E_1^{s,d} = 0$ for $s < 0$ (or $s < s_0$ for some fixed integer $s_0$) by the (lower half-plane) condition that $E_1^{s,d} = 0$ and $\bar E_1^{s,d} = 0$ for $d > 0$ (or $d > d_0$ for some fixed integer $d_0$).

This adjustment is often enough. Do you need to refer to other half-planes than those bounded by a horizontal or a vertical line?

EDIT: The OP added some questions, including one about the case of a right half-plane cohomological bicomplex $(B^{i,j}, d_h, d_v)$, where $B^{i,j} = 0$ for $i<0$, $d_h : B^{i,j} \to B^{i+1,j}$ and $d_v : B^{i,j} \to B^{i,j+1}$. Suppose that $Z^{i,j} \subset B^{i,j}$ is such that $d_h(Z^{i,j}) = 0$, and filter the total complex $$ (C, d) = (\bigoplus_{i,j} B^{i,j}, d_h + d_v) $$ by $$ F^s = \bigoplus_{j-i>s} B^{i,j} \oplus \bigoplus_{j-i=s} Z^{i,j} $$ for all integers $s$. Here $C$ and $F^s$ are graded, with $B^{i,j}$ and $Z^{i,j}$ in degree $i+j$. Then $(F^s, d)$ is a subcomplex of $(C, d)$, and contains $(F^{s+1}, d)$ as a further subcomplex. We get an exact couple in the usual way, with $A^{s,t}$ and $E_1^{s,t}$ equal to the degree $s+t$ parts of $H(F^s, d)$ and $H(F^s/F^{s+1}, d)$, respectively.

I claim that $\lim_s F^s = 0$ and $\lim^1_s F^s = 0$. This can be checked one degree at a time, since $F^s = 0$ in degrees less than $s$. It follows (see Boardman's Theorem 9.2) that $\lim_s A^s = 0$ and $\lim^1_s A^s = 0$, so the spectral sequence is conditionally convergent to the colimit $G = H(C, d)$.

Furthermore, $F^s/F^{s+1}$ and $E^1_s$ are concentrated in degrees $\ge s$, corresponding to $t\ge0$, so this is an upper half-plane cohomological spectral sequence with exciting differentials. Hence it is strongly convergent to $G$, by the modified form of Boardman's Theorem 6.1 that I mentioned above.

To prove the modified form, one does as Boardman says. Let $F^s G$ be the image of $H(F^s, d)$ in $G$. One must check that the filtration $\{F^s G\}_s$ of $G$ is complete Hausdorff and exhaustive, and that the natural inclusion $F^s G/F^{s+1} G \to E_\infty^s$ is an isomorphism. Both claims can be checked one degree at a time, and for each degree the proof of Theorem 6.1(a) carries over. (I do not have Cartan--Eilenberg at hand: I do not recall if they spelled this out.)

$\endgroup$
  • $\begingroup$ Thank you very much for your answer! Switching $s$ and $d$ helps for showing strong convergence of the horizontal filtration of a cohomological right half-plane bicomplex. But I have another example: Consider the direct-sum total complex of a cohomological half-plane bicomplex $B^{ij}$ (I draw it in the right half-plane, the vertical differential points upstairs and the horizontal to the right), and let $Z^{ii} \subset B^{ii}$ be subspaces closed under the horizontal differential. Let $F_k = \bigoplus_{i + j > k} B^{ij} \oplus \bigoplus_i Z^{ii}$ be the diagonal filtration. What now? $\endgroup$ – Pavel Jul 23 at 18:43
  • $\begingroup$ Also, please, could you comment on the proof of your statement? Is it, in fact, important that the groups in a half-plane vanish or is the vanishing of the differentials fundamental? P.S. Thanks for the better reference! $\endgroup$ – Pavel Jul 23 at 18:47
  • $\begingroup$ (1) By "let $Z^{ii} \subset B^{ii}$ be subspaces closed under the horizontal differential", do you mean that the horizontal differential maps $Z^{ii}$ to zero? (2) In the definition of $F_k$, is the direct sum with $Z^{ii}$ for all $i$ intended to mean the internal sum in $\bigoplus_{i,j} B^{i,j}$? (3) It looks as if the vertical differential might not map $F_k$ to itself. Don't you want $F_k$ to be a subcomplex with respect to the total differential? $\endgroup$ – John Rognes Jul 23 at 22:04
  • $\begingroup$ (1) I mean $d_h Z^{ii} = 0$. (2) It is the internal sum (I imagine replacing the diagonal with Z^{ii} and discarding all $B^{ij}$'s below). (3) The vertical differential points upstairs, and hence it preserves $F_k$, as far as I can see. Therefore, it should be invariant under the total differential. $\endgroup$ – Pavel Jul 23 at 22:08
  • $\begingroup$ Your answers to (2) and (3) suggest that $i+j>k$ is not the intended condition in the definition of $F_k$. Do you mean $j > i > k$ or something like that? $\endgroup$ – John Rognes Jul 23 at 22:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.