Local Characterizations of Saddle Points and Their Morse Indices
Local Characterizations of Saddle Points and Their Morse Indices
Yongxin Li
( u) = 0
as an operator J
( u)
is invertible as a linear operator J
Department of Mathematics, Texas A& M University, College Station, TX 77843. Supported in part by
NSF Grant DMS 96-10076.
1
and called saddle points, that is, critical points u
in
H contains points v, w s.t. J(v) < J(u
( u) is
negative denite; the nullity of a critical point u is the dimension of the null-space of J
( u).
So for a non-degenerate critical point, if its MI = 0, then it is a local minimizer and a stable
solution, and if its MI > 0, then it is a min-max type and unstable solution.
Multiple solutions with dierent performance and instability exist in many nonlinear
problems in natural and social sciences [21, 18, 13, 22, 12]. Stability is one of the main
concerns in system design and control theory. Morse index of a solution can be used to
measure its instability for some variational problems [4,20]. However in many applications,
performance or maneuverability is more desirable, in particular, in system design or control
of emergency or combat machineries. Usually instable solutions have much higher maneu-
verability or performance indices. For providing choice or balance between stability and
maneuverability or performance, it is important to solve for multiple solutions and their
Morse indices.
When cases are variational, they become multiple critical point problems. So it is impor-
tant for both theory and applications to numerically solve for multiple critical points and
their Morse indices in a stable way. So far, little is known in the literature to devise such
a feasible numerical algorithm. Minimax principle is one of the most popular approaches
in critical point theory. However, most minimax theorems in the literature (See [1], [15],
[16], [18], [22]), such as the mountain pass, various linking and saddle point theorems, re-
quire one to solve a two-level global optimization problem and therefore not for algorithm
implementation.
In [10], motivated by the numerical works of Choi-McKenna (6) and Ding-Costa-Chen (8),
the Morse theory and the idea to dene a solution submanifold, new local minimax theorems
which characterize a critical point as a solution to a two-level local optimization problem
are established. Based on the local characterization, a new numerical minimax method for
nding multiple critical points is devised. The numerical method is implemented successfully
2
to solve a class of semilinear elliptic PDE on various domains for multiple solutions [10].
Although Morse index has been printed for each numerical solution obtained in [10], their
mathematical verications have not been established. In [2], by using a global minimax
principle, A. Bahri and P.L. Lions established some lower bound estimates for Morse indices
of solutions to a class of semilinear elliptic PDE. There are also some eorts in the literature
to numerically compute the Morse index of a solution to a class of semilinear elliptic PDE. In
addition to nding a saddle point, one has to solve a corresponding linearized elliptic PDE
at the solution for its rst a few eigen-values. It is very expensive.
Since Morse index reects local structure of a critical point, in this paper we show that our
local minimax characterization enables us to establish more precise estimates for the Morse
index of a saddle point in a more general setting. By our results the Morse index of a saddle
point based on the local minimax method can be estimated even before we numerically
compute the saddle point. So no extra work is required in addition to computation for
the saddle point. In the last section, new local characterization of saddle points which are
more general than minimax solutions and bound estimates for their Morse indices will be
developed. In the rest of this section, we introduce some notations and theorems from [10]
for future use.
For any subspace H
H, let S
H
= v[v H
.
Let L be a closed subspace in H, called a base space, and H = L
be the orthogonal
decomposition where L
if for any v S
L
, P(v) is the set of all local maximum points of J on [L, v].
A single-valued mapping p: S
L
H is a peak selection of J w.r.t. L if
p(v) P(v) v S
L
.
For a point v S
L
, we say J has a local peak selection w.r.t. L at v, if there is a
neighborhood ^(v) of v and a mapping p: ^(v) S
L
H such that
p(u) P(u) u ^(v) S
L
.
