))a. (4) If V' is a nowhere vanishing section of then W = CV'©V'"'") and a{Tp, •) defines a bundle isomorphism between M^; © tp-^ and T*Y ® iM. Denote by Ay the configuration space of (y, s) consisting of pairs (A, ip) with the completion under L^-norm. The gauge group of automorphisms of the Spin'^-bundle W is Qy = M.ap{Y,U{l)) with L2-completion. Qy acts on Ay by and the Seiberg-Witten equations are invariant under this action. Denote by By the quotient space of Ay by the gauge group action. By is an infinite dimensional Hilbert (4) (7(^,-0) = -p-''^{{lp®1p*)o) = -p-'^{lp®1p- = -j^Im{p{ei)i},i})e'- ,* u{A,il:) = {A — 2u ^du,utp), 5 manifold except at points where the spinor part is zero, which are called reducible points. Otherwise, points {A,tp) with ip ^ arc called irreducible. As noted in [13], the Seiberg-Witten equations on (1", s, g) arc the equations for the critical points of the following Chern-Simons-Dirac functional on Ay- C^{A, V') = - Aq) a {Fa + - 2 * /x) + ^(V, ^A^)dvolY, (5) where is a fixed connection on detiW). Note that C^^ descends to a circle- valued function on By- The set of critical points of C^^ on By is denoted by A^y,^(s), its irreducible critical point set is denoted by A1y_^(s)*. For any critical point (A, ^) on Ay, the infinitesimal action of Qy and the derivative of grad{Cij) at (A, ip) define a complex 2 (y, iM) ^"^^ (y, i^)®Ll{w) "'^^^^ {y, iR)®L\w), (6) where the maps G(^A,tp) -^(A.V') ^® given by G(A,v>)(/) = (-2d/,/^), H(A,tjj){a, 4>) = {*da - 2a{ip, 4>), ^a4> + -^Oi-i^)- We say that [A, ■0] is a non-degenerate critical point of C^^ on By if the middle cohomology of (6) is zero: KerH^A,i,^/ImG^A,i,) = 0. At the smooth points of By, this definition is the same as saying that the derivative of grad{C^) at a critical point is non-degenerate. The gradient of C^^ can be viewed as an L^-tangent vector field on By, a section of the L^-tangent bundle over By, while the tangent space of By at [A, ■0] is the -completion of -f^erG^^ = {(q, + ilm{^), 0) = 0.} The covariant derivative of grad{C^), denoted by H^a,iI)\-i defines a operator on KerG*^j^^^y sending {a,4>) G KerG*^^^^^^ to {*da - 2a{iP, (t>) - 2df, ^a

)-
Note that H\^a,^ is a closed, unbounded, essentially self-adjoint, Fredholm operator
on the L^-completion of KerG*,. , its eigenvectors form an L^-complete orthonormal
6
basis, its L^-spectrum forms a discrete subset of the real line with no accumulation
points. Hence, as in [20], the spectral flow of i^jy^ ,^], along a path connecting two critical
points defines a relative index on Aiy^i^) ^ -^y^(^)- This relative index depends only
on the homotopy class of the connecting path for non-torsion Spin'^ structure (Cf.
Remark 4.5 in [20] and Definition 3.6 in [7]).
The following properties about the critical points of are now standard (See [20]
[8] [11] [14] [16]).
Proposition 2.1. There exists a Baire set of co-closed 1-form fi G ^l^j^2{Y,iM.) such
that all the critical points in A^y^^(s) are non-degenerate. Moreover, if bi{Y) > 0,
My,ij,{s) consists of only finitely many irreducible points in By; if Y is a rational
homology 3-sphere, assume that a generic fj, satisfies Ker^Q = (where 9 is the unique
reducible point in A1y,^(s), that is, ^Fq = fJ,), then My,ij,{s)* = My,ij,{s) — {0} consists
of only finitely many irreducible points.
In this paper and sequel work, it is convenient to use a perturbation with support
contained in a fixed open set, so that Proposition 2.1 still holds for perturbations with
compact support contained in a fixed open set. The first such statement was made in
Proposition 7.1 [32] by Taubes, who kindly communicated the proof to us.
Proposition 2.2. Fix a non-empty open set U in Y and a Spin'^ structure s on
Y, if bi(Y) > and ci{det{s)) = 0, we require that U is chosen so that the map
ii"^(y, R) — ^ H'^(U,W) is non-zero. Then there exists a Baire set of co-closed imaginary
valued 1 -forms fx with compact support in U such that all the critical points of on
By are non-degenerate.
Proof. We first study the family version of the critical points of on By, where fi
is from a set of imaginary valued co-closed 1-forms on Y with compact support in U.
Denote this set of perturbations as Z{U, iM). Let [fi, A, tp] be a critical point of C^. We
need to show that the derivative of the gradient of {C^}^eZ(c/,iR) is surjective. Namely,
consider
KerGl^^^^x Z{U,iR) ^ KerGl^^^^,
which sends (a,0,/xi) to
{*da - 2a{ip, (p) + /xi, ^a4> + ^"•V')-
Suppose that (a, (p) is orthogonal to the image of the above map, then (a, cp) satisfies
7
the following equations:
(1)
(2)
(3)
(4)
d*a + ilm{4!, 4>) = 0,
*da — 2a{tp, (j)) = 0,
^a4> + \ot.'4) = 0,
a is exact when restricted to U.
