The Salamander Lemma

At last we come to it: the greatest diagram lemma of all time.

The Salamander Lemma

For each piece of the double complex of the form

we obtain the following salamander-shaped diagram of mural maps:

Don't think this looks like a salamander? Fair enough. How about if we adjust it a bit ...

Still doesn't look like a salamander? One more try:

Okay, it's definitely a stretch. But I didn't name this lemma, so don't blame me! In any case, here's the (in)famous lemma:

The Salamander Lemma

If a diagram

is part of a double complex in an abelian category, then there is a 6-term exact sequence

ABhorBCChorD,

where the first morphism is the composition ABBhor and the last morphism is the composition ChorCD.

Similarly, if a diagram

is part of a double complex in an abelian category, then there is a 6-term exact sequence

ABvertBCCvertD,

where the first morphism is the composition ABBvert and the last morphism is the composition CvertCD.

I'll include the full proof of this at some point, but for now here is an official reference. Notice how it's not very long!

To assuage my guilt, I'll include at least a part of the proof. This is really one-eighth of the full proof, although all eight parts are very similar.

Suppose the diagram below is part of a double complex:

By composing the extramural map AB (which is given on representatives by ver) with the intramural map BBhor (which is the identity on coset representatives), we obtain a map ABhor. Composing that map with the intramural map BhorB, we obtain the 3-term sequence

ABhorB.

For absolute clarity, we note that if [a]A, then its image under the first map is [vert(a)]Bhor, and the image under the second map is [vert(a)]B. The bracket notation for cosets is hiding the different types of cosets involved in these maps.

In any case, let's prove this sequence is exact at Bhor. First suppose an element [b]Bhor is in the kernel of the intramural map BhorB. Recall that this intramural map is simply the identity on coset representatives, so we are now assuming the coset [b]B is the zero coset. By the definition of the donor, this implies b is the sum of elements in the images of vert and hor mapping to B. In other words, there are elements aA and eE (where EB is the map to the left of B) such that b=vert(a)+hor(e).

Next, recall that we began with an element [b]Bhor. These coset representatives are only defined up to elements in the image of EB, so [b]=[vert(a)].

Now notice that the element bker(hor), so applying hor to the previous equation gives

0=horvert(a)+horhor(e).

The last term on the right vanishes (since our diagram is part of a double-complex), so we can conclude horvert(a)=0. In other words, aker(outdiag), and so [a]A.

We've therefore proven [b]Bhor is in the image of ABhor. So, we have shown that the kernel of BhorB is contained in the image of ABhor.

Conversely, take any [a]A. Then the image of [a] under the map ABhor is [vert(a)]. The element vert(a) is (by definition!) in the image of vert and hence is zero in the donor B. Thus, the image of ABhor is contained in the kernel of BhorB.

Does this proof seem terrible to you?

If you read through the above proof, you probably agree that it's terribly presented. I think this is a case in which "clever" notation is actually obscuring things and making "easy" proofs appear convoluted. Some day I will re-do this proof and attempt to make things more streamlined and clear.

Extramural isomorphisms

Now let's see some corollaries of the Salamander Lemma. First let's focus on situations in which the mural maps are isomorphisms.

Extramural isomorphisms

If for some horizontal arrow hor:BC in the double complex one has Bhor=0=Chor, then the horizontal extramural map BC is an isomorphism.

Similarly, if for some vertical arrow vert:BC in the double complex one has Bvert=0=Cvert, then the vertical extramural map

BC

is an isomorphism.

The above is an immediate corollary of the corollary of the Salamander Lemma. For instance, the first statement follows from the exactness of the sequence below:

A0BC0D.

Intramural isomorphisms

As for the intramural maps, we have the following results. If it looks intimidating, note that it's covering four very similar situations, which we've described by which "edge" of the double-complex we're considering.

Intramural isomorphisms (Left Edge)

Suppose the diagram below is horizontally exact at B, i.e., ker(f)=0:

Then we have the following isomorphisms

Let's prove the above statement. First notice that we can extend the lower row to the left by inserting a zero object and arrow. Then, by our assumption, that bottom row is horizontally exact at 0 and B:

00BfC

By our extramural isomorphism them, we then have a horizontal isomorphism 0B. This proves B0, which we will use shortly.

Next consider the following part of the double-complex, where the indicates whatever object happens to be in that position. (Remember that we can always add zero objects and zero arrows, if necessary.)

By the Salamander Lemma, we have a six-term exact sequence

0hor0AAhorB.

Since 0B0, the end of this exact sequence is the 4-term exact sequence

0AAhor0.

This proves AAhor.

Similarly, now consider the following part of the double-complex:

By the Salamander Lemma, we have a six-term exact sequence

0AvertABBvert.

Once again, since 0B0, the beginning of this exact sequence is the 4-term exact sequence

0AvertA0.

This proves AvertA.


The following three analogous situations have similar proofs:

Intramural isomorphisms (Top Edge)

Suppose the diagram below is vertically exact at B, i.e., ker(f)=0:

Then we have the following isomorphisms

Intramural isomorphisms (Right Edge)

Suppose the diagram below is horizontally exact at A, i.e., im(f)=A:

Then we have the following isomorphisms

Intramural isomorphisms (Bottom Edge)

Suppose the diagram below is vertically exact at B, i.e., im(f)=A:

Then we have the following isomorphisms

Let's see how a multitude of famous "diagram lemmas" now all immediately follow.


Suggested next note

Diagram lemmas