3
Most minimax theorems in critical point theory require one to solve a two-level global mini-
max problem and not for algorithm implementation. Our local minimax algorithm requires
one to solve only unconstrained local maximizations at the rst level. As pointed in [10],
numerically it is great. However, theoretically, it causes three major problems: (a) for some
v S
L
, P(v) may contain multiple local maxima in [L, v]. In particular, P may contain
multiple branches, even U-turn or bifurcation points; (b) p may not be dened at some points
in S
L
; (c) the limit of a sequence of local maximum points may not be a local maximum
point. So the analysis involved becomes much more complicated. We have been devoting
great eorts to solve these three problems. We solve (a) and (b) by using a local peak se-
lection. Numerically it is done by following certain negative gradient ow and developing
some consistent strategies to avoid jumps between dierent branches of P. As for Problem
(c), numerically we showed in [11] that as long as a sequence generated by the algorithm
converges, the limit yields a saddle point. New local characterization of saddle points in this
paper will further help us to solve those problems.
The following two local characterizations of saddle points are established in [10] and
played important role in our local theory. We then provide some bound estimates of Morse
indices of solutions based upon these two local characterizations.
Lemma 1.1 Let v
S
L
be a point. If J has a local peak selection p w.r.t. L at v
such that
p is continuous at v
and dis(p(v
(p(v
)) = 0 or
for any > 0 with |J
(p(v
)) < |v(s) v
|
for any 0 < s < s
0
and
v(s) =
v
+ sd
|v
+ sd|
, d = J
(p(v
)).
The above result indicates that v(s) dened in the lemma represents a direction for certain
negative gradient ow of J(p()) from v. So it is clear that if p(v
0
) is a local minimum point
of J on any subset containing the path p(v
0
(s)) for some small s > 0 then J
(p(v
0
)) = 0. In
particular, when we dene a solution manifold
/ =
p(v) : v S
L
,
4
we have p(v(s)) /. A solution submanifold was rst introduced by Nehari in a study
of a dynamic system [14] and then by Ding-Ni in study of semilinear elliptic PDE [16] and
they prove that a global minimum point of a generic energy function J on the solution
submanifold / w.r.t. L = 0 is a saddle point basically with MI= 1. It is clear that our
denition generalizes the notion of solution (stable) submanifold.
Theorem 1.1 Let v
0
S
L
be a point. If J has a local peak selection p w.r.t. L at v
0
s.t.
(i) p is continuous at v
0
,
(ii) dis(p(v
0
), L) > 0 and
(iii) v
0
is a local minimum point of J(p(v)) on S
L
.
Then p(v
0
) is a critical point of J.
The following PS condition will be used to replace the usual compact condition.
Denition 1.2 A function J C
1
(H) is said to satisfy the Palais-Smale (PS) condition, if
any sequence u
n
H with J(u
n
) bounded and J
(u
n
) 0 has a convergent subsequence.
2 Bound Estimates for Morse Index
Morse index provides understanding of the local structure of a critical point and is used as
an instability index for an unstable solution. It is an important notion in stability analysis
[4]. Although we have printed Morse index for each numerical solution computed by our
minimax method in [10], their mathematical justications have not been veried. In this
section, we establish several bound estimates for the Morse index of a critical point based
on our minimax method.
Lemma 2.1 Let v
0
S
L
be a point. If there exist a neighborhood ^(v
0
) of v
0
and a locally
dened mapping p : ^(v
0
) S
L
H such that p(v) L, v v ^(v
0
) S
L
. If p is
dierentiable at v
0
and u
0
= p(v
0
) / L, then
p
(v
0
)(L, v
0
) +L, v
0
= H.
5
Proof. For any w L, v
0
, |w| = 1, denote v
s
=
v
0
+sw
v
0
+sw
. Then there exists s
0
> 0
such that when [s[ < s
0
, we have v
s
^(v
0
) S
L
.
Consider the one dimensional vector function (s) = P
L
(p(v
s
)), where P
L
is the projec-
tion onto L
. Since p is dierentiable at v
0
and v
s
smoothly depends on s, is dierentiable
at 0 and
(0) = P
L
(p
(v
0
)(
v
s
s
)) = P
L
(p
(v
0
)(w)).
On the other hand, p(v
s
) L, v
s
, we have (s) = t
s
v
s
, where t
0
= (s), v
s
) is dieren-
tiable. So
(0) = t
0
v
0
+ t
0
w, where due to our assumption that u
0
= p(v
0
) / L, we have
t
0
,= 0. The two dierent expressions of
(0) imply
P
L
(p
(v
0
)(w)) = t
s
(0)v
0
+ t
0
w.
