Lectures on Mean Curvature Flow (MAT 1063 HS) I. M. Sigal

Transcription

Lectures on Mean Curvature Flow (MAT 1063 HS) I. M. Sigal
Lectures on Mean Curvature Flow (MAT 1063 HS)
I. M. Sigal
Dept of Mathematics, Univ of Toronto
October 26, 2014
Abstract
The mean curvature flow arises material science and condensed matter physics and has been recently
successfully applied to topological classification of surfaces and submanifolds. It is closely related to the
Ricci and inverse mean curvature flow.
The most interesting aspect of the mean curvature flow is formation of singularities, which is the
main theme of these lectures.
Background on geometry of surfaces and some technical statements are given in appendices.
These are rough, unedited notes, based on the course given a few years back and will changed and
further developed in the course of the present course. My thanks go to Greg Fournodavlos, Dan Ginsberg
and Wenbin Kong for the help with the material of the notes.
Contents
1 The mean curvature flow and its general properties
3
1.1
Mean curvature flow for level sets and graphs . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.2
Special solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.3
Different form of the mean curvature flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.4
Mean curvature flow and the volume functional . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.5
Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
2 Self-similar surfaces and rescaled MCF
9
2.1
Solitons - Self-similar surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2.2
Rescaled MCF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
2.3
Classification of mean-convex self-similar surfaces . . . . . . . . . . . . . . . . . . . . . . . . .
12
3 Global Existence vs. Singularity Formation
1
13
MCF Lectures, October 26, 2014
4 Normal hessians.
2
15
4.1
Spectra of hessians for spheres and cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
4.2
Connection of the spectrum and geometry of self-similar surfaces . . . . . . . . . . . . . . . .
19
4.3
Linearized stability of self-similar surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
4.4
Explicit expression for the normal hessians . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
4.5
Hessians for spheres and cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
4.6
Minimal and self-similar submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
5 Stability of elementary solutions I. Spherical collapse
24
5.1
Rescaled equation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
5.2
The graph represetation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
5.3
Reparametrization of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
5.4
Lyapunov-Schmidt decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
5.5
Linearized operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
5.6
Lyapunov functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
5.7
Proof of Theorem 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
6 Stability of elementary solutions II. Neckpinching
32
6.1
Existing Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
6.2
Rescaled surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
6.3
Linearized map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
6.4
Equations for the graph function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
6.5
Reparametrization of the solution space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
6.6
Lyapunov-Schmidt splitting (effective equations) . . . . . . . . . . . . . . . . . . . . . . . . .
38
6.7
General case. Transformed immersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
6.8
Translational and rotational zero modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
6.9
Collapse center and axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
6.10 Equations for the graph function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
7 Volume-preserving mean curvature flow and isoperimetric problem (to do)
44
8 Submanifolds (to do)
44
MCF Lectures, October 26, 2014
3
9 Ricci flow (to do)
44
10 Evolution of graphs (to do)
44
11 Local existence (to do)
44
A Elements of Theory of Surfaces
44
A.1 Mean curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
A.2 Matrix representation of the first and second fundamental forms . . . . . . . . . . . . . . . .
46
A.3 Integration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
A.4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
A.5 Various differential operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
A.6 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
A.7 Riemann curvature tensor for hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
n
A.8 Mean curvature of normal graphs over the round sphere S and cylinder C
n+1
. . . . . . . . .
53
B Elliptic maximum principle
55
C Some general results using the parabolic maximum principle
55
D Elements of spectral theory
56
D.1 A characterization of the essential spectrum of a Schr¨odinger operator . . . . . . . . . . . . .
56
D.2 Perron-Frobenius Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
57
E Proof of the Feynmann-Kac Formula
1
60
The mean curvature flow and its general properties
The mean curvature flow, starting with a hypersurface S0 in Rn+1 , is the family of hypersurfaces S(t) given
by immersions ψ(·, t) which satisfy the evolution equation
∂ψ
= −H(ψ)ν(ψ)
∂t
(1.1)
ψ|t=0 = ψ0
where ψ0 is an immersion of S0 , H(x) and ν(x) are mean curvature and the outward unit normal vector
at x ∈ S(t), respectively. Recall that a map ψ : U → S, where U ⊂ Rn is open, is called the immersion,
iff dψ(u) is one-to-one (i.e. of the maximal rank) for every u ∈ U , and the mean curvature is defined as
H(x) = Tr dν(x), where ν : S → S n is the Gauss map given by ν : x → ν(x). (Wx := dν(x) : Tx S → Tν(x) S n
MCF Lectures, October 26, 2014
4
is the Weingarten map, or the shape operator. One can show that Wx : Tx S → Tx S, see Appendix A, where
the notions used above are defined and discussed.)
In this lecture we describe some general properties of the mean curvature flow, (1.1). We begin with
writing out (1.1) for various explicit representations for surfaces St .
1.1
Mean curvature flow for level sets and graphs
We rewrite out (1.1) for the level set and graph representation of S. Below, all differential operations, e.g.