(7)
The elliptic regularity implies that (a, 4>) is smooth. Prom (4) and (2) of equations
(7), we know that a{tl^,(j)) = on U. The following Lemma due to Taubes [33] will
ensure that cr^tp, 0) = on Y. Hence, there is a real-valued smooth function f on Y,
such that (p = if'fp- Using (3) of (7), we obtain
on Y. Note that ip ^{0) does not disconnect any domain in Y (the unique continuation
principle for Dirac operator (see page 57-58 [10])). Therefore, / = which implies that
Prom the Sard-Smale theorem, there is a Baire set of /x G Z{U,i'R) such that all
critical points of C^^ in By are non-degenerate for a generic fi.
Now we need to prove that the reducible critical point of is also non-degenerate.
By the assumption, admits reducible critical point if and only if y is a rational
homology 3-sphere. From the analysis in [20], we know that, in order to achieve the
non-degenerate condition at reducible critical point, /i is required to be away from
the codimension one subset Z{U,iM.) where the corresponding Dirac operator has non-
trivial kernel. This completes the proof of the Proposition. Now we give the proof of
Taubes' Lemma.
Lemma 2.3. (Taubes) Let {A,tp) and {oi,4>) o,^ above, where {A,i/j) is a solution to
the Seiberg-Witten equation (4) and {a,(f)) satisfies (l)-(3) of the equations (7). Then
q = (f)) obeys an equation of the form
at all points where ip ^ 0. Here A is the Laplacian on differential 1- forms and H and
K are linear maps that depend implicitly on ip. The set of points where ip ^ is path
connected open dense set in Y . The unique continuation principle applies to q so that
q cannot vanish on U without vanishing everywhere on Y.
on Y, which leads to a = —2idf on Y. By the equation (1) in (7), we get
2d*df + =
(«,) =0.
Aq = H ■ q + K -Vq
8
Proof of Taubes' Lemma : Apply the Laplacian to q = a{ilj,(f)), we have the following
expression of Aq:
Aq = a{A^, (t>) + (7(^, A4>) + 2(7({Va^, Va4>}t*y), (8)
here A acting on spinors is V^V^, and {VaV') ^ a4'}t*y is the pairing using the metric
on T*Y . Now invoke the Weitzenbock formula for the Dirac operator,
where n is the scalar curvature on Y. Thus, from the Dirac equations for and 0, we
obtain
A(/> = f - \{*Fa)4 + \{d*a)il) - \{*da)4 - V^V,
here V^ip = {a,V^V}T*y- Plug these two equations into (8), and note that
(7(V', ^{d*a)'ip) = and cr('i/', -^{*da).ip) = \tp\'^q. We get
i1 + m^)q + *FAiRe{^,(^))
(9)
+2a({VAV', Vyi<^}T-y) + -V^V)-
Write (f) = iril) + A where r is a real- valued function on Y and i?e(A, = 0, then
Re{'il^,(t)) = Re{ip,X),
sufficiently small. Consider the flat metric on given by go = A*g. There exists
a constant c and a smooth path g{s) (s G of flat metrics on with the following
properties:
(i) g{s) = jjgo, for all s < e and g{s) = gi for all s > 1 - e;
(ii) 9i = 5(1) is a metric of the form gi = kidv? + k2dv'^ with positive constants ki;
(Hi) The scalar curvature of the metric g := g{s) + ds^ on x M is non-negative;
(iv) The metric gi can be extended to a metric inside the solid torus ^{K) with
positive scalar curvature.
Proof of Nicolaescu's Lemma : Choose a unit vector du with respect to the metric ^^Oj
and complete it to an oriented orthonormal frame. Let {<^i,<^2} C O-'^(r^) be the dual
coframe. This is related to {du, dv} by
ifi = du + aodv Lp2 = k dv,
for some positive constant k > 0.
The path g[s) is defined by requiring that the coframe
cpi{s) = du + a{s)dv (p2{s) = kdv
be orthonormal with respect to g{s), where a(s) is a smooth function satisfying a(s) =
for all s > 1 — e and a{s) = ao for all s < e. The only conditions that need to be verified
are (iii) and (iv).
We have an orthonormal coframe {ipQ,(pi,ip2} on X = x M, with respect to
the metric g, with (po = ds, and a corresponding orthonormal frame {60,61,62}. The
Levi-Civita connection is of the form
/ X y\
r= -X z x,y,z en^{x).
\-y -z /
The Cartan structural equation gives d(p = T /\ 0, with = {(po,(pi,(f2)- By the
expression of (pi , we have
a
dipo = dip2 = 0, dtpi = -(fo A (p2,
44
hence we obtain
a; A (^1 + y A <^2 = 0,
Xf\ipQ + Zf\'f2
-y Aipo - z Aifi = 0.
These equations imply
^ ip2
-ip2 ipo
\ -^1 -^0
Thus, we can compute the scalar curvature of ^ = g{s) + ds on X M which is 3(d/fc)
by direct calculation.
Claim (iv) then follows by noticing that any diagonal metric of the form
realizes the torus metrically as the product of two circles of different radii. Each can
bound a hemisphere, endowed with a positive scalar curvature metric, thus extending
gi to a metric on a solid torus, with positive scalar curvature.
4.4 Lines in
In this subsection wc justify why it is sufficient to consider the gluing map in Theorem
Lemma 4.12. The intersection points d^oMv H x{^{^))j "f^^th x{i^{K)) C the
circle of reducibles for v{K) in either Y , Yi, or Yq, are contained in x{T'^)\Uq, for
some neighborhood Uq of the singular point 6. Thus, the gluing of Theorem 1.3 happens
away from the reducible point.
Proof. The torus inside Y inherits from the trivial Spin structure of Y the non-
trivial Spin structure in which both circles (longitude and meridian) bound, that is,
the one determined by the element (1,1) in H^{T'^,Z2). Introduce the coordinates u
and V on iif^(r^,M), defined by the property that, under the projection to x(r^), they
satisfy the following condition. For [A] G ""([^D is the holonomy around the
meridian m and w([^]) is the holonomy around the longitude I. Under this coordi-
nate system, the singular point Q is given by (1, 1) in x(T^), and the reducible circle
x(F) = My'^ is given by {u = 0}. Also for the unperturbed Seiberg-Witten equations
gi = kidu"^ + k2dv'^
□
1.3 away from the singular point 6 G x(^^)-
45
on i^iK) C Y, with the metric of non-negative scalar curvature, strictly positive away
from the boundary, the reducible circle xii^i^)jY) is given by
Ly :={v = 0}CxoiT\Y).
Similarly, in the case of Yi, choose a metric with a long cylinder [— r, r] x T^, which
agrees with the original metric on Y when restricted to the knot complement V, and
such that the induced metric in the torus neighbourhood i^(i^) is as described in Lemma
4.11, then the reducible circle x{'^{K),Yi) is given by
Ly, := {v-u = l}Cxo{T\Yi).
We might have expected that the reducible case would be Ly^ := {v — u = 0} from
the -|-l-surgery manifold Yi. The shift is due to the fact that -|-l-surgery changes the
underlying Spin structure (1, 1) by tensoring with a flat Z2-bundle of class (0, 1) in
H^{T^,Z2). In the case of Iq, the reducible circle x{^{^):Yo) for the unperturbed
equations is mapped to a circle {n = 0} C x(r^). This is because
x(t2, u{K)) = x{T\ V) = x{T\ Fo) = K X S\
so that x{^{K)i ^) consist of a Z-family of circles given by {-u = 2A;, /c G Z} where u is
the coordinate of M in M x 5^ . The gluing map on the fiber product
-^y,!^, ^x{T\Yo) {u = 2k}
would correspond to the moduli space A^i^, (^fc) where is the Spin'^ structure with
Ci{det5k) G H^(Yq,Z). For the trivial Spin'^ structure Sq, there would be a circle of
reducible monopoles in Alyo(so) resulting from gluing the reducibles x(^) Y)i^{u = 0},
we need to introduce a small perturbation inside iy{K) such that x{^{K), Yq) = {u = rj}
where rj is small number in M for the trivial Spin"^ structure So-
Clearly, in all the cases, x(i^(-f^)) does not go through the singular points {7r~^(0)},
hence there is no need to consider the gluing map at the singular point Q. Note that,
after perturbing the metric inside a compact set on the manifold V with an infinite
cylindrical end, we can make the open ends in Mv not limit to any intersection points
of x(^) = My'^ = {n = 0} with any circle xit^iK)) for either Y,Yi or Yq.
□
Thus, with our choice of metric as in the previous subsection, and with the choice
of perturbation as in Theorem 3.10, we see that the gluing result stated in Theorem
1.3 holds for the manifolds Y,Yi, and Yq. This completes the proof of Theorem 1.3.
Remark 4.13. Note that gluing the reducible m,onopoles on V and v{K) with matching
boundary condition just gives the extension of the flat connections to the whole manifold
46
(after a possible gauge transformation). We call this the trivial gluing. The unique
reducible point Oy in My is obtained by the trivial gluing of the unique intersection point
between the lines Ly = {v = 0} and {u = 0} = tt^^{x{V)) C x{T'^i^)- The unique
reducible point Oy^ in My-^ is obtained by the trivial gluing of the unique intersection
point between the lines Ly = {v — u = 1} and {n = 0} = TT^^(xi^)) ^ x(r^, Yq).
5 The geometric triangle and proof of Theorem 1.1
In the previous section, we showed that the moduH spaces for irreducible monopoles
on Y, Yi and Yq are given by the gluing maps on the following fiber products:
^*Yir) = Mv,Y XxiT^Y) = 0}'
■^k(r) = ^V,Yi Xx(T2,n) {V-U= 1},
^*Yo{r)(^''^-^V,Yo>