Lemma 17
Let
${\omega}_{n}\in {V}_{{m}_{n}}^{\prime}$
for
$n=1,2,...$
, where
${m}_{n}\to \infty $
. Then there exists for each
$n$
a representative
${S}_{n}$
of
${\omega}_{n}$
such that a subsequence of
${S}_{n}$
converges locally in
${L}_{1}^{p}$
over
$X$
to a smooth configuration representing an element of
$\overline{G}$
.
Proof. By assumption there exist for each
$n$
a representative
${\mathsf{S}}_{n}$
of
${\omega}_{n}$
and a configuration
${\overline{\mathsf{S}}}_{n}$
over
$X$
representing an element of
$\overline{G}$
such that
$$\begin{array}{c}{d}_{{m}_{n}}({\mathsf{S}}_{n},{\overline{\mathsf{S}}}_{n})<\frac{1}{{m}_{n}}.\end{array}$$ 
(11)

After passing to a subsequence we may assume (since
$\overline{G}$
is compact) that
$\left[{\overline{\mathsf{S}}}_{n}\right]$
converges in
$\overline{G}$
to some element
$\left[\overline{S}\right]$
, and we can choose
$\overline{S}$
smooth. Since
${M}_{\mathfrak{b}}={M}_{\mathfrak{b}}^{*}$
, the local slice theorem guarantees that for large
$n$
we can find
${u}_{n}\in {\mathcal{G}}_{\mathfrak{b}}$
such that
${\overline{S}}_{n}={u}_{n}\left({\overline{\mathsf{S}}}_{n}\right)$
satisfies
$$\parallel {\overline{S}}_{n}\overline{S}{\parallel}_{Lp,w1}\to 0.$$
Set
${S}_{n}={u}_{n}\left({\mathsf{S}}_{n}\right)$
, which is again a representative of
${\omega}_{n}$
. Let
$\overline{A},{\overline{A}}_{n}$
be the connection parts of
$\overline{S},{\overline{S}}_{n}$
, respectively. Then ( 11 ) implies that
${\overline{S}}_{n}{S}_{n}\to 0$
and
${\nabla}_{{\overline{A}}_{n}}({\overline{S}}_{n}{S}_{n})\to 0$
locally in
${L}^{p}$
over
$X$
, hence also
${S}_{n}\to \overline{S}$
locally in
${L}^{p}$
over
$X$
. Now
$${\nabla}_{\overline{A}}({S}_{n}\overline{S})={\nabla}_{{\overline{A}}_{n}}({S}_{n}{\overline{S}}_{n})+{\nabla}_{\overline{A}}({\overline{S}}_{n}\overline{S})+(\overline{A}{\overline{A}}_{n})({S}_{n}{\overline{S}}_{n}),$$
and each of the three terms on the right hand side converges to
$0$
locally in
${L}^{p}$
over
$X$
(the third term because of the continuous multiplication
${L}_{1}^{p}\times {L}^{p}\to {L}^{p}$
in
$\mathbb{R}4$
for
$p>4$
). Hence
${S}_{n}\to \overline{S}$
locally in
${L}_{1}^{p}$
over
$X$
.
Corollary 1
For sufficiently large
$n$
one has that
${R}_{K}\left({V}_{n}^{\prime}\right)\subset V$
.
Lemma 18
Let
${\omega}_{n}\in {V}_{{m}_{n}}^{\prime}$
for
$n=1,2,...$
, where
${m}_{n}\to \infty $
. Suppose
$q\left({\omega}_{n}{}_{K}\right)$
converges in
${M}_{\mathfrak{b}}$
to an element
$g$
as
$n\to \infty $
. Then
$g\in \overline{G}$
, and there exists for each
$n$
a representative
${S}_{n}$
of
${\omega}_{n}$
such that the sequence
${S}_{n}$
converges locally in
${L}_{1}^{p}$
over
$X$
to a smooth configuration representing
$g$
.
Proof. Let
${\mathsf{S}}_{n},{\overline{\mathsf{S}}}_{n}$
be as in the proof of Lemma 17 . First suppose that
$\left[{\overline{\mathsf{S}}}_{n}\right]$
converges in
$\overline{G}$
to some element
$\left[\overline{S}\right]$
, where
$\overline{S}$
is smooth. Choosing
${S}_{n},{\overline{S}}_{n}$
as in that proof we find again that
${S}_{n}\to \overline{S}$
locally in
${L}_{1}^{p}$
over
$X$
, hence
$$g={lim}_{n}q\left({S}_{n}{}_{K}\right)=q\left(\overline{S}{}_{K}\right)=\left[\overline{S}\right].$$
We now turn to the general case when
$\left[{\overline{\mathsf{S}}}_{n}\right]$
is not assumed to converge.
Because
$\overline{G}$
is compact, every subsequence of
$\left[{\overline{\mathsf{S}}}_{n}\right]$
has a convergent subsequence whose limit must be
$g$
by the above argument. Hence
$\left[{\overline{\mathsf{S}}}_{n}\right]\to g$
.
Suppose we are given a sequence
$\{{m}_{n}{\}}_{n=1,2,...}$
of natural numbers tending to infinity, and for each
$n$
an
$r$
tuple
$T\left(n\right)$
of real numbers such that
$$\stackrel{\u02c7}{T}\left(n\right):={min}_{j}{T}_{j}\left(n\right)>{m}_{n}.$$
Define
${G}^{\left(T\right(n\left)\right)}$
and
${\mathbf{f}}_{n}$
as in Theorem 1 , with
${K}^{\prime}=X:{m}_{n}$
and
${V}^{\prime}={V}_{{m}_{n}}^{\prime}$
.
Lemma 19
For
$n=1,2,...$
suppose
${S}_{n}$
is a smooth configuration over
${X}^{\left(T\right(n\left)\right)}$
representing an element
${\omega}_{n}\in {G}^{\left(T\right(n\left)\right)}$
, and such that
$${\mathbf{f}}_{n}\left({\omega}_{n}\right)\to ({\omega}_{0},z)\in \overline{G}\times \text{U}(1{)}^{{r}_{0}}$$
as
$n\to \infty $
. There exists a constant
$C14<\infty $
such that for sufficiently large
$\tau $
the following holds for sufficiently large
$n$
. Let the map
$\hat{\zeta}={\hat{\zeta}}_{n}$
be defined as above and set
${\hat{S}}_{n}={\hat{\zeta}}_{n}\left(0\right)$
. Then there exists a smooth gauge transformation
${u}_{n}\in {\mathcal{G}}_{\mathfrak{b}}^{\prime}$
such that
$$\parallel {u}_{n}\left({S}_{n}\right){\hat{S}}_{n}{\parallel}_{Lp,\kappa 1}\le C14{e}^{(3\sigma \lambda )\tau}.$$
Note: This constant
$C14$
depends on
$({\omega}_{0},z)$
but not on the sequence
${S}_{n}$
.
Before proving the lemma, we will use it to show that
${\mathbf{f}}_{n}$
is injective for some
$n$
. This will prove Proposition 2 . Suppose
${\omega}_{n},{\omega}_{n}^{\prime}\in {G}^{\left(T\right(n\left)\right)}$
and
${\mathbf{f}}_{n}\left({\omega}_{n}\right)={\mathbf{f}}_{n}\left({\omega}_{n}^{\prime}\right)$
,
$n=1,2,...$
. After passing to a subsequence we may assume that
${\mathbf{f}}_{n}\left({\omega}_{n}\right)$
converges to some point
$({\omega}_{0},z)\in \overline{G}\times \text{U}(1{)}^{{r}_{0}}$
.
Combining Lemmas
16 , 19 and the assumption
$6\sigma <\lambda $
we conclude that if
$\tau $
is sufficiently large then for sufficiently large
$n$
we can represent
${\omega}_{n}$
and
${\omega}_{n}^{\prime}$
by configurations
$\hat{\zeta}\left({x}_{n}\right)$
and
$\hat{\zeta}\left({x}_{n}^{\prime}\right)$
, respectively, where
${x}_{n},{x}_{n}^{\prime}\in {B}_{{\epsilon}^{\prime}}$
.
