Complements Of Locally Flat Submanifolds Are Finite CW Complexes

Abstract.

We show that if $Y$ is a compact topological manifold and $X$ is a locally flat submanifold, then the complement $Y-X$ is homotopy equivalent to a finite CW complex. This is a direct proof, and does not rely on much of the theory of topological manifolds.

1. Introduction

If $Y$ is a smooth manifold while $X$ is a compact smooth submanifold, then $Y-X$ is homotopy equivalent to a finite CW complex: equipping $Y$ with a Riemannian metric $g$, the normal vector bundle of $X$ yields a closed disk bundle $D$ of $X$ such that $Y-D^{\mathsf{o}}$ is a compact smooth manifold (thus homotopy equivalent to a finite CW complex, by Theorem 3.5 of [Mil63]) while $Y-X$ deformation retracts onto $Y-D^{\mathsf{o}}$.

Attempting to generalize this fact to the case where $X,Y$ are topological manifolds, one requires that $X$ be a locally flat (or tame) submanifold of $Y$. The reason is that if one takes $X$ to be the Alexander Horned Sphere along with its interior and $Y=S^{3}$, then $\pi_{1}(Y-X)$ is not finitely generated (by [BF50]), and so $Y-X$ cannot be homotopy equivalent to a finite CW complex. The main result to be proven in this paper is:

Theorem 1.1.

Suppose $Y^{n}$ is a compact topological manifold while $X^{d}$ is a compact locally flat submanifold. Then $Y-X$ is homotopy equivalent to a finite CW complex.

It is indeed true that if $X$ has a closed disk bundle, then $Y-X$ would be homotopy equivalent to a finite CW complex. This fact will be proven in Section 5, and is listed below:

Proposition 1.2.

Assume the same hypotheses as Theorem 1.1. If $X$ has a closed disk bundle in $Y$, then $Y-X$ is homotopy equivalent to a finite CW complex.

The proof of Theorem 1.1 thus seems to reduce to proving Proposition 1.2 and showing that $X$ has a closed disk bundle. However, closed disk bundles need not exist in general, seen by the following examples:

• [RS67] shows that if one takes $Y=S^{29}$, then there exists a compact locally flat submanifold $X$ that does not admit a normal bundle.

• Theorems 1, 3 of [Bro66] gives an example where $X$ is locally flat and has a normal bundle, but does not have a closed disk bundle.

• Theorem 4 of [Hir67] gives examples of locally flat $X^{4}\subset S^{7}$ and $S^{4}\subset M^{7}$ that do not admit closed disk bundles.

Even though $X$ need not have a closed disk bundle, the stable uniqueness theorem of [KS77] (also see (B) of [Hir66] and Section 5 of [Mil68]) shows that for some $N\in\mathbb{N}$, $X\times 0$ has a normal bundle in $Y\times\mathbb{R}^{N}$; by this the corollary of Proposition 2.2 of [Bro66] shows that, after incrementing $N$ by $1$, $X\times 0$ has a closed disk bundle in $Y\times\mathbb{R}^{N}$ where $Y\times\mathbb{R}^{N}$ deformation retracts onto $Y\times[0,1]^{N}$ while fixing $X-0$. The key of the proof of Theorem 1.1 thus lies in arguing that the case for the pair $(Y,X)$ reduces to that for the pair $(Y\times[0,1]^{N},X\times 0)$. This is done by using the sum formula of Wall’s finiteness obstruction discussed in [Sie65], where regular open neighbourhoods discussed in [Sie73] are used to show that $Y-X$ is finitely dominated, allowing the consideration of Wall’s finiteness obstruction.

Remark 1.3.

Our proof of Theorem 1.1 avoids a potential proof relying on several cases and much of the theory of topological manifolds. An outline of this proof is:

• When $n-d\leq 2$: we use the same argument as presented in Section 2.1.

• When $n-d\geq 3$ and $n\geq 5$: we show that $X$ admits a mapping cylinder neighborhood $\text{Cyl}(f)=((W\times[0,1])\sqcup X)/\sim$ with $(w,1)\sim f(x)$ where $f:W\to X$. For $n\geq 6$, this is already done in [Ped77] and [Qui79], and that of [Qui79] can supposedly be generalized to $n\geq 5$ using [Qui82].