∇, ∆, are defined in the corresponding Euclidian space (either Rn+1 or Rn ).
1) Level set representation S = {ϕ(x, t) = 0}. Then, by (A.2) of Appendix A, we have
∇ϕ
∇ϕ
,
H(x) = div
.
ν(x) =
|∇ϕ|
|∇ϕ|
∇ϕ
∂ϕ
∂ϕ
∇ϕ
∂x
We compute 0 = dϕ
=
∇
ϕ
·
+
and
therefore
=
∇ϕ
·
div
x
dt
∂t
∂t
∂t
|∇ϕ|
|∇ϕ| , which gives
∂ϕ
= |∇ϕ| div
∂t
∇ϕ
|∇ϕ|
(1.2)
.
(1.3)
2) Graph representation: S = graph of f over U ⊂ Rn . Here f : U → R. In this case S is the zero level
set of the function ϕ(x) = xn+1 − f (u), where u = (x1 , . . . , xn ) and x = (u, xn+1 ), and using (1.3) with
this function, we obtain
!
p
∂f
∇f
2
= |∇f | + 1 div p
.
(1.4)
∂t
|∇f |2 + 1
Denote by Hess f the standard euclidean hessian, Hess f :=
∂2f
∂ui ∂uj
. Then we can rewrite (1.4) as
∂f
∇f Hess f ∇f
= ∆f −
.
∂t
|∇f |2 + 1
(1.5)
We can also consider graphs over another surface, say, the sphere, S n , i.e. ψ : S n → Rn+1 , with ψ given
by
ψ(ω, t) = f (ω, t)ω, where f : S n → R,
or the cylinder, C n = R × S n , i.e. ψ : C n → Rn+1 , with ψ given by
ψ(x, ω, t) = (x, f (x, ω, t)ω), where f : C n → R
1.2
Special solutions
Static solutions: Static solutions satisfy the equation H(x) = 0 on S, which, by Proposition 2 below, is the
equation for critical points of the volume functional V (ψ) (the Euler - Lagrange equation for V (ψ)). Thus,
static solutions of the MCF (1.1) are minimal surfaces.
MCF Lectures, October 26, 2014
5
Explicit solutions:
a) Sphere x(t) = R(t)ˆ
ˆ = x/kxk, or {ϕ(x, t) = 0}, where ϕ(x, t) := |x|2 − R(t)2 , or graph f ,
p x, where x
2
2
where f (u, t) = R(t) − |u| . (SR = RS n , where S n is the unit n-sphere.) Then we have
∇ϕ
n
H(x) = div
= div(ˆ
x) =
|∇ϕ|
R
p
n
and therefore we get R˙ = − R
which implies R = R02 − 2nt. So this solution shrinks to a point.
n−1
˙
b) Cylinder x(t) = (R(t)xˆ0 , x00 ), where x = (x0 , x00 ) ∈ Rk+1 × Rm . Then H(x) = n−1
R and R = − R
p
2
which implies R = R0 − 2(n − 1)t. (In the implicit function representation the cyliner is given by
P
1
n−1 2 2
{ϕ = 0}, where ϕ := r − R, with r =
x
.)
i
i=1
Not explicit example: Motion of torus (H. M. Soner and P. E. Souganidis).
1.3
Different form of the mean curvature flow
Multiplying the equation (1.1) in Rn+1 by ν(ψ), we obtain the equation
hν(ψ),
∂ψ
i = −H(ψ).
∂t
(1.6)
In opposite direction we have
Proposition 1. (i) The equations (1.1) and (1.6) are equivalent in the sense that if ψ satisfies (1.6), then
it satisfies (1.1) and, if ψ satisfies (1.6), then there is a (time-dependent) reparametrization ϕ of S, s.t.
ψ ◦ ϕ satisfies (1.1). (ii) The equation (1.6) is invariant under reparametrization. (One can say that (1.1)
involves fixing a specific parametrization (fixing the gauge).)
∂ψ
∂x
T
Proof. Denote ( ∂ψ
:= ∂ψ
˙ =
∂t )
∂t − hν, ∂t iν (the projection of ∂t onto Tψ S) and let ϕ satisfy the ODE ϕ
∂ψ
∂
∂
−1
T
−(dψ) ( ∂t ◦ ϕ) (parametrized by u ∈ U ). Then ∂t (ψ ◦ ϕ) = ( ∂t ψ) ◦ ϕ + dψ ϕ.
˙ Substituting ϕ˙ =
∂ψ
∂ψ
∂ψ N
∂ψ T
∂ψ N
T
T
◦
ϕ)
◦
ϕ)
=
(
)
−(dψ)−1 ( ∂ψ
into
this
and
using
that
hν,
(
i
=
0
and
+
(
=
∂t
∂t
∂t
∂t
∂t ) , where ( ∂t )
∂ψ
∂
∂
hν, ∂t iν, we obtain ∂t (ψ ◦ ϕ) = hν, ( ∂t ψ) ◦ ϕiν ◦ ϕ = −(Hν) ◦ ϕ.
