%!TEX root = ../blob1.tex

\section{Action of \texorpdfstring{$\CH{X}$}{$C_*(Homeo(M))$}}

\label{sec:evaluation}

\nn{should comment at the start about any assumptions about smooth, PL etc.}

Let $CH_*(X, Y)$ denote $C_*(\Homeo(X \to Y))$, the singular chain complex of

the space of homeomorphisms

between the $n$-manifolds $X$ and $Y$ (any given singular chain extends a fixed homeomorphism $\bd X \to \bd Y$).

We also will use the abbreviated notation $CH_*(X) \deq CH_*(X, X)$.

(For convenience, we will permit the singular cells generating $CH_*(X, Y)$ to be more general

than simplices --- they can be based on any linear polyhedron.

\nn{be more restrictive here?  does more need to be said?})

\begin{prop}  \label{CHprop}

For $n$-manifolds $X$ and $Y$ there is a chain map

\eq{

    e_{XY} : CH_*(X, Y) \otimes \bc_*(X) \to \bc_*(Y)

}

such that

\begin{enumerate}

\item on $CH_0(X, Y) \otimes \bc_*(X)$ it agrees with the obvious action of 

$\Homeo(X, Y)$ on $\bc_*(X)$ (Property (\ref{property:functoriality})), and

\item for any compatible splittings $X\to X\sgl$ and $Y\to Y\sgl$, 

the following diagram commutes up to homotopy

\begin{equation*}

\xymatrix@C+2cm{

      CH_*(X, Y) \otimes \bc_*(X)

        \ar[r]_(.6){e_{XY}}  \ar[d]^{\gl \otimes \gl}   &

            \bc_*(Y)\ar[d]^{\gl} \\

     CH_*(X\sgl, Y\sgl) \otimes \bc_*(X\sgl) \ar[r]_(.6){e_{X\sgl Y\sgl}}   & 	\bc_*(Y\sgl)  

}

\end{equation*}

\end{enumerate}

Up to (iterated) homotopy, there is a unique family $\{e_{XY}\}$ of chain maps

satisfying the above two conditions.

\end{prop}

Before giving the proof, we state the essential technical tool of Lemma \ref{extension_lemma}, 

and then give an outline of the method of proof.

Without loss of generality, we will assume $X = Y$.

\medskip

Let $f: P \times X \to X$ be a family of homeomorphisms (e.g. a generator of $CH_*(X)$)

and let $S \sub X$.

We say that {\it $f$ is supported on $S$} if $f(p, x) = f(q, x)$ for all

$x \notin S$ and $p, q \in P$. Equivalently, $f$ is supported on $S$ if 

there is a family of homeomorphisms $f' : P \times S \to S$ and a ``background"

homeomorphism $f_0 : X \to X$ so that

\begin{align*}

	f(p,s) & = f_0(f'(p,s)) \;\;\;\; \mbox{for}\; (p, s) \in P\times S \\

\intertext{and}

	f(p,x) & = f_0(x) \;\;\;\; \mbox{for}\; (p, x) \in {P \times (X \setmin S)}.

\end{align*}

Note that if $f$ is supported on $S$ then it is also supported on any $R \sup S$.

(So when we talk about ``the" support of a family, there is some ambiguity,

but this ambiguity will not matter to us.)

Let $\cU = \{U_\alpha\}$ be an open cover of $X$.

A $k$-parameter family of homeomorphisms $f: P \times X \to X$ is

{\it adapted to $\cU$} 

if the support of $f$ is contained in the union

of at most $k$ of the $U_\alpha$'s.

\begin{lemma}  \label{extension_lemma}

Let $x \in CH_k(X)$ be a singular chain such that $\bd x$ is adapted to $\cU$.

Then $x$ is homotopic (rel boundary) to some $x' \in CH_k(X)$ which is adapted to $\cU$.

Furthermore, one can choose the homotopy so that its support is equal to the support of $x$.

\end{lemma}

The proof will be given in \S\ref{sec:localising}.

\medskip

Before diving into the details, we outline our strategy for the proof of Proposition \ref{CHprop}.

Let $p$ be a singular cell in $CH_k(X)$ and $b$ be a blob diagram in $\bc_*(X)$.

We say that $p\ot b$ is {\it localizable} if there exists $V \sub X$ such that

\begin{itemize}

\item $V$ is homeomorphic to a disjoint union of balls, and

\item $\supp(p) \cup \supp(b) \sub V$.