Now recall that
$\hat{\Xi}\circ \hat{\zeta}=I$
, and that the components
${\hat{\Xi}}_{1},{\hat{\Xi}}_{2}$
are
${\mathcal{G}}_{\mathfrak{b}}^{\prime}$
invariant whereas
${\hat{\Xi}}_{3}$
is the Seiberg–Witten map. Comparing the definitions of
${\mathbf{f}}_{n}$
and
$\hat{\Xi}$
we conclude that
$${x}_{n}=\hat{\Xi}\left(\hat{\zeta}\right({x}_{n})=\hat{\Xi}(\hat{\zeta}\left({x}_{n}^{\prime}\right)={x}_{n}^{\prime},$$
hence
${\omega}_{n}={\omega}_{n}^{\prime}$
for large
$n$
. To complete the proof of Proposition 2 it therefore only remains to prove Lemma 19 .
Proof of Lemma 19 : In this proof, constants will be independent of the sequence
${S}_{n}$
(as well as of
$\tau $
as before).
By Lemma
18 we can find for each
$n$
an
${L}_{2,\text{loc}}^{p}$
gauge transformation
${v}_{n}$
over
$X$
with
${v}_{n}{}_{\mathfrak{b}}=1$
such that
${S}_{n}^{\prime}={v}_{n}\left({S}_{n}\right)$
converges locally in
${L}_{1}^{p}$
over
$X$
to a smooth configuration
${S}^{\prime}$
representing
${\omega}_{0}$
. A moment's thought shows that we can choose the
${v}_{n}$
smooth, and we can clearly arrange that
${S}^{\prime}={S}_{0}$
.
Then for any
$t\ge 0$
we have
$$\begin{array}{c}{limsup}_{n}\parallel {S}_{n}^{\prime}{\hat{S}}_{n}{\parallel}_{Lp,\kappa 1(X:t)}={limsup}_{n}\parallel {S}_{0}{\hat{S}}_{n}{\parallel}_{Lp,\kappa 1(X:t)}\le \text{const}\cdot {e}^{(2\sigma \lambda )\tau}\end{array}$$ 
(12)

when
$\tau $
is so large that Lemma 9 applies.
For
$t\ge 0$
and any smooth configurations
$S$
over
$X:t$
consider the functional
$$E(S,t)={\sum}_{j=1}^{r}{\lambda}_{j}\left(\vartheta \left(S{}_{\left\{t\right\}\times {Y}_{j}}\right)+\vartheta \left(S{}_{\left\{t\right\}\times ({Y}_{j})}\right)\right)+{\sum}_{j=1}^{{r}^{\prime}}{\lambda}_{j}^{\prime}\left(\vartheta \left(S{}_{\left\{t\right\}\times {Y}_{j}^{\prime}}\right)\vartheta \left({\alpha}_{j}^{\prime}\right)\right),$$
where in this formula
$\left\{t\right\}\times (\pm {Y}_{j})$
has the boundary orientation inherited from
$X:t$
. (Recall that the ChernSimonsDirac functional
$\vartheta $
changes sign when the orientation of the
$3$
manifold in question is reversed.) The assumption on
${\lambda}_{j},{\lambda}_{j}^{\prime}$
and
${\stackrel{~}{\eta}}_{j},{\stackrel{~}{\eta}}_{j}^{\prime}$
in Theorem 1 implies that
$E(S,t)$
depends only on the gauge equivalence class of
$S$
. Since
$\vartheta $
is a smooth function on the
${L}_{1/2}^{2}$
configuration space by [
10,Lemma 3.1]
, we obtain
$$E({S}_{n},t)=E({S}_{n}^{\prime},t)\to E({S}_{0},t)$$
as
$n\to \infty $
. By our exponential decay results (see the proof of [
10,Theorem 6.1]
),
$E({S}_{0},t)<\text{const}\cdot {e}^{2\lambda t}\text{for}t\ge 0\text{.}$
It follows that
$E({S}_{n},t)<\text{const}\cdot {e}^{2\lambda t}\text{for}n>N\left(t\right)\text{}$
for some positive function
$N$
. By assumption the perturbation parameters
$\stackrel{\u20d7}{\mathfrak{p}},{\stackrel{\u20d7}{\mathfrak{p}}}^{\prime}$
are admissible, hence there is a constant
$C<\infty $
such that when
$\stackrel{\u02c7}{T}\left({m}_{n}\right)>C$
, each of the
$(r+{r}^{\prime})$
summands appearing in the definition of
$E({S}_{n},t)$
is nonnegative. Explicitly, this yields
$$0\le \vartheta \left({S}_{n}{}_{\{{T}_{j}(n)+t\}\times {Y}_{j}}\right)\vartheta \left({S}_{n}{}_{\left\{{T}_{j}\right(n)t\}\times {Y}_{j}}\right)<\text{const}\cdot {e}^{2\lambda t},0\le \vartheta \left({S}_{n}{}_{\left\{t\right\}\times {Y}_{j}^{\prime}}\right)\vartheta \left({\alpha}_{j}^{\prime}\right)<\text{const}\cdot {e}^{2\lambda t},$$
where the first line holds for
$0\le t\le {T}_{j}\left(n\right)$
and
$j=1,...,r$
, the second line for
$t\ge 0$
and
$j=1,...,{r}^{\prime}$
, and in both cases we assume
$\stackrel{\u02c7}{T}\left(n\right)>C$
and
$n>N\left(t\right)$
.
In the following we will ignore the ends
$\mathbb{R}+\times {Y}_{{j}^{\prime}}^{\prime}$
of
$X$
, ie we will pretend that
$X\#$
is compact. If
${\alpha}_{{j}^{\prime}}^{\prime}$
is irreducible then the argument for dealing with the end
$\mathbb{R}+\times {Y}_{{j}^{\prime}}^{\prime}$
is completely analogous to the one given below for a neck
$[{T}_{j},{T}_{j}]\times {Y}_{j}$
, while if
${\alpha}_{{j}^{\prime}}^{\prime}$
is reducible it is simpler. (Compare the proof of [
10,Proposition 6.3(ii)]
.) For the remainder of the proof of this lemma we will focus on one particular neck
$[{T}_{j}(n),{T}_{j}(n\left)\right]\times {Y}_{j}$
where
$1\le j\le r$
. To simplify notation we will therefore mostly omit
$j$
from notation and write
$T\left(n\right),Y,\alpha $
etc instead of
${T}_{j}\left(n\right),{Y}_{j},{\alpha}_{j}$
.
For
$0\le t\le T\left(n\right)$
set
$${B}_{t}=[T(n)+t,T(n)t]\times Y,$$
regarded as a subset of
$X\left(T\right(n\left)\right)$
. By the above discussion there is a constant
${t}_{1}>0$
such that when
$n$
is sufficiently large,
${S}_{n}$
will restrict to a genuine monopole over the band
${B}_{{t}_{1}+3}$
by [
10,Lemmas 4.1,4.2,4.3]
and will have small enough energy over this band for [
10,Theorem 6.2]
to apply. That theorem then provides a smooth
$${\stackrel{~}{v}}_{n}:{B}_{{t}_{1}}\to \text{U}\left(1\right)$$
such that
${S}_{n}^{\prime \prime}={\stackrel{~}{v}}_{n}\left({S}_{n}{}_{{B}_{{t}_{1}}}\right)$
is in temporal gauge and
$$\parallel {S}_{n}^{\prime \prime}\underline{\alpha}{\parallel}_{Lp,\kappa 1\left({B}_{t}\right)}\le \text{const}\cdot {e}^{(\sigma \lambda )t},t\ge {t}_{1}.$$
Writing
$${S}_{n}^{\prime \prime}{\hat{S}}_{n}=({S}_{n}^{\prime \prime}\underline{\alpha})+(\underline{\alpha}\stackrel{~}{S})+(\stackrel{~}{S}{\hat{S}}_{n})$$
we get
$$\begin{array}{c}{limsup}_{n\to \infty}\parallel {S}_{n}^{\prime \prime}{\hat{S}}_{n}{\parallel}_{Lp,\kappa 1\left({B}_{t}\right)}\le \text{const}\cdot \left({e}^{(2\sigma \lambda )\tau}+{e}^{(\sigma \lambda )t}\right)\end{array}$$ 
(13)

when
$t\ge {t}_{1}$
and
$\tau $
is so large that Lemma 9 applies.