Then it leads to w p
(v
0
)(w), L, v
0
. Since w is an arbitrary vector in L, v
0
, it follows
that
p
(v
0
)(L, v
0
) +L, v
0
= H. (2.1)
Lemma 2.2 Let v
0
S
L
be a point. If there exist a neighborhood ^(v
0
) of v
0
and a
locally dened mapping p : ^(v
0
) S
L
H such that p(v) L, v v ^(v
0
) S
L
.
Assume that p is dierentiable at v
0
and u
0
= p(v
0
) / L. If u
0
is a critical point of J with
MI(u
0
) > dimL + 1, then
p
(v
0
)(L, v
0
) H
,= 0.
Proof. Denote H
(u
0
) and k = dimL+1. Then dimH
> k.
By applying Lemma 2.1, there exit linearly independent vectors e
0
, e
1
, . . . , e
k
H
which can
be represented as e
i
= g
i
+ f
i
with g
i
p
(v
0
)(L, v
0
) and f
i
L, v
0
. f
0
, f
1
, . . . , f
k
have
to be linearly dependent because k = dimL + 1. So we can nd real numbers a
0
, a
1
, . . . , a
k
such that
k
i=0
a
2
i
,= 0 and
k
i=0
a
i
f
i
= 0. Therefore
k
i=0
a
i
e
i
=
k
i=0
a
i
g
i
p
(v
0
)(L, v
0
) H
.
Because, e
0
, e
1
, . . . , e
k
are linearly independent,
k
i=0
a
i
e
i
,= 0. Thus, the conclusion of the
lemma is veried.
6
Theorem 2.1 Let v
0
S
L
be a point. If J has a local peak selection p w.r.t. L at v
0
such
that p is dierentiable at v
0
and u
0
= p(v
0
) / L. If v
0
is a local minimum point of J p on
S
L
, then u
0
is a critical point of J with MI(u
0
) dimL + 1.
Proof. Since p is a local peak selection of J w.r.t. L at v
0
, there exists a neighborhood
^(v
0
) of v
0
such that p(v) L, v, v ^(v
0
) S
L
. By applying Lemma 2.1, we have
p
(v
0
)(L, v
0
) +L, v
0
= H
or
codim(p
(v
0
)(L, v
0
)) dimL + 1.
Now suppose that MI(u
0
) > dimL + 1. Denote H
(u
0
). By
Lemma 2.2, we have
p
(v
0
)(L, v
0
) H
,= 0. (2.2)
Choose any w L, v
0
(v
0
)(w) H
. Around u
0
= p(v
0
), we have
the second Taylor expansion
J(u) = J(u
0
) +
1
2
J
(u
0
)(u u
0
), u u
0
) + o(|u u
0
|
2
) (2.3)
Denote v
s
=
v
0
+sw
v
0
+sw
, we have v
s
^(v
0
) S
L
for [s[ small and then
dvs
ds
= w. So it follows
p(v
s
) = u
0
+ sp
(v
0
)(w) + o([s[). (2.4)
Combining the above two estimates (2.3) and (2.4), we obtain
J(p(v
s
))
= J(u
0
) +
1
2
J
(u
0
)(sp
(v
0
)(w) + o([s[)), sp
(v
0
)(w) + o([s[)) + o(|sp
(v
0
)(w) + o([s[)|
2
)
= J(u
0
) +
1
2
s
2
J
(u
0
)(p
(v
0
)(w)), p
(v
0
)(w)) + o(s
2
)
< J(u
0
),
where the last strict inequality holds for [s[ suciently small, because p
(v
0
)(w) H
.
Since v
s
^(v
0
) S
L
and u
0
= p(v
0
), the above contradicts the assumption that v
0
is
a local minimum point of J p on S
L
. Therefore MI(u
0
) dimL + 1.
7
Theorem 2.2 If p is a local peak selection of J w.r.t. L at v
0
S
L
and u
0
= p(v
0
) is a
non-degenerate critical point of J, then MI(u
0
) dimL + 1.
Proof. Assume that k = MI(u
0
) < dimL +1. By our assumption, u
0
is non-degenerate,
i.e., J
(u
0
) is invertible, we have H = H
+
where H
+
is the maximum positive subspace
and H
(u
0
). It follows that codim(H
+
) = dim(H
(u
0
) is invertible, then MI(u
0
) = dimL + 1.