To show that the equation (1.6) is invariant under reparametrization, we observe that, by the above,
∂
∂
ψ) ◦ ϕ + dψ ◦ ϕ ϕi
˙ = hν, ( ∂t
ψ)i ◦ ϕ.
hν ◦ ϕ, ( ∂t
1.4
Mean curvature flow and the volume functional
We show that the mean curvature arises from the first variation of the surface area functional. We do this
for surfaces which are graphs of some function, which is locally the case for any surface.
MCF Lectures, October 26, 2014
6
Recall that locally the surface volume functional can be written as
Z
√ n
gd u,
V (S ∩ U ) =
U
or V (ψ) =
functional.
R √
ψ
gdn u, where g := det(gij ). We show that the MCF is a gradient flow of the surface area
Recall that, if E : M → R is a functional on an infinite dimensional manifold M , then the Gˆateaux
derivative dE(u), u ∈ M , of E(u), is a linear functional on X defined as
dE(u)ξ =
∂
E(us )|s=0 ,
∂λ
(1.7)
where us is a path on M satisfying us |s=0 = u and ∂s us |s=0 = ξ, with ξ ∈ Tu M , if the latter derivative
exists.
If M is a Riemannian manifold, with an inner product (Riemann metric) h(ξ, η), then there exists a
vector field, gradg E(u), such that
h(gradh E(u), ξ) = dE(u)ξ.
(1.8)
This vector is called the gradient of E(u) at u w.r.t. the metric h.
We want to compute the Gˆ
ateaux derivative of the area functional V . Any infinitesimal variation
(vector field on the space of surfaces), η, can be decomposed into the normal and tangential components,
η = η N + η T , where η N is the projection onto the normal vector-filed ν, η N = hν, ηiν. As we have shown,
the tangential component, η T , generates a reparametrization of the surface. Since V is independent of
reparametrization, it suffices to look only at normal variations, ψs , of the immersion ψ, i.e. generated by
vector fields η, directed along the normal ν: η = f ν. We begin with
Proposition 2. For a surface S given locally by an immersion ψ and normal variations, η = f ν, we have
Z
√
dV (ψ)η =
Hν · η gdn u.
(1.9)
U
Proof. We want to show for g ≡ g(ψ) := det(gij ) that
√
√
d gη = H gν · η.
(1.10)
To this end we denote gˆ := (gij ) and use the representation det gˆ = eTr ln gˆ , which gives dgη = g Tr[ˆ
g −1 dˆ
g η]
and therefore
dgη = gg ji dgij η.
Next, we compute dgij η = 2
D
∂ψ ∂η
∂ui , ∂uj
E
∂
= 2 ∂u
j
D
∂ψ
∂ui , η
E
−2
D
∂
∂uj
(1.11)
E
∂ψ
,
η
. For normal variations, η = f ν, the
i
∂u
2
ψ
last relation and the relation bij = −h ∂u∂i ∂u
, νi (see Lemma 38 of Appendix A) give
j
dgij η = 2bij (ν · η),
(1.12)
MCF Lectures, October 26, 2014
7
where bij are the matrix elements of the second fundamental form. This relation, together with the relation
dg(ψ)η = g(ψ)g ji dgij η, proven above, and, implies dgη = 2(ν · η)gg ji bij , which, due to the relation bij =
2
ψ
−h ∂u∂i ∂u
, νi, gives
j
dgη = 2gH(ν · η).
√
1 −1/2
dgη, gives in turn (1.10).
This, together with d gη = 2 g
Corollary 3. Critical points of V (S), which are called the minimal surfaces satisfy the equation (the EulerLagrange equation for V (S)) H(x) = 0.
We define the normal gradient, gradN V (ψ), as grad V (ψ), restricted to the normal vector fields, ξ =
f ν. Then the equation (1.9) and the definition (1.8) imply that the (normal) gradient of V (ψ) at ψ is
gradN V (ψ) = H(ψ)ν(ψ), which implies the following
R
Theorem 4. The mean curvature flow is a gradient flow with the functional V and the metric S ξ · η:
∂t ψ = − gradN V (ψ).
Corollary 5. If S evolves according to (1.1), then V (S) decreases. In fact, ∂t V (S) = −
R
S
H 2 (x) < 0.
Proof. The assertion on ψ(U ) follows from (1.9) as
Z
Z
∂t V (ψ) =
Hν · ∂t ψ = − H 2 (x),
U
ψ
which after using a partition of unity proves the statement.
As a result, the area of a closed surface shrink under the mean curvature flow. Note also that the
equation (1.12) implies ∂t gij = 2bij (ν · ∂t ψ), which, together with (1.6), gives the following equation for the
evolution of the metric
∂t gij = −2bij H.