To complete the proof of the lemma we interpolate between
${v}_{n}$
and
${\stackrel{~}{v}}_{n}$
in the overlap region
${\mathcal{O}}_{\tau}=X:\tau \cap {B}_{\tau 1}$
. (This requires
$\tau \ge {t}_{1}+1$
.) The choice of this overlap region is somewhat arbitrary but simplifies the exposition.
Define
$${w}_{n}={\stackrel{~}{v}}_{n}v{1}_{n}:{\mathcal{O}}_{\tau}\to \text{U}\left(1\right).$$
Then
${w}_{n}\left({S}_{n}^{\prime}\right)={S}_{n}^{\prime \prime}\text{on}{\mathcal{O}}_{\tau}\text{.}$
Set
${x}^{\pm}=\gamma (\pm (T\tau \left)\right)$
, where
$\gamma ={\gamma}_{j}$
is the path introduced in the beginning of this section. If
$\alpha $
is reducible then by multiplying each
${\stackrel{~}{v}}_{n}$
by a constant and redefining
${w}_{n},{S}_{n}^{\prime \prime}$
accordingly we can arrange that
${w}_{n}\left({x}^{+}\right)=1$
for all
$n$
.
These changes have no effect on the estimates above.
Lemma
19 is a consequence of the estimates ( 12 ), ( 13 ) together with the following sublemma (see the proof of [
10,Proposition 6.3(ii)]
.)
Sublemma 1
There is a constant
$C15<\infty $
such that if
$\tau \ge C15$
then
$${limsup}_{n\to \infty}\parallel {w}_{n}1{\parallel}_{{L}_{2}^{p}\left({\mathcal{O}}_{\tau}\right)}\le C15{e}^{(2\sigma \lambda )\tau}.$$
Proof of sublemma: If
$\alpha $
is irreducible then the sublemma follows from inequalities ( 12 ), ( 13 ) and [
10,Lemmas 6.9,6.11]
. (In this case the sublemma holds with
$C15{e}^{(\sigma \lambda )\tau}$
as upper bound.) Now suppose
$\alpha $
is reducible. We will show that
$$\begin{array}{c}{limsup}_{n\to \infty}\left{w}_{n}\right({x}^{})1\le \text{const}\cdot {e}^{(2\sigma \lambda )\tau}\end{array}$$ 
(14)

for large
$\tau $
. Granted this, we can prove the sublemma by applying [
10,Lemma 6.9]
and [
10,Lemma 6.10(ii)]
to each component of
${\mathcal{O}}_{\tau}$
.
In the remainder of the proof of the sublemma we will omit
$n$
from subscripts. To prove ( 14 ), define intervals
$${J}_{0}=[T1,T+\tau ],{J}_{1}=[T+\tau ,T\tau ],{J}_{2}=[T\tau ,T+1]$$
and for
$k=0,1,2$
set
$\gamma \left(k\right)=\gamma {}_{{J}_{k}}$
. Let
$\text{Hol}\left(k\right)$
denote holonomy along
$\gamma \left(k\right)$
in the same sense as ( 1 ), ie
$\text{Hol}\left(k\right)$
is the result of replacing the domain of integration
${I}_{j}$
in that formula with
${J}_{k}$
. Define
$\delta \left(k\right)\in \mathbb{C}$
by
$$\begin{array}{cc}\text{Hol}\left(k\right)\left(\hat{S}\right)& =\text{Hol}\left(k\right)\left({S}^{\prime}\right)(1+\delta (k\left)\right),k=0,2,\end{array}$$  
$$\begin{array}{cc}\text{Hol}\left(1\right)\left(\hat{S}\right)& =\text{Hol}\left(1\right)\left({S}^{\prime \prime}\right)(1+\delta (1\left)\right)\end{array}$$  
$$\begin{array}{}\end{array}$$  
where as usual we mean holonomy with respect to the connection parts of the configurations. For large
$\tau $
the estimates ( 12 ) and ( 13 ) give
$$\left\delta \right(k\left)\right\le \text{const}\cdot {e}^{(2\sigma \lambda )\tau}$$
when
$n$
is sufficiently large.
Writing
$b={\prod}_{k=0}^{2}(1+\delta (k\left)\right)$
we have
$$z=\text{Hol}\left(\hat{S}\right){=}^{2}{\prod}_{k=0}\text{Hol}\left(k\right)\left(\hat{S}\right)=b\text{Hol}\left(0\right)\left({S}^{\prime}\right)\text{Hol}\left(1\right)\left({S}^{\prime \prime}\right)\text{Hol}\left(2\right)\left({S}^{\prime}\right).$$
Now, by the definition of holonomy,
$$\text{Hol}\left(1\right)\left({S}^{\prime \prime}\right)=\frac{\stackrel{~}{v}\left({x}^{+}\right)}{\stackrel{~}{v}\left({x}^{}\right)}\text{Hol}\left(1\right)\left(S\right),$$
and there are similar formulas for
$\text{Hol}\left(k\right)\left({S}^{\prime}\right)$
. Because
$w\left({x}^{+}\right)=1$
we obtain
$$z=b\text{Hol}\left(S\right)w\left({x}^{}\right)1.$$
Setting
$a=\text{Hol}\left(S\right)z1$
we get
$$w\left({x}^{}\right)1=ab1=(a1)b+b1.$$
Since by assumption
$a\to 1$
as
$n\to \infty $
, we have
$$\leftw\right({x}^{})1\le \text{const}\cdot \left(a1+{\sum}_{k}\left\delta \right(k\left)\right\right)\le \text{const}\cdot {e}^{(2\sigma \lambda )\tau}$$
for large
$n$
, proving the sublemma and hence also Lemma 19 .
This completes the proof of Proposition
2 and thereby also the proof of Theorem 1 .
5 A model application
In this section we will show in a model case how the gluing theorem may be applied in combination with the compactness results of [
10]
. Here we only consider gluing along irreducible critical points. Examples of gluing along reducible critical points will appear in [
9]
. The main result of this section, Theorem 2 , encompasses both the simplest gluing formulae for Seiberg–Witten invariants (in situations where reducibles are not encountered) and, as we will see in Section 6 , the formula
$d\circ d=0$
for the standard Floer differential.
Recall that the Seiberg–Witten invariant of a closed
$\text{spin}c$
$4$
–manifold (with
${b}_{2}^{+}>1$
) can be defined as the number of points (counted with sign) in the zeroset of a generic section of a certain vector bundle over the moduli space. To obtain a gluing formula, this vector bundle and its section should be expressed as the pullback of a vector bundle
$E\to {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
with section
$s$
, where
$K\subset X$
. In the proof of Theorem 2 below we will see how the section
$s$
gives rise in a natural way to a map
$q$
as in Theorem 1 . Thus, the section
$s$
is being incorporated into the equations that the gluing map is required to solve. (We owe this idea to [
4,p 99]
.) We will now describe the setup for our model application. Let
$X$
be as in [
10,Subsection 1.4]
with
$r=1$
and
${r}^{\prime}\ge 0$
, and set
$Y={Y}_{1}$
. In other words, we will be gluing one single pair of ends
${\mathbb{R}}_{+}\times (\pm Y)$
of
$X$
, but
$X$
may have other ends
${\mathbb{R}}_{+}\times {Y}_{j}^{\prime}$
not involved in the gluing. We assume
${X}^{\#}$
is connected, which means that
$X$
has one or two connected components.
For
$j=1,...,{r}^{\prime}$
fix a critical point
${\alpha}_{j}^{\prime}\in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{{Y}_{j}^{\prime}}$
. Let
$\mu $
be a
$2$
form and
$\mathfrak{p}$
a perturbation parameter for
$Y$
, and let
${\mu}_{j}^{\prime},{\mathfrak{p}}_{j}^{\prime}$
be similar data for
${Y}_{j}^{\prime}$
. Let each
$\mathfrak{p},{\mathfrak{p}}_{j}^{\prime}$
have small
${C}^{1}$
norm. To simplify notation we write, for
$\alpha ,\beta \in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}$
,
$${M}_{\alpha ,\beta}=M(X;\alpha ,\beta ,{\stackrel{\u20d7}{\alpha}}^{\prime}),M\left(T\right)=M\left(X\right(T);{\stackrel{\u20d7}{\alpha}}^{\prime}).$$
We make the following assumptions:

∙
(Compactness) At least one of the conditions (B1), (B2) of [10] holds,

∙
(Regularity) All moduli spaces over
$\mathbb{R}\times Y$
,
$\mathbb{R}\times {Y}_{j}^{\prime}$
, and
$X$
contain only regular points, and

∙
(No reducibles) Given
${\alpha}_{1},{\alpha}_{2}\in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}$
and
${\alpha}_{j}^{\prime}\in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{{Y}_{j}^{\prime}}$
, if there exist a broken gradient line over
$\mathbb{R}\times Y$
from
${\alpha}_{1}$
to
${\alpha}_{2}$
and for each
$j$
a broken gradient line over
$\mathbb{R}\times {Y}_{j}^{\prime}$
from
${\alpha}_{j}^{\prime}$
to
${\beta}_{j}^{\prime}$
then
$M(X;{\alpha}_{1},{\alpha}_{2},{\stackrel{\u20d7}{\alpha}}^{\prime})$
contains no reducible. (It then follows by compactness that
$M\left(T\right)$
contains no reducible when
$T$
is large.)
The regularity condition is stronger than necessary, because there are energy constraints on the moduli spaces that one may encounter in the situation to be considered, but we will not elaborate on this here.
Note that we have so far only developed a full transversality theory in the case when
$Y$
and each
${Y}_{j}^{\prime}$
are rational homology spheres; in the remaining cases the discussion here is therefore somewhat theoretical at this time.
Let
$K\subset X$
be a compact codimension
$0$
submanifold which intersects every component of
$X$
. When
$T\gg 0$
then
$K$
may also be regarded as a submanifold of
$X\left(T\right)$
, and we have restriction maps
$${R}_{\alpha ,\beta}:{M}_{\alpha ,\beta}^{*}\to {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right),{R}^{\prime}:M\left(T\right)\to {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right).$$
These take values in
${\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
rather than just in
$\stackrel{\u02d8}{\mathcal{\mathcal{B}}}\left(K\right)$
because of the unique continuation property of harmonic spinors.
Suppose
$E\to {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
is an oriented smooth real vector bundle whose rank
$d$
is equal to the (expected) dimension of
$M\left(T\right)$
. Choose a smooth section
$s$
of
$E$
such that the pullback section
${s}_{\alpha ,\beta}={R}_{\alpha ,\beta}^{*}s$
is transverse to the zerosection of the pullback bundle
${E}_{\alpha ,\beta}={R}_{\alpha ,\beta}^{*}E$
over
${M}_{\alpha ,\beta}^{*}$
for each pair
$\alpha ,\beta $
. (Here the Sobolev exponent
$p>4$
should be an even integer to ensure the existence of smooth partitions of unity.) Set
${s}^{\prime}=({R}^{\prime}{)}^{*}s$
, which is a section of
${E}^{\prime}=({R}^{\prime}{)}^{*}E$
. We write
${M}_{\alpha}={M}_{\alpha ,\alpha}={M}_{\alpha ,\alpha}^{*}$
and
${s}_{\alpha}={s}_{\alpha ,\alpha}$
etc.
Let
${\hat{M}}_{\alpha}$
,
$\hat{M}\left(T\right)$
denote the zerosets of
${s}_{\alpha}$
,
${s}^{\prime}$
respectively. By index theory we have
$$0=dim\hat{M}\left(T\right)=dim{\hat{M}}_{\alpha}+{n}_{\alpha},$$
where
${n}_{\alpha}=0$
if
$\alpha $
is irreducible and
${n}_{\alpha}=1$
otherwise. Thus,
${\hat{M}}_{\alpha}$
is empty if
$\alpha $
is reducible.
Lemma 20
If
${\omega}_{n}\in {\hat{M}}^{\left(T\right(n\left)\right)}$
for
$n=1,2,...$
, where
$T\left(n\right)\to \infty $
, then a subsequence of
${\omega}_{n}$
chainconverges to an element of
${\hat{M}}_{\alpha}$
for some
$\alpha \in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}^{*}$
.
Moreover, if
${\omega}_{n}=\left[{S}_{n}\right]$
chainconverges to
$\left[S\right]\in {\hat{M}}_{\alpha}$
then there exists for each
$n$
a smooth
${u}_{n}:X\left(T\right(n\left)\right)\to \text{U}\left(1\right)$
whose restriction to each end
$\mathbb{R}+\times {Y}_{j}^{\prime}$
is nullhomotopic and such that the sequence
${u}_{n}\left({S}_{n}\right)$
cconverges over
$X$
to
$S$
.
Proof. The statement of the first sentence follows from [10,Theorem 1.4] by dimension counting. Such maps
${u}_{n}$
exist in general for chainconvergent sequences when the
${\omega}_{n}$
all have the same asymptotic limits over the ends
$\mathbb{R}+\times {Y}_{j}^{\prime}$
, see [10] . Let
$J\subset {H}^{1}(Y;\mathbb{Z})$
be the subgroup consisting of elements of the form
$z{}_{Y}$
where
$z$
is an element of
${H}^{1}({X}^{\#};\mathbb{Z})$
satisfying
$z{}_{{Y}_{j}^{\prime}}=0$
for
$j=1,...,{r}^{\prime}$
.
This group
$J$
acts on the disjoint union
$${\hat{M}}^{u}={\bigcup}_{\alpha \in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}^{*}}{\hat{M}}_{\alpha},$$
permuting the summands.
Lemma 21
The quotient
$\hat{M}={\hat{M}}^{u}/J$
is a finite set.
Proof. By [
10,Theorem 1.3]
any sequence
${\omega}_{n}\in {\hat{M}}_{{\alpha}_{n}}$
,
$n=1,2,...$
has a chainconvergent subsequence, and for dimensional reasons the limit (welldefined up to gauge equivalence) must lie in some moduli space
${\hat{M}}_{\beta}$
. Furthermore, if
${\omega}_{n}$
chainconverges to an element in
${\hat{M}}_{\beta}$
then
${\hat{M}}_{{\alpha}_{n}}$
is contained in the orbit
$J\cdot {\hat{M}}_{\beta}$
for
$n\gg 0$
. Therefore, each
${\hat{M}}_{\alpha}$
is a finite set, and only finitely many orbits
$J\cdot {\hat{M}}_{\alpha}$
are nonempty. This is equivalent to the statement of the lemma.
Note that
$J$
is the largest subgroup of
${H}^{1}(Y;\mathbb{Z})$
which acts on
${\hat{M}}^{u}$
in a natural way. On the other hand, if
${\hat{M}}^{u}$
is nonemtpy then, since
${H}^{1}(Y;\mathbb{Z})$
acts freely on
${\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}$
, only subgroups
${J}^{\prime}\subset J$
of finite index have the property that
${\hat{M}}^{u}/{J}^{\prime}$
is finite.