• When $n=4$ and $d=1$: We use [FQ90] to obtain a normal vector bundle of $X$ in $Y$, by which we obtain a closed disk bundle of $X$ in $Y$ and apply Proposition 1.2.

Henceforth we assume that $Y$ is connected, as it only has finitely many components by compactness. To be able to use the sum formula in [Sie65], we require $n-d\geq 3$ so that the inclusion $Y-X\xhookrightarrow{}Y$ induces an isomorphism of fundamental groups. Thus, in Section 2, we first show that Theorem 1.1 holds for $n-d\leq 2$, and then discuss the assumption that $n-d\geq 3$ to prepare for the proof of Theorem 1.1 under this assumption, which the later sections of the paper concern. It should be noted that the proof of Proposition 1.2 in Section 5 does not rely on any consideration of codimension, so we will be able to use Proposition 1.2 in the proof of Theorem 1.1 in low codimension.

It is worth noting that the aforementioned stable existence theorem requires a lack of boundary of $X,Y$. We will thus first assume until Section 5 that $\partial X=\emptyset$, and in Section 5, we will address the case where $\partial X\neq\emptyset$ and $\partial X\subseteq\partial Y$.

Acknowledgements

We thank Alexander Kupers for helpful discussions and comments.

Contents

1. 1 Introduction
2. 2 Proof for $n-d\leq 2$ and preparation for $n-d\geq 3$
3. 3 $Y-X$ is finitely dominated
4. 4 Reduction to $(Y\times\mathbb{R}^{N})-(X\times 0)$
5. 5 Assembly of Proof for Theorem 1.1

2. Proof for $n-d\leq 2$ and preparation for $n-d\geq 3$

2.1. Proof of Theorem 1.1 when $1\leq n-d\leq 2$

When $n-d=1$, the fact that $X$ is locally flat in $Y$ implies that $X$ admits a bicollar, by Theorem 3 of [Bro62]. This bicollar is then a closed disk bundle, since $[-1,1]$ is the $1$-dimensional closed unit disk, and so Proposition 1.2 finishes the proof of Theorem 1.1.

For the case where $n-d=2$: it suffices to show that $X$ admits a normal vector bundle in $Y$, since that normal vector bundle gives us a closed disk bundle, by which Proposition 1.2 proves Theorem 1.1. If $n\neq 4$, then the existence of a normal vector bundle is given by Theorem A of [KS74]; if $n=4$, then the existence of a normal vector bundle is given by Section 9.3 of [FQ90] (Freedman-Quinn defines a “normal bundle” as a normal vector bundle).

2.2. Proof of Theorem 1.1 when $n-d=0$

In this case, $\partial X$ is compact locally flat submanifold of $Y$, with codimension $1$. Since we have proven Theorem 1.1 for codimension $1$, we thus deduce that $Y-\partial X$ is homotopy equivalent to a finite CW complex, so that the path components of $Y-\partial X$ are all homotopy equivalent to finite CW complexes. Since $Y-X$ and $X$ only have finitely many components, let us denote the path components of $Y-X$ as $C_{1},...,C_{N}$ and denote the path components of $X$ as $D_{1},...,D_{M}$. It then suffices to show that $D^{\mathsf{o}}_{1},...,D^{\mathsf{o}}_{M},C_{1},...,C_{N}$ are the path components of $Y-\partial X$, in which case we would be able to conclude that $Y-X$ is homotopy equivalent to a finite CW complex.

We first note that the union of $D^{\mathsf{o}}_{1},...,D^{\mathsf{o}}_{M},C_{1},...,C_{N}$ is $Y-\partial X$, and this union is disjoint. Additionally, these sets are open in $Y-\partial X$ due to the identity $n=d$ along with local flatness of $X$. As $D^{\mathsf{o}}_{1},...,D^{\mathsf{o}}_{M}$ are the path components of the interior of $M$, we deduce that $D^{\mathsf{o}}_{1},...,D^{\mathsf{o}}_{M},C_{1},...,C_{N}$ are path connected in $Y-\partial X$, and so we are left with showing that they are maximally path-connected in $Y-\partial X$.

Since $C_{1},...,C_{N}$ are the path components of $Y-X$ while $D^{\mathsf{o}}_{1},...,D^{\mathsf{o}}_{M}$ are the path components of the interior of $X$ while $Y-\partial X$ is the disjoint union of these $M+N$ sets, it suffices to show that given $p\in C_{i}$ and $q\in D_{j}$, there cannot exist a path in $Y-\partial X$ between $p$ and $q$. Indeed, supposing for contradiction that $\gamma$ were such a path with $\gamma(0)=p$ and $\gamma(1)=q$, noting that $Y-X$ and the interior of $X$ are open in $Y$, by connectedness of $[0,1]$ we deduce that $\gamma$ must intersect $\partial X$ somewhere, a contradiction.

2.3. $n-d\geq 3$: first facts

We show preliminary facts for the proof of Theorem 1.1 when $n-d\geq 3$. $Y$ and $X$ are given in more generality here, with the compactness assumption removed.

Lemma 2.1.

Suppose $Y^{n}$ is a connected topological manifold and $X^{d}$ is a locally flat submanifold. If $n-d\geq 2$, then the inclusion $Y-X\xhookrightarrow{}Y$ is $\mathrm{0}$-connected.

Proof.

Since $Y$ is connected, the induced map $\pi_{0}(Y-X)\to\pi_{0}(Y)$ is immediately surjective. Injectivity of this map follows from Theorem 2 of [EW42], where we take $([0,1],\{0,1\},Y-X,Y)$ in our notation as $(B,Z,A,X)$ in the notation of [EW42]; in our notation, we have $\overline{Y-X}=Y$ since $X$ has positive codimension (one applies either invariance of domain or local flatness), and the relevant hypothesis of Theorem 2 of [EW42] also follows by local flatness of our $X\subseteq Y$ along with the fact that $n-d\geq 2$. We also want to show that the induced map $\pi_{1}(Y-X)\to\pi_{1}(Y)$ is surjective, and this also follows from Theorem 2 of [EW42] with $(S^{1}\times[0,1],S^{1},Y-X,Y)$ in our notation taken as $(B,Z,A,X)$; the same comment as our first application of Theorem 2 of [EW42]. ∎

Lemma 2.2.

Suppose $Y^{n}$ is a connected topopological manifold and $X^{d}$ is a locally flat submanifold. If $n-d\geq 3$, then the inclusion $Y-X\xhookrightarrow{}Y$ induces an isomorphism of fundamental groups.

Proof.

The inclusion is $0$-connected by Lemma 2.1, so it suffices to show that the induced surjection $\pi_{1}(Y-X)\to\pi_{1}(Y)$ is injective. We apply Theorem 2 of [EW42], taking $(D^{2},S^{1},Y-X,Y)$ in our notation as $(B,Z,A,X)$ in the notation of [EW42]; in our notation we have $\overline{Y-X}=Y$, and the relevant hypothesis follows by local flatness and $n-d\geq 3$. ∎

3. $Y-X$ is finitely dominated

To show that $Y-X$ is finitely dominated, we use regular open neighbourhoods developed in [Sie73], [SGH73], and [SGH74]. In particular, we note the following terminology used:

• $I$-compressibility and the axiom $I$-$\operatorname{Comp}(Y,X)$, introduced in [Sie73] right after Theorem 1.3.

• A pinch of $A$ in $U$ with respect to $(Y,X)$, introduced in Definition 2.1 of [Sie73].

• The pinching axiom $P(x)$, introduced in Definition 2.1 and Theorem 2.2 of [Sie73].

• An anti-$I$-nest $\{F_{n}\}$, appearing in [SGH73] and [SGH74] only. It is introduced in Definition 3.1 of [SGH73] and originally named “anti-$I$-gigogne”.

To show that $Y-X$ is finitely dominated, we will first show:

Proposition 3.1.

The axiom $I$-$\operatorname{Comp}(Y,X)$ holds.

Since $Y$ is locally compact and $X$ is compact in $Y$, by Proposition 5.7 of [SGH73], it suffices to show that the pinching axiom $P(x)$ holds for every $x\in X$; recall the pinching axiom $P(x)$ means that for every neighborhood $x\in U\subseteq Q$, there existence a pinch of $x$ in $U$ with respect to $(Y,X)$. In order to show this, we consider a simplification below.

Lemma 3.2.

To prove Proposition 3.1, it suffices to show existence a pinch of $\mathrm{0}$ in $B(0,10)$ with respect to $(\mathbb{R}^{n},\mathbb{R}^{d}\times 0)$.

Proof.

Fixing $x\in X$ and considering a neighborhood $x\in U\subseteq Y$, to prove Proposition 3.1 we want the existence of a pinch of $x$ in $U$ with respect to $(Y,X)$. By local flatness of $X$, we choose an open neighbourhood $x\in V\subseteq Y$ and a homeomorphism $\varphi:V\to\mathbb{R}^{n}$ such that $\varphi(V\cap X)=\mathbb{R}^{d}\times 0$.

A pinch of $x$ in an open subset of $U$ (containing $x$) will also be a pinch of $x$ in $U$. The first implication is that we can take $U$ to be an open neighborhood of $x$. Furthermore, by intersecting $U$ with $V$ and the preimage of an open ball in $\mathbb{R}^{n}$ under $\varphi$, we can assume that $U\subseteq V$ and that $\varphi(U)$ is an open ball in $\mathbb{R}^{n}$. By scaling and translating $\varphi$, we can also assume that $\varphi(U)=B(0,10)$ and that $\varphi(x)=0$, where $B(0,10)$ denotes the open ball centered at $0$ with radius $10$.

We then see that it suffices to show the existence a pinch of $0$ in $B(0,10)$ with respect to $(\mathbb{R}^{n},\mathbb{R}^{d}\times 0)$. Given such a pinch $F_{t}:\mathbb{R}^{n}\to\mathbb{R}^{n}$, the homotopy $\varphi^{-1}\circ F_{t}\circ\varphi:V\to V$ would be a pinch of $x$ in $\varphi^{-1}(B(0,10))=U$ with respect to $(V,X\cap V)$, and since (by the definition of a pinch) this homotopy is supported in the compact subset $\varphi^{-1}(\overline{B(0,20)})$ where $\overline{B(0,20)}$ denotes a closed ball, we can extend it outside of $V$ to obtain a pinch of $X$ in $U$ with respect to $(Y,X)$. ∎

Due to Lemma 3.2, the proof of Proposition 3.1 reduces to the proof of Proposition 3.3.

Proposition 3.3.

There exists a pinch of $\mathrm{0}$ in $B(0,10)$ with respect to $(\mathbb{R}^{n},\mathbb{R}^{d}\times 0)$.

Proof.

We first define the homotopy $H_{t}:[0,\infty)\to[0,\infty)$ by letting $H_{t}$ fix $[1,\infty)$ and having its graph on $[0,1]$ be the linear interpolation of the four points

$$(0,0),(\frac{1-t}{4},\frac{1-t}{4}),(\frac{1}{2},\frac{1-t}{2}),(1,1)$$

$H_{0}$ is the identity, and since $\frac{1-t}{4}=0$ as when $t=1$, the graph of $H_{1}$ on $[0,1]$ is the linear interpolation of $(0,0),(\frac{1}{2},0),(1,1)$. Addtionally, note that $H_{t}\leq H_{0}$. Figure $1$ illustrates the described homotopy.

Figure 1. The homotopy $H_{t}$.

Let $\psi:\mathbb{R}^{d}\to[0,1]$ be a continuous function that is constant $1$ in the closed ball $\overline{B(0,1)}$ and supported in the closed ball $\overline{B(0,2)}$. We now define the homotopy $F_{t}:\mathbb{R}^{n}\to\mathbb{R}^{n}$ with

$$F_{t}(u,v)=\begin{cases}(u,\frac{H_{t\psi(u)}(\lvert v\rvert)}{\lvert v\rvert}v)&v\neq 0\\ (u,0)&v=0\end{cases}$$

For $(u,v)\in\mathbb{R}^{n}$ where $u\in\mathbb{R}^{d}$ and $v\in\mathbb{R}^{n-d}$; continuity is seen by noting that $H_{t\psi(u)}(\lvert v\rvert)\leq H_{0}(\lvert v\rvert)=\lvert v\rvert$.

$F_{0}$ is the identity. $F_{1}$ maps $B(0,1)\times B(0,\frac{1}{2})$ into $\mathbb{R}^{d}\times 0$ since $\psi$ is constant $1$ on $\overline{B(0,1)}$ while $H_{1}$ is constant $0$ on $[0,\frac{1}{2}]$. We also see that $F_{t}(u,v)=(u,v)$ for all $t\in[0,1]$ and $(u,v)\in\mathbb{R}^{n}-B(0,10)$, since we would either have $\lvert u\rvert>2$ so that $\psi(u)=0$ hence $H_{t\psi(u)}(\lvert v\rvert)=H_{0}(\lvert v\rvert)=\lvert v\rvert$, or $\lvert v\rvert>1$ so that $H_{t\psi(u)}(\lvert v\rvert)=\lvert v\rvert$ since $H_{t\psi(u)}$ fixes $[1,\infty)$. Finally we note that for each $\tau\in[0,1)$, the family $\{F_{t}:t\in[0,\tau]\}$ is an isotopy that fixes $\mathbb{R}^{d}\times B(0,\frac{1-\tau}{4})$: for $t\in[0,\tau]$ and $\lvert v\rvert<\frac{1-\tau}{4}$ we have $\frac{1-\tau}{4}\leq\frac{1-t}{4}\leq\frac{1-t\psi(u)}{4}$ thus $H_{t\psi(u)}(\lvert v\rvert)=\lvert v\rvert$, and the family $\{H_{s}:s\in[0,\tau]\}$ is an isotopy. This completes the proof. ∎

We have thus proven Proposition 3.1, showing that the axiom $I$-$\operatorname{Comp}(Y,X)$ holds. By the definition of $I$-$\operatorname{Comp}(Y,X)$, we choose neighborhoods $E_{0},E_{1}$ of $X$ such that $E_{0}\subseteq E_{1}$ and that $E_{0}$ is $I$-compressible to $X$ in $E_{1}$. Then since $I$-$\operatorname{Comp}(Y,X)$ holds, by Proposition 3.4 of [SGH73] we choose an anti-$I$-nest $\{F_{n}\}$ such that, denoting $B=\cap_{n=0}^{\infty}F_{n}$, we have

$$E_{0}\subseteq B\subseteq F_{0}\subseteq E_{1}.$$

To prepare for the proof that $Y-X$ is finitely dominated, we first show the below fact.

Lemma 3.4.

$Y-B$ is precompact in $Y-X$. In other words, the relative closure $\operatorname{closure}_{Y-X}(Y-B)$ is compact in $Y-X$.

Proof.

It suffices to show that $\overline{Y-B}$ (the closure in $X$) is contained in $Y-X$. The reason is that we would have $\operatorname{closure}_{Y-X}(Y-B)=\overline{Y-B}\cap(Y-X)=\overline{Y-B}$, and since $\overline{Y-B}$ is compact in $Y$ (as $Y$ is compact), we deduce that $\overline{Y-B}$ is compact in $Y-X$, which is what we wanted to conclude. Indeed,

$$\overline{Y-B} =Y-B^{\mathsf{o}} Y$$ (The interior is with respect to )

$$\subseteq Y-X E_{0}\subseteq B E_{0} X$$ (Since while is a neighborhood of )

As required. ∎

Since $Y$ is metrizable, $X$ as a closed subset is also a $G_{\delta}$ set. So by Example 3.3.(ii) and Theorem 3.6 of [SGH73], the inclusion $Y-B\xhookrightarrow{\iota}Y-X$ is a homotopy equivalence. With this result, we finally show that $Y-X$ is finitely dominated.

Proposition 3.5.

$Y-X$ is finitely dominated.

Proof.

$Y-X$ is a topological manifold, and thus an absolute neighborhood retract (ANR); see [Mil59] (Corollary 1) and [Han50] (Theorem 3.3). Due to Lemma 3.4, we can apply Lemma 1.2 of [SGH74] to choose a finite CW complex $L$ and continuous maps $i:Y-B\to L$ and $p:L\to Y-X$ where the following diagram commutes up to homotopy:

$$\begin{tikzcd}$$

Letting $F:Y-X\to Y-B$ denote a homotopy inverse of $\iota$, we then have

$$p\circ(i\circ F)=(p\circ i)\circ F\cong\iota\circ F\cong 1_{Y-X}.$$

Since $L$ is a finite CW complex, we conclude that $Y-X$ is finitely dominated. ∎

4. Reduction to $(Y\times\mathbb{R}^{N})-(X\times 0)$

In this section we aim to show the following:

Proposition 4.1.

For every $N\in\mathbb{N}$, if $(Y\times\mathbb{R}^{N})-(X\times 0)$ is homotopy equivalent to a finite CW complex then $Y-X$ is homotopy equivalent to a finite CW complex.

Proposition 4.1 follows by Lemma 4.2 and Proposition 4.3:

Lemma 4.2.

If $(Y\times\mathbb{R})-(X\times 0)$ being homotopy equivalent a finite CW complex implies $Y-X$ being homotopy equivalent a finite CW complex, then Proposition 4.1 holds.

Proof.

We will use the assumption to prove Proposition 4.1 via induction on $N\in\mathbb{N}$. The base case is precisely the assumption, so we now verify the induction step. Supposing that $(Y\times\mathbb{R}^{N+1})-(X\times 0)$ is homotopy equivalent to a finite CW complex $K$, we define the compact topological manifolds $Y^{\prime},X^{\prime}$ with $X^{\prime}$ being a locally flat submanifold $Y^{\prime}$, by the following:

$$Y^{\prime}=Y\times[-1,1]^{N}\hskip 14.22636ptX^{\prime}=X\times 0.$$

$Y^{\prime}\times\mathbb{R}-(X^{\prime}\times 0)$ is homotopy equivalent to $(Y\times\mathbb{R}^{N+1})-(X\times 0)$ by considering a deformation retract of $\mathbb{R}^{N}$ onto $[-1,1]^{N}$. Therefore, $Y^{\prime}\times\mathbb{R}-(X^{\prime}\times 0)$ is homotopy equivalent to $K$.

Since the dimension of $Y^{\prime}$ is $n+N$ and the dimension of $X^{\prime}$ is $d$, the codimension of $X^{\prime}$ in $Y^{\prime}$ is $n+N-d\geq n-d=3$. So since $Y^{\prime}\times\mathbb{R}-(X^{\prime}\times 0)$ is homotopy equivalent to $K$, we apply the base case to deduce that $Y^{\prime}-X^{\prime}$ is homotopy equivalent to a finite CW complex. Now again considering a deformation retract of $\mathbb{R}^{N}$ onto $[-1,1]^{N}$, we see that $Y^{\prime}-X^{\prime}$ is homotopy equivalent to $(Y\times\mathbb{R}^{N})-(X\times 0)$. This means that $(Y\times\mathbb{R}^{N})-(X\times 0)$ is homotopy equivalent to a finite CW complex, and so applying the induction hypothesis, we conclude that $Y-X$ is homotopy equivalent to a finite CW complex. ∎

Proposition 4.3.

If $(Y\times\mathbb{R})-(X\times 0)$ is homotopy equivalent to a finite CW complex, then $Y-X$ is homotopy equivalent to a finite CW complex.

Proof.

By the hypothesis, $(Y\times[-1,1])-(X\times 0)$ is homotopy equivalent to a finite CW complex. Define

$$U=(Y\times(0,1])\cup((Y-X)\times 0)\hskip 14.22636ptV=(Y\times[-1,0))\cup((Y-X)\times 0)$$

Where $(Y-X)\times 0$ is a topological manifold with dimension $n$. Note that $U,V$ are homotopy equivalent to $Y$, as they both deformation retract onto $Y\times 1$. Also note that $U\cup V=(Y\times[-1,1])-(X\times 0)$, which is a topological manifold, and connected by Lemma 2.1. By the hypothesis, $U\cup V$ is homotopy equivalent to a finite CW complex. We also note that $U\cap V=(Y-X)\times 0$, so that the inclusion $U\cap V\xhookrightarrow{}Y\times 0$ induces an isomorphism of fundamental groups by Lemma 2.2. Finally, since $Y\times 0\xhookrightarrow{}Y\times[-1,1]$ induces an isomorphism of fundamental groups while the inclusion $U\cup V\xhookrightarrow{}Y\times[-1,1]$ induces an isomorphism of fundamental groups by Lemma 2.2, by considering the following commutative diagram of inclusions

$$\begin{tikzcd}$$

We deduce that the inclusion $U\cap V\xhookrightarrow{}U\cup V$ induces an isomorphism of fundamental groups.

By Theorems 2.4-2.5 of [Hir15], we choose CW approximations $\operatorname{CW}(U\cup V),\operatorname{CW}(U),\operatorname{CW}(V),\operatorname{CW}(U\cap V)$ such that each of the four topological spaces is weakly homotopy equivalent to their CW approximation, that the other three CW approximations are subcomplexes of $\operatorname{CW}(U\cup V)$, that $\operatorname{CW}(U\cap V)=\operatorname{CW}(U)\cap\operatorname{CW}(V)$, and that we obtain the following commutative diagram:

$$\begin{tikzcd}$$

$U,V,U\cap V,U\cup V$ are connected topological manifolds up to homotopy, while topological manifolds are homotopy equivalent to CW complexes (Corollary 1 of [Mil59]), so by Whitehead’s theorem, all weak homotopy equivalences here are homotopy equivalences. In particular, as $U\cap V=(Y-X)\times 0$ is finitely dominated by Proposition 3.5, it suffices to show that $\sigma(\operatorname{CW}(U)\cap\operatorname{CW}(V))$ vanishes.

Since $U,V$ are homotopy equivalent to the compact connected topological manifold $Y$, where compact topological manifolds are homotopy equivalent to finite CW complexes by [KS69], $U,V$ are in particular finitely dominated so that their finiteness obstructions are well defined, with $\sigma(\operatorname{CW}(U))=\sigma(\operatorname{CW}(V))=0$. Additionally, since $U\cup V$ is homotopy equivalent to a finite CW complex, its finiteness obstruction is also well-defined with $\sigma(\operatorname{CW}(U\cup V))=0$. Thus by the sum formula for finiteness obstruction (Theorem 6.5 of [Sie65]), we have

$$0=0+0-j_{\ast}\sigma(\operatorname{CW}(U)\cap\operatorname{CW}(V))$$

Where $j_{\ast}$ is induced by the inclusion $j:\operatorname{CW}(U)\cap\operatorname{CW}(V)\xhookrightarrow{}\operatorname{CW}(U\cup V)$. But $j$ induces an isomorphism of fundamental groups due to the previous commutative diagram where the inclusion $U\cup V\xhookrightarrow{}U\cap V$ induces an isomorphism of fundamental groups, and so we necessarily have $\sigma(\operatorname{CW}(U)\cap\operatorname{CW}(V))=0$. ∎

5. Assembly of Proof for Theorem 1.1

We finish the proof of Theorem 1.1, by proving Proposition 1.2 and considering the case where $\partial X\neq\emptyset$ (and $\partial X\subseteq\partial Y$) and showing that $X$ still admits a normal bundle in $Y\times\mathbb{R}^{N}$ for some $N\in\mathbb{N}$. Recall that Proposition 1.2 states that the existence of a closed disk bundle $B$ for $X$ implies that $Y-X$ is homotopy equivalent to a finite CW complex.

5.1. Proof of Proposition 1.2

By local triviality of the closed disk bundle $B$, the complement $Y-B^{\mathsf{o}}$ is a compact manifold, thus homotopy equivalent to a finite CW complex. Therefore, it suffices to show that $Y-B^{\mathsf{o}}$ and $Y-X$ are homotopy equivalent.

We note that $\partial B$ is a neighbourhood deformation retract (NDR) of $Y-B^{\mathsf{o}}$: local triviality of $B$ along with Satz 2 of [Dol68] shows that $(Y-B^{\mathsf{o}},\partial B)$ is closed cofibration, by which we apply Proposition 5.4.4 of [Die08]. Additionally, attaching $B-X$ to $Y-B^{\mathsf{o}}$ along $\partial B$ yields $Y-X$, giving us the pushout diagram:

$$\begin{tikzcd}$$

We now show that the inclusion $\partial B\xhookrightarrow{}B-X$ is a homotopy equivalence. Recalling that $B$ is a closed disk bundle over $X$, we let $k$ denote the dimension of this closed disk.

$$\begin{tikzcd}$$

Since $B$ is a $D^{k}$ bundle over $X$, we see that $\partial B$ and $B-X$ are $S^{k-1}$ and $D^{k}-\{0\}$ bundles over $X$ respectively, where the inclusion $S^{k-1}\xhookrightarrow{}D^{k}-\{0\}$ is a homotopy equivalence. By local triviality of $\partial B$ and $B-X$, along with paracompactness of $X$ and the inclusion $S^{k-1}\xhookrightarrow{}D^{k}-\{0\}$ being a homotopy equivalence, we choose a numerable open cover $\{U_{j}:j\in J\}$ of $B$ such that each induced map $\pi_{1}^{-1}(U_{j})\to\pi_{2}^{-1}(U_{j})$ is a fiberwise homotopy equivalence. By Theorem 13.3.3 of [Die08], we thus deduce that the inclusion $\partial B\xhookrightarrow{}B-X$ is a homotopy equivalence.

Having shown that the inclusion $\partial B\xhookrightarrow{}B-X$ is a homotopy equivalence, we then deduce that $Y-B^{\mathsf{o}}\xhookrightarrow{}Y-X$ is a homotopy equivalence by applying Proposition 5.3.3 of [Die08] with the following commutative diagram in mind:

$$\begin{tikzcd}$$

where the three labelled diagonal maps are homotopy equivalences, and the inner and outer squares are homotopy cocartesian. Having shown that $Y-B^{\mathsf{o}}\xhookrightarrow{}Y-X$ is a homotopy equivalence, we conclude that $Y-X$ is homotopy equivalent to a finite CW complex, completing the proof of Proposition 1.2.

5.2. Consideration of Boundary

Here, we suppose $\partial X\neq\emptyset$ and $\partial X\subseteq\partial Y$. Since $Y$ is compact while $X$ is compact and locally flat in $Y$, we see that $\partial X$ is locally flat in $\partial Y$ with $\partial X,\partial Y$ both being closed topological manifolds. Their lack of boundaries allow us to apply the stable existence theorem from [KS77], so that $\partial X\times 0$ admits a normal microbundle in $\partial Y\times\mathbb{R}^{N}$ for some $N\in\mathbb{N}$.

We apply generalization A.4(b) of [KS77] with $(Y\times\mathbb{R}^{N},\{\partial Y\times\mathbb{R}^{N}\},X\times 0)$ in our notation as $(W,\{W_{\alpha}\},M)$ in the notation of [KS77], to obtain a map $f:\partial Y\times\mathbb{R}^{N}\times[0,\infty)\to Y\times\mathbb{R}^{N}$ such that $\partial X\times 0\times[0,\infty)=f^{-1}(X\times 0)$. This thus allows us to extend the normal bundle of $\partial X\times 0$ in $\partial Y\times\mathbb{R}^{N}$ to a normal bundle of $\partial X\times 0$ in $Y\times\mathbb{R}^{N}$; applying the stable existence theorem from [KS77] again, after possibly incrementing $N$, we obtain a normal bundle of $X\times$ in $Y\times\mathbb{R}^{N}$, which is what we wanted. We have thus dealt with the case $\emptyset\neq\partial X\subseteq\partial Y$.