(1.13)
Graph representation. Note that, if S = graph f, f : U → R, then we can take ψ(u) = (u, f (u)) and
the proof of (1.9) simplifies. In this case we can take ψ(u) = (u, f (u)) and the surface area functional is
given by
Z p
V (f ) =
1 + |∇f |2 dn u.
U
we compute, using that ξ|∂U = 0,
∇f
· ∇ξdn u,
2
1
+
|∇f
|
U
∇f
which, after integrating by parts and using the equation H(x) = − div √
(see (A.3) of Appendix
2
Z
dV (f )ξ = ∂s |s=0 V (f + sξ) =
p
1+|∇f |
A), gives
Z
dV (f )ξ =
U
Hξdn u.
(1.14)
MCF Lectures, October 26, 2014
8
Using (1.14) we find
Z
dV (f )ξ = −
n
Hξd u = −
U
So we have grad V (f ) = −
p
Z p
U
1 + |∇f |2 Hξ p
1 + |∇f |2 H(x) in the metric hf (ξ, η) :=
dn u
1 + |∇f |2
R
U
ξη √
.
dn u
,
1+|∇f |2
which together with
(1.4) gives ∂t f = − grad V (f ).
1.5
Symmetries
Proposition 6. Mean curvature is invariant under translations (S → S + h ∀h ∈ Rn+1 ), rotations (S →
g −1 S ∀g ∈ O(n + 1)) and scaling transformations,
S(t) → λS(λ−2 t)
⇔
x→
λx,
t → λ−2 t.
The invariance under translations and rotations is obvious. To prove the invariance under scaling, we
first show that
H(λx) = λ−1 H(x).
D
E
2
∂ψ ∂ψ
ψ
·
ν
and
g
:=
Indeed, looking at the definitions bij = ∂u∂i ∂u
,
, we see that
i
j
ij
∂u ∂u
j
bij → λbij
and gij → λ2 gij
if
(1.15)
ψ → λψ.
Hence g ij bji → λ−1 g ij Bji , which due to the equation H = g ij bji (see Lemma 38 of Appendix A), implies
(1.15).
Now, let τ := λ−2 t, and ψ λ (t) := λψ(τ ). Then
∂t ψ λ = λλ−2 (∂τ ψ)(τ ) = −λ−1 H(x(τ ))ν(ψ(τ )).
On the other hand, ν λ ≡ ν(ψ λ (τ )) = ν ≡ ν(ψ(τ )) and H λ ≡ H(ψ λ (τ )) = λ−1 H(ψ(τ )) ≡ λ−1 H (by (1.15)),
which gives
∂t ψ λ = −H λ ν λ .
(1.16)
We can also use the scaling ψ(u, t) → λψ(λ−1 u, λ−2 t), together with the fact that H → H, under
u → λ−1 u. Indeed, bij → λ−2 bij and gij → λ−2 gij if ψ(u) → ψ(λ−1 u), and therefore g ij bji → g ij bji .
For S = graph f , f (u) → λf (λ−1 u) ⇔ ψ(u) → λψ λ−1 u , where ψ(u) := (u, f (u)). Indeed,
ψ(u) := (u, f (u)) → u, λf (λ−1 u) = λ λ−1 u, f (λ−1 u) =: λψ λ−1 u .
Spherically symmetric and axi-symmetric (equivariant) solutions:
Convexity
To-do
MCF Lectures, October 26, 2014
2
9
Self-similar surfaces and rescaled MCF
2.1
Solitons - Self-similar surfaces
Consider solutions of the MCF of the form S(t) ≡ S λ(t) := λ(t)S (standing waves), or ψ(u, t) = λ(t)ϕ(u),
˙ = −λ−1 H(ϕ)ν(ϕ). Multiwhere λ(t) > 0. Plugging this into (1.1) and using H(λy) = λ−1 H(y), gives λϕ
plying this by λ and ν(ϕ), we obtain
H(ϕ) = ahν(ϕ), ϕi,
(2.1)
and λλ˙ =p−a. Since H(ϕ) is independent of t, then so should be λλ˙ = −a. Solving the last equation, we
find λ = λ20 − 2at.
i) a > 0 ⇒ λ → 0 as t → T :=
λ20
2a
⇒ S λ is a shrinker.
ii) a < 0 ⇒ λ → ∞ as t → ∞ ⇒ S λ is an expander.
iii) a = 0 ⇒ λ =const and S λ is a minimal surface.
The equation (2.1) has the solutions: a is time-independent and x is one of the following
a) Sphere x = Rˆ
x, where R =
pn
a.
b) Cylinder x = (Rxˆ0 , x00 ), where x = (x0 , x00 ) ∈ Rk+1 × Rm , where R =
q
k
a.
As stated in the following theorem, these solutions are robust.
Theorem 7 (Huisken). Let S satisfy H = ax · ν, a > 0, and H ≥ 0. We have
p
(i) If n ≥ 2, and S is compact, then S is a sphere of radius na .
(ii) If n = 2 and S is a surface of revolution, then S is the cylinder of radius
q
n−1
a .
A surface S given by an immersion ϕ will be called the self-similar surface.