Proof. Since under the conditions, we have proved that u
0
= p(v
0
) is a non-degenerate
critical point of J. The conclusion follows by combining the last two theorems.
Theorem 2.4 Let v
0
S
L
be a point. If there exist a neighborhood ^(v
0
) of v
0
and a
locally dened mapping p: ^(v
0
)S
L
H such that p(v) L, v, J
(p(v)) L, v, v
^(v
0
) S
L
and p dierentiable at v
0
. If v
0
S
L
is a local minimum point of J p on S
L
with u
0
= p(v
0
) / L, then u
0
is a critical point of J with MI(u
0
) dimL + 1.
Proof. We rst prove that u
0
= p(v
0
) is a critical point of J. The second part of the
theorem follows from a similar proof of Theorem 2.1.
For any w L, v
0
, denote
v(s) =
v
0
+ sw
|v
0
+ sw|
.
We have v(s) ^(v
0
) S
L
for [s[ small and
dv(s)
ds
= w. Therefore
p(v(s)) = p(v
0
) + sp
(v
0
)
dv(s)
ds
+ o([s[)
= u
0
+ sp
(v
0
)(w) + o([s[).
8
It follows that
J(p(v(s))) = J(p(v
0
)) + J
(p(v
0
))(p(v(s)) p(v
0
)) + o(|p(v(s)) p(v
0
)|)
= J(u
0
) + sJ
(u
0
)p
(v
0
)(w) + o([s[).
If
J
(u
0
)p
(v
0
)(w) ,= 0
for some w L, v
0
(u
0
)p
(v
0
)(L, v
0
) = 0.
Since by our assumption
J
(u
0
)(L, v
0
) = 0
and by Lemma 2.1
p
(v
0
)(L, v
0
) +L, v
0
= H,
it follows that
J
(u
0
)u = 0 u H,
i.e., u
0
= p(v
0
) is a critical point of J.
It is worthwhile indicating that if p is a local peak selection of J at v
0
S
L
, then
p(v) [L, v] and J
(u
0
) [L, v
0
]. So such a locally dened mapping
generalized the notion of a local peak selection and resolved the problem that a limit of a
sequence of local maximum points may not be a local maximum point. This generalization
has a potential to design a new type of local algorithm for nding multiple saddle points
that are not necessarily of a minimax type.
Theorem 2.5 If u
0
/ L is a non-degenerate critical point of J such that u
0
is not a local
minimum point of J along any direction v L, u
0
, then
MI(u
0
) dimL + 1.
9
Proof. Assume that k = MI(u
0
) < dimL +1. By our assumption, u
0
is non-degenerate,
i.e., J
(u
0
) is invertible, we have H = H
+
+H
where H
+
is the maximum positive subspace
and H
(u
0
). It follows that codim(H
+
) = dim(H
0
f(x, t)dt.
10
(h4) says that f is superlinear, which implies that there exist positive numbers a
3
and a
4
such that for all x
and R
F(x, ) a
3
[[
a
4
. (3.5)
The variational functional associated to the Dirichlet problem (3.1) is
J(u) =
1
2
[u(x)[
2
dx
F(x, u(x))dx, u H H
1
0
(), (3.6)
where we use an equivalent norm |u| =
[u(x)[
2
dx for the Sobolev space H = H
1
0
().
It is well known [18] that under Conditions (h1) through (h4), J is C
1
and satisfy (PS)
condition. A critical point of J is a weak solution, and also a classical solution of (3.1). 0 is
a local minimum point of J and so a trivial solution. Moreover, in any nitely dimensional
subspace of H, J goes to negative innity uniformly. Therefore, for any nite dimensional
subspace L, the peak mapping P of J w.r.t. L is nonempty.
We need one more hypothesis, that is
(h5)
f(x,)
||
is increasing w.r.t. , or
(h5) f(x, ) is C
1
w.r.t. and f
(x, )
f(x,)
> 0.
It is clear that (h5) implies (h5). If f(x, ) is C
1
in , then (h5) and (h5) are equivalent.
All the power functions of the form f(x, ) = [[
k
with k > 0, satises (h1) through (h5),
and so do all the positive linear combinations of such functions. Under (h5) or (h5), J has
only one local maximum point in any direction, or, the peak mapping P of J w.r.t. L = 0
has only one selection. In other words, P = p. The proof can be found in [16] and [14].
Theorem 3.1 Under the hypothesis (h1) through (h5), if the peak mapping P of J w.r.t. a
nitely dimensional subspace L is singleton at v
0
S
L
and for any v S
L
around v
0
, a
peak selection p(v) is a global maximum point of J in [L, v], then p is continuous at v
0
.
Proof. See [10].
Theorem 3.2 Assume that Conditions (h1) (h5) are satised and that there exist positive
constants a
5
and a
6
s.t.
[f
(x, )[ a
5
+ a
6
[[
s1
(3.7)
11
where s is specied in (h2). Then the only peak selection p of J w.r.t. L = 0 is C
1
.
Proof. See [10].
Since w
0
= 0 is the local minimum point of J and along each direction v H, J has
only one maximum point p(v), we have J(p(v)) > 0, v S. If for each v S
L
, p(v) is a
local maximum point of J in [L, v], then p(v) is the only local maximum point of J along
the direction u =
p(v)
p(v)
, Therefore we have J(p(v)) > 0, v S
L
. As a composite function
J(p()) is bounded from below by 0. So Ekelands variational principle can be applied. With
the PS condition, existence result can also be established.
Theorem 3.3 Under the hypothesis of (h1) through (h5), and that there exist positive con-
stants a
5
and a
6
such that
[f
(x, )[ a
5
+ a
6
[[
s1
,
where s is specied in (h2), if v
0
= arg min
vS
H
J(p(v)) then u
0
= p(v
0
) is a critical point
with MI(u
0
) = 1.
Proof. Assume u
0
= p(v
0
) = t
0
v
0
. Consider the 1-dimensional function
g(t) = J(tv
0
) =
1
2
t
2
[v
0
(x)[
2
dx
F(x, tv
0
(x)) dx.
We have
g
(t) = t
[v
0
(x)[
2
dx
f(x, tv
0
(x))v
0
(x) dx.
So
1 =
f(x, t
0
v
0
(x))
t
0
v
0
(x)
v
2
0
(x) dx.
Meanwhile we have
g
(t) =
[v
0
(x)[
2
dx
(x, tv
0
(x))v
2
0
(x) dx
g
(t
0
) = 1
(x, t
0
v
0
(x))v
2
0
(x) dx
< 1
f(x, t
0
v
0
(x))
t
0
v
0
(x)
v
2
0
(x) dx (ref. (h5)
= 0,
12
which implies that
H
0
L, v
0
= 0,
where H
0
is the nullspace of J
at u
0
and L = 0. By Theorem 2.1, we obtain
MI(u
0
) = dim(L) + 1 = 1.
4 Some New Saddle Point Theorems
As our convergence results in [11] indicate that our algorithm can be used to nd a non-
minimax critical point, e.g., a Monkey saddle point. Thus the argument already exceeded
the scope of a minimax approach. So far the only results we found in the critical point
theory which are more general than a minimax principle are two theorems proved by S. I.
Pohozaev in [9] or [17]. The following results are interesting generalizations. The rst one
is an embedding result. It is general but lacks of characterization. The second result has
potential applications in devising a new numerical algorithm.
Theorem 4.1 Given L = spanw
1
, ..., w
k
in H and let
J(t, v, t
1
, .., t
k
) J(tv + t
1
w
1
+ ... + t
k
w
k
).
If (t
0
, v
, t
1
, .., t
k
) is a conditional critical point of
J subject to v S
L
with t
,= 0, then
t
0
v
+ t
1
w
1
+ ... + t
k
w
k
is a critical point of J.
Proof: By the Lagrange Multiplier Theorem, there exist , ,
1
, ..,
k
with
2
+
2
+
2
1
+
... +
2
k
,= 0 such that the Lagrange functional
L(t, v, t
1
, ..., t
k
) =
J(t, v, t
1
, ..., t
k
) + |v| +
k
i=1
i
w
i
, v)
has a critical point at (t
0
, v
, t
1
, .., t
k
). So we have
L
t
= 0
J
t
(t
0
, v
, t
1
, ..., t
k
) = 0; (4.8)
L
v
= 0
J
v
(t
0
, v
, t
1
, ..., t
k
) + |v
+
k
i=1
i
w
i
= 0; (4.9)
L
t
i
= 0
J
t
i
(t
0
, v
, t
1
, .., t
k
) = 0 or
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
), w
i
) = 0, (i = 1, .., k). (4.10)
13
From (4.9), we have
J
v
(t
0
, v
, t
1
, .., t
k
), v) + |v
, v) +
k
i=1
i
w
i
, v) = 0 v H. (4.11)
In particular
J
v
(t
0
, v
, t
1
, .., t
k
), v
) + |v
, v
) +
k
i=1
i
w
i
, v
) = 0. (4.12)
Since |v
, v
) = 1 and w
i
, v
J
v
(t
0
, v
, t
1
, .., t
k
), v
) + = 0. (4.13)
So = 0 will lead to = 0 and then
i
= 0 by choosing v = w
i
in (4.10). It contradicts
2
+
2
+
2
1
+ ... +
2
k
,= 0. Therefore ,= 0 and (4.8) gives
t
(t
0
, v
, t
1
, ..., t
k
) = 0
or
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
), v
) = 0.
It leads to
v
(t
0
, v
, t
1
, ..., t
k
), v
) = t
0
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
), v
) = 0.
(4.13) then yields = 0. Taking any v [w
1
, ..., w
k
] in (4.11), we obtain
v
(t
0
, v
, t
1
, ..., t
k
), v) = 0
or
t
0
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
), v) = 0.
Since t
0
,= 0, it leads to
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
), v) = 0 v [w
1
, ..., w
k
].
Taking (4.10) into account, we have
J
(t
0
v
+ t
1
w
1
+ ... + t
k
w
k
) = 0.
Theorem 4.1 reduces to Pohozaevs embedding result in [9] or [17] if we set L = 0.
14
Theorem 4.2 Let v
S
L
be a point. If J has a local peak selection p w.r.t. L at v
and
u
= p(v
) such that
(i) p is Lipschitz continuous at v
,
(ii) dis(u
, L) > 0,
(iv) u
is a critical point of J.
Proof: Suppose that |J
(u
)| > 0. Set =
1
2
|J
(u
(s))) J(p(v
)) < dis(u
, L)|v
(s) v
| 0 < s < s
0
where
v
(s) =
v
+ sd
|v
+ sd|
^(v
) S
L
, d = J
(p(v
)).
Then we have
J(p(v
(s))) J(p(v
))
|p(v
(s)) u
|
|p(v
(s)) u
|
|v
(s) v
|
< dis(u
) is a neighborhood of v
and u
(s)) u
|
|v
(s) v
|
is bounded and
J(p(v
(s))) J(p(v
))
|p(v
(s)) u
|
0 as s 0.
So the left hand side of (4.14) goes to zero, which leads to a contradiction.
Theorem 4.3 Let v
0
S
L
be a point. If J has a local peak selection p w.r.t. L such that p
is dierentiable at v
0
and u
0
= p(v
0
) / L. If u
0
is a conditional critical point of J on p(S
L
)
and v
0
is not a local maximum point of J p along the projection of any direction v on S
L
,
then u
0
is a critical point of J with MI(u
0
) dimL+1. In addition, if u
0
is nondegenerate,
then MI(u
0
) = dimL + 1.
15
Proof. By Theorem 4.2, we obtain that u
0
is a critical point of J. Then following a
similar argument in the proof of Theorem 2.1, until we have
J(p(v
s
))
= J(u
0
) +
1
2
J
(u
0
)(sp
(v
0
)(w) + o([s[)), sp
(v
0
)(w) + o([s[)) + o(|sp
(v
0
)(w) + o([s[)|
2
)
= J(u
0
) +
1
2
s
2
J
(u
0
)(p
(v
0
)(w)), p
(v
0
)(w)) + o(s
2
)
< J(u
0
),
where
v
s
=
v
0
+ sw
|v
0
+ sw|
^(v
0
) S
L
and w [L, v
0
]
, |w| = 1.
So the last strict inequality contradicts to our assumption that v
0
is not a local maximum
point of J p along the projection of any direction v on S
L
. Thus MI(u
0
) dimL + 1.
If in addition, u
0
is nondegenerate, we can apply Theorem 2.5 to conclude that MI(u
0
) =
dimL + 1.
Condition (iv) in Theorem 4.2 is clearly satised if w