# Comparing Spectral Sequences

There is a comparison theorem for spectral sequnces in Weibel's book (5.2.12) stating;

Assume $E_{p,q}$ and $\bar E_{p,q}$ converge to $H_*$ $\bar H_*$ respectively. Furthermore we have given a map $h: H_{*} \to \bar H_{*}$ compatible with a morphism $f$ of spectral sequences.

If $f^r: E^r_{p,q} \to \bar E^r_{p,q}$ is an isomorphism for all $p,q$ and some $r$ then $h$ is an isomorphism.

What I want to ask is what happens if we have a milder situation than isomorphism. For example if they just differ on the border?

To be precise let $E^2_{p,q}$ and $\bar E^2_{p,q}$ are two first quadrant spectral sequences converging to $H_*$ $\bar H_*$ respectively. Also there is a map $h$ compatible with a morphism $f$ of spectral sequences as above. Assume $E^2_{p,q} \cong \bar E^2_{p,q}$ if $q\neq0$ and $E^2_{p,0}$ vanishes. Can we calculate kernel and cokernel of $h$?

Thanks for your help.

-
Is it a first quadrant SS? –  Fernando Muro Aug 14 '12 at 10:06
Sorry, I fixed the question. –  Grilo Aug 14 '12 at 10:54
Consider the mapping cone of a chain map. It may be viewed as the total complex of a double complex with two rows. Say $\bar{E}^r_{p,q}$ is the spectral sequence for this double complex, living in the region $q\leq1$. Take for $E^r_{p,q}$ the spectral sequence corresponding with the $q=1$ row. Then the question is about the long exact sequence of homology for a mapping cone. A lot can happen. –  Wilberd van der Kallen Aug 14 '12 at 15:40
Wilberd -- I am not quite sure this works: the spectral sequence you mention (I usually think of it as having two non-zero columns rather than rows) is mapped to by the 1-column spectral sequence that computes the homology of the source chain complex shifted by 1, and is maps to by the 1-column spectral sequence that computes the homology of the target chain complex. In neither case is the 0-th column of $E_{p,q}$ (0-th row in your version) zero. –  algori Aug 14 '12 at 16:21
Grilo -- where do the differentials of your spectral sequences go? Since you use $E_{pq}$, I presume you consider something like homology (and not cohomology) spectral sequences of fibrations with differential $d^r$ going from $E_{pq}$ to $E_{p-r,q+r-1}$; but is this case it is a bit strange that it is the source $E_{pq}$ and not the target $\bar E_{pq}$ spectral sequence whose 0-th row is 0. –  algori Aug 14 '12 at 16:47

The cokernel is entirely due to $\bar{E}^\infty_{*,0}$ but the kernel is more mysteriuous.

First observe that if $g:C_\cdot\to D_\cdot$ is a chain map and $i$ is such that $g_i$ is surjective and $g_{i-1}$ is injective, then $H_i(g)$ is surjective and $H_{i-1}(g)$ is injective. (Exercise.)

Using this, one sees by induction on $r$ that $f^r_{p,q}$ is surjective for $q\geq1$ and injective (hence bijective) for $q\geq r-1\geq1$.

Now take $r=\infty$. One sees that $h$ hits all layers except the top one in the filtration of $\bar{H}_*$.

Let me include Grilo's formulas as I now believe they should read: We have exact sequences

$$0\to{\bar{E}}^{r+2}_{n+1,0} \to\bar{E}^{r+1}_{n+1,0}\to E^{r+2}_{n-r,r}\to\bar{E}^{r+2}_{n-r,r}\to0$$

and then $$0\to{\bar{E}}^{r+2}_{n+1,0} \to\bar{E}^{r+1}_{n+1,0}\to E^{\infty}_{n-r,r}\to\bar{E}^{\infty}_{n-r,r}\to0$$

Putting $r=n$ it becomes

$$0\to{\bar{E}}^{\infty}_{n+1,0} \to\bar{E}^{n+1}_{n+1,0}\to E^{\infty}_{0,n+1}\to\bar{E}^{\infty}_{0,n+1}\to0$$

or

$$H_{n+1}\to \bar{H}_{n+1} \to\bar{E}^{n+1}_{n+1,0}\to E^{\infty}_{0,n+1}\to\bar{E}^{\infty}_{0,n+1}\to0$$

So far so good.

-
Wilberd - Thanks a lot for the input. I think I also calculated the kernel. Since I am new to SS it is probable that I might missed something. Anyways here it goes: First note that $E^3_{n-1,1} = Ker (d^2_{n-1,1})$ and ${\bar E}^3_{n-1,1} = Ker ({\bar d}^2_{n-1,1})/ Im ({\bar d}^2_{n+1,0})$ and since $d^2_{n,1}$'s are the same we have the following exact sequence \begin{eqnarray} \bar E^2_{n+1,0} \to E^2_{n-1,1} \to {\bar E}^2_{n-1,1} \to 0 \nonumber \end{eqnarray} –  Grilo Aug 17 '12 at 6:55
Finally if we look at the last one we get \begin{eqnarray} 0 \to \bar E^\infty_{n+1,0} \to \bar E^{n+1}_{n+1,0} \to E^\infty_{0,n} \to {\bar E}^\infty_{0,n} \to 0 \nonumber \end{eqnarray} since $\bar E^\infty_{n+1,0} = \bar E^{n+2}_{n+1,0}$. Combining these one gets an exact sequence \begin{eqnarray} 0 \to \bar E^\infty_{n+1,0} \to \bar E^{2}_{n+1,0} \to H_n \to \bar H_n \nonumber \end{eqnarray} –  Grilo Aug 17 '12 at 6:57
And the interesting result comes combining with what you write about cokernel: we have a long exact sequence \begin{eqnarray} .... \to H_{n-1} \to \bar H_{n-1} \to \bar E^{2}_{n+1,0} \to H_n \to \bar H_n \to \bar E^{2}_{n,0} \to ... \nonumber \end{eqnarray} –  Grilo Aug 17 '12 at 7:02
I can not follow. You mean $\bar{E}^2_{n+1,0}\to E^3_{n-1,1}\to \bar{E}^3_{n-1,1}$? The map $E^2_{n-1,1}\to \bar{E}^2_{n-1,1}$ is an isomorphism after all. –  Wilberd van der Kallen Aug 17 '12 at 8:30
And how do you get from $\bar{E}^{r+1}_{n-r+2,0}$ to $\bar{E}^{n+1}_{n+1,0}$? –  Wilberd van der Kallen Aug 17 '12 at 8:54