\end{itemize}

(Recall that $\supp(b)$ is defined to be the union of the blobs of the diagram $b$.)

Assuming that $p\ot b$ is localizable as above, 

let $W = X \setmin V$, $W' = p(W)$ and $V' = X\setmin W'$.

We then have a factorization 

\[

	p = \gl(q, r),

\]

where $q \in CH_k(V, V')$ and $r \in CH_0(W, W')$.

We can also factorize $b = \gl(b_V, b_W)$, where $b_V\in \bc_*(V)$ and $b_W\in\bc_0(W)$.

According to the commutative diagram of the proposition, we must have

\[

	e_X(p\otimes b) = e_X(\gl(q\otimes b_V, r\otimes b_W)) = 

				gl(e_{VV'}(q\otimes b_V), e_{WW'}(r\otimes b_W)) .

\]

Since $r$ is a  0-parameter family of homeomorphisms, we must have

\[

	e_{WW'}(r\otimes b_W) = r(b_W),

\]

where $r(b_W)$ denotes the obvious action of homeomorphisms on blob diagrams (in

this case a 0-blob diagram).

Since $V'$ is a disjoint union of balls, $\bc_*(V')$ is acyclic in degrees $>0$ 

(by Properties \ref{property:disjoint-union} and \ref{property:contractibility}).

Assuming inductively that we have already defined $e_{VV'}(\bd(q\otimes b_V))$,

there is, up to (iterated) homotopy, a unique choice for $e_{VV'}(q\otimes b_V)$

such that 

\[

	\bd(e_{VV'}(q\otimes b_V)) = e_{VV'}(\bd(q\otimes b_V)) .

\]

Thus the conditions of the proposition determine (up to homotopy) the evaluation

map for localizable generators $p\otimes b$.

On the other hand, Lemma \ref{extension_lemma} allows us to homotope 

arbitrary generators to sums of localizable generators.

This (roughly) establishes the uniqueness part of the proposition.

To show existence, we must show that the various choices involved in constructing

evaluation maps in this way affect the final answer only by a homotopy.

Now for a little more detail.

(But we're still just motivating the full, gory details, which will follow.)

By Lemma \ref{extension_lemma} we can restrict our attention to $k$-parameter families 

$p$ of homeomorphisms such that $\supp(p)$ is contained in the union of $k$ $\gamma$-balls.

For fixed blob diagram $b$ and fixed $k$, it's not hard to show that for $\gamma$ small enough

$p\ot b$ must be localizable.

On the other hand, for fixed $k$ and $\gamma$ there exist $p$ and $b$ such that $p\ot b$ is not localizable,

and for fixed $\gamma$ and $b$ there exist non-localizable $p\ot b$ for sufficiently large $k$.

Thus we will need to take an appropriate limit as $\gamma$ approaches zero.

The construction of $e_X$, as outlined above, depends on various choices, one of which 

is the choice, for each localizable generator $p\ot b$, 

of disjoint balls $V$ containing $\supp(p)\cup\supp(b)$.

Let $V'$ be another disjoint union of balls containing $\supp(p)\cup\supp(b)$,

and assume that there exists yet another disjoint union of balls $W$ containing 

$V\cup V'$.

Then we can use $W$ to construct a homotopy between the two versions of $e_X$ 

associated to $V$ and $V'$.

If we impose no constraints on $V$ and $V'$ then such a $W$ need not exist.

Thus we will insist below that $V$ (and $V'$) be contained in small metric neighborhoods

of $\supp(p)\cup\supp(b)$.

Because we want not mere homotopy uniqueness but iterated homotopy uniqueness,

we will similarly require that $W$ be contained in a slightly larger metric neighborhood of 

$\supp(p)\cup\supp(b)$, and so on.

\medskip

\begin{proof}[Proof of Proposition \ref{CHprop}.]

We'll use the notation $|b| = \supp(b)$ and $|p| = \supp(p)$.

Choose a metric on $X$.

(e.g.\ $\ep_i = 2^{-i}$).

Choose another sequence of positive real numbers $\delta_i$ such that $\delta_i/\ep_i$

converges monotonically to zero (e.g.\ $\delta_i = \ep_i^2$).

Let $\phi_l$ be an increasing sequence of positive numbers

satisfying the inequalities of Lemma \ref{xx2phi} below.