Now we consider self-similar surfaces in the graph representation. Let
f (u, t) = λχ(λ−1 u),
λ depends on
˙ we find
Substituting this into (1.4) and setting y = λ−1 u and a = −λλ,
q
1 + |∇y χ|2 H(χ) = a(y∂y − 1)χ.
t.
(2.2)
(2.3)
We see that the minimal surface equation H(ϕ) = 0 is a special case of (2.1) and minimal surfaces form
a special class of self-similar surfaces.
MCF Lectures, October 26, 2014
10
Translation solitons. These are solutions of the MCF of the form S(t) ≡ S + h(t) (traveling waves),
or ψ(u, t) = ϕ(u) + h(t), where h(t) ∈ Rn+1 . Plugging this into (1.1) and using H(y + h) = H(y), gives
h˙ = −H(ϕ)ν(ϕ), or
H(ϕ) = hν(ϕ), vi,
and
h˙ = v.
(2.4)
Since H(ϕ) is independent of t, then so should be h˙ = v. (more to come)
Breathers.
2.2
MCF periodic in t.
Rescaled MCF.
First we note that the self-similar surfaces are static solutions of the rescaled MCF,
(∂τ ϕ)N = −H(ϕ) + aϕN ,
(2.5)
˙ which is obtained by rescaling the surface ψ
where (∂τ ϕ)N := h∂τ ϕ, ν(ϕ)i, ϕN := hϕ, ν(ϕ)i and a = −λλ,
and time t as
Z t
ϕ(u, τ ) := λ−1 (t)ψ(u, t), τ =
λ−2 (s)ds,
(2.6)
0
resc
and then reparametrizing the obtained surface S (τ ) := λ(t)S(t) as in the proof of Proposition 1. Indeed,
˙ −2 ψ + λ−1 ∂t ψ), (2.6), (1.1), λH(ϕ) = H(λ−1 ϕ) and ν(ψ) = ν(ϕ), we find
using ∂τ ϕ = λ2 ∂t ϕ = λ2 (−λλ
˙
∂τ ϕ = −λλϕ
− λH(ψ)ν(ψ) = aϕ − H(ϕ)ν(ϕ).
Multiplying this by ν(ϕ) gives (2.5).
˙ = −a, we obtain the parabolic scaling:
For static solutions, a =const. Then solving the equation λλ
p
1
(2.7)
λ = 2a(T − t) and τ (t) = − ln(T − t),
2a
where T := λ20 /2T , which was already discussed in connection with the scaling solitons.
Note that the equation (2.1) for self-similar surfaces is the equation for static solutions to (2.5). (This
is another analogy with minimal surfaces.)
Now, we know that minimal surfaces are critical points of the volume functional V (ψ) (by Proposition
2, the equation H(x) = 0 on S is the Euler - Lagrange equation for V (ψ)). Are self-similar surfaces
critical points of some modification of the volume functional? (Recall that, because of reparametrization
(see Proposition 1), it suffices to look only at normal variations, ψs , of the immersion ψ, i.e. generated by
vector fields η, directed along the normal ν: η = f ν.) The answer to this question is yes and is given in the
following
R
2
a
Proposition 8. Let ρ(x) = e− 2 |x| and Va (ϕ) := S λ ρ (conformal area of S λ ). For a surface S given locally
by an immersion ψ and normal variations, η = f ν, we have
Z
dVa (ϕ)η =
(H − aϕN )hν, ηi ρ.
(2.8)
Sλ
MCF Lectures, October 26, 2014
11
p
R
2
a
Proof. The definition of Va (ϕ) gives Va (ϕ) = U ρ(ϕ) g(ϕ)dn u, where ρ(ϕ) = e− 2 |ϕ| . We have dρ(ϕ)η =
√
√
−aρ(ϕ)hν, ηi. Using this formula and the equation d gη = H ghν, ηi proven above (see (1.10)) and the
fact that we are dealing with normal variations, η = f ν, we obtain (2.8).
By the definition (1.8), the definition of the normal gradient and the formula (2.8), we have
N
gradN
h Va (ϕ) = (H − aϕ )ν
R
in the Riemann metric h(ξ, η) := S λ ξ · ηρ, where ξ · η = hξ, ηi is the Euclidean inner product in Rn+1 . This
and the equation (2.5) and Proposition 8 imply
Corollary 9. Assume a in (2.5) is constant (i.e. the rescaling (2.6) is parabolic, (2.7)). Then
a) Assume a is constant. Then the modified area functional Va (ϕ) is monotonically decaying under the
rescaled flow (2.5), more precisely,
Z
∂τ Va (ϕ) = −
ρ|H − aϕN |2 .
(2.9)
Sλ
b) The renormalizedRflow (2.5) is a gradient flow for the modified area functional Va (ϕ) and the Riemann
metric h(ξ, η) := S λ ξ · ηρ:
ϕ˙ = − gradh Va (ϕ).
