Talk:Connected sum of topological manifolds (Ex)

From Manifold Atlas
Jump to: navigation, search

Part 1

Let M^n be an (oriented) topological n-manifold. Choose any two locally flat embeddings f,g:S^{n-1}\to M^n so that f(S^{n-1}) and g(S^{n-1}) bound topological balls in M^n. We will show that f(S^{n-1}) is topologically isotopic to g(S^{n-1}). This fact implies that connected sum of topological manifolds is well-defined, as the connected sum of M^n and N^n can then be defined by deleting a topological ball from each and then gluing the boundary, respecting the orientations on M^n and N^n (the gluing map is determined up to homotopy by the orientations, as an orientation-preserving automorphism of S^{n-1}. We will show this in Part 2).

So now let x and y be the center points of the balls bounded by F:=f(S^{n-1}) and G:=g(S^{n-1}). Let B_F and B_G be balls bounded by F/var/www/vhost/map.mpim-bonn.mpg.de/tmp/AppWikiTex/tex_jnhban and G respectively. Since F and G are locally flat, they have collar neighborhoods. Then by extending B_F and B_G along these collars, we find open balls U_F and U_G containing F and G (respectively).

Our current goal is to ambiently isotope B_G so that x and y agree. Since M^n is a manifold, it is path-connected. Then there exists some continuous map \gamma:[0,1]\to M^n so that \gamma(0)=y and \gamma(1)=x. Pick charts \{(O_\alpha,f_\alpha)\}_{\alpha\in\Lambda} covering M, where f_\alpha:O_\alpha\to\mathbb{R}^n is a homeomorphism. For each p\in [0,1], Let I_p be an open (in [0,1]) subinterval of [0,1] containing p so that \gamma(I_p) is contained in a single O_{\alpha} (call this O_p). Then \{I_p\}_{p\in[0,1]} is an open cover of [0,1], so from compactness there exists a finite collection p_1<\ldots<p_k\in[0,1] so that [0,1]=\cup_{i=1}^k I_{p_i}. If I_{p_i}\subset I_{p_j} for i\neq j, delete p_i from the collection \{p_1,\ldots, p_k\} and relabel. Let p_{k+1}:=1. In O_{p_1}\cong\mathbb{R}^4, choose a path \eta_{p_1} from x to \gamma(p_1) which has a tubular neighborhood \eta_{p_1}\times \mathring{D}^{n-1}. Ambiently isotope B_G along \eta_{p_1} so that \gamma(p_1)\in B_G. Then ambiently isotope the interior of B_G to take y along \eta_{p_1} so that y agrees with \gamma(p_1). Repeat for each p_i for i=2,\ldots, k until y agrees with \gamma(p_{k+1})=\gamma(1)=x.

Shrink U_G so that B_G\subset U_G\subset B_F. So in U_F\cong\mathbb{R}^n, F and G are nested disjoint (n-1)-spheres. By the annulus conjecture, F and G cobound an S^{n-1}\times I in U_F. Therefore, we may isotope F in U_F along the I direction of this product until F and G agree.

Part 2

Finally, we must show that there is a unique orientation-preserving automorphism of S^n up to topological isotopy (for each n). First we remark that if f_1:D^n\to D^n is a continuous map fixing \partial D^n pointwise, then f_1 is isotopic rel boundary to the identity map f_0:D^n\to D^n. (This is called Alexander's trick.) This can be shown constructively; consider the isotopy f_t(x)=\begin{cases}tf_1(x/t)&|x|\le t\\x&|x|>t,\end{cases} where D^n has radius 1.

Now let h:S^n\to S^n be an orientation-preserving automorphism. We will proceed by induction; as a base case consider n=0. There are two automorphisms of S^0, one of which is orientation-preserving. Then the claim holds trivially.

Now let n be arbitrary. Let C be a locally flat copy of S^{n-1} embedded in S^n which splits S^n into two n-balls. Then h(C) is another such copy of S^{n-1}. By the argument we used above to isotope F and G, we may isotope h (and hence h(C)) so that C and h(C) are disjoint and contained in an n-ball. By the annular embedding theorem, C and h(C) cobound an annulus, so we may isotope h to fix C setwise (isotoping h(C) through the I direction of the annulus). Call the closures of the n-ball components of S^n\setminus C A and B. If h reverses the orientation of C, isotope h to isotope one hemisphere of h(C) through A and the other through B so that h fixes C setwise and preserves the orientation of C.

By induction hypothesis, isotope h to fix C pointwise. Since h is orientation preserving, h(A)=A and h(B)=B. Then by Alexander's trick, h can be isotoped rel boundary to be the identity within A and B. Thus, h is isotopic to the identity.
Personal tools
Namespaces
Variants
Actions
Navigation
Interaction
Toolbox