Given a generator $p\otimes b$ of $CH_*(X)\otimes \bc_*(X)$ and non-negative integers $i$ and $l$

define

\[

	N_{i,l}(p\ot b) \deq \Nbd_{l\ep_i}(|b|) \cup \Nbd_{\phi_l\delta_i}(|p|).

\]

In other words, for each $i$

we use the metric to choose nested neighborhoods of $|b|\cup |p|$ (parameterized

by $l$), with $\ep_i$ controlling the size of the buffers around $|b|$ and $\delta_i$ controlling

the size of the buffers around $|p|$.

Next we define subcomplexes $G_*^{i,m} \sub CH_*(X)\otimes \bc_*(X)$.

Let $p\ot b$ be a generator of $CH_*(X)\otimes \bc_*(X)$ and let $k = \deg(p\ot b)

= \deg(p) + \deg(b)$.

We say $p\ot b$ is in $G_*^{i,m}$ exactly when either (a) $\deg(p) = 0$ or (b)

there exist codimension-zero submanifolds $V_0,\ldots,V_m \sub X$ such that each $V_j$

is homeomorphic to a disjoint union of balls and

\[

	N_{i,k}(p\ot b) \subeq V_0 \subeq N_{i,k+1}(p\ot b)

\]

and further $\bd(p\ot b) \in G_*^{i,m}$.

We also require that $b$ is splitable (transverse) along the boundary of each $V_l$.

Note that $G_*^{i,m+1} \subeq G_*^{i,m}$.

As sketched above and explained in detail below, 

$G_*^{i,m}$ is a subcomplex where it is easy to define

the evaluation map.

The parameter $m$ controls the number of iterated homotopies we are able to construct

(see Lemma \ref{m_order_hty}).

The larger $i$ is (i.e.\ the smaller $\ep_i$ is), the better $G_*^{i,m}$ approximates all of

$CH_*(X)\ot \bc_*(X)$ (see Lemma \ref{Gim_approx}).

Next we define a chain map (dependent on some choices) $e_{i,m}: G_*^{i,m} \to \bc_*(X)$.

(When the domain is clear from context we will drop the subscripts and write

simply  $e: G_*^{i,m} \to \bc_*(X)$).

Let $p\ot b \in G_*^{i,m}$.

If $\deg(p) = 0$, define

\[

	e(p\ot b) = p(b) ,

\]

where $p(b)$ denotes the obvious action of the homeomorphism(s) $p$ on the blob diagram $b$.

For general $p\ot b$ ($\deg(p) \ge 1$) assume inductively that we have already defined

$e(p'\ot b')$ when $\deg(p') + \deg(b') < k = \deg(p) + \deg(b)$.

Choose $V = V_0$ as above so that 

\[

	N_{i,k}(p\ot b) \subeq V \subeq N_{i,k+1}(p\ot b) .

\]

Let $\bd(p\ot b) = \sum_j p_j\ot b_j$, and let $V^j$ be the choice of neighborhood

of $|p_j|\cup |b_j|$ made at the preceding stage of the induction.

For all $j$, 

\[

	V^j \subeq N_{i,k}(p_j\ot b_j) \subeq N_{i,k}(p\ot b) \subeq V .

\]

(The second inclusion uses the facts that $|p_j| \subeq |p|$ and $|b_j| \subeq |b|$.)

We therefore have splittings

\[

	p = p'\bullet p'' , \;\; b = b'\bullet b'' , \;\; e(\bd(p\ot b)) = f'\bullet f'' ,

\]

where $p' \in CH_*(V)$, $p'' \in CH_*(X\setmin V)$, 

$b' \in \bc_*(V)$, $b'' \in \bc_*(X\setmin V)$, 

$f' \in \bc_*(p(V))$, and $f'' \in \bc_*(p(X\setmin V))$.

(Note that since the family of homeomorphisms $p$ is constant (independent of parameters)

near $\bd V$, the expressions $p(V) \sub X$ and $p(X\setmin V) \sub X$ are

unambiguous.)

We have $\deg(p'') = 0$ and, inductively, $f'' = p''(b'')$.

%We also have that $\deg(b'') = 0 = \deg(p'')$.

Choose $x' \in \bc_*(p(V))$ such that $\bd x' = f'$.

This is possible by Properties \ref{property:disjoint-union} and \ref{property:contractibility}  and the fact that isotopic fields

differ by a local relation.

Finally, define

\[

	e(p\ot b) \deq x' \bullet p''(b'') .

\]

Note that above we are essentially using the method of acyclic models.

For each generator $p\ot b$ we specify the acyclic (in positive degrees) 

target complex $\bc_*(p(V)) \bullet p''(b'')$.

The definition of $e: G_*^{i,m} \to \bc_*(X)$ depends on two sets of choices:

The choice of neighborhoods $V$ and the choice of inverse boundaries $x'$.

The next lemma shows that up to (iterated) homotopy $e$ is independent

of these choices.

(Note that independence of choices of $x'$ (for fixed choices of $V$)

is a standard result in the method of acyclic models.)

%\begin{lemma}

%Let $\tilde{e} :  G_*^{i,m} \to \bc_*(X)$ be a chain map constructed like $e$ above, but with

%different choices of $x'$ at each step.

%(Same choice of $V$ at each step.)

%Then $e$ and $\tilde{e}$ are homotopic via a homotopy in $\bc_*(p(V)) \bullet p''(b'')$.

%Any two choices of such a first-order homotopy are second-order homotopic, and so on, 

%to arbitrary order.

%\end{lemma}

%\begin{proof}

%This is a standard result in the method of acyclic models.

%\nn{should we say more here?}

%\nn{maybe this lemma should be subsumed into the next lemma.  probably it should.}

%\end{proof}

\begin{lemma} \label{m_order_hty}

Let $\tilde{e} :  G_*^{i,m} \to \bc_*(X)$ be a chain map constructed like $e$ above, but with

different choices of $V$ (and hence also different choices of $x'$) at each step.

If $m \ge 1$ then $e$ and $\tilde{e}$ are homotopic.

If $m \ge 2$ then any two choices of this first-order homotopy are second-order homotopic.

Continuing, $e :  G_*^{i,m} \to \bc_*(X)$ is well-defined up to $m$-th order homotopy.

\end{lemma}

\begin{proof}

We construct $h: G_*^{i,m} \to \bc_*(X)$ such that $\bd h + h\bd = e - \tilde{e}$.

The chain maps $e$ and $\tilde{e}$ coincide on bidegrees $(0, j)$, so define $h$

to be zero there.

Assume inductively that $h$ has been defined for degrees less than $k$.

Let $p\ot b$ be a generator of degree $k$.

Choose $V_1$ as in the definition of $G_*^{i,m}$ so that

\[

	N_{i,k+1}(p\ot b) \subeq V_1 \subeq N_{i,k+2}(p\ot b) .

\]

There are splittings

\[

	p = p'_1\bullet p''_1 , \;\; b = b'_1\bullet b''_1 , 

			\;\; e(p\ot b) - \tilde{e}(p\ot b) - h(\bd(p\ot b)) = f'_1\bullet f''_1 ,

\]

where $p'_1 \in CH_*(V_1)$, $p''_1 \in CH_*(X\setmin V_1)$, 

$b'_1 \in \bc_*(V_1)$, $b''_1 \in \bc_*(X\setmin V_1)$, 

$f'_1 \in \bc_*(p(V_1))$, and $f''_1 \in \bc_*(p(X\setmin V_1))$.

Inductively, $\bd f'_1 = 0$ and $f_1'' = p_1''(b_1'')$.

Choose $x'_1 \in \bc_*(p(V_1))$ so that $\bd x'_1 = f'_1$.

Define 

\[

	h(p\ot b) \deq x'_1 \bullet p''_1(b''_1) .

\]

This completes the construction of the first-order homotopy when $m \ge 1$.

The $j$-th order homotopy is constructed similarly, with $V_j$ replacing $V_1$ above.

\end{proof}

Note that on $G_*^{i,m+1} \subeq G_*^{i,m}$, we have defined two maps,

$e_{i,m}$ and $e_{i,m+1}$.

An easy variation on the above lemma shows that 

the restrictions of $e_{i,m}$ and $e_{i,m+1}$ to $G_*^{i,m+1}$ are $m$-th 

order homotopic.

Next we show how to homotope chains in $CH_*(X)\ot \bc_*(X)$ to one of the 

$G_*^{i,m}$.

Choose a monotone decreasing sequence of real numbers $\gamma_j$ converging to zero.

Let $\cU_j$ denote the open cover of $X$ by balls of radius $\gamma_j$.