Lemma 22
There is a compact codimension
$0$
submanifold
${K}_{0}\subset X$
such that the restriction map
$\hat{M}\to \mathcal{\mathcal{B}}\left({K}_{0}\right)$
is injective.
Proof. Let
$\left[{S}_{j}\right]\in {\hat{M}}_{{\beta}_{j}}$
,
$j=1,2$
, where each
${S}_{j}$
is in temporal gauge over the ends of
$X$
(and therefore decays exponentially). Suppose there exists a sequence of smooth gauge transformations
${u}_{n}:X:{t}_{n}\to \text{U}\left(1\right)$
where
${t}_{n}\to \infty $
, such that
${u}_{n}\left({S}_{1}\right)={S}_{2}$
over
$X:{t}_{n}$
. After passing to a subsequence we can arrange that
${u}_{n}$
cconverges over
$X$
to some gauge transformation
$u$
with
$u\left({S}_{1}\right)={S}_{2}$
. If
$t\gg 0$
then
$u{}_{\left\{t\right\}\times (\pm Y)}$
will both be homotopic to a smooth
$v:Y\to \text{U}\left(1\right)$
with
$v\left({\alpha}_{1}\right)={\alpha}_{2}$
. Hence
${\hat{M}}_{{\alpha}_{1}},{\hat{M}}_{{\alpha}_{2}}$
lie in the same
$J$
–orbit, and
${S}_{1},{S}_{2}$
represent the same element of
$\hat{M}$
by [
10,Proposition 2.5 (iii)]
.
Thus we can take
${K}_{0}=X:t$
for
$t\gg 0$
.
Now fix
${K}_{0}$
as in Lemma 22 and with
$K\subset {K}_{0}$
. Let
$\{{b}_{1},...,{b}_{m}\}$
be the image of the restriction map
${R}_{{K}_{0}}:\hat{M}\to \stackrel{\u02d8}{\mathcal{\mathcal{B}}}\left({K}_{0}\right)$
. Choose disjoint open neighbourhoods
${W}_{j}\subset \stackrel{\u02d8}{\mathcal{\mathcal{B}}}\left({K}_{0}\right)$
of the points
${b}_{j}$
. If
$T\gg 0$
then
$${R}_{{K}_{0}}\left(\hat{M}\right(T\left)\right)\subset {\bigcup}_{j}{W}_{j}$$
by Lemma 20 . For such
$T$
we get a natural map
$$g:\hat{M}\left(T\right)\to \hat{M}.$$
It is clear that if
${g}^{\prime}$
is the map corresponding to a different choice of
${K}_{0}$
and neighbourhoods
${W}_{j}$
then
$g={g}^{\prime}$
for
$T$
sufficiently large.
Theorem 2
For sufficiently large
$T$
the following hold:

(i)
Every element of
$\hat{M}\left(T\right)$
is a regular point in
$M\left(T\right)$
and a regular zero of
${s}^{\prime}$
.

(ii)
$g$
is a bijection.
Proof. If
$\hat{M}$
is empty then, by Lemma 20 ,
$\hat{M}\left(T\right)$
is empty as well for
$T\gg 0$
, and there is nothing left to prove.
We now fix
${b}_{j}$
and for the remainder of the proof omit
$j$
from notation.
(Thus
$b={b}_{j}$
,
$W={W}_{j}$
etc.) We will show that for
$T\gg 0$
the set
$$\hat{B}\left(T\right)=\{\omega \in \hat{M}(T):\omega {}_{{K}_{0}}\in W\}$$
consists of precisely one element, and that this element is regular in the sense of (i). This will prove the theorem.
By definition,
$b$
is the restriction of some
${\omega}_{0}\in {\hat{M}}_{\alpha}$
. Choose an open neighbourhood
$V\subset {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
of
$b{}_{K}$
and a smooth map
$$\pi :E{}_{V}\to {\mathbb{R}}^{d}$$
which restricts to a linear isomorphism on every fibre. Choose an open neighbourhood
${V}_{0}\subset W$
of
$b$
such that
${R}_{K}\left({V}_{0}\right)\subset V$
. Let
${G}_{+}\subset {M}_{\alpha}$
be a precompact open neighbourhood of
${\omega}_{0}$
such that
${R}_{{K}_{0}}\left({\overline{G}}_{+}\right)\subset {V}_{0}$
. The assumption that
${\omega}_{0}$
be a regular zero of
${s}_{\alpha}$
means that the composite map
$${G}_{+}\to {R}_{K}V\to \pi \circ s{\mathbb{R}}^{d}$$
is a local diffeomorphism at
${\omega}_{0}$
. We can then find an injective smooth map
$$p:{\mathbb{R}}^{d}\to {M}_{\alpha}$$
such that
$p\circ \pi \circ s\circ {R}_{K}=\text{id}$
in some open neighbourhood
$G\subset {G}_{+}$
of
${\omega}_{0}$
.
In particular,
$p1\left({\omega}_{0}\right)=\left\{0\right\}$
and
$p$
is a local diffeomorphism at
$0$
. Set
$$q=p\circ \pi \circ s:V\to {M}_{\alpha}.$$
By Theorem 1 there is a kvpair
$({K}^{\prime},{V}^{\prime})\le ({K}_{0},{V}_{0})$
such that if
$T\gg 0$
then
$G\left(T\right)$
consists only of regular monopoles and
$$\mathbf{f}=q\circ {R}_{K}:G\left(T\right)\to G$$
is a diffeomorphism. By Lemma 20 one has
$$\hat{B}\left(T\right)=G\left(T\right)\cap \left({s}^{\prime}\right)1\left(0\right)=\mathbf{f}1\left({\omega}_{0}\right)$$
for
$T\gg 0$
. For such
$T$
the set
$\hat{B}\left(T\right)$
consists of precisely one point, and this point is regular in the sense of (i).
6 The Floer differential
Consider the situation of [
10,Subsection 1.2]
. Suppose a perturbation parameter
$\mathfrak{p}$
of small
${C}^{1}$
norm has been chosen for which all moduli spaces
$M({\alpha}_{1},{\alpha}_{2})$
over
$\mathbb{R}\times Y$
are regular. (This is possible at least when
$Y$
is a rational homology sphere, see [
10]
.) Fix
${\alpha}_{1},{\alpha}_{2}\in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}^{*}$
with
$$dimM({\alpha}_{1},{\alpha}_{2})=2.$$
We will show that the disjoint union
$$\stackrel{\u02c7}{M}:={\bigcup}_{\beta \in {\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}^{*}\backslash \{{\alpha}_{1},{\alpha}_{2}\}}\stackrel{\u02c7}{M}({\alpha}_{1},\beta )\times \stackrel{\u02c7}{M}(\beta ,{\alpha}_{2})$$
is the boundary of a compact
$1$
–manifold. (In other words, the standard Floer differential
$d$
satisfies
$d\circ d=0$
at least with
$\mathbb{Z}/2$
coefficients.) To this end we will apply Theorem 2 to the case when
$X$
consists of two copies of
$\mathbb{R}\times Y$
, say
$$X=\mathbb{R}\times Y\times \{1,2\},$$
and we glue
$\mathbb{R}+\times Y\times \left\{1\right\}$
with
$\mathbb{R}\times Y\times \left\{2\right\}$
. Thus
$r=1,{r}^{\prime}=2$
. We take
$K={K}_{1}\cup {K}_{2}$
, where
${K}_{j}=[0,1]\times Y\times \left\{j\right\}$
. In this case,
${\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
is the quotient of
${\mathcal{C}}^{*}\left(K\right)$
by the nullhomotopic gauge transformations. The bundle
$E$
over
${\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)$
will be the product bundle with fibre
${\mathbb{R}}^{2}$
. To define the section
$s$
of
$E$
, choose
${\delta}_{1},{\delta}_{2}>0$
such that
$\vartheta $
has no critical value in the set
$$\left(\vartheta \right({\alpha}_{2}),\vartheta ({\alpha}_{2})+{\delta}_{2}]\cup \left[\vartheta \right({\alpha}_{1}){\delta}_{1},\vartheta ({\alpha}_{1}\left)\right).$$
This is possible because we assume Condition (O1) of [
10]
. For any configuration
$S$
over
$[0,1]\times Y$
set
$${s}_{j}\left(S\right)={\int}_{0}^{1}\vartheta \left({S}_{t}\right)dt\vartheta \left({\alpha}_{j}\right)(1{)}^{j}{\delta}_{j}.$$
Note that
${s}_{j}\left(S\right)$
does not change if we apply a nullhomotopic gauge transformation to
$S$
.
A configuration over
$K$
consists of a pair
$({S}_{1},{S}_{2})$
of configurations over
$[0,1]\times Y$
. Define a smooth function
$s:{\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}^{*}\left(K\right)\to {\mathbb{R}}^{2}$
(ie a section of
$E$
) by
$$s\left(\right[{S}_{1}],[{S}_{2}\left]\right)=\left({s}_{1}\right({S}_{1}),{s}_{2}({S}_{2}\left)\right).$$
If
$\left[S\right]$
belongs to some moduli space
$M({\beta}_{1},{\beta}_{2})$
over
$\mathbb{R}\times Y$
with
${\beta}_{1}\ne {\beta}_{2}$
in
${\stackrel{~}{\mathcal{\mathcal{R}}}}_{Y}$
then
$\frac{d}{dt}\vartheta \left({S}_{t}\right)<0$
for all
$t$
by choice of
$\mathfrak{p}$
. Since
$J=0$
, the natural map
$\hat{M}\to \stackrel{\u02c7}{M}$
is therefore a bijection. Let
${s}_{j}^{\prime}$
be the pullback of
${s}_{j}$
to
$M\left(T\right)$
. Here
$M\left(T\right)$
is defined using [
10,Equation 17]
with
$\mathfrak{q}=0$
, and so can be identified with
$M({\alpha}_{1},{\alpha}_{2})$
. By [
10,Theorem 1.3]
the set
$$Z\left(T\right)=\{\omega \in M(T):{s}_{1}^{\prime}(\omega )=0,{s}_{2}^{\prime}(\omega )\le 0\}$$
is compact for all
$T>0$
. If
$T\gg 0$
then, by Theorem 2 ,
$Z\left(T\right)$
is a smooth submanifold of
$M\left(T\right)$
, and the composition of the two bijections
$$\partial Z\left(T\right)=\hat{M}\left(T\right)\to g\hat{M}\to \stackrel{\u02c7}{M}$$
yields the desired identification of
$\stackrel{\u02c7}{M}$
with the boundary of a compact
$1$
manifold.
7 Parametrized moduli spaces
We will now introduce certain parametrized moduli spaces that are useful for the construction of Floer homology groups. A natural setting here would involve certain fibre bundles whose fibres are
$4$
manifolds. We feel, however, that gauge theory for such bundles in general deserves a separate treatment, and will therefore limit ourselves, at this time, to the case of a product bundle over a vector space.
As in [
10,Subsection 1.3]
let
$X$
be a
$\text{spin}c$
$4$
manifold with Riemannian metric
$\overline{g}$
and tubular ends
$\overline{\mathbb{R}}+\times {Y}_{j}$
,
$j=1,...,r$
. Let
$\mathsf{W}$
be a finitedimensional Euclidean vector space and
$\mathbf{g}=\{{g}_{\mathsf{w}}{\}}_{\mathsf{w}\in \mathsf{W}}$
a smooth family of Riemannian metrics on
$X$
all of which agree with
$\overline{g}$
outside
$X:0$
. We then have a principal
$\text{SO}\left(4\right)$
bundle
$P\text{SO}(\mathbf{g})\to X\times \mathsf{W}$
whose fibre over
$(x,\mathsf{w})$
consists of all positive
${g}_{\mathsf{w}}$
orthonormal frames in
${T}_{x}X$
.
In the notation of [
10,Subsection 3.1]
let
$P\text{GL}c\to P\text{GL}+$
be the
$\text{spin}c$
structure on
$X$
. Denote by
$P\text{Spin}c(\mathbf{g})$
the pullback of
$P\text{SO}(\mathbf{g})$
under the projection
$P\text{GL}c\times \mathsf{W}\to P\text{GL}+\times \mathsf{W}$
. Then
$P\text{Spin}c(\mathbf{g})$
is a principal
$\text{Spin}c\left(4\right)$
bundle over
$X\times \mathsf{W}$
.
For
$j=1,...,r$
let
${\alpha}_{j}\in \mathcal{C}\left({Y}_{j}\right)$
be a nondegenerate smooth monopole.
Let
$\mathcal{C}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})$
denote the
$Lp,w1$
configuration space for the metric
${g}_{\mathsf{w}}$
and limits
${\alpha}_{j}$
, where
$p,w$
are as in [
10,Subsection 3.4]
. We will provide the disjoint union
$$\mathcal{C}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})={\bigcup}_{\mathsf{w}\in \mathsf{W}}\mathcal{C}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})\times \{\mathsf{w}\}$$
with a natural structure of a (trivial) smooth fibre bundle over
$\mathsf{W}$
. Let
$$\begin{array}{c}\mathsf{v}:P\text{Spin}c(\mathbf{g})\to P\text{Spin}c\left({g}_{0}\right)\times \mathsf{W}\end{array}$$ 
(15)

be any isomorphism of
$\text{Spin}c\left(4\right)$
bundles which covers the identity on
$X\times \mathsf{W}$
and which outside
$X:1$
is given by the identification
$P\text{Spin}c\left({g}_{\mathsf{w}}\right)=P\text{Spin}c\left({g}_{0}\right)$
.
There is then an induced isomorphism of
$\text{SO}\left(4\right)$
bundles
$P\text{SO}(\mathbf{g})\to P\text{SO}\left({g}_{0}\right)\times \mathsf{W}$
, since these are quotients of the corresponding
$\text{Spin}c\left(4\right)$
bundles by the
$\text{U}\left(1\right)$
action. Such an isomorphism
$\mathsf{v}$
can be constructed by means of the holonomy along rays of the form
$\left\{x\right\}\times \overline{\mathbb{R}}+\mathsf{w}$
where
$(x,\mathsf{w})\in X\times \mathsf{W}$
, with respect to any connection in
$P\text{Spin}c(\mathbf{g})$
which outside
$X:1$
is the pullback of a connection in
$P\text{Spin}c\left({g}_{0}\right)$
. Then
$\mathsf{v}$
induces a
$\mathcal{G}(X;\stackrel{\u20d7}{\alpha})$
equivariant diffeomorphism
$$\begin{array}{c}\mathcal{C}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})\to \mathcal{C}(X,{g}_{0};\stackrel{\u20d7}{\alpha})\end{array}$$ 
(16)

for each
$\mathsf{w}$
, where the map on the spin connections is obtained by identifying these with connections in the respective determinant line bundles and applying the isomorphism between these bundles induced by
$\mathsf{v}$
. Putting together the maps ( 16 ) for all
$\mathsf{w}$
yields a bijection
$${\mathsf{v}}_{*}:\mathcal{C}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})\to \mathcal{C}(X,{g}_{0};\stackrel{\u20d7}{\alpha})\times \mathsf{W}.$$
If
$\stackrel{~}{\mathsf{v}}$
is another isomorphism as in ( 15 ) then
${\mathsf{v}}_{*}\left({\stackrel{~}{\mathsf{v}}}_{*}\right)1$
is smooth, hence we have obtained the desired structure on
$\mathcal{C}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
. Furthermore, because of the gauge equivariance of
${\mathsf{v}}_{*}$
we also get a similar smooth fibre bundle structure on
$$\begin{array}{c}{\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})={\bigcup}_{\mathsf{w}\in \mathsf{W}}{\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})\times \{\mathsf{w}\}\end{array}$$ 
(17)

for any finite subset
$\mathfrak{b}\subset X$
.
The monopole equations on
$\mathcal{C}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
that we will consider will involve an additional class of perturbations which are useful for handling deformations of
$\mathfrak{p}$
perturbations. Such a perturbation will depend on the choice of an isomorphism
$\mathsf{v}$
as above, a nonnegative constant
$\mathfrak{t}$
, and a smooth map
$$\mathfrak{o}:\mathcal{C}(X:\mathfrak{t},{g}_{0})\times \mathsf{W}\to {L}^{p}(X:\mathfrak{t},{g}_{0};{\Lambda}^{2})$$
whose image consists of smooth
$2$
forms supported in the interior of
$X:\mathfrak{t}$
.
We make two further assumptions on
$\mathfrak{o}$
. Firstly,
$\mathfrak{o}$
should be gauge invariant in the restricted sense that
$\mathfrak{o}\left(u\right(S\left)\right)=\mathfrak{o}\left(S\right)$
for all
$S\in \mathcal{C}(X:\mathfrak{t},{g}_{0})\times \mathsf{W}$
and all smooth
$u:X:\mathfrak{t}\to \text{U}\left(1\right)$
that restrict to
$1$
on
$X:\mathfrak{t}\cap \mathfrak{b}$
. Here
$u$
acts trivially on
$\mathsf{W}$
. Secondly, for every nonnegative integer
$k$
the image of
$\mathfrak{o}$
should be a bounded set with respect to the
${C}^{k}$
norm.
Having made these choices, for any
$\mathsf{w}\in \mathsf{W}$
we denote by
${\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})$
the set of all elements
$[A,\Phi ]\in {\mathcal{\mathcal{B}}}_{\mathfrak{b}}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})$
that satisfy the perturbed monopole equations
$$\begin{array}{c}\begin{array}{cc}{\left({\hat{F}}_{A}+\mathfrak{m}\left({\mathsf{v}}_{*}\right(A,\Phi ,\mathsf{w}\left)\right)\right)}^{+}Q(\Phi )=0& {D}_{A}\Phi =0\\ \end{array}\end{array}$$ 
(18)

where
$$\mathfrak{m}(A,\Phi ,\mathsf{w})=i\mu +i\mathfrak{q}{\sum}_{j=1}^{r}{h}_{A,\Phi ,{\mathfrak{p}}_{j}}+\mathfrak{o}(A,\Phi ,\mathsf{w}).$$
Here
$\mu $
and the term involving
${\mathfrak{p}}_{j}$
should be of the same kind as in [
10,Equation 18]
, and of course the spin bundles and Dirac operators are those corresponding to
$P\text{Spin}c\left({g}_{\mathsf{w}}\right)$
. Taking the disjoint union of these moduli spaces for all
$\mathsf{w}$
yields the parametrized moduli space
$${\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})={\bigcup}_{\mathsf{w}\in \mathsf{W}}{\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X,{g}_{\mathsf{w}};\stackrel{\u20d7}{\alpha})\times \{\mathsf{w}\}.$$
To get a more concrete description of the parametrized monopole equations note that the left hand side of ( 18 ) defines a fibrepreserving smooth map from
$\mathcal{C}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
into a Banach vector bundle over
$\mathsf{W}$
whose fibre over
$\mathsf{w}$
is
$Lp,w(X,{g}_{\mathsf{w}};i{\Lambda}^{+}\oplus {\mathbb{S}}^{})$
. If we conjugate this monopole map by the diffeomorphisms induced by
$\mathsf{v}$
we obtain the smooth
${\mathcal{G}}_{\mathfrak{b}}(X;\stackrel{\u20d7}{\alpha})$
equivariant map
$$\Theta :\mathcal{C}(X,{g}_{0};\stackrel{\u20d7}{\alpha})\times \mathsf{W}\to Lp,w(X,{g}_{0};i{\Lambda}^{+}\oplus {\mathbb{S}}^{}),(A,\Phi ,\mathsf{w})\mapsto \left(\left({\mathsf{v}}_{\mathsf{w}}\right({\hat{F}}_{A}+\mathfrak{m}(A,\Phi ,\mathsf{w})){)}^{+}Q(\Phi ),{\sum}_{j}{\mathsf{v}}_{\mathsf{w}}({e}_{j})\cdot {\nabla}_{{e}_{j}}^{A+{\mathsf{a}}_{\mathsf{w}}}(\Phi )\right).$$
Here
${\mathsf{v}}_{\mathsf{w}}$
denotes the isomorphism that
$\mathsf{v}$
induces from the Clifford bundle of
$(X,{g}_{\mathsf{w}})$
to the Clifford bundle of
$(X,{g}_{0})$
, and
$\left\{{e}_{j}\right\}$
is a local
${g}_{\mathsf{w}}$
orthonormal frame on
$X$
. Finally, if we temporarily let
${\nabla}^{(\mathsf{w})}$
denote the
${g}_{\mathsf{w}}$
Riemannian connection in the tangent bundle of
$X$
then
$${\mathsf{a}}_{\mathsf{w}}={\mathsf{v}}_{\mathsf{w}}\left({\nabla}^{(\mathsf{w})}\right){\nabla}^{\left(0\right)}.$$
Note that
${\mathsf{a}}_{\mathsf{w}}$
is supported in
$X:1$
.
By construction,
${\mathsf{v}}_{*}$
induces a homeomorphism
$${\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})\to \approx \Theta 1\left(0\right)/{\mathcal{G}}_{\mathfrak{b}}(X;\stackrel{\u20d7}{\alpha}).$$
A point in
${\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
is called regular if the corresponding zeros of
$\Theta $
are regular (a regular zero being one where the derivative of
$\Theta $
is surjective).
This notion is independent of
$\mathsf{v}$
. By the local slice theorem, the set of regular points in
${\mathcal{\mathcal{M}}}_{\mathfrak{b}}^{*}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
is a smooth submanifold of
${\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}(X,\mathbf{g};\stackrel{\u20d7}{\alpha})$
.
8 The gluing theorem in the parametrized case
We continue the discussion of the previous section, but we now specialize to the case when the ends of
$X$
are
${\mathbb{R}}_{+}\times (\pm {Y}_{j})$
,
$j=1,...,r$
and
${\mathbb{R}}_{+}\times {Y}_{j}^{\prime}$
,
$j=1,...,{r}^{\prime}$
, with nondegenerate limits
${\alpha}_{j}$
over
${\mathbb{R}}_{+}\times (\pm {Y}_{j})$
and
${\alpha}_{j}^{\prime}$
over
${\mathbb{R}}_{+}\times {Y}_{j}^{\prime}$
, as in Section 2 . The family of metrics
$\mathbf{g}$
on
$X$
defines, in a natural way, a smooth family of metrics
$\mathbf{g}\left(T\right)=\left\{g\right(T,\mathsf{w}){\}}_{\mathsf{w}\in \mathsf{W}}$
on
$X\left(T\right)$
for any
$T$
. To simplify notation we write
$\mathcal{C}=\mathcal{C}(X,{g}_{0};\stackrel{\u20d7}{\alpha},\stackrel{\u20d7}{\alpha},{\stackrel{\u20d7}{\alpha}}^{\prime})$
and
${\mathcal{C}}^{\prime}=\mathcal{C}\left(X\right(T),g(T,0);{\stackrel{\u20d7}{\alpha}}^{\prime})$
. The corresponding orbit spaces will be denoted
${\mathcal{\mathcal{B}}}_{\mathfrak{b}},{\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{\prime}$
, respectively. The parametrized moduli space over
$X\left(T\right)$
is denoted
$$\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}={\mathcal{\mathcal{M}}}_{\mathfrak{b}}(X\left(T\right),\mathbf{g}\left(T\right);{\stackrel{\u20d7}{\alpha}}^{\prime}).$$
We digress briefly to point out that the notions of chainconvergence and admissible perturbations have natural generalizations to the parametrized situation, and that the compactness theorem [
10,Theorem 1.4]
carries over to sequences
$[{A}_{n},{\Phi}_{n},{\mathsf{w}}_{n}]\in {\mathcal{\mathcal{M}}}_{\mathfrak{b}}\left(X\right(T\left(n\right)),{\mathbf{g}}^{\left(T\right(n\left)\right)};{\stackrel{\u20d7}{\alpha}}_{n}^{\prime})$
provided the sequence
${\mathsf{w}}_{n}$
is bounded (and similarly for [
10,Theorem 1.3]
). The only new ingredient in the proof is the following simple fact: Suppose
$B$
is a Banach space,
$E,F$
vector bundles over a compact manifold,
$L,{L}^{\prime}:\Gamma \left(E\right)\to \Gamma \left(F\right)$
differential operators of order
$d$
, and
$K:\Gamma \left(E\right)\to B$
a linear operator. If
$L$
satisfies an inequality
$$\parallel f{\parallel}_{{L}_{k}^{p}}\le C\left(\parallel Lf{\parallel}_{{L}_{kd}^{p}}+\parallel Kf{\parallel}_{B}\right)$$
and
$L,{L}^{\prime}$
are sufficiently close in the sense that
$$\parallel (L{L}^{\prime})f{\parallel}_{{L}_{kd}^{p}}\le \epsilon \parallel f{\parallel}_{{L}_{k}^{p}}$$
for some constant
$\epsilon >0$
with
$\epsilon C<1$
, then
${L}^{\prime}$
obeys the inequality
$$\parallel f{\parallel}_{{L}_{k}^{p}}\le (1\epsilon C)1C\left(\parallel {L}^{\prime}f{\parallel}_{{L}_{kd}^{p}}+\parallel Kf{\parallel}_{B}\right).$$
This ends our digression, and we continue the main discussion.
Let the paths
${\gamma}_{j}^{\pm},{\gamma}_{j}$
and
$\mathfrak{b}\subset X$
be as in Section 2 , so that
$\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}={\left(\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}\right)}^{*}$
. As before, a choice of reference configuration in
$\mathcal{C}$
gives rise to a reference configuration in
${\mathcal{C}}^{\prime}$
and a holonomy map
$${\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{\prime}\to \text{U}(1{)}^{{r}_{0}}.$$
Composing this with the map
$\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}\to {\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{\prime}$
defined by the chosen isomorphism
$\mathsf{v}$
yields a holonomy map
$$\text{Hol}:\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}\to \text{U}(1{)}^{{r}_{0}}.$$
Fix an open subset
$G\subset {\mathcal{\mathcal{M}}}_{\mathfrak{b}}$
whose closure is compact and contains only regular points.
By a kvpair we mean as before a pair
$(K,V)$
, where
$K\subset X$
is a compact codimension
$0$
submanifold which contains
$\mathfrak{b}$
and intersects every component of
$X$
, and
$V$
is an open neighbourhood of
${R}_{K}\left(\overline{G}\right)$
in
$${\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}_{\mathfrak{b}}(K,\mathbf{g})={\bigcup}_{\mathsf{w}\in \mathsf{W}}{\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}_{\mathfrak{b}}(K,{g}_{\mathsf{w}})\times \{\mathsf{w}\}.$$
Now fix a kvpair
$(K,V)$
satisfying similar additional assumptions as before: firstly, that
$V\subset {\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}_{\mathfrak{b}}^{*}(K,\mathbf{g})$
; secondly, that if
${X}_{e}$
is any component of
$X$
which contains a point from
$\mathfrak{b}$
then
${X}_{e}\cap K$
is connected.
Suppose
$$q:V\to {\mathcal{\mathcal{M}}}_{\mathfrak{b}}$$
is a smooth map such that
$q\left(\omega {}_{K}\right)=\omega $
for all
$\omega \in G$
. Choose
${\lambda}_{j},{\lambda}_{j}^{\prime}>0$
.
Theorem 3
Theorem 1 holds in the present situation if one replaces
$M(T{)}_{\mathfrak{b}}$
by
$\mathcal{\mathcal{M}}(T{)}_{\mathfrak{b}}$
.
Proof. The proof of Theorem 1 carries over without any substantial changes.
A Existence of maps
$q$
Let
$G\subset {M}_{\mathfrak{b}}$
be as in Section 2 . In this appendix we will show that there is always a map
$q$
as in ( 2 ) provided
$K$
is sufficiently large. It clearly suffices to prove the same with
${\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}\left(K\right)$
in place of
${\stackrel{\u02d8}{\mathcal{\mathcal{B}}}}_{\mathfrak{b}}^{*}\left(K\right)$
.
Throughout this appendix
$K$
will be as in Theorem 1 , ie a compact codimension
$0$
submanifold of
$X$
containing
$\mathfrak{b}$
such that if
${X}_{e}$
is any component of
$X$
which contains a point from
$\mathfrak{b}$
then
${X}_{e}\cap K$
is connected. The restriction map
${R}_{K}:{M}_{\mathfrak{b}}\to {\mathcal{\mathcal{B}}}_{\mathfrak{b}}\left(K\right)$
then takes values in
${\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}\left(K\right)$
.
Proposition 3
If
$K$
is sufficiently large then there exist an open neighbourhood
$V\subset {\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}\left(K\right)$
of
${R}_{K}\left(\overline{G}\right)$
and a smooth map
$q:V\to {M}_{\mathfrak{b}}$
such that
$q\left(\omega {}_{K}\right)=\omega $
for all
$\omega \in G$
.
We first prove two lemmas. Choose a precompact open subset
${G}_{0}\subset {M}_{\mathfrak{b}}$
containing
$\overline{G}$
.
Lemma 23
If
$K$
is sufficiently large then
${R}_{K}:{M}_{\mathfrak{b}}\to {\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}\left(K\right)$
restricts to an immersion on an open neighbourhood of
${\overline{G}}_{0}$
.
By `immersion' we mean the same as in [
12]
. Since a finitedimensional subspace of a Banach space is always complemented, the condition in our case is simply that the derivative of the map be injective at every point.
Proof. Fix
$\omega =\left[S\right]\in {\overline{G}}_{0}$
. We will show that
${R}_{K}$
is an immersion at
$\omega $
(hence in a neighbourhood of
$\omega $
) when
$K$
is large enough. Since
${\overline{G}}_{0}$
is compact, this will prove the lemma.
Let
${\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}$
be the ambient configuration space containing
${M}_{\mathfrak{b}}$
as defined in the beginning of Section 3 .
Let
$W\subset Lp,w1$
be a linear subspace such that the derivative at
$S$
of the projection
$S+W\to {\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}$
is a linear isomorphism onto the tangent space of
${M}_{\mathfrak{b}}$
at
$\omega $
. Let
$\delta $
denote that derivative. For
$t\ge 0$
so large that
$\mathfrak{b}\subset X:t$
let
${\delta}_{t}$
be the derivative at
$S$
of the natural map
$S+W\to {\mathcal{\mathcal{B}}}_{\mathfrak{b}}^{*}(X:t)$
. We claim that
${\delta}_{t}$
is injective for
$t\gg 0$
. For suppose
$\left\{{w}_{n}\right\}$
is a sequence in
$W$
such that
$\parallel {w}_{n}{\parallel}_{Lp,w1}=1$
and
${\delta}_{{t}_{n}}\left({w}_{n}\right)=0$
for each
$n$
, where
${t}_{n}\to \infty $
. Set
${K}_{n}=X:{t}_{n}$
.
Then