The relation (2.9) is the Huisken monotonicity formula (earlier results of this type were obtained by
Giga and Kohn and by Struwe).
Most interesting minimal surfaces are not just critical points of the volume functional V (ψ) but are
minimizers for it. What about self-similar surfaces? We see that
(i) inf Va (ϕ) = 0 and Va (ϕ) is minimized by any sequence shrinking to a point.
(ii) Va (ϕ) is unbounded from above. This is clear for a < 0. To see this for a > 0, we construct a sequence
of surfaces lying inside a fixed ball in Rn+1 and folding tighter and tighter.
Thus self-similar surfaces are neither minimizers nor maximizers of Va (ϕ). We conjecture that they are
saddle points satisfying min-max principle: supV inf ϕ:V (ϕ)=V Va (ϕ). One can try use this principle (say in
the form of the mountain pass lemma) to find solutions of (2.1).
Appendix: Graph representation. As an exercise, we prove (2.8) for a local patch, S λ ∩ W given by a
graph of a function f λ : U λ → R, S λ ∩ W = graph f λ which is the rescaling
Z t
f λ (v, τ ) = λ−1 f (u, t), v = λ−1 u, τ =
λ−2 (s)ds,
(2.10)
0
MCF Lectures, October 26, 2014
12
p
p
˙ λ − λy
˙ · ∇v f λ f λ +
of S =graph f . In this case, using that 1 + |∇f λ |2 λ−1 H λ = 1 + |∇f |2 H and ∂t f = λf
−2
λ
λ
λλ ∂τ f , we find that f (v, τ ) satisfies the equation
!
q
∇f λ
λ
λ
2
∂τ f = 1 + |∇f | div p
− a(v∂v − 1)f λ .
(2.11)
1 + |∇f λ |2
In the rest of the derivation we omit the superindex λ. Let x = (v, f (v)). Since |x|2 = |v|2 + |f (v)|2 , we can
write in local coordinates
Z
Z
2
2 2p
a
Va (f ) :=
ρ=
e− 2 (v +f ) 1 + |∇f |2 dn v.
ψ
We compute dVa (f )ξ =
U
ρ −af ξ + √∇f ·∇ξ
U
R
1+|∇f |2
. Integration by parts of the second term gives
"
p
∇f
− div
dVa (f )ξ =
ρ −af 1 + |∇f |2 + a(v + f ∇f ) p
1 + |∇f |2
Uλ
Z
!#
∇f
ξdn v.
p
1 + |∇f |2
The first two terms on the
from differentiating ρ. Reducing them to the common denominator
r.h.s. come
and using that H = div √ ∇f 2 , the mean curvature, we find
1+|∇f |
(v · ∇ − 1)f
dVa (f )ξ = −
ρ H − ap
1 + |∇f |2
U
Z
!
ξdn v.
(2.12)
−v·∇f +f
Now ν = √(−∇f,1) 2 and x = (v, f (v)). Hence we have ν · x = √
, which gives (2.8).
2
1+|∇f |
1+|∇f |
As before, (2.12) implies
R that (2.11) is the gradient flow for the functional Va :=
and the metric h(ξ, η) := U λ ξη √ ρ λ 2 :
a
R
Uλ
e− 2 (v
2
+f 2 )2
p
1 + |∇f |2 dn v
1+|∇f |
f˙λ = − gradh Ja (f λ ).
Indeed, he equation (2.12) implies
(y · ∇ − 1)f λ
gradh Va (f ) = − H − p
1 + |∇f λ |2
λ
!q
1 + |∇f λ |2 .
which gives the result.
2.3
Classification of mean-convex self-similar surfaces
Theorem 10 (Huisken ([16, 17]), Colding - Minicozzi (removed Huisken’s restriction on the boundedness
of |A| in the non-compact case)). The only smooth, complete, embedded self-similar surfaces in Rn+1
q, with
a > 0, polynomial growth and H(ϕ) > 0, are S k × Rn−k , where S k is the round k−sphere of radius
k
a.
MCF Lectures, October 26, 2014
13
Some steps of the proof in the compact case following [16]. Let {ej } be an orthonormal frame on S and ∇j
denote the covariant derivatives w.r.to {ej }. As usual W denotes the Weingarten map and |W | its HilbertSchmidt norm, |W |2 := Tr(W 2 ). Differentiating the self-similarity equation H(ϕ) = hν(ϕ), ϕi, where we set
a = 1, one derives the following equations
(a) ∆H = H − H|W |2 + hϕ, ej i∇j H.
(b) ∆f = 2H −4 |T |2 − vi ∇i f , where f := |W |2 /H 2 , Tijk := bij ∇k W − ∇k bij W and vi := 2H −1 ∇i H −
hϕ, ej i.