Let $h_j: CH_*(X)\to CH_*(X)$ be a chain map homotopic to the identity whose image is 

spanned by families of homeomorphisms with support compatible with $\cU_j$, 

as described in Lemma \ref{extension_lemma}.

Recall that $h_j$ and also the homotopy connecting it to the identity do not increase

supports.

Define

\[

	g_j \deq h_j\circ h_{j-1} \circ \cdots \circ h_1 .

\]

The next lemma says that for all generators $p\ot b$ we can choose $j$ large enough so that

$g_j(p)\ot b$ lies in $G_*^{i,m}$, for arbitrary $m$ and sufficiently large $i$ 

(depending on $b$, $\deg(p)$ and $m$).

%(Note: Don't confuse this $n$ with the top dimension $n$ used elsewhere in this paper.)

\begin{lemma} \label{Gim_approx}

Fix a blob diagram $b$, a homotopy order $m$ and a degree $n$ for $CH_*(X)$.

Then there exists a constant $k_{bmn}$ such that for all $i \ge k_{bmn}$

there exists another constant $j_{ibmn}$ such that for all $j \ge j_{ibmn}$ and all $p\in CH_n(X)$ 

we have $g_j(p)\ot b \in G_*^{i,m}$.

\end{lemma}

For convenience we also define $k_{bmp} = k_{bmn}$

and $j_{ibmp} = j_{ibmn}$ where $n=\deg(p)$.

Note that we may assume that

\[

	k_{bmp} \ge k_{alq}

\]

for all $l\ge m$ and all $q\ot a$ which appear in the boundary of $p\ot b$.

Additionally, we may assume that

\[

	j_{ibmp} \ge j_{ialq}

\]

for all $l\ge m$ and all $q\ot a$ which appear in the boundary of $p\ot b$.

\begin{proof}

There exists $\lambda > 0$ such that for every  subset $c$ of the blobs of $b$ $\Nbd_u(c)$ is homeomorphic to $|c|$ for all $u < \lambda$ .

We need to consider all such $c$ because all generators appearing in

iterated boundaries of $p\ot b$ must be in $G_*^{i,m}$.)

Let $r = \deg(b)$ and 

\[

	t = r+n+m+1 = \deg(p\ot b) + m + 1.

\]

Choose $k = k_{bmn}$ such that

\[

	t\ep_k < \lambda

\]

and

\[

	n\cdot (2 (\phi_t + 1) \delta_k) < \ep_k .

\]

Let $i \ge k_{bmn}$.

Choose $j = j_i$ so that

\[

	\gamma_j < \delta_i

\]

and also so that $\phi_t \gamma_j$ is less than the constant $\rho(M)$ of Lemma \ref{xxzz11}.

Let $j \ge j_i$ and $p\in CH_n(X)$.

Let $q$ be a generator appearing in $g_j(p)$.

Note that $|q|$ is contained in a union of $n$ elements of the cover $\cU_j$,

which implies that $|q|$ is contained in a union of $n$ metric balls of radius $\delta_i$.

We must show that $q\ot b \in G_*^{i,m}$, which means finding neighborhoods

$V_0,\ldots,V_m \sub X$ of $|q|\cup |b|$ such that each $V_j$

is homeomorphic to a disjoint union of balls and

\[

	N_{i,n}(q\ot b) \subeq V_0 \subeq N_{i,n+1}(q\ot b)

			\subeq V_1 \subeq \cdots \subeq V_m \subeq N_{i,t}(q\ot b) .

\]

Recall that

\[

	N_{i,a}(q\ot b) \deq \Nbd_{a\ep_i}(|b|) \cup \Nbd_{\phi_a\delta_i}(|q|).

\]

By repeated applications of Lemma \ref{xx2phi} we can find neighborhoods $U_0,\ldots,U_m$

of $|q|$, each homeomorphic to a disjoint union of balls, with

\[

	\Nbd_{\phi_{n+l} \delta_i}(|q|) \subeq U_l \subeq \Nbd_{\phi_{n+l+1} \delta_i}(|q|) .

\]

The inequalities above guarantee that 

for each $0\le l\le m$ we can find $u_l$ with 

\[

	(n+l)\ep_i \le u_l \le (n+l+1)\ep_i

\]

such that each component of $U_l$ is either disjoint from $\Nbd_{u_l}(|b|)$ or contained in 

$\Nbd_{u_l}(|b|)$.

This is because there are at most $n$ components of $U_l$, and each component

has radius $\le (\phi_t + 1) \delta_i$.

