Local Degree of a map between n-spheres

We are talking about singular homology.

Let $f: S^n \rightarrow S^n$ be a map. The degree $deg(f)$ is the unique integer such that under the identi cation $H_n(S^n) \cong \mathbb{Z}$, the map $f_*$ is given by multiplication by $deg(f)$. Generally speaking, I have the following problem: I can understand the general-abstract theorems (excision, the long exact sequence for relative homology, Mayer-Vietoris), but I don't know how to use them in practice. I can't see in many proofs that involve these theorems, why this generator $1$ goes to $1$ via the map $\partial$ of MV or why this element goes to the other via the map of excision.

Having that said, specifically, here, I have a problem understanding the following proof and consequently to compute the local degree in practice. I know that the upper-right iso is derived from LES for relative homology, the lower-right iso from excision theorem and the lower-left iso from the formula that relates the homology of a space with the homologies of its path-components (in the version for relative homology). I can also see why the generator $1$ of the upper-left $H_d(S^d)$ goes to deg(f) at the upper-right $H_d(S^d)$. But I can't see how we derive the rest of the relations.

Could please someone explain to me how we pass from the general theorem to the specific maps, in this setting?

I have spent many hours thinking of that without any significant progress and I would really appreciate your help.

To make sense of a proof of this nature, it isn't enough to know that there exists a map from $H_d(V, V \backslash \{ y \})$ to $H_d(S^d, S^d \backslash \{ y \})$, and that there exists an isomorphism between $H_d(S^d)$ and $H_d(S^d, S^d \backslash \{ y \})$, and so on. You really also need understand how all of these maps act on the cycles in the various homology groups.

Sometimes (but not always), it is possible to describe how a map acts on cycles in homology groups by identifying the map on homology as the map induced by a continuous function between the relevant topological spaces. Since we usually have a good idea of how to compose continuous functions between topological spaces, this will often enable us to work out how to compose the maps on the homology groups. In certain situations, this will also enable us to prove that certain maps between homology groups are identity maps, or zero maps.

As it happens, all of the maps in your diagram are induced by continuous functions between topological spaces:

• The horizontal maps are all induced by $f$ (or restrictions of $f$ to the appropriate spaces).

• The top-left vertical map is induced by the identity map $S^d \to S^d$. The same applies to the top-right vertical map. This is just how these maps are defined.

• The bottom-left vertical map is induced by the various identity maps $U_i \to U_i$.

• The middle-left vertical map is induced by the various inclusion maps $U_i \to S^d$. The bottom-right vertical map is induced by the inclusion map $V \to S^d$. The fact that these maps are induced by the relevant inclusion maps is a part of the statement of the excision theorem, and it is worth noting this!

It's clear that the diagram in your book really is a commutative diagram, because the corresponding maps between topological spaces all commute appropriately!

Now, each $H_d(U_i, U_i \backslash \{ x_i \} )$ is isomorphic to $\mathbb Z$, since $$H_d(U_i, U_i \backslash \{ x_i \} ) \cong H_d(S^d, S^d \backslash \{ x_i \}) \cong H_d (S^d) \cong \mathbb Z,$$ where the first equality is by excision and the second equality is by the LES for the pair $(S^d, S^d \backslash \{ x_i \})$.

Therefore, $$H_d (S^d, S^d \backslash \{ x_1, \dots, x_k \} ) \cong \oplus_i H_d(U_i , U_i \backslash \{ x_i \}) \cong \mathbb Z^{\oplus k}.$$

Now let us define the cycle $$(0, \dots, 1, \dots, 0) \in H_d (S^d, S^d \backslash \{ x_1, \dots, x_k \} )$$ (with the $1$ in the $i$th position) to be generator coming from the generator $1 \in H_d(U_i, U_i \backslash \{ x_i \} ) \cong \mathbb Z$.

Here's an important question we must address, if we're to make any progress: Given a cycle $$(a_1, \dots, a_k) \in H_d (S^d, S^d \backslash \{ x_1, \dots, x_k \} ),$$ is there a simple way to determine numbers $a_1, \dots, a_k$, if we don't already know them?

And here's my proposal: For each $i \in \{ 1, \dots, k \}$, define a natural map $$p_i : H_d(S^d, S^d \backslash \{ x_1, x_2, \dots, x_k \}) \to H_d(S^d, S^d \backslash \{ x_i \}),$$ to be the map on homology induced by the identity map $S^d \to S^d$.

Then for each $i$, $$p_i(a_1, \dots, a_k) = a_i \in H_d(S^d, S^d \backslash \{ x_i \}).$$

[Let's prove this carefully. First, take $$(1,0,\dots , 0) \in H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} )$$ and map it to $H_d(S^d, S^d \backslash \{ x_1 \})$ via $p_1$.

Since $(1,0,\dots, 0)$ is itself the image of the generator $1 \in H_d(U_1, U_1 \backslash \{ x_1 \})$, we know that $p_1(1,0,\dots, 0)$ is the image of $1 \in H_d(U_1, U_1 \backslash \{ x_1 \})$ under the composition, $$H_d(U_1,U_1 \backslash \{ x_1 \}) \to H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} ) \to H_d(S^d, S^d \backslash \{ x_1 \}).$$ The first map is induced by the inclusion $U_1 \to S^d$ and the second map is induced by the identity map $S^d \to S^d$, so the composition is induced by the inclusion $U_1 \to S^d$.

But we know that the map $H_d(U_1,U_1 \backslash \{ x_1 \}) \to H_d(S^d, S^d \backslash \{ x_1 \})$ induced by the inclusion $U_1 \to S^d$ is an isomorphism, by excision! So we conclude that $$p_1(1,0, \dots, 0) = 1 \in H_d(S^d, S^d \backslash \{ x_1 \}).$$

Okay, how about we take $$(1,0,\dots , 0) \in H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} )$$ as before, but this time, we map it via $p_2$ to $H_d(S^d, S^d \backslash \{ x_2 \})$?

The image $p_2(1,0 \dots, 0)$ is the same as the image of $1 \in H_d(U_1, U_1 \backslash \{ x_1 \})$ under the composition, $$H_d(U_1,U_1 \backslash \{ x_1 \}) \to H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} ) \to H_d(S^d, S^d \backslash \{ x_2 \}).$$ But if you think about it, this composition is the same as the composition, $$H_d(U_1,U_1 \backslash \{ x_1 \}) \to H_d(U_1,U_1) \to H_d(S^d, S^d \backslash \{ x_2 \}),$$ where the first map is induced by the identity $U_1 \to U_1$ and the second map is induced by the inclusion $U_1 \to S^d$. (Note that since $U_1 \subset S^d \backslash \{ x_2 \}$, the second map really is well-defined.) And why are these two compositions equal? Because both of these compositions are the maps on homology induced by the inclusion map $U_1 \to S^d$!

Of course, $H_d(U_1, U_1) = 0$, so it clear that $$p_2(1,0,\dots 0 ) = 0 \in H_d(S^d, S^d \backslash \{ x_2 \}).$$ This completes the proof of my claim.]

Right. Having done all this hard work, we're going to prove that the image of $1 \in H_d(S^d)$ under the top-left map in the diagram is the element $$(1,1, \dots 1) \in H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} ).$$ I believe this is the part of the proof that you weren't sure about.

By the claim that we just proved, we only have to verify that the image of $1 \in H_d(S^d)$ under the composition $$H_d (S^d) \to H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \} ) \overset{p_i} {\to} H_d(S^d, S^d \backslash \{ x_i \})$$ is the element $$1 \in H_d(S^d, S^d \backslash \{ x_i \}).$$

This is straightforward to show. The composition I wrote down is the map on homology induced by the identity map $S^d \to S^d$. But the map $H_d (S^d) \to H_d(S^d, S^d \backslash \{ x_i \})$ induced by the identity map $S^d \to S^d$ is precisely the map appearing in the LES for the pair $(S^d, S^d \backslash \{ x_i \})$, and this map is an isomorphism.

So $1 \in H_d(S^d)$ maps to $1 \in H_d(S^d, S^d \backslash \{ x_i \})$, and we're done.

To finish off, ${\rm deg}f$ is the image of $1 \in H_d(S^d)$ under the map $f_\star$. By the commutative diagram, this is the same as the image of $(1,1,\dots, 1) \in H_d(S^d, S^d \backslash \{ x_1, \dots, x_k \})$ under $f_\star$. And this is the same as the sums of the images of $1 \in H_d( U_i, U_i \backslash \{ x_i \})$ under $f_\star$.

Thus we have shown that $${\rm deg} f = \sum_i {\rm deg} f_i.$$

• [1/2] Wow, thank you very very much for your amazing answer! I went through of it but of course, I have to do it again in order to digest it in depth. I have two questions: 1) You are using at some points of this illustration that an isomorphism from $\mathbb{Z} \rightarrow \mathbb{Z}$ sends $1$ to $1$. Why can't be the case that sends $1$ to $-1$? – perlman Mar 27 '17 at 21:52
• [2/2] 2) Your arguments are very solid and clear and I think I can follow them step by step. However, I think I couldn't grasp the bigger picture especially because we introduce these $p_i$'s. Do we have to argue in this way in similar cases in order to make a solid argument? Additionally, is there a way I could have come up to define these $p_i$'s? Thank you very much again! – perlman Mar 27 '17 at 21:52
• Hi @MathewJames, yes, an isomorphism from $\mathbb Z \to \mathbb Z$ can either send $1 \mapsto 1$ or $1 \mapsto -1$. But these are really the same maps, up to a change of basis. Said another way, when you decide that it sends $1 \mapsto 1$ rather than $1 \mapsto -1$, you're picking a choice of orientation for your cycles. – Kenny Wong Mar 27 '17 at 22:02
• As for those $p_i$'s - it's like if you have a 3d vector $v = \sum c_i e_i$ and you want to get the coefficients, then you get them by projecting onto the coordinate axes. I tried to come up with a "projection" that works. and those $p_i$'s just happened to be what I came up with. – Kenny Wong Mar 27 '17 at 22:05
• In a way, those $p_i$ are quite natural. For example, the natural map $H_d(U_1, U_1 \backslash \{ x_1 \} \to H_d(S^d, S^d \backslash \{ x_1 \} )$ is clearly an isomorphism by excision, and the natural map $H_d(U_1, U_1 \backslash \{ x_1 \} \to H_d(S^d, S^d \backslash \{ x_2 \} )$ is clearly the zero map, because the $U_1$ gets completely "swallowed up" by the $S^d \backslash \{ x_2 \}$. – Kenny Wong Mar 27 '17 at 22:07