By the second equation, since S is compact, we see by the maximum principle (see Appendix B) that f
is a constant. Hence there is a number α s.t. |W |2 = α/H 2 . Moreover, T = 0. Decomposing T into the
symmetric and antisymmetric parts, T sym and T anti−sym , and using that |T |2 = |T sym |2 + |T anti−sym |2 , we
anti−sym
conclude that T anti−sym = 0. Now writing out Tijk
= 12 (bij ∇k W − bik ∇j W − (∇k bij W − ∇j bik W ))
and using Codazzi’s equality ∇k bij ∇k = ∇j bik , we find bij ∇k W − bik ∇j W = 0. (to continue)
There is a considerable literature on stable minimal surfaces. Much of it related to the Bernstein
conjecture:
The only entire minimal graphs are linear functions.
It was shown it is true for n ≤ 7:
(a) Bernstein for n = 2;
(b) De Georgi for n = 3;
(c) Almgren for n = 4; .
(d) Simons for n = 5, 6, 7.
(b) and (d) used in part results of Fleming on minimal cones. These results were extended by Schoen, Simon
and Yau.
Bombieri, De Georgi and Giusti constructed a contra example for n > 7.
3
Global Existence vs. Singularity Formation
The next two results indicate some possible scenarios for the long-time behavior of surface under mean
curvature flow. Definition: S is uniformly convex if all curvatures of S ≥ δ for some δ > 0.
Theorem 11. (Ecker-Huisken) If S0 is an entire graph over a hyperplane in Rn+1 , satisfying certain growth
conditions, then the MCF St , starting at S0 , exists ∀t and converges to a a hyperplane.
Theorem 12. (Huisken) Let S0 be a compact p
uniformly convex hypersurface. Then there is T ∈ (0, ∞) s.t.
(for a sequence tn → T ) λ−1 S → S∞ , for λ = 2n(T − t), where S∞ is a sphere.
MCF Lectures, October 26, 2014
14
In the both cases, the rescaled flow converges to a surface S∞ , which satisfies H = −ν · x in the first
case and H = ν · x in the second.
The next result establishes a general relation between a bound on the singularity (Type I collapse) and
the behaviour near collapse time. Let W be the Weingarten map of the surface and let |W |2 := T r(W 2 ) =
g ij g kl bik bjl , where recall bik are the matrix elements of W in the natural basis, i.e. (bik ) is the second
fundamental form. (Thus |W |2 is the square root of the sum of squares of principal curvatures.)
Theorem 13. (Huisken) Let S0 be a compact and satisfy supS |W |2 ≤ T c−t and n ≥ 2. Then there is
p
T ∈ (0, ∞) s.t. (for a sequence tn → T ) λ−1 S → S∞ , for λ = 2n(T − t), where S∞ satisfies H = ν · x.
Main steps of proof of Theorem 13. We conduct the proof in four steps.
Step #1: Rescaling. To understand the asymptotic behavior of the surface, we rescale it so that the rescaled
surface converges to a limit (similarly as in traveling wave and blowup problems). Consider the new surface
S λ given by (blowup variables) (2.6) Its immersion ϕ satisfies the equation
∂τ ϕ = −H(ϕ)ν(ϕ) + aϕ,
(3.1)
˙
where a = −λλ.
Step #2: Lyapunov functional. On this step one proves Huisken monotonicity formula (2.9), which we recall
here
Z
Z
∂τ
ρ=−
ρ|H − aν · ϕ|2 .
(3.2)
Sλ
Sλ
Step #3: Compactness. By the local existence theorem there T > 0, s.t. the flow exists on the interval
[0, T ). Assume that T is the maximal
p existence time. Let T < ∞. We want to show that there is a sequence
tn → T s.t. λ−1 S → S∞ , for λ = 2n(T − t). To this end, we use the following key result:
Theorem 14 (Compactness theorem: Langer, n = 3). Given constants A > 0, E > 0, p > 2, the set Ω of
immersed surfaces ψ : S → Rn satisfying
• V (ψ) =
R
S
dvol < A;
• Ep (ψ) := S Tr |W |p dA < E (p???);
R
• S ψdvol = 0 (center of gravity at 0);
R
is compact in the sense that ∀ {ψ n } ⊂ Ω, ∃ surface diffeomorphisms φn such that a subsequence of {ψ n ◦ φn }
converges in the C 1 topology to an immersion ψ in Ω. Here S is compact and without boundary.
It is shown in [14, 16] that for compact, uniformly convex hypersurfaces, S0 , there is T ∈ (0, ∞) s.t. the
Weingarten map W λ of the rescaled surface S λ satisfies
sup |W λ | ≤ c.
S
Next, we have the following estimate which follows from (2.9): for any x0 ,
Z
Z
λ
aR2 /2
aR2 /2
V (S ∩ BR (x0 )) ≤ e
ρx0 ≤ e
ρx0 |t=0 .
(3.3)
MCF Lectures, October 26, 2014
15
Then by Langer’s theorem, the family S λ (τ ) ∩ BR (x0 ), 0 ≤ τ < T, is compact (up to the closure) for each
0
R > 0 in the C 1 topology. Hence S λ (T ) := limτ 0 →T S λ (τ ) exists on each ball BR (x0 ), for a subsequence
0
0
{τ }, τ → T, so we can take S λ (T ) for the new initial condition and continue the flow beyond T . This leads
to the contradiction. Hence T = ∞.