It follows that

\[

	V_l \deq \Nbd_{u_l}(|b|) \cup U_l

\]

is homeomorphic to a disjoint union of balls and satisfies

\[

	N_{i,n+l}(q\ot b) \subeq V_l \subeq N_{i,n+l+1}(q\ot b) .

\]

The same argument shows that each generator involved in iterated boundaries of $q\ot b$

is in $G_*^{i,m}$.

\end{proof}

In the next three lemmas, which provide the estimates needed above, we have made no effort to optimize the various bounds.

(The bounds are, however, optimal in the sense of minimizing the amount of work

we do.  Equivalently, they are the first bounds we thought of.)

We say that a subset $S$ of a metric space has radius $\le r$ if $S$ is contained in

some metric ball of radius $r$.

\begin{lemma}

Let $S \sub \ebb^n$ (Euclidean $n$-space) have radius $\le r$.  

Then $\Nbd_a(S)$ is homeomorphic to a ball for $a \ge 2r$.

\end{lemma}

\begin{proof} \label{xxyy2}

Let $S$ be contained in $B_r(y)$, $y \in \ebb^n$.

Note that if $a \ge 2r$ then $\Nbd_a(S) \sup B_r(y)$.

Let $z\in \Nbd_a(S) \setmin B_r(y)$.

Consider the triangle

with vertices $z$, $y$ and $s$ with $s\in S$.

The length of the edge $yz$ is greater than $r$ which is greater

than the length of the edge $ys$.

It follows that the angle at $z$ is less than $\pi/2$ (less than $\pi/3$, in fact),

which means that points on the edge $yz$ near $z$ are closer to $s$ than $z$ is,

which implies that these points are also in $\Nbd_a(S)$.

Hence $\Nbd_a(S)$ is star-shaped with respect to $y$.

\end{proof}

If we replace $\ebb^n$ above with an arbitrary compact Riemannian manifold $M$,

the same result holds, so long as $a$ is not too large:

\nn{replace this with a PL version}

\begin{lemma} \label{xxzz11}

Let $M$ be a compact Riemannian manifold.

Then there is a constant $\rho(M)$ such that for all

subsets $S\sub M$ of radius $\le r$ and all $a$ such that $2r \le a \le \rho(M)$,

$\Nbd_a(S)$ is homeomorphic to a ball.

\end{lemma}

\begin{proof}

Choose $\rho = \rho(M)$ such that $3\rho/2$ is less than the radius of injectivity of $M$,

and also so that for any point $y\in M$ the geodesic coordinates of radius $3\rho/2$ around

$y$ distort angles by only a small amount.

Now the argument of the previous lemma works.

\end{proof}

\begin{lemma} \label{xx2phi}

Let $S \sub M$ be contained in a union (not necessarily disjoint)

of $k$ metric balls of radius $r$.

Let $\phi_1, \phi_2, \ldots$ be an increasing sequence of real numbers satisfying

$\phi_1 \ge 2$ and $\phi_{i+1} \ge \phi_i(2\phi_i + 2) + \phi_i$.

For convenience, let $\phi_0 = 0$.

Assume also that $\phi_k r \le \rho(M)$,

where $\rho(M)$ is as in Lemma \ref{xxzz11}.

Then there exists a neighborhood $U$ of $S$,

homeomorphic to a disjoint union of balls, such that

\[

	\Nbd_{\phi_{k-1} r}(S) \subeq U \subeq \Nbd_{\phi_k r}(S) .

\]

\end{lemma}

\begin{proof}

For $k=1$ this follows from Lemma \ref{xxzz11}.

Assume inductively that it holds for $k-1$.

Partition $S$ into $k$ disjoint subsets $S_1,\ldots,S_k$, each of radius $\le r$.

By Lemma \ref{xxzz11}, each $\Nbd_{\phi_{k-1} r}(S_i)$ is homeomorphic to a ball.

If these balls are disjoint, let $U$ be their union.

Otherwise, assume WLOG that $S_{k-1}$ and $S_k$ are distance less than $2\phi_{k-1}r$ apart.

Let $R_i = \Nbd_{\phi_{k-1} r}(S_i)$ for $i = 1,\ldots,k-2$ 

and $R_{k-1} = \Nbd_{\phi_{k-1} r}(S_{k-1})\cup \Nbd_{\phi_{k-1} r}(S_k)$.