On the $h$-cobordism category. I
Extracted source text ยท arXiv 1805.04395
\documentclass{amsart}
\usepackage[centertags]{amsmath}
\usepackage{amssymb,amsthm}
\usepackage[all]{xy}
\usepackage{hyperref}
\usepackage{graphicx}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\title{On the $h$-cobordism category. I}
\author{George Raptis}
\address{\newline
G. Raptis \newline
Fakult\"{a}t f\"{u}r Mathematik, Universit\"{a}t Regensburg, D-93040 Regensburg, Germany}
\email{georgios.raptis@mathematik.uni-regensburg.de}
\author{Wolfgang Steimle}
\address{\newline
W. Steimle \newline
Institut f\"ur Mathematik,
Universit\"at Augburg,
D-86135 Augsburg, Germany}
\email{wolfgang.steimle@math.uni-augsburg.de}
%\date{\today}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\DeclareMathOperator{\asmb}{asmb}
\DeclareMathOperator{\aut}{aut}
\DeclareMathOperator{\bary}{bary}
\DeclareMathOperator{\can}{can}
\DeclareMathOperator{\card}{card}
\DeclareMathOperator{\cent}{cent}
\DeclareMathOperator{\ch}{ch}
\DeclareMathOperator{\class}{class}
\DeclareMathOperator{\coker}{coker}
\DeclareMathOperator{\colim}{colim}
\DeclareMathOperator{\con}{con}
\DeclareMathOperator{\cone}{cone}
\DeclareMathOperator{\cyl}{cyl}
\DeclareMathOperator{\Diff}{Diff}
\DeclareMathOperator{\Emb}{Emb}
\DeclareMathOperator{\FIB}{FIB}
\DeclareMathOperator{\fib}{fib}
\DeclareMathOperator{\func}{func}
\DeclareMathOperator{\GL}{GL}
\DeclareMathOperator*{\hocolim}{hocolim}
\DeclareMathOperator{\hofib}{hofib}
\DeclareMathOperator*{\holim}{holim}
\DeclareMathOperator{\hfd}{hfd}
\DeclareMathOperator{\id}{id}
\DeclareMathOperator{\im}{im}
\DeclareMathOperator{\inc}{inc}
\DeclareMathOperator{\ind}{ind}
\DeclareMathOperator{\Idem}{Idem}
\DeclareMathOperator{\map}{map}
\DeclareMathOperator{\mor}{mor}
\DeclareMathOperator{\Nil}{Nil}
\DeclareMathOperator{\NWh}{NWh}
\DeclareMathOperator{\pt}{pt}
\DeclareMathOperator{\pr}{pr}
\DeclareMathOperator{\res}{res}
\DeclareMathOperator{\sd}{sd}
\DeclareMathOperator{\sing}{sing}
\DeclareMathOperator{\sintop}{sing^{top}}
\DeclareMathOperator{\simp}{simp}
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\topo}{Top}
\DeclareMathOperator{\tors}{tors}
\DeclareMathOperator{\trf}{trf}
\DeclareMathOperator{\Wh}{Wh}
\DeclareMathOperator{\Wall}{Wall}
\DeclareMathOperator{\coass}{\mathrm{\nabla}}
\DeclareMathOperator{\scan}{\mathrm{scan}}
\DeclareMathOperator{\Imm}{Imm}
\DeclareMathOperator{\inter}{int}
\DeclareMathOperator{\Bun}{Bun}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\cob}{\mathcal{C}_{d}}
\newcommand{\cobx}{\mathcal C}
\newcommand{\cobd}{\mathcal{C}_{d}}
\newcommand{\cobdn}{\mathcal{C}_{d, n}}
\newcommand{\cobdstr}{\mathcal{C}_\theta}
\newcommand{\cobdnstr}{\mathcal{C}_{\theta,n}}
\newcommand{\cobtriad}{\mathcal C_{d,\dell^0,\dell^1,\dell^+}}
\newcommand{\corn}[2]{\RR^{#1}_{\langle #2\rangle}}
\renewcommand{\sin}{\mathrm{sing}}
\newcommand{\ob}{\operatorname{ob}}
\newcommand{\mto}{\mathrm{MTO}}
\newcommand{\Th}{\mathrm{Th}}
\newcommand{\Gr}{\mathrm{Gr}}
\newcommand{\brac}[1]{\langle #1 \rangle}
\newcommand{\eps}{\varepsilon}
\newcommand{\WS}{\mathtt{S}}
\newcommand{\Ab}{\mathrm{Ab}}
\newcommand{\RMod}{R\mathrm{-mod}}
\newcommand{\Sp}{\mathrm{Spectra}}
\newcommand{\OSp}{\Omega\mathrm{-Spectra}}
\newcommand{\Cat}{\mathrm{Cat}}
\newcommand{\cat}{\mathbf{cat}}
\newcommand{\CGHaus}{\mathbf{CGHaus}}
\newcommand{\op}{^{op}}
\newcommand{\inv}{^{-1}}
\newcommand{\fibr}[2]{\begin{pmatrix} {#1} \\ \downarrow \\ {#2}\end{pmatrix}}
\newcommand{\hcobd}{\mathcal{C}_{\theta}^{\sim}}
\newcommand{\hcobdno}{\mathcal{C}_d^{\sim}}
\newcommand{\hcobdn}{\mathcal{C}_{\theta,n}^{\sim}}
\newcommand{\cobdk}{\mathcal{C}_{d, \langle k \rangle}}
\newcommand{\dell}{\partial}
\newcommand{\taud}{\tau_{\partial}}
\newcommand{\RR}{\mathbb{R}}
\newcommand{\ZZ}{\mathbb{Z}}
\newcommand{\NN}{\mathbb{N}}
\newcommand{\QQ}{\mathbb{Q}}
\newcommand{\CC}{\mathbb{C}}
\newcommand{\calR}{\mathcal{R}}
\newcommand{\calU}{\mathcal{U}}
\newcommand{\calI}{\mathcal{I}}
\newcommand{\calC}{\mathcal{C}}
\renewcommand{\hom}{\underline{\mathrm{Hom}}}
\renewcommand{\cyl}{\mathrm{Cyl}}
\newcommand{\Rfd}{\mathcal{R}^{hf}}
\newcommand{\Rd}{\mathcal{R}^{\delta}}
\newcommand{\Rdd}{\overline{\mathcal{R}}^{\delta}}
\newcommand{\tRfd}{\tilde{\mathcal{R}}^{hf}}
\newcommand{\fRfd}{\mathcal{R}^{hf}_{\mathrm{fib}}}
\newcommand{\Map}{\mathrm{Map}}
\newcommand{\Hdiff}{\mathcal{H}}
\newcommand{\HdiffM}{\mathcal{H}^{\theta}_M}
\newcommand{\HdiffMno}{\mathcal{H}_M}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcounter{commentcounter}
\newcommand{\g}[1]{\marginpar{\raggedleft\sffamily{\scriptsize George: #1}}}%
\newcommand{\w}[1]{\marginpar{\raggedleft\sffamily{\scriptsize Wolfgang: #1}}}%
%------------------------------------------------------------------------
%-------------------------------------------------------------------------
\begin{document}
\theoremstyle{plain}
\newtheorem{thm}{Theorem}
\newtheorem{cor}[thm]{Corollary}
\newtheorem{lem}[thm]{Lemma}
\newtheorem{prop}[thm]{Proposition}
\newtheorem{claim}[thm]{Claim}
\theoremstyle{definition}
\newtheorem{defn}[thm]{Definition}
\newtheorem{obs}[thm]{Observation}
\newtheorem{constr}[thm]{Construction}
\theoremstyle{remark}
\newtheorem{rem}[thm]{Remark}
\setcounter{secnumdepth}{2}
\numberwithin{thm}{section}
\SelectTips{eu}{10}
\renewcommand{\theenumi}{\roman{enumi}}
\renewcommand{\labelenumi}{\textup{(}\theenumi\textup{)}}
\renewcommand{\theequation}{\arabic{equation}}
\begin{abstract}
We consider the topological category of $h$-cobordisms between manifolds with boundary and compare its homotopy type
with the standard $h$-cobordism space of a compact smooth manifold.
\end{abstract}
\maketitle
\setcounter{tocdepth}{1}
\tableofcontents
\section{Introduction}
The spaces of $d$-dimensional smooth cobordisms between closed manifolds organize themselves into a topological category where the composition is given by concatenating cobordisms. The homotopy type of the
classifying space of this cobordism category was identified with the infinite loop space of a Thom spectrum \cite{GMTW(2009)}. Analogous results have been obtained for variants of this cobordism category,
defined either by considering general tangential structures, or by allowing cobordisms between manifolds with boundaries, corners, or Baas-Sullivan singularities (see, for example, \cite{Genauer(2008)} and \cite{Perlmutter}). These
results have turned out to be of fundamental importance in the study of diffeomorphism groups, characteristic classes, invertible field theories, and positive scalar curvature metrics.
The composition of two $h$-cobordisms is again an $h$-cobordism, so the spaces of $d$-dimensional smooth $h$-cobordisms form a subcategory of the standard cobordism category. In this
paper, we study the homotopy type of (the loop space of) the $h$-cobordism category, in the setting of smooth manifolds with boundary, and compare this with the standard
(stable) parametrized $h$-cobordism space. The latter space is closely connected with the algebraic $K$-theory of spaces as a consequence of the celebrated stable parametrized $h$-cobordism theorem \cite{WJR}.
\medskip
Recall that if $M^{d-1}$ and $N^{d-1}$ are manifolds with boundary, a \emph{cobordism} from $M$ to $N$ is a compact manifold $W^d$ with boundary together with a decomposition of its boundary
into three submanifolds of codimension $0$,
\[\dell W = M \cup \dell_h W \cup N,\]
such that $M$ and $N$ are disjoint and $M\sqcup N$ meets $\dell_h W$ in the common boundary $\dell M \amalg \dell N = \dell (\dell_h W)$. Such a cobordism is an \emph{$h$-cobordism} if
each of the inclusions
\[M \hookrightarrow W \hookleftarrow N, \quad \dell M \hookrightarrow \dell_h W \hookleftarrow \dell N\]
is a homotopy equivalence. As usual, there is a generalization to the case where the manifolds are endowed with tangential $\theta$-structures. This leads to the definition of the
\emph{$\theta$-structured $h$-cobordism category}, denoted by $\hcobd$. (See section \ref{sec:h_cob_cat} for the details. For notational simplicity, we suppress from the notation any
symbol to indicate the existence of ``horizontal'' boundaries.)
We emphasize that in our definition, the horizontal boundary $\dell_h W$ can be an arbitrary $h$-cobordism between the closed manifolds $\dell M$ and $\dell N$. This is in contrast to
the $h$-cobordisms considered in the definition of the standard $h$-cobordism space, where one has
$\dell_h W = \dell M\times [0,1]$ and therefore $\dell M = \dell N$. An $h$-cobordism of this special form will be called a \emph{$\dell$-trivial} $h$-cobordism in this paper. The
additional flexibility of our definition will be of great importance for our comparison between the classifying space of the $h$-cobordism category and the $h$-cobordism space. More precisely,
the $h$-cobordism space $\Hdiff(M)$ of a compact smooth $(d-1)$-manifold $M$ is the classifying space for bundles of $\dell$-trivial $h$-cobordism which agree with the trivial
$M$-bundle on one side of the parametrized family of $\partial$-trivial $h$-cobordisms, while the restriction to the other side is an arbitrary bundle of smooth compact manifolds (see, for example, \cite{WJR}). Our main result compares,
in the case $d\geq 7$, the homotopy type of $B\hcobd$ with the homotopy types of $\Hdiff(M)$ for different $M$.
\begin{thm}\label{thm:main1}
Let $M^{d-1}$ be an object of $\hcobd$, $d\geq 7$. For any $\dell$-trivial $h$-cobordism $(W;M,N)$ on $M$, there is a homotopy fiber sequence
\[\Emb_\theta^\sim(M,N) \to \Omega_M B\hcobd \to \Hdiff(M)\]
where the loop space is based at $M$, and the homotopy fiber is taken over $W$.
\end{thm}
Here $\Emb_\theta^\sim(M,N)$ denotes the \emph{space of $\theta h$-embeddings} from $M$ into $N$. These have the property that both inclusions of $\dell M$ and $\dell N$ into the
complement $N-\inter(M)$ are homotopy equivalences (so that, in particular, the embedding itself is a homotopy equivalence), and there is an identification between the $\theta$-structure on $N$
induced by $(W; M, N)$ and the one induced by the embedding. See section \ref{sec:h_cob_space} for the precise definition.
\medskip
An interesting consequence of Theorem \ref{thm:main1} is in the case where the tangent bundle of $M$ is regarded as a $\theta$-structure and $M$ is a compact smooth $(d-1)$-manifold
of lower handle dimension. Let $\theta_M = (\eps \oplus TM \to M)$, where $\eps$ denotes the trivial line bundle.
\begin{thm}\label{thm:main2}
Let $M$ be a compact connected smooth $(d-1)$-manifold of handle dimension $k$, regarded as an object of $\mathcal{C}^{\sim}_{\theta_M}$, where $d \geq 7$. Then there is a $(d-2k-2)$-connected map
\[
\Omega_M B\mathcal{C}^{\sim}_{\theta_M} \xrightarrow{(d-2k-2)} \Hdiff(M).
\]
\end{thm}
Therefore, after stabilizing both sides by taking products with $I$ (and straightening the resulting corners), we conclude the following corollary as a consequence of
Theorem \ref{thm:main2} and the stable parametrized $h$-cobordism theorem \cite{WJR}.
\begin{cor}\label{thm:cor}
Let $M$ be a compact connected smooth $(d-1)$-manifold. There is a homotopy equivalence
\[\hocolim_n \Omega_{M \times D^n} B\mathcal{C}^{\sim}_{\theta_{M \times D^n}} \simeq \Omega \Wh^{\operatorname{diff}}(M).\]
\end{cor}
Here $\Wh^{\operatorname{diff}}(M)$ denotes the smooth Whitehead space of $M$, defined by means of algebraic $K$-theory (see \cite{WJR}). We note that we work entirely in the
smooth setting in this paper.
\medskip
The strategy for the proof of these results is as follows. First, by the classical $h$-cobordism theorem, the category $\hcobd$ is a non-unital groupoid in a homotopical sense (Section \ref{sec:invertibility_hcob}).
This is indeed the only place where we use the dimension restriction -- in particular, our techniques apply in any dimension if we restrict to the cobordism category of \emph{invertible} cobordisms instead.
Using standard homotopical techniques discussed in Section \ref{sec:non_unital_categories}, it follows that the classifying space $B \hcobd$ is homotopy equivalent to a disjoint union of spaces of endomorphisms, one
from each component. In Section \ref{sec:h_cob_space}, we analyze geometrically the homotopy type of an endomorphism space. The main idea is to trade the $\dell$-triviality of the $h$-cobordisms
in $\Hdiff(M)$ with the condition that an $h$-cobordism in an endomorphism space has both ends fixed. This leads to Theorem \ref{thm:main1}. Theorem \ref{thm:main2} follows by analyzing
the homotopy types of the $\theta h$-embedding spaces under consideration.
\medskip
In a sequel to this work, we aim to carry out a different analysis of the homotopy type of $B\hcobd$ and study directly its connection with algebraic $K$-theory.
\section{The $h$-cobordism category}\label{sec:h_cob_cat}
In this section, we recall the definition of the cobordism category $\cobdstr$ \cite{GMTW(2009), Genauer(2008)} and then define the $h$-cobordism category $\hcobd$ to be the subcategory of $h$-cobordisms. The definition
of the cobordism category here uses a slight variation of the definition of tangential structure as defined in \cite{GMTW(2009)} and \cite{Genauer(2008)}. We emphasize that \emph{objects (morphisms)
are allowed to have boundaries (corners)}. Moreover, we will need to allow arbitrary tangential structures as these will play an important role in the proofs of our main results.
\medskip
Let $\RR_+:=[0,\infty)$, and $\corn{n}{k}:= \RR_+^k \times \RR^{n-k}\subset \RR^n.$
We call a subset $M\subset \corn{d-1+n}1$ a $(d-1)$-\emph{dimensional neatly embedded submanifold} if the following hold:
\begin{enumerate}
\item $M\subset \corn{d-1+n}1$ is locally diffeomorphic to $\corn{d-1}1\subset \corn{d-1+n}1$. In particular, $\dell M = M \cap \dell\corn{d-1+n}1$.
\item There is $\epsilon>0$ such that
\[M \cap \bigl([0, \epsilon) \times \RR^{d-2+n}\bigr)= [0, \epsilon) \times \dell M,\]
\end{enumerate}
with the appropriate interpretation of the product on the right hand side as a subspace of $\corn{d-1+n}1$.
\medskip
Fix a $d$-dimensional vector bundle $\theta = (V \to X)$ over a space $X$. A \emph{$\theta$-structure} on a smooth $(d-1)$-dimensional manifold $M$ (possibly with boundary)
is a map of vector bundles $l_M\colon \eps\oplus TM\to \theta$, where $\eps$ denotes the trivial line bundle.
\medskip
Note that the boundary of the manifold $[0,a]\times \corn n 1$ consists of three boundary faces, which we denote by
\begin{align*}
\dell_0([0,a]\times \corn n 1) & := \{0\}\times \corn n 1,\\
\dell_1([0,a]\times \corn n 1) &:= \{a\}\times \corn n 1,\\
\dell_h([0,a]\times \corn n 1) &:= [0,a]\times \dell \corn n 1.
\end{align*}
We call a subset $W\subset [0,a]\times \corn{d-1+n}1$ a $d$-\emph{dimensional neatly embedded submanifold} if the following hold:
\begin{enumerate}
\item $W\subset [0,a]\times \corn{d-1+n}1$ is locally diffeomorphic to $\corn d 2 \subset \corn {d+n}2$. In particular, $W$ is a smooth $d$-manifold, possibly with corners.
\item There is $\epsilon>0$ such that
\begin{enumerate}
\item $W \cap ([0, \epsilon) \times \corn{d-1 + n}{1}) = [0, \epsilon) \times \dell_0 W$,
\item $W \cap ((a - \epsilon, a] \times \corn{d-1 + n}{1}) = (a - \epsilon, a] \times \dell_1 W$, and
\item $W \cap ([0, a] \times [0, \epsilon) \times \RR^{d-2+n})= [0, \epsilon) \times \partial_h W$.
\end{enumerate}
\end{enumerate}
Here we use the notation
\[\dell_i W := W\cap \dell_i([a,b]\times \corn n 1), \quad (i\in \{0,1,h\})\]
and the appropriate interpretation of the products on the right.
We call $\dell_h W$ the \emph{horizontal} boundary of $W$ and $\dell_0 W\sqcup \dell_1 W$ the \emph{vertical} boundary of $W$.
\medskip
A $\theta$-structure on smooth $d$-dimensional manifold $W$ (possibly with corners) is a map of vector bundles $l_W\colon TW\to \theta$. Note that such
a $\theta$-structure induces a $\theta$-structure $l_{\dell_i W}$ on $\dell_i W$, $i\in \{0,1,h\}$, by means of the canonical splittings
\begin{align*}
TW\vert_{\dell_0 W} &= T[a_0, a_0+\epsilon)\vert_{a_0} \times T(\dell_0 W) \cong \eps\oplus T(\dell_0 W),\\
TW\vert_{\dell_1 W} &= T(a_1 - \epsilon, a_1]\vert_{a_1} \times T(\dell_1 W) \cong \eps\oplus T(\dell_1 W),\\
TW\vert_{\dell_h W} &= T[0, \epsilon)\vert_{0} \times T(\dell_h W) \cong \eps\oplus T(\dell_h W),
\end{align*}
induced by the standard trivialization of $T\RR$, sending $\dell/\dell x$ to 1.
\begin{defn} Let $\theta = (V \to X)$ be a vector bundle on a space $X$ of rank $d$. The \emph{$\theta$-cobordism category $\cobdnstr$ in $\corn{d+n}1$} is defined as follows:
\begin{itemize}
\item An object is a pair $(M, l_M)$ where $M\subset \corn{d-1+n}1$ is a compact $(d-1)$-dimensional neatly embedded submanifold; $l_M$ is a $\theta$-structure on $M$.
\item A morphism is a triple $(a, W, l_W)$ where $a>0$ is a real number; $W\subset [0, a]\times \corn{d-1+n}1$ is a compact $d$-dimensional neatly embedded submanifold;
$l_W$ is a $\theta$-structure on $W$.
\item The source and target of a morphism $(a,W, l_W)$ are given by $(\dell_i W, l_{\dell_i W})$ for $i=0$ and $i=1$, respectively.
\item Composition of morphisms is defined by
\[(a', W', l') \circ (a, W, l) := (a+a', W\cup \mathrm{sh}_a(W'), L)\]
with $\mathrm{sh}_a(W')$ denotes the embedding of $W'$ shifted by $\mathrm{sh}_a \colon \RR \to \RR$, $x \mapsto x + a$, in the first coordinate; $L$ is defined by the conditions $l\vert_{W}= l$ and
$l\vert_{\mathrm{sh}_a(W')} = l' \circ D\mathrm{sh}_a\inv$.
\end{itemize}
\end{defn}
We use the abbreviation $\cobdstr := \underset{n \to \infty}{\colim} \ \cobdnstr.$ In this case, note that $n$ is arbitrarily large and not part of the structure.
We will use the notation $\cobdn$ and $\mathcal{C}_d$, respectively, when no
$\theta$-structures are considered in the definition of the objects and the morphisms of the category. We will generally regard this as a specific case of
$\cobdstr$ since it corresponds up to homotopy equivalence to the case of the
universal vector bundle of rank $d$.
\medskip
Next we recall the definition of the topology on the spaces of objects. For notational simplicity, we restrict only to the case $n=\infty$. Given a compact smooth $(d-1)$-manifold $M$ with $\eps$-collared boundary, let $\mathrm{Emb}^\eps(M, \RR_+ \times \RR^{\infty})$ denote the space $\eps$-neat
embeddings with the $C^\infty$-topology. Here ``$\eps$-neat embedding'' means a diffeomorphism onto a smooth neatly embedded submanifold, which is cylindrical in the collar coordinate inside $[0, \eps) \times \RR^{\infty}$. Then we define $\mathrm{Emb}(M, \RR_+ \times \RR^{\infty})$ as the colimit of these spaces as $\eps \to 0$.
There is a canonical map
\[ \mathrm{Emb}(M, \RR_+\times \RR^{\infty})\times \Bun(\eps\oplus TM, \theta)\to \ob \cobdstr,\]
which sends $(e,l)$ to $(e(M), l\circ (\id\oplus De\inv))$, and the topology on the space of objects is defined by the quotient topology with respect to the collection of these maps,
one for each $M$, where one representative of each diffeomorphism class suffices.
Similarly, let $\Diff_{\eps}(M)$ denote the group of diffeomorphisms which are cylindrical near the collared boundary, and let $\Diff(M) := \colim_{\eps \to 0} \Diff_{\eps}(M)$. Note
that $\Diff(M)$ acts freely on $\mathrm{Emb}(M, \RR_+\times \RR^{\infty})\times \Bun(\eps\oplus TM, \theta)$, by precomposition in both factors. We denote by $B_\theta(M)$ the
quotient of this action; then we have a homeomorphism
\[\ob \cobdstr \cong \coprod_{M} B_\theta(M)\]
where the coproduct ranges over compact smooth $(d-1)$-manifolds $M$ with collared boundary, one from each diffeomorphism class. This defines the topology on the space of objects.
The analogous construction for compact $d$-dimensional cobordisms, possibly with corners which are collared, defines the topology of the space of morphisms. In this case, we have a homeomorphism
\[\mor\cobdstr \cong \coprod_{W} (0,\infty) \times B_\theta(W)\]
where $B_\theta(W)$ is the quotient under the $\Diff(W)$-action on the product
\[\Emb(W, [0,1]\times \RR_+\times \RR^\infty)\times \Bun(TW, \theta).\]
\begin{rem}
Our definition of the cobordism category differs from \cite{Genauer(2008)} or \cite{Perlmutter} in several respects. Firstly, we use a different definition of tangential structure which does
not require a model for the classifying map of the tangent bundle. Secondly, we use a reduced version of the cobordism category as suggested in \cite[Remark 2.1]{GMTW(2009)}. These technical
modifications do not affect the homotopy type of the spaces of objects or morphisms, or of the classifying space. Finally, the category $\cobdnstr$ has no identity morphisms, and we shall
consider it here as a \emph{non-unital category}, while in \cite{GMTW(2009)} the identity morphisms are formally added. Again, the homotopy type of the classifying space is not affected by this
difference (see also Section \ref{sec:non_unital_categories}).
\end{rem}
\begin{defn}
Let $M$ and $N$ be compact smooth $(d-1)$-manifolds, possibly with boundary. A cobordism (with corners) $W = (W; \partial_v W = M \sqcup N, \partial_h W)$ from $M$ to $N$ is an
$h$-\emph{cobordism} if all of the inclusions
\[M\hookrightarrow W\hookleftarrow N, \quad \dell M \hookrightarrow \dell_h W \hookleftarrow \dell N\]
are homotopy equivalences. The category $\hcobd\subset \cobdstr$ is the subcategory which has the same objects as $\cobdstr$, and where the morphisms are
$h$-cobordisms. We will also denote by $\hcobdno \subset \cobd$ the corresponding subcategory where no $\theta$-structures are considered.
\end{defn}
\section{Non-unital categories and semi-simplicial spaces} \label{sec:non_unital_categories}
Let $\Delta_{<} \subseteq \Delta$ denote the subcategory of the simplex category which consists of the injective maps. A functor $X_{\bullet}: \Delta_{<}^{\mathrm{op}} \to \mathrm{Spaces}$ is a \emph{semi-simplicial space}. Similarly to simplicial spaces, $X_{\bullet}$ has a geometric realization $\|X_\bullet\|$. This is isomorphic to the geometric realization of the simplicial space which is associated to $X_\bullet$ by freely adding degeneracies (i.e., by taking the left
Kan extension of $X_\bullet$ along the inclusion $\Delta_{<} \subseteq \Delta$.)
The geometric realization of a semi-simplicial space admits a skeletal filtration
$$\varnothing = \|X_{\bullet}\|_{-1} \subseteq \| X_{\bullet} \|_0 \subseteq \| X_{\bullet} \|_1 \subseteq \cdots \subseteq \| X_{\bullet}\|_n \subseteq \cdots \subseteq \| X_\bullet\| = \bigcup_{n \geq 0} \|X_{\bullet}\|_n$$
which is defined inductively as follows: $\| X_{\bullet}\|_0 : = X_0$ and there are successive pushout squares
\[
\xymatrix{
X_k \times \partial \Delta^{k} \ar[r] \ar[d] & \|X_{\bullet}\|_{k-1} \ar[d] \\
X_k \times \Delta^k \ar[r] & \|X_{\bullet}\|_k
}
\]
where the top map is defined using the simplicial operators on $X_{\bullet}$. It follows that the geometric realization of a
degreewise weak equivalence of semi-simplicial spaces is again a weak equivalence of spaces. We also refer to \cite{ERW} for a detailed treatment of the homotopy theory of semi-simplicial spaces.
\medskip
\noindent The nerve of a non-unital topological category $\cobx$ defines a semi-simplicial space
$$N_\bullet \cobx: \Delta_{<}^{\mathrm{op}} \to \mathrm{Spaces}, \ [n] \mapsto N_n \cobx := \underbrace{\mor \cobx \times_{\ob \cobx} \cdots \times_{\ob \cobx} \mor \cobx}_{n}.$$
Note that $N_0 \cobx = \ob \cobx$. The geometric realization of $N_{\bullet} \cobx$ is the \emph{classifying space} of $\cobx$ and
will be denoted by $B \cobx$. We can associate a unital topological category
$\cobx \oplus \mathbf{1}$ to a non-unital topological category $\cobx$, simply by formally adding identities (disjointly in the topological sense). Then the classifying space of $\cobx \oplus \mathbf{1}$, as a unital topological category, is isomorphic to the classifying space of $\cobx$, as a non-unital topological category.
\medskip
The following proposition is a variation of well-known results but formulated in the context of non-unital categories. Given a space $X$ and points $x,y \in X$, we write $\Omega_{x, y} X$ for the space of paths in $X$ starting at $x$ and ending at $y$. If $\cobx$ is a (non-unital, topological) category and $X, Y$ are objects of $\cobx$,
then every morphism from $X$ to $Y$ defines an element in $\Omega_{X, Y} B \cobx$,
and thus we obtain a continuous map
$$i_{X,Y}: \cobx(X, Y) \to \Omega_{X,Y} B \cobx.$$
For an object $X$ in a (non-unital) category $\cobx$, there is a `transport' (topological) category $X \wr \cobx$ associated to
the functor which is represented by the object $X$. The objects of
$X \wr \cobx$ are the morphisms in $\cobx$ with source $X$. This set
of objects is topologized as a subspace of the space of all morphisms
in $\cobx$. The morphisms are given by morphisms in $\cobx$ that
make the obvious triangle commute. The topology is defined as the pullback topology
using the commutative square of spaces
\begin{equation} \label{transport-cat}
\xymatrix{
\mor{(X \wr \cobx)} \ar[r] \ar[d] & \ob{(X \wr \cobx)} \ar[d] \\
\mor{(\cobx)} \ar[r] & \ob{(\cobx)}
}
\end{equation}
where the horizontal maps evaluate at the source. (This data can be used as the definition of the transport category, see \cite[pp. 235-236]{GMTW(2009)}.)
Note that the topological category $X \wr \cobx$ has an initial
object if $\cobx$ is unital: this is given by the identity of $X$.
In this case, it follows that its classifying space is contractible.
In general, there is a canonical functor
$(X \wr \cobx) \oplus \mathbf{1} \to X \wr (\cobx \oplus \mathbf{1})$
which is not an equivalence of categories.
For every object $Y$ in $\cobx$, there is a continuous map
$$j_{X, Y}: \cobx(X, Y) \to B(X \wr \cobx)$$
which is given by the inclusion of $0$-simplices.
\begin{prop} \label{technical-lemma}
Let $\cobx$ be a non-unital topological category and let $X$ be an object of $\cobx$ such that the projection map
\[\ob(X\wr \cobx) \to \ob(\cobx)\]
is a Serre fibration, and for every morphism $f: Y \to Z$, the induced map
$$f_*: \cobx(X, Y) \to \cobx(X, Z)$$
is a weak equivalence of spaces. Then, for any object $Y$ of $\cobx$, the canonical map
$$(i_{X, Y}, j_{X, Y}): \cobx(X, Y) \to \Omega_{X, Y} B \cobx \times B(X \wr \cobx)$$
is a weak equivalence.
\end{prop}
\begin{proof}
There is a projection functor $p: X \wr \cobx \to \cobx$, $(X \to Y) \mapsto Y$. The induced map between classifying spaces is homotopic to the constant map at $X$. To see this, it suffices to note that the associated functor
$$p \oplus \mathbf{1} \colon (X \wr \cobx) \oplus \mathbf{1} \to \cobx \oplus \mathbf{1}$$
admits a natural tranformation from the constant functor at $X$.
Since the map $\ob(X \wr \cobx) \to \ob(\cobx)$ is a Serre fibration, the assertion will follow if we show that the square
\[
\xymatrix{
\ob{(X \wr \cobx)} \ar[r] \ar[d] & B(X \wr \cobx) \ar[d] \\
\ob{(\cobx)} \ar[r] & B(\cobx)
}
\]
is a homotopy pullback. For this, it suffices to show inductively that the square
\begin{equation} \label{pullback-sq}
\xymatrix{
\parallel N_{\bullet}(X \wr \cobx) \parallel_{k-1} \ar[r] \ar[d] & \parallel N_{\bullet} (X \wr \cobx) \parallel_k \ar[d] \\
\parallel N_{\bullet} \cobx \parallel_{k-1} \ar[r] & \parallel N_{\bullet} \cobx \parallel_k
}
\end{equation}
is a homotopy pullback for $k \geq 0$. The claim is obvious for $k = 0$. For the inductive step, note that the right vertical map in \eqref{pullback-sq} is the pushout of the vertical maps in the following diagram
\[
\xymatrix{
\parallel N_{\bullet}( X \wr \cobx) \parallel_{k-1} \ar[d] & N_k (X \wr \cobx) \times \partial \Delta^{k} \ar[r] \ar[l] \ar[d] & N_k(X \wr \cobx) \times \Delta^k \ar[d] \\
\parallel N_{\bullet} \cobx \parallel_{k-1} & N_k(\cobx) \times \partial \Delta^k \ar[r] \ar[l] & N_k(\cobx) \times \Delta^k
}
\]
The right square is clearly a homotopy pullback. Note that the projection
$$N_k(X \wr \cobx) \to N_k(\cobx)$$
is a Serre fibration since it is obtained as a pullback of $\ob(X \wr \cobx) \to \ob(\cobx)$. We assume inductively that the left vertical map in \eqref{pullback-sq} is a quasi-fibration.
To see that the left square in the last diagram is a homotopy pullback, it suffices to show that the maps between the vertical (homotopy) fibers are weak equivalences.
Each of these maps is either the identity or it is given by post-composition with a morphism in $\cobx$. By assumption, these maps are weak equivalences, therefore the left square above
is a homotopy pullback. Thus the square \eqref{pullback-sq} is also a homotopy pullback and the right vertical map in \eqref{pullback-sq} is again a quasi-fibration.
This concludes the proof.
\end{proof}
\begin{rem}
Proposition \ref{technical-lemma} holds more generally for classes of weak equivalences between spaces that are closed under homotopy colimits.
For example, if the maps $f_*$ are homology equivalences, then the resulting map $(i_{X, Y}, j_{X, Y})$ is also a homology equivalence.
The proof is essentially the same. See also \cite[Appendix A]{Raptis}.
\end{rem}
\section{Basic properties of the $h$-cobordism category} \label{sec:invertibility_hcob}
In this section, we prove some basic results about the homotopy type of the classifying space of the $h$-cobordism category $\hcobd \subseteq \cobdstr$. The main result (Proposition
\ref{hcod-is-a-grpd}) shows that the homotopy type of each component reduces to the homotopy type of a space of endomorphisms. This is a consequence of the invertibility of
$h$-cobordisms by the classical $h$-cobordism theorem, combined with the results of Section \ref{sec:non_unital_categories}.
\medskip
We will need the following preliminary lemmas.
\begin{lem} \label{etale-category}
The source-target map
\[(s,t): \mor \cobdstr \to \ob \cobdstr \times \ob \cobdstr\]
is a Serre fibration. As a consequence, the same is true for $\hcobd$.
\end{lem}
\begin{proof}
It suffices to show that for each cobordism $(W; M_0, M_1)$, the map that restricts to the (vertical) boundary
\begin{equation} \label{Serre-fibration}
B_{\theta}(W) \longrightarrow B_{\theta}(M_0) \times B_\theta(M_1)
\end{equation}
is a Serre fibration. Consider the commutative square
$$
\xymatrix{
\Emb(W, [0,1] \times \corn\infty1) \times \Bun(TW,\theta) \ar[r] \ar[d] & B_{\theta}(W) \ar[d]\\
\Emb(M_0 \sqcup M_1, \{0, 1\} \times \corn \infty 1)\times \Bun(\eps \oplus (TM_0\sqcup TM_1), \theta) \ar[r] & B_{\theta}(M_0) \times B_{\theta}(M_1)
}
$$
The left vertical restriction map is a Serre fibration in each factor. This can be shown for the first factor following the methods of \cite{Lima}; for the second factor this is true because we restrict along a cofibration. Secondly, the horizontal surjective
quotient maps are also Serre fibrations. This can be obtained as a special case of the main result in \cite[Appendix A]{Perlmutter}. Then
it follows that \eqref{Serre-fibration} is a Serre fibration.
\end{proof}
\begin{lem} \label{thin-identities}
Let $M = (M, l_M)$ be an object of $\cobdstr$. Then $B(M\wr\cobdstr)$ and $B(M \wr \hcobd)$ are weakly contractible.
\end{lem}
\begin{proof}
Consider $M\times [0,1]$ as a $\theta$-manifold with the cylindrical $\theta$-structure induced from $l_M$. For any object $N=(N,l_N)$, the composition map
\[\cobdstr(M, N)\to \cobdstr(M, N), \quad W\mapsto W\circ (M\times [0,1])\]
is a weak equivalence. It follows, using Lemma \ref{etale-category}, that the functor
\[F\colon M \wr \cobdstr \to M \wr \cobdstr\]
induced by precomposition with $M\times [0,1]$ is a weak equivalence on each level of the nerve, and hence on the geometric realization.
On the other hand, the map $B(F)$ is homotopically constant because $F\oplus \mathbf 1$ receives a natural transformation from the constant functor
at $M\times [0, 1]$. This shows that $B(M\wr\cobdstr)$ is contractible. The same argument applies to $B(M\wr\hcobd)$.
\end{proof}
The following proposition establishes one of the main properties of $\hcobd$ as a consequence of the $h$-cobordism theorem. To state it, denote by $\pi_0 \hcobd$ the ordinary category which has the same objects as $\hcobd$ (with the discrete topology), and morphism spaces
\[(\pi_0 \hcobd)(M, N):= \pi_0 (\hcobd(M, N)).\]
Note that this turns out to be a unital category, where $M\times [0,1]$ (with the cylindrical $\theta$-structure) is the identity morphism of $M$.
\begin{defn}
A $\theta$-structured $h$-cobordism $W\in \mor \hcobd$ is called \emph{invertible} if it defines an isomorphism in $\pi_0 \hcobd$.
\end{defn}
\begin{prop} \label{hcod-is-a-grpd}
For $d\geq 7$, any morphism in $\hcobd$ is invertible.
\end{prop}
\begin{proof}
A morphism in $\hcobd$ is represented by a compact smooth neatly embedded $d$-manifold with corners, depicted as follows,
\[
\xymatrix{
\dell(\dell_h W) \ar[r] \ar[d] & \dell_h W \ar[d]\\
M\amalg N \ar[r] & W
}
\]
together with a bundle map $l_W\colon TW\to \theta$ inducing $\theta$-structures $l_M$ on $\dell_0 W=M$ and $l_N$ on $\dell_1 W=N$.
It is enough to show that any such morphism admits a right inverse. We first construct an (abstract) $h$-cobordism $U$ from $N$ to $M$, such that $W\circ U$ is diffeomorphic, relative to both ends, to $N\times [0,1]$.
In the case where $\dell_h W$ is empty or diffeomorphic to a product cobordism, and $d\geq 6$, the existence of such a $U$ is a well-known consequence of the $s$-cobordism theorem.
To construct $U$ in the general case, we first construct a cobordism $U'$ with target $M$ such that the horizontal boundary $\dell_h(W\circ U')$ is diffeomorphic to a product cobordism. To do this, we choose a right inverse $(\dell_h W)\inv$ of $\dell_h W$. Consider the manifold with corners
$$Z:= [0,1] \times M$$
and choose an embedding of $(\dell_h W)\inv$ into $(0,1] \times \dell M$. After rearranging the boundary pieces of $Z$, we obtain a cobordism $U'$ which ends at $\{1\} \times M$
and starts from
$$\dell_0 U' = \{0\} \times M \cup \overline{([0,1] \times \dell M - (\dell_h W)\inv)},$$
with horizontal boundary $\dell_h U' = (\dell^- W)\inv$. This is again an
$h$-cobordism.
This completes the construction of $U'$. Since $\dell_h(W \circ U')$ is diffeomorphic to a product cobordism, it has a right inverse $U''$. Then, $U:=U'\circ U''$ and $W \circ U$ will be
diffeomorphic, relative to both ends, to $N \times [0,1]$. This completes the construction of $U$.
Next choose a collar for $U$ and a neat embedding of $U$ into $[0,1]\times \corn{\infty}1$; this is possible because the space of neat embeddings is contractible \cite[Theorem 2.7]{Genauer(2008)}.
For the same reason, the induced embeddings of $M\cong \dell_1 U$ and $N\cong \dell_0 U$ into $\corn{\infty}1$ will be isotopic to the inclusion maps. We have already used in the proof of Lemma \ref{etale-category} that the restriction map
\[\Emb(U, [0,1]\times \corn{\infty}1)\to \Emb(N\sqcup M, \{0,1\}\times \corn{\infty}1)\]
is a Serre fibration; hence after possibly changing the neat embedding of $U$ by an isotopy, we can assume that $\dell_1 U=M$ and $\dell_0 U = N$ as subsets of $\corn{\infty}1$.
Next we construct a suitable $\theta$-structure on $U$. By the $h$-cobordism condition, any bundle map $\eps\oplus TM\to \theta$ extends to a bundle map $l'_U$ on $TU$, inducing
a $\theta$-structure $l'_N$ on $N=\dell_0 U$. The composite $\theta$-structure on $W\circ U\cong N\times [0,1]$ is then a homotopy of bundle maps between $l_N$ and $l'_N$.
We insert a backwards copy of this bundle homotopy in a cylinder near $N\times \{0\}$. This defines a new $\theta$-structure $l_U$ on $U$ which induces $l_N$ on $N=\dell_0 U$. Composing
$l_U$ with $l_W$ defines a $\theta$-structure on $W \circ U \cong N \times [0,1]$ which is homotopic, relative to both ends, to the cylindrical $\theta$-structure.
Then the morphism $U=(1, U, l_U)$ is a morphism from $(N, l_N)$ to $(M, l_M)$, and $W \circ U=N\times [0,1]$ in $\pi_0 \hcobd(N,N)$.
\end{proof}
We can now prove our main result in this section.
\begin{prop} \label{h-cob-cat}
Let $d \geq 7$ and $M$ and $N$ be objects of $\hcobd$. Then, the canonical map
$$i_{M,N}: \hcobd(M, N) \to \Omega_{M, N} B \hcobd$$
is a weak equivalence.
\end{prop}
\begin{proof}
We check that the assumptions of Proposition \ref{technical-lemma} are satisfied. The projection $\ob (M\wr\hcobd) \to \ob(\hcobd)$ is a Serre fibration because the source-target map $(s,t)$ is a Serre fibration by Lemma \ref{etale-category}.
Moreover, by Lemma \ref{thin-identities}, the space $B(M \wr \hcobd)$ is weakly contractible. Lastly, any $h$-cobordism $W\colon N\to P$ in $\hcobd$ is invertible by Lemma \ref{hcod-is-a-grpd}, and it follows easily that the induced map
\[W_*\colon \hcobd(M, N)\to \hcobd(M,P)\]
is a homotopy equivalence of spaces, with right (left) homotopy inverse induced by a right (left) inverse of $W$.
\end{proof}
\section{Comparison with the $h$-cobordism space}\label{sec:h_cob_space}
\subsection{\normalfont \textit{Outline}}
We start by giving an informal definition of the terms appearing in Theorem \ref{thm:main1} and an outline of the strategy of proof. We fix a vector bundle $\theta = (V \to X)$ of rank $d$.
Given a $(d-1)$-dimensional compact manifolds $M$, $N$ with $\theta$-structures $l_M$ and $l_N$, a \emph{$\theta$-embedding} from $M$ into $N$ is an embedding $e\colon M\to N$ together with a bundle homotopy
from $l_M$ to $l_N\circ (\id \oplus De)$. It is called a \emph{$\theta$-$h$-embedding} if, moreover, the complement $N-e(\inter M)$ is an $h$-cobordism from $e(\dell M)$ to $\dell N$.
Given a $\dell$-trivial $h$-cobordism $(W;M,N)$ and a $\theta$-structure on $M$, we first note that $W$ and hence $N$ inherit $\theta$-structures from the one on $M$. Then given
a $\theta$-$h$-embedding $e$ from $M$ into $N$, we can construct an endomorphism of $M$ in $\hcobd$ by introducing corners in $W$ at $\dell M$ and $e(\dell M)$. This construction
defines the map
\[\Emb^\sim_\theta(M,N)\to \hcobd(M,M) \xrightarrow[4.5]{\simeq} \Omega_M B\hcobd\]
appearing in the statement of Theorem \ref{thm:main1}.
On the other hand, given any morphism $W\colon M\to N$ in $\hcobd$, after straightening the corners and forgetting the $\theta$-structure, we may view $W$ as an $h$-cobordism on $M$, whose loose end is $\dell_h W\cup N$. Applying this construction to the case $N=M$ defines the second map of Theorem \ref{thm:main1},
\[\Omega_M B\hcobd \xleftarrow[4.5]{\simeq} \hcobd(M,M) \to \Hdiff(M).\]
To make sure that these constructions are continuous and prove the theorem, we replace $\Omega_M B\hcobd\simeq \hcobd(M,M)$ by yet another space $\HdiffM(M)$ whose points are $\dell$-trivial $h$-cobordisms $W$ on $M$ (for technical reasons, this is assumed to have codimension 2 corners at $\dell M$),
together with a $\theta$-structure on $W$ extending the one on $M$, and an additional $\theta$-$h$-embedding from $M$ into the loose end of $W$. The construction from above refines to a map
\[\hcobd(M,M)\to \HdiffM(M),\]
and in Proposition \ref{homotopy-equivalence} we show that this map is a homotopy equivalence. The homotopy inverse map corresponds to the operation of introducing corners at the
boundary of the additional embedding. The homotopy fiber of the forgetful map $\HdiffM(M)\to \Hdiff(M)$, over some $(W;M,N)$ is easily seen to be equivalent to $\Emb^\sim_\theta(M,N)$,
so Theorem \ref{thm:main1} follows.
\subsection{\normalfont \textit{The $h$-cobordism space $\Hdiff(M)$}} Let $M$ be a compact smooth $(d-1)$-manifold with boundary $\partial M$. A \emph{$\dell$-trivial $h$-cobordism on $M$} is a compact smooth
$d$-manifold $W$ whose boundary splits as a union
$$\dell W = M\cup N$$
along a smooth submanifold of codimension one which is the common boundary of $M$ and $N$. Following \cite{WJR}, the \emph{$h$-cobordism space} $\Hdiff(M)_{\bullet}$ is a simplicial set where a
$p$-simplex is a smooth bundle $\pi \colon W \to \Delta^p$ with a trivial subbundle $M \times \Delta^p \subseteq E$ such that for each $x \in \Delta^p$, $\pi^{-1}(x)$ is a $\partial$-trivial
$h$-cobordism on $M \cong M \times \{x\}$. This is a classifying space for bundles of $\partial$-trivial $h$-cobordisms on $M$, i.e., there is a weak equivalence
$$|\Hdiff(M)_{\bullet}| \simeq \coprod_{W} B \Diff(W; M)$$
where $W$ is a $\partial$-trivial $h$-cobordism on $M$, one from each diffeomorphism class. Here $\Diff(W; M)$ denotes the group of diffeomorphisms of $W$ that fix $M$ pointwise.
It will be convenient to use the following geometric model for the $h$-cobordism space. Fix a neat embedding $M_0 = (M, e_M \colon M \hookrightarrow \{0\} \times \RR_+ \times \RR^N)$. We define a model for $|\Hdiff(M)_{\bullet}|$ by
$$\Hdiff(M_0) := \coprod_W \Hdiff(M_0)_W := \coprod_W \Emb(W, \corn{\infty}{2}; M_0)/ \Diff(W;M)$$
where:
\begin{itemize}
\item[(i)] the coproduct ranges over the diffeomorphism classes of compact smooth $d$-manifolds with corners $(W; M, N, \partial M = \partial N)$, where $W$ is collared as manifold with corners at $\partial M$, and the inclusions $M \hookrightarrow W \hookleftarrow N$ are homotopy equivalences. We refer to these as \emph{cornered $\partial$-trivial $h$-cobordisms on $M$}. This index set agrees with the set of diffeomorphism classes of $\partial$-trivial
$h$-cobordisms on $M$ by the process of straightening corners.
\item[(ii)] $\Emb(W, \corn{\infty}{2}; M_0)$ denotes the space of smooth neat embeddings into $\RR_+ \times \RR_+ \times \RR^{\infty}$
that restrict to the chosen embedding $e_M$ on $M$.
\item[(iii)] $\Diff(W; M)$ denotes the subgroup of elements in $\Diff(W)$ that fix $M$ pointwise and are cylindrical at the bicollared boundary. (This is not the same as $\Diff(W; M)$ for a
manifold $W$ with boundary but no corners, in the sense used earlier, but the operation of straightening the corners defines a homotopy equivalence between these two models.)
\end{itemize}
This defines a model for the same classifying space by straightening corners, using the methods described in the Apppendix, and by the contractibilty of the spaces of collars
(cf. \cite[1.1.1(b)]{WJR}).
\subsection{\normalfont \textit{The $h$-cobordism space $\HdiffM(M_0)$}}
We compare the $h$-cobordism space $\Hdiff(M_0)$ with the loop space at $M_0$ of the classifying space of the $h$-cobordism category. More specifically, using Proposition \ref{h-cob-cat},
we will compare the $h$-cobordism space directly with the corresponding mapping space in the $h$-cobordism category. Note that the main differences between these two spaces is
that the $h$-cobordisms in $\hcobd(M_0, M_0)$ are $\theta$-structured and both of their ends are fixed and given by $M$, while in $\Hdiff(M_0)$ only one end of the $h$-cobordisms is
fixed and identified with $M$. Our main result in this section will produce a highly connected map connecting these two spaces when appropriate $\theta$-structures are considered.
\medskip
For the general comparison of these two spaces, we introduce an auxiliary intermediate space of $h$-cobordisms denoted by $\HdiffM(M_0)$. We will show
that this is a model up to homotopy equivalence for the space $\hcobd(M_0, M_0)$ which is convenient for the comparison with $\Hdiff(M_0)$.
We consider first the case without $\theta$-structures. In this case, the space $\HdiffMno(M_0)$
is a variation of the $h$-cobordism space $\Hdiff(M_0)$ in which each cornered $\partial$-trivial $h$-cobordism comes with an additional embedding of $M$ into the boundary of the cobordism and this embedding satisfies certain conditions. The precise definition is as follows: for each cornered $\partial$-trivial $h$-cobordism $W$ on $M$, we consider the space
\begin{equation*}
\begin{aligned}
\HdiffMno(M_0)_W := \Emb(W, \corn{\infty}{2}; M_0) \times_{\Diff(W;M)} \Emb^{\sim}(M, \partial W - M)
\end{aligned}
\end{equation*}
where:
\begin{itemize}
\item[(i)] $\Emb(W, \corn{\infty}{2}; M_0)$ denotes the space of smooth neat embeddings which extend the embedding $e_M$ of $M$.
\item[(ii)] $\Emb^{\sim}(M, \partial W - M)$ denotes the \emph{space of h-embeddings}, that is, smooth collared embeddings $\iota \colon M \to \partial W - M$ such that
$\big(\partial W - \inter(M)\big) - \inter(\iota(M))$ is an $h$-cobordism from $\partial M$ to itself. In particular, the embedding $\iota$ is a homotopy equivalence.
\item[(iii)] The group $\Diff(W;M)$ acts on $\Emb(M, \partial W - M)$ by post-composition, and on $\Emb(W, \corn{\infty}{2}; M_0)$ by precomposition.
\end{itemize}
We define $\HdiffMno(M_0) \colon = \coprod_W \HdiffMno(M_0)_W$ where the coproduct ranges over the diffeomorphism classes of cornered $\partial$-trivial $h$-cobordisms on $M$. This space differs from $\hcobdno(M_0, M_0)$ essentially in two respects: first,
given an element $$(\psi \colon W \hookrightarrow \corn{\infty}{2}, \quad \iota \colon M \hookrightarrow \partial W - M),$$ the induced embedding $\psi \circ \iota$ is not
fixed, and secondly, such pairs appear as endomorphisms in $\hcobdno(M_0, M_0)$ after some of their corner points are straightened. For dealing with the first difference, it will be convenient to
consider the following map in order to compare $\HdiffMno(M_0)$ with the endomorphisms in the $h$-cobordism category,
$$
\HdiffMno(M_0)_W \xrightarrow{r_W} \Emb(M, \corn{\infty}{1}), \quad r_W(\psi, \iota)= \psi \circ \iota.$$
\begin{lem} \label{crazy-map}
The map $r_W$ is a Serre fibration.
\end{lem}
\begin{proof}
It suffices to show that the map
$$\widetilde{r}_{W, n} \colon \Emb(W, \corn{n}{2}; M_0) \times \Emb^{\sim}(M, \partial W - M) \longrightarrow \Emb(M, \corn{n-1}{1})$$
is a Serre fibration for $n$ large enough. Let $j \colon M \hookrightarrow \corn{n-1}{1}$ be an embedding. By \cite{Lima}, there is an open neighborhood
$U_j$ containing $j$ and a map $z \colon U_j \to \Diff(\corn{n}{2})$ such
that $j' = z(j') \circ j$ for each $j' \in U_j$. Then the map
$$\widetilde{r}_{W, n}^{-1}(U_j) \rightarrow \widetilde{r}_{W, n}^{-1}(j) \times U_j, \quad (\psi, \iota) \mapsto (z(\psi \circ \iota)^{-1} \circ \psi, \iota, \psi \circ \iota)$$
is a homeomorphism over $U_j$ with inverse given by $(\psi, \iota, j') \mapsto (z(j') \circ \psi, \iota)$. Therefore $\widetilde{r}_{W, n}$ is a fiber bundle
and the required result follows.
\end{proof}
The homotopy equivalence between $\hcobdno(M_0, M_0)$ and $\HdiffMno(M_0)$ is based on the following observations. Firstly, as a consequence of Lemma \ref{crazy-map} and
the fact that $\Emb(M, \corn{\infty}{1})$ is contractible, the inclusion of a fiber $r_W^{-1}(j) \hookrightarrow \HdiffMno(M_0)_W$ is a weak equivalence for each embedding
$j \colon M \hookrightarrow \corn{\infty}{1}$. Thus, we may restrict to the subspace of pairs $(\psi, \iota) \in \HdiffMno(M_0)_W$ for which $\psi \circ \iota$ is a fixed embedding
without changing the homotopy type. Secondly, an endomorphism in $\hcobdno(M_0, M_0)$ gives rise to a pair in this subspace by straightening the corners at the outgoing boundary.
Moreover, this procedure can also reversed up to homotopy by introducing corners at the embedded $\iota(\partial M)$. This is the rough outline of the comparison between these two spaces which we follows next in more detail.
\smallskip
There is a comparison map between the spaces $\hcobdno(M_0, M_0)$ and $\HdiffMno(M_0)$, $$\mathcal{J} \colon \hcobdno(M_0, M_0) \to \HdiffMno(M_0),$$
which is defined as follows. For each $h$-cobordism $(a, W) \in \hcobdno(M_0, M_0)$, let $\Phi(W)$ denote the cornered $\partial$-trivial $h$-cobordism
on $M$ that is obtained by straightening the corner at $\{a\} \times \{0\} \times \RR^{\infty}$ -- the map $\Phi$ that straightens corners is discussed in detail in the Appendix. Then, for each such
$W$, the map $\mathcal{J}$ is induced by the map (ignoring the auxiliary $a > 0$):
$$\Emb(W, [0,1] \times \RR_+ \times \RR^{\infty}; \partial_v W) \to \Emb(\Phi(W), \corn{\infty}{2}; M_0) \times \Emb^{\sim}(M, \partial \Phi(W) - M)$$
which sends a neat embedding of the $h$-cobordism $(W; M_0, M_1)$:
$$\psi \colon W \to [0,1] \times \RR_+ \times \RR^{\infty}$$
to the pair $(\Phi(\psi), \iota_{M_1})$, where:
$$\Phi(\psi) \colon \Phi(W) \xrightarrow{\Phi^{-1}} W \xrightarrow{\psi} [0,1] \times \RR_+ \times \RR^{\infty} \xrightarrow{\Phi \times \mathrm{id}} \corn{\infty}{2}$$
$$\iota \colon M = M_1 \stackrel{\Phi}{\cong} \Phi(M_1) \subseteq \partial \Phi(W) - M_0.$$
Note that this map preserves the respective actions of the diffeomorphism groups
$\Diff(W; \partial_v W)$ and $\Diff(\Phi(W); M_0)$.
\begin{prop} \label{homotopy-equivalence}
The map $\mathcal{J}$ is a homotopy equivalence.
\end{prop}
\begin{proof}
Since $r_W$ is a fibration and $\Emb(M, \RR_+ \times \RR^{\infty})$ is contractible, it follows that the space $\HdiffMno(M_0)_{\Phi(W)}$ is homotopy equivalent
to the fiber of $r_{\Phi(W)}$ at the point $\Phi(e_M)$, denoted by $F$. By definition, the map $\mathcal{J}$ takes values in this
fiber and we claim that the restriction of this map
$$\mathcal{J}_W \colon \hcobdno(M_0, M_0)_W \to F$$
is a weak equivalence. To see this, it suffices to identify these two spaces as classifying spaces of certain types of bundles, and see that these types of bundles
correspond to each other up to isomorphism bijectively. For a CW-complex $X$, the set of maps
$$X \to \hcobdno(M_0, M_0)_W (\simeq B \Diff(W; \partial_v W))$$
corresponds to the set $S_1$ of pairs $(E, \pi \colon E \to X)$, where:
\begin{itemize}
\item $E$ is a subspace of $X \times I \times \corn{\infty}{1}$ such that the projection $\pi \colon E \to X$ is a bundle with fiber $W$, as smooth $h$-cobordism with corners, and the inclusion of each fiber to $I \times \corn{\infty}{1}$ is a neat embedding,
\item the restriction of $\pi \colon E \to X$ to the fiberwise vertical boundary
$\partial^{\pi}_v E \to X$ is identified with $X \times (M \sqcup M) \subseteq X \times I \times \corn{\infty}{1}$ (using $e_{M,i} \colon M \hookrightarrow \{i\} \times \corn{\infty}{1}$ for $i=0,1$).
\end{itemize}
On the other hand, the set of maps $X \to F$ corresponds to the set $S_2$ of triples $(V, \xi \colon V \to X, \iota)$, where:
\begin{itemize}
\item $V$ is a subspace of $X \times \corn{\infty}{2}$ such that the projection $\xi \colon V \to X$ is a bundle with fiber $\Phi(W)$, as $\partial$-trivial $h$-cobordism on $M$, and the inclusion of each fiber to $\corn{\infty}{2}$ is a neat embedding,
\item the restriction to the fiberwise incoming boundary $\partial^{\xi}_0 V \to X$ is given by $X \times M \subseteq X \times \{0\} \times \corn{\infty}{1}$ (using $e_M$),
\item there is an embedding $\iota \colon X \times M \hookrightarrow (\partial^{\xi} V - \partial^{\xi}_0 V)$ which is fiberwise an $h$-embedding. Moreover, the image of this embedding agrees with the subspace
$X \times M \subseteq X \times \RR_+ \times \{0\} \times \RR^{\infty}$ (using $\Phi(e_M))$.
\end{itemize}
The process of fiberwise straightening the corners at the boundary $\partial M$ of the fiberwise outgoing boundary of $\pi \colon E \to X$ yields a map $S_1 \to S_2$. Conversely, the process of introducing corners to $\xi \colon V \to X$, fiberwise at $\iota(\partial M \times X)$, defines a map in the other direction $S_2 \to S_1$. (These operations can be made using the map $\Phi$ and the methods of the Appendix.) These two operations induce inverse bijections after passing to the concordance classes of elements in $S_1$ and $S_2$, respectively. Lastly, an analogous arguments shows that the map $\mathcal{J}$ is $\pi_0$-surjective.
\end{proof}
\smallskip
Next we discuss the definition and properties of $\HdiffM(M_0)$ for general $\theta$-structures. We consider a fixed vector bundle $\theta = (V \to X)$ of rank $d$. In addition, $(M, e_M, l_M)$ will denote
a fixed $\theta$-structure on $M_0$ (assuming this exists), so that this can be regarded as an object in $\hcobd$. Informally, the space $\HdiffM(M_0)$ is a variation of
the space $\HdiffMno(M_0)$ in which the $h$-cobordisms are also endowed with a $\theta$-structures together with an identification of $l_M$ with the $\theta$-structure
that is induced by the additional embedding of $M$. The precise definition is as follows. For each cornered $\partial$-trivial
$h$-cobordism $W$ on $M$, we consider the space
\begin{equation*}
\mathcal{E}_{\theta, W} := \big(\Emb(W, \corn{\infty}{2}; M_0) \times \Bun(TW, \theta ; \eps \oplus TM)\big) \times_{\Diff(W;M)} \Emb^{\sim}(M, \partial W - M)
\end{equation*}
where:
\begin{itemize}
\item[(i)] $\Emb(W, \corn{\infty}{2}; M_0)$ denotes the space of neat embeddings which extend the embedding $e_M$ of $M$.
\item[(ii)] $\Emb^{\sim}(M, \partial W - M)$ denotes the space of h-embeddings, i.e., smooth collared embeddings $\iota \colon M \to \partial W - M$ such that $\big(\partial W - \inter(M)\big) - \inter(\iota(M))$
is an $h$-cobordism from $\partial M$ to itself. (In particular, the embedding $\iota$ is a homotopy equivalence.)
\item[(iii)] $\Bun(TW, \theta; \eps \oplus TM)$ denotes the space of bundle
maps $TW \to \theta$ which restrict to the chosen $\theta$-structure on the collared boundary $M$.
\item[(iv)] The group $\Diff(W;M)$ acts on $\Emb(M, \partial W - M)$ by post-composition, and on $\Emb(W, \corn{\infty}{2}; M_0) \times \Bun(TW, \theta; \eps \oplus TM)$ by pre-composition.
\end{itemize}
\medskip
We consider the restriction map
$$
\mathcal{E}_{\theta, W} \xrightarrow{c_W} \Bun(\eps\oplus TM, \theta), \quad c_W(\psi, l_W , \iota )= l_W \circ (\id \oplus D(\iota)).$$
\begin{defn}
Let $W$ be a cornered $\partial$-trivial $h$-cobordism on $M$. We define
$\HdiffM(M_0)_W$ to be the homotopy fiber of the map $c_W$
taken over the point $l_M \in \Bun(\eps \oplus TM, \theta)$. We write
$$\HdiffM(M_0) \colon = \coprod_W \HdiffM(M_0)_W$$
where the coproduct ranges over cornered $\partial$-trivial $h$-cobordisms on $M$, one from each diffeomorphism class.
\end{defn}
Observe that this definition agrees up to homotopy equivalence with our previous definition of $\HdiffMno(M_0)$, in the case where there are no $\theta$-structures,
as a consequence of Lemma \ref{crazy-map}.
There is a comparison map between the spaces $\hcobd(M_0, M_0)$ and $\HdiffM(M_0)$, $$\mathcal{J}_{\theta} \colon \hcobd(M_0, M_0)
\to \HdiffM(M_0),$$
which is defined as follows.
For each $h$-cobordism $(a, W, l_W) \in \hcobd(M_0, M_0)$, the map $\mathcal{J}_{\theta}$
is induced by the following map (for $\epsilon$ sufficiently small):
\newpage
\begin{equation} \label{tricky-map}
\Emb^{\epsilon}(W, [0,1] \times \RR_+ \times \RR^{\infty}; \partial_v W) \times \Bun(TW, \theta; \eps \oplus TM)
\end{equation}
\begin{center}$\big\downarrow $ \end{center}
\begin{equation*} \Emb(\Phi(W), \corn{\infty}{2}; M_0) \times \Bun(T\Phi(W), \theta; \eps \oplus TM) \times \Emb^{\sim}(M, \partial \Phi(W) - M_0)
\end{equation*}
which sends a $\epsilon$-neat embedding of the $\theta$-structured $h$-cobordism $(W; M_0, M_1)$:
$$\psi \colon W \to [0,1] \times \RR_+ \times \RR^{\infty}, \quad TW \xrightarrow{l} \theta$$
to the triple $(\Phi(\psi), \Phi(l), \iota_{M_1})$ defined as follows:
$$\Phi(\psi) \colon \Phi(W) \xrightarrow{\Phi^{-1}} W \xrightarrow{\psi} [0,1] \times \RR_+ \times \RR^{\infty} \xrightarrow{\Phi \times \mathrm{id}} \corn{\infty}{2}$$
$$\Phi(l) \colon T\Phi(W) \xrightarrow{(T\Phi^{-1}, \simeq)} TW \xrightarrow{l} \theta$$
$$\iota \colon M = M_1 \stackrel{\Phi}{\cong} \Phi(M_1) \subseteq \partial \Phi(W) - M_0.$$
Here $(T \Phi^{-1}, \simeq)$ stands for a fixed \emph{choice} of an isomorphism which covers the map $\Phi^{-1} \colon \Phi(W) \to W$. This isomorphism is supposed to agree with the bundle map induced by $\Phi^{-1}$ outside a fixed $\epsilon$-small neighborhood of the corner points, and comes together with a bundle homotopy which identifies it with this bundle map (induced by $\Phi^{-1}$) away from the corner points.
Then note that the map \eqref{tricky-map} preserves the respective actions of the diffeomorphism groups
$\Diff^\epsilon(W; \partial_v W)$ and $\Diff(\Phi(W); M_0)$. This correspondence defines a map to $\mathcal{E}_{\theta, W}$ which produces canonically up homotopy
a map to the homotopy fiber $\HdiffM(M_0)_W$ by using the choice of the bundle homotopy for $(T\Phi^{-1}, \simeq)$.
\begin{prop} \label{homotopy-equivalence-2}
The map $\mathcal{J}_{\theta}$ is a homotopy equivalence.
\end{prop}
\begin{proof}
Note that for each $W$ there is a square, commutative up to a preferred homotopy,
$$
\xymatrix{
\hcobd(M_0, M_0)_W \ar[r]^{\mathcal{J}_{\theta}} \ar[d]_{p_W} & \HdiffM(M_0)_{\Phi(W)} \ar[d]^{q_{\Phi(W)}} \\
\hcobdno(M_0, M_0)_W \ar[r]^{\mathcal{J}}_{\simeq} & \HdiffMno(M_0)_{\Phi(W)}
}
$$
where the vertical maps are given by forgetting the $\theta$-structures. By Proposition \ref{homotopy-equivalence}, it suffices to show that this is a homotopy
pullback. Consider the following homotopy commutative diagram
$$
\xymatrix{
\mathrm{hofib}(q_{\Phi(W)}) \ar[r] \ar[dd] & \Bun(T\Phi(W), \theta; \eps \oplus TM) \ar[dd]^{i} \ar[rd] & \\
&& \Bun(\eps \oplus TM, \theta) \\
\HdiffM(M_0)_{\Phi(W)} \ar[r] \ar[rd]_{q_{\Phi(W)}} & \mathcal{E}_{\theta, \Phi(W)} \ar[d] \ar[ru]^{c_{\Phi(W)}} \\
& \HdiffMno(M_0)_{\Phi(W)} &&&&
}
$$
where both columns are homotopy fiber sequences, taken over a point $(\psi, \iota) \in \HdiffMno(M_0)_{\Phi(W)}$, and both rows define homotopy fiber sequences by definition.
This diagram identifies the homotopy fiber of $q_{\Phi(W)}$ with the homotopy fiber of the map
$$c_{\Phi(W)} \circ i \colon \Bun(T\Phi(W), \theta; \eps \oplus TM) \to \Bun(\eps \oplus TM, \theta)$$
$$(l \colon T\Phi(W) \to \theta) \mapsto (\eps \oplus TM \xrightarrow{\id \oplus D(\iota)} T \Phi(W) \xrightarrow{l} \theta).$$
Then the canonical map between the homotopy fibers $\mathrm{hofiber}(p_W) \to \mathrm{hofib}(q_{\Phi(W)})$ is identified with the canonical
weak equivalence
$$\Bun(TW, \theta; \eps \oplus T \partial_v W) \simeq \mathrm{hofiber} \big( \Bun(T\Phi(W), \theta; \eps \oplus TM) \xrightarrow{c_{\Phi(W)} \circ i} \Bun(\eps \oplus TM, \theta)\big)$$
and the result follows.
\end{proof}
\medskip
Next we compare the space $\HdiffM(M_0)$ with the $h$-cobordism space $\Hdiff(M_0)$. For each $\partial$-trivial $h$-cobordism $W$ on $M$, there is a comparison map
$$\mathcal{Q}_{\theta, W} \colon \HdiffM(M_0)_W \to \Hdiff(M_0)_W$$
which simply forgets the additional embedding of $M$ and the $\theta$-structures.
\begin{defn}
Let $N$ be a smooth $(d-1)$-dimensional manifold and let $\tau \colon \eps^1 \oplus TN \to \theta$ be a bundle map endowing $N$ with a $\theta$-structure. We define
the \emph{space of} $\theta h$-\emph{embeddings} $\Emb^{\sim}_{\theta}(M, N)$ to be the homotopy fiber at $(l_M \colon \eps \oplus TM \to \theta)$ of the map
$$s_{\theta, N} \colon \Emb^{\sim}(M, N) \longrightarrow \Bun(\eps \oplus TM, \theta), \quad (\iota \colon M \to N) \mapsto \big(\tau \circ (\id \oplus D(\iota))\big).$$
\end{defn}
\begin{prop} \label{fiber-sequence}
Let $W$ be a cornered $\partial$-trivial $h$-cobordism on $M$ together with a $\theta$-structure $l_W \colon TW \to \theta$ which extends $l_M$. There is a homotopy fiber sequence
$$\Emb^{\sim}_{\theta}(M, \partial W - M) \to \HdiffM(M_0)_W
\xrightarrow{\mathcal{Q}_{\theta, W}} \Hdiff(M_0)_W.$$
\end{prop}
\begin{proof}
The map $s_{\theta, N}$ induces a map as follows,
$$\mathcal{E}_{\theta, W} \xrightarrow{g} \mathcal{E}'_{\theta, W}:= \Emb(W, \corn{\infty}{2}; M_0)\times_{\Diff(W;M)} \bigl(\Bun(TW, \theta; \eps \oplus TM)\times \Bun(\eps \oplus TM, \theta)\bigr).$$
Since $W$ is an $h$-cobordism, the space $\Bun(TW, \theta; \eps \oplus TM)$ is contractible. Therefore the homotopy fiber of this map $g$ is identified with $\Emb^{\sim}_{\theta}(M, \partial W - M)$
(taken over the point which is defined by $l_M$). Moreover, since $\Diff(W; M)$ acts trivially on $\Bun(\eps \oplus TM, \theta)$, we have
$$\mathcal{E}'_W \simeq \Hdiff(M_0)_W \times \Bun(\eps \oplus TM, \theta).$$
Then consider the following homotopy commutative diagram:
\[
\xymatrix{
\HdiffM(M_0)_W \ar[r] \ar[dd]_{\mathcal{Q}_{\theta, W}} & \mathcal{E}_W \ar[rd]^{c_W} \ar[dd]^g \\
& & \Bun(\eps \oplus TM, \theta) \\
\Hdiff(M_0)_W \ar[r] & \mathcal{E}'_W \ar[ru]_p
}
\]
where $p$ is identified with the canonical projection. By definition, the top row is a homotopy fiber sequence at $l_M$. The lower row is also a homotopy fiber sequence by the previous observations. Hence the left square is a homotopy pullback and the result follows.
\end{proof}
\noindent \textbf{Proof of Theorem \ref{thm:main1}.} Combine Propositions \ref{homotopy-equivalence-2} and \ref{fiber-sequence} with Proposition \ref{h-cob-cat}. \qed
\subsection{\normalfont \textit{Tangent bundles as $\theta$-structures}} In this subsection, we specialize our previous results to the case where the $\theta$-structure corresponds to the tangent bundle of a compact smooth $(d-1)$-manifold.
\smallskip
Let $M$ be a smooth connected compact $(d-1)$-manifold of handle dimension
$k < d - 3$. Let $\theta_M$ denote the vector bundle $\eps \oplus TM$ over
$M$ of rank $d$. Note that a $\theta_M$-structure on a compact smooth $d$-cobordism $W$ is a bundle map $TW \to \eps \oplus TM.$
Clearly $M$ admits a $\theta_M$-structure which is given by the identity map $\id_{\eps \oplus TM}$.
\begin{prop} \label{connectivity-result}
Let $W$ be a cornered $\partial$-trivial $h$-cobordism on $M$ together with a $\theta_M$-structure $l_W \colon TW \to \eps \oplus TM$ which extends the identity structure on $M$. Then the
space $\Emb^{\sim}_{\theta_M}(M, \partial W - M)$ is $(d-2k-3)$-connected.
\end{prop}
\begin{proof}
Let $N = \partial W - M$. First consider the inclusion map $\Emb(M, N) \to \Imm(M, N)$ into the space of immersions. This map
is $(d-2k-2)$-connected by \cite[p.~70]{Weiss(99)}, \cite[Corollary 2.5]{GW}.
Note that for such homotopical assertions, it suffices to work with the interior of the manifold $M$. By our assumptions on the handle dimension $k < d-3$, it follows that $\partial M$ is also connected and $\pi_1(\partial M) \cong \pi_1(M)$. As a consequence, the space $\Emb^{\sim}(M, N)$ consists exactly of those smooth embeddings which are homotopy equivalences. By restricting to
the appropriate set of path components, we obtain also a $(d-2k-2)$-connected map $\Emb^{\sim}(M, N) \to \Imm^{\sim}(M,N)$.
By immersion theory and the Hirsch-Smale theorem, the space of immersions $\Imm(M, N)$ is homotopy equivalent to the space $\Bun(TM, TN)$. Note that $M$ has no closed
components because it is homotopy equivalent to a space of smaller dimension. As a consequence, we also have $\Imm^{\sim}(M, N) \simeq \Bun^{\sim}(TM, TN)$ where the latter space
contains
those bundles maps which are homotopy equivalences between the bases.
The space $\Bun(TM, TN)$ includes into the space $\Bun(\eps \oplus TM , \eps \oplus TN)$ by stabilization. This map is $(d-k-2)$-connected since the inclusion
$O_{d-1} \to O_{d}$ is $(d-2)$-connected and $M$ is $k$-dimensional up to homotopy equivalence.
Lastly, the map $\Bun(\eps \oplus TM, \eps \oplus TN) \to \Bun(\eps \oplus TM, \eps \oplus TM)$, given by composition with the $\theta_M$-structure on $N$,
is a homotopy equivalence. This is because this $\theta_M$-structure is a homotopy equivalence between the bases since it is restricted from $l_W$ and $W$ is an
$h$-cobordism on $M$.
As a consequence, the map $s_{\theta_M, N} \colon \Emb^{\sim}(M, N) \longrightarrow \Bun^{\sim}(\eps \oplus TM, \eps \oplus TM)$
is $(d-2k-2)$-connected. Therefore its homotopy fiber $\Emb^{\sim}_{\theta_M}(M, \partial W - M)$ is $(d-2k-3)$-connected.
\end{proof}
\noindent \textbf{Proof of Theorem \ref{thm:main2}.} As before, we fix an embedding $e_M \colon M \hookrightarrow \{0\} \times \RR_+ \times \RR^{\infty}$ and
consider $M_0 = (M, e_M, \id_{\eps \oplus TM})$ as an object of $\hcobd$. Then combine Proposition \ref{connectivity-result} with Theorem \ref{thm:main1}. \qed
\appendix
\section{Straightening corners}
The goal of this appendix is to give a rigorous model for the straightening procedure used in this article.
Let $I:=[0,1]$ and $\RR_+:=[0,\infty)$. Choose, once and for all, a homeomorphism
\begin{equation}\label{eq:bending_map}
\Phi\colon I\times \RR_+ \to \RR_+\times \RR_+
\end{equation}
which is a diffeomorphism except at the corner point $(1,0)$, and which is prescribed in a neighborhood of the boundary as follows:
% \begin{enumerate}
% \item\label{item:straight1} At $[0,\eps)\times \RR_+$ and $[\eps/2, 1-\eps/2]\times [0, \eps)$, $\Phi$ is the identity;
% \item\label{item:straight2} at $(1-\eps, 1]\times [\eps/2, \infty)$, $\Phi$ is rotation by $\pi/4$;
% \item\label{item:straight3} in the $\eps$-ball around the corner point $(1,0)$, $\Phi$ preserves the radial coordinate and scales the angle by a diffeomorphism
% \[\alpha\colon [0,\pi/2] \to [0,\pi] \]
% where $\alpha(t)=t$ on $[0,\pi/6]$ and $\alpha(t)=t+\pi/2$ on $[\pi/3, \pi/2]$.
% \end{enumerate}
\begin{enumerate}
\item\label{item:straight1} in a neighborhood of $\{0\} \times \RR_+ $ and of $[0,1)\times \{0\}$, $\Phi$ is the identity;
\item\label{item:straight2} in a neighborhood of $\{1\}\times (0,\infty)$, $\Phi$ is rotation by $\pi/2$; \item\label{item:straight3} in a neighborhood of the corner point $(1,0)$, $\Phi$ preserves the radial coordinate and scales the angle by a diffeomorphism
\[\alpha\colon [0,\pi/2] \to [0,\pi] \]
where $\alpha(t)=t$ on $[0,\pi/6]$, and $\alpha(t)=t+\pi/2$ on $[\pi/3, \pi/2]$. Here the angle is measured clockwise from (1,0).
\end{enumerate}
\begin{rem}
A more naive condition in (\ref{item:straight3}) would be that $\alpha$ just doubles the angle. But this is inconsistent with conditions (\ref{item:straight1}) and (\ref{item:straight2}).
\end{rem}
We will show at the end of this section that such a map exists, and, moreover, that the space of all such maps, equipped with a suitable topology, is contractible.
\smallskip
For the ease of notation, we will also write $\Phi$ for any map of the type $\Phi\times \id_{\RR^n}$. With this convention in mind, if $W\subset I\times \RR_+\times \RR^{n-2}$ is a compact neat submanifold, then
\[\Phi(W)\subset \RR_+\times \RR_+ \times \RR^{n-2}\]
is a again neat submanifold. Note that $\Phi(W)$ and $W$ are homeomorphic. They are diffeomorphic except at the region $\{1\}\times \{0\} \times \RR^{n-2}$, where $W$ has corner points but $\Phi(W)$ does
not. Moreover, if
\[e\colon W\to I\times\RR_+\times \RR^{n-2}\]
is a neat embedding, then
\[\Phi\circ e\circ \Phi\inv\colon \Phi(W)\to \RR_+\times \RR_+\times \RR^{n-2}\]
is also a smooth neat embedding.
\begin{lem}\label{lem:bending_continuous}
For compact smooth manifold $W$, possibly with corners, the induced map
\begin{equation}\label{eq:bending_map_on_embedding}
\Phi\circ - \circ \Phi\inv\colon \Emb(W, I\times \RR_+\times \RR^{n-2}) \to \Emb(\Phi(W), \RR_+\times \RR_+\times \RR^{n-2})
\end{equation}
is continuous.
\end{lem}
Recall that the topology on the spaces of (neat) embedding under consideration is such that a map from the embedding space is continuous
if and only if it is continuous on the restriction to the subspace of $\eps$-neat embeddings, for every $\eps>0$.
\begin{proof}
\newcommand{\halfinter}{\mathrm{int}'}
We denote by
\[\halfinter W:=W\cap ([0,1)\times (0,\infty)\times \RR^{n-2}).\]
For any $\delta>0$, there is an (injective, continuous) restriction map
\[R\colon \Emb^\delta(\Phi(W), \RR_+\times \RR_+\times \RR^{n-2}) \to \Emb(\Phi(\halfinter W), \RR_+\times (0,\infty)\times \RR^{n-2})\]
where both the domain and the target have the compact-open $C^\infty$-topology.
\smallskip
\noindent \textbf{Claim.} The first space has the subspace topology of the second.
\smallskip
This claim will imply the Lemma: choosing $\delta>0$ small enough, we have a commutative square
\[\xymatrix{
\Emb^\eps(W, I\times \RR_+\times \RR^{n-2}) \ar[rr]^R \ar[d]^{\Phi\circ - \circ \Phi\inv}
&& \Emb(\halfinter W, [0,1)\times (0,\infty)\times \RR^{n-2}) \ar[d]^{\Phi\circ - \circ \Phi\inv}
\\
\Emb^\delta(\Phi(W), \RR_+\times \RR_+\times \RR^{n-2}) \ar[rr]^R
&& \Emb(\Phi(\halfinter W), \RR_+\times (0,\infty)\times \RR^{n-2}))
}\]
where the right vertical map is continuous, for it is given by pre- and post-composition with a smooth map; then the left vertical map is continuous as well, by the definition of the subspace topology.
To prove the claim, we recall that the topology on the domain of $R$ is given by the family of semi-norms
\[\Vert e\Vert_{\beta,\alpha}:= \sup_{x\in U_\beta} \big\Vert \frac{\partial^\alpha (e\circ h_\beta)}{\partial x_{\alpha}} (x) \big\Vert\]
where $\beta$ runs through the indexing set of a cover of $\Phi(W)$ by compact charts
\[h_\beta\colon \RR_+\times \RR_+\times \RR^{d-2} \supset U_\beta \to V_\beta\subset \Phi(W),\]
and $\alpha$ runs through the set of multi-indices in $\{1, \dots, n\}$. We may choose such a cover as follows:
the charts that meet a $\delta/2$-neighborhood of $\Phi(W-\halfinter W)$, are of the form $\bar h_\beta\times \id_{[0,\delta/2]}$ for some chart $\bar h_\beta$ on the boundary.
In contrast, the topology on the target of $R$ is given by a similar family of semi-norms $\Vert e\Vert_{\beta', \alpha}$ where $\beta'$ now runs through the indexing set of a cover of $\halfinter \Phi(W)$ by compact charts. Here we use the following cover: the charts that do not meet the $\delta/2$-neighborhood of $\Phi(W-\halfinter W)$ are precisely
the same as above; each chart $\bar h_\beta\times \id_{[0,\delta/2]}$ as above is covered by the countably many charts $\bar h_\beta\times \id_{[\delta/n, \delta/2]}$.
Now assume that $e$ is in the image of $R$. Then, for any $\alpha$, the semi-norms of such a countable family agree with the seminorm of the chart $\bar h_\beta\times \id_{[0,\delta/2]}$, for $e$ is $\delta$-neat. This shows that the family of seminorms on the image of $R$ is equivalent to the family of seminorms on the domain of $R$.
\end{proof}
Let $\theta:= (V \to X)$ be a vector bundle of rank $d$ on a space $X$. Next we discuss how $\Phi(W)$ inherits a $\theta$-structure from $W$. The map
\[\Phi\colon W\to \Phi(W)\]
is a diffeomorphism except at the set corner points, which we denote by $\dell_{01}W$. This induces a bundle isomorphism
\[D\Phi \colon TW\vert_{W-\dell_{01}W} \to T\Phi(W)\vert_{\Phi(W-\dell_{01}W)}\]
covering $\Phi$. This bundle map does not extend to a bundle map on $TW$ (unless $\dell_{01}W$ is empty). Thus, instead of $\theta$-structures on $W$, it will
be convenient to work here
%We consider the following variation of the notion of $\theta$-structure on $W$:
with bundle maps
\[l_W\colon TW \vert_{\inter W} \to \theta.\]
Such a bundle map determines, uniquely up to a contractible choice, a bundle map on all of $TW$, which then defines a $\theta$-structure on $W$. As a consequence,
a $\theta$-structure $l_W$ on $W$ induces (canonically up to a contractible space of choices) a $\theta$-structure on $\Phi(W)$ that corresponds to the bundle map
\[l_{\Phi(W)}\colon T\Phi(W)\vert_{\inter W} \xrightarrow{D\Phi\inv} TW\vert_{\inter W} \xrightarrow{l_W} \theta.\]
We will not explicitly write down the corresponding (zigzag) continuous map between the respective spaces of embeddings of manifolds with $\theta$-structures.
\begin{rem}
It should be noted that with this weaker definition of $\theta$-structure on the morphism space of $\cobdn$, the resulting space of $\theta$-structured cobordisms will not be the morphism
space of a topological category anymore (there is no canonical source-target map anymore).
\end{rem}
Next we show that a map $\Phi$ as postulated in \eqref{eq:bending_map} exists. Choose first some diffeomorphism $\alpha$ as required in condition (\ref{item:straight3}),
and some $0<\eps<1$. Denote by $B_\eps$ the open $\eps$-ball around $(1,0)\in I\times \RR_+$, and define $\Phi$ first on $B_\eps$ by rescaling the angle by $\alpha$. Then
we note that we can extend this map smoothly by the identity to the entire open $\eps/2$-neighborhood of $I\times \RR_+$ and by rotation by $\pi/2$ to the entire open
$\eps/2$-neighborhood of the ray $\{1\}\times \RR_+$. Using such an extension, we may clearly further extend $\Phi$ to some neighborhood $N$ of $\dell(I\times \RR_+)$ in such
a way so that conditions (\ref{item:straight1}) and (\ref{item:straight3}) hold.
To extend $\Phi$ to all of $I\times \RR_+$, we use an isotopy extension argument. In more detail, consider the isotopy
\[\alpha_s\colon [0,\pi/2]\to [0,\pi], \quad t\mapsto t\cdot (1-s)+\alpha(t)\cdot s \quad (s\in [0,1]).\]
Then is an obvious isotopy between the inclusion $N\to \RR_+\times \RR_+$ and $\Phi\vert_N$, which scales the angle at time $s$ by $\alpha_s$. This is a smooth isotopy except at the corner point $(1,0)$ and hence comes equipped with a flow vector field outside $(1,0)$, which we may extend to a smooth vector field on all $\RR_+\times \RR_+-\{(1,0)\}$. We can certainly also arrange that the slopes of the vectors stay bounded as $y\to \infty$, decreasing $N$ if necessary. In this case the flow from every point in $I\times \RR_+$ exists at least to the time $s=1$. Evaluating the flow at time $s=1$ then defines a map $\Phi$ as required.
\smallskip
If $\Phi$ and $\Phi'$ are two maps as in \eqref{eq:bending_map}, then
\[H:= \Phi'\circ \Phi\inv\colon \RR_+\times \RR_+\to \RR_+\times \RR_+\]
is a homeomorphism, diffeomorphism except at $(1,0)$, and there is an $\eps>0$ such that the restriction of $H$ to $\RR_+\times [0,\eps)$ is
as follows:
\begin{enumerate}
\item In polar coordinates around $(1,0)$, $H$ preserves the radius and rescales the angle by some self-diffeomorphism $\alpha$ of $[0,\pi/2]$ which is the identity on $[0,\pi/6]$ and $[\pi/3, \pi/2]$.
\item Outside the cone formed by the rays of angle $\pi/6$ and $\pi/3$ starting at $(1,0)$, $H$ is the identity.
\end{enumerate}
Denote by $D^\eps$ the space of all such self-homeomorphisms of $\RR_+\times \RR_+$, equipped with the locally convex topology induced by the semi-norms
\[ \Vert H\Vert_{K, \alpha} := \sup_{x\in K} \big\Vert \frac{\partial^\alpha H}{\partial x_{\alpha}} (x)\big\Vert,\]
where $K$ ranges over the compact subsets of the interior of $\RR_+\times \RR_+$, and $\alpha$ ranges over the multi-indices in $\{1,2\}$. Finally, we let $D=\colim_{\eps\to 0} D^\eps$.
\begin{lem}
$D$ is contractible.
\end{lem}
\begin{proof}
The inclusion $D^{2\eps}\to D^\eps$ is homotopic to a map which lands inside the subspace $D^\eps_\dell$ of those maps where $\alpha=\id$. The homotopy is given by taking the straight line isotopy between $\alpha$ and the identity, and applying it in the radial coordinate around $(1,0)$.
Now note that any element of $D^\eps_\dell$ defines just as well a diffeomorphism of the manifold $K$ obtained from $\RR_+ \times \RR_+$ by smoothing the corners.
In more detail, let $K$ be the manifold with boundary (but without corners) obtained from $\RR_+\times \RR_+$ by straightening the corners, and let $\Diff^\eps_\dell(K)$ denote the space of diffeomorphisms of $K$ which are the identity in an $\eps$-neighborhood of the boundary, which the compact-open $C^\infty$-topology. Then the inclusion of $D^\eps_\dell$ into the colimit factors canonically through some $\Diff^{\eps'}_\dell$. As $K$ is diffeomorphic to $\RR_+\times \RR$, we have $\Diff^\eps_\dell(K)\simeq *$, by the usual rescaling isotopy.
Hence the inclusion $D^{2\eps}\to D$ is nullhomotopic, for any $\eps$.
\end{proof}
\begin{thebibliography}{1}
\bibitem{ERW}
J.~Ebert, O.~Randal-Williams, \emph{Semi-simplicial spaces}, Preprint (arXiv). \url{https://arxiv.org/abs/1705.03774}
\bibitem{GMTW(2009)}
S.~Galatius, U.~Tillmann, I.~Madsen, and M.~Weiss, \emph{The homotopy type of
the cobordism category.} Acta Math. 202 (2009), no.~2, 195--239.
\bibitem{Genauer(2008)}
J.~Genauer, \emph{Cobordism categories of manifolds with corners.}
Trans. Amer. Math. Soc. 364 (2012), no.~1, 519--550.
\bibitem{GW}
T.~Goodwillie, M.~Weiss, \emph{Embeddings from the point of view of immersion theory. II.}
Geom. Topol. 3 (1999), 103--118.
\bibitem{Lima}
E.~L.~Lima, \emph{On the local triviality of the restriction map for embeddings.}
Comment. Math. Helv. 38 (1964), 163--164.
\bibitem{Perlmutter}
N.~Perlmutter, \emph{Cobordism category of manifolds with Baas-Sullivan singularities.}
M\"unster~J.~Math. 8 (2015), no.~1, 119--167.
\bibitem{Raptis}
G.~Raptis, \emph{On the homotopy type of certain cobordism categories of surfaces.} Int. Math. Res. Notices 8 (2014), 2037--2089.
\bibitem{WJR}
F.~Waldhausen, B.~Jahren, J.~Rognes, \emph{Spaces of PL manifolds and categories of simple maps.}
Annals of Mathematics Studies, 186. Princeton University Press, Princeton, NJ, 2013.
\bibitem{Weiss(99)}
M.~Weiss, \emph{Embeddings from the point of view of immersion theory. I.}
Geom. and Topology 3 (1999), 67--101
\end{thebibliography}
\end{document}