Since the family S λ (τ ), 0 ≤ τ < ∞, is still compact by Langer’s theorem, there exists a subsequence
0
0
{τ }, τ → ∞, s.t. S∞ := limτ 0 →∞ S λ (τ ) exists on each ball BR (x0 ).
0
Step #4: Identification of S∞ . To show that if S λ (τ ) → S∞ , then S∞ satisfies
H(y) = aν · y,
(3.4)
one uses the monotonicity formula. Indeed, integrating (3.2), we obtain
Z ∞ Z
Z
Z
Z
dτ
ρ|H − aν · x|2 =
ρ|τ =0 −
ρ|τ =∞ ≤
0
This gives
R
Sλ
Sλ
Sλ
Sλ
ρ|τ =0 .
Sλ
ρ|H λ − aν λ · xλ |2 → 0 as τ → ∞. So we get (3.4).
By the Huisken classification
p theorem, stating that if n ≥ 2, H ≥ 0, and if S compact and satisfies (3.4),
then S is the sphere of radius na , we conclude that S is a sphere in the second case.
In the original (non-rescaled) variables the estimate (3.3) becomes supS |W | ≤ T c−t . A collapsing solution
1
is called of type I if |W | is bounded as |W | ≤ C(t∗ − t)− 2 . Otherwise, it is called of type II (see Huisken). It
was conjectured that the generic collapse is of type I. Indeed, all collapses investigated in the papers above
are of type I.
MCF of submanifolds. For hypersurfaces immersed in manifolds, the theory of collapse and neckpinching
are expected to go through with minimal modification. The reason for this is that they are local theories.
The volume preserving flow is affected by the geometry of ambient space in a more substantial way, see
Section ??.
4
Normal hessians.
Recall that the hessian of a functional E(ϕ) is defined as Hess E(ϕ) := d grad E(ϕ). Note that unlike the
Gˆ
ateaux derivative, d, the gradient grad and therefore the hessian, Hess, depends on the Riemann metric on
the space on which E(ϕ) is defined.
The fact that geometrical quantities are independent of the parameterization, e.g. Ha (ϕ ◦ α) = Ha (ϕ),
where Ha (ϕ) := H(ϕ) − ahϕ, ν(ϕ)i, and that the tangential variations lead to reparametrization of the
surface, have the following spectral consequence
Proposition 15. The full hessian, Hess Va (ϕ), of the modified volume functional Va (ϕ), has the eigenvalue
0 with the eigenfunctions which are tangential vector fields on S
Proof. Given a (tangential) vector field ξ on the surface S, denote by ϕs a family of deformations of ϕ
satisfying ϕ0 = ϕ and ∂s ϕs |s=0 = ξ. (Consequently, there is a family αs of diffeormorphisms of U , s.t.
MCF Lectures, October 26, 2014
16
ϕ ◦ αs = ϕs and, in particular, with α0 = 1 and ∂s ϕ ◦ αs |s=0 = ξ). Since H(ϕ) and hϕ, ν(ϕ)i are independent
of reparametrization, ϕs satisfies again the soliton equation, Ha (ϕs ) = 0. Differentiating the latter equation
w.r.to s at s = 0 and using that ∂s ϕ ◦ αs |s=0 = ξ, is a tangential vector field, we obtain
dHa (ϕ)ξ = 0,
(4.1)
which proves the proposition.
Since the tangential variations lead to reparametrization of the surface to the eigenvalue 0 of the hessian,
H, we concentrate, in what follows, on normal variations, η = f ν. (In physics terms, specifying normal
variations is called fixing the gauge.) We use the following notation for a linear operator, A, on normal
vector fields on S: AN f = A(f ν). For instance, HessN E(ϕ)f = Hess E(ϕ)(f ν) and
dN F (ϕ)f = dF (ϕ)(f ν).
(4.2)
We consider the hessian of the modified volume functional Va (ϕ), at a self-similar ϕ (i.e. H(ϕ) R= aϕN )
and in the normal direction (i.e. for normal variations, η = f ν) in the Riemann metric h(ξ, η) := S λ ξηρ.
In what follows, we call this hessian the normal hessian and denote it by HessN Va (ϕ). We want to describe
it spectrum.
Theorem 16. The hessian, HessN Va (ϕ), of the modified volume functional Va (ϕ), at a self-similar ϕ (i.e.
H(ϕ) = aϕ · ν), which is not a plane (i.e. ϕ ≡
6 0), and in the normal direction (i.e. for normal variations,
η = f ν), has
• the eigenvalue −2a with the eigenfunction hϕ, ν(ϕ)i,
• the eigenvalue −a with the eigenfunctions ν j (ϕ), j = 1, . . . , n + 1,
• the eigenvalue 0 of the multiplicity 21 n(n − 1), if ϕ 6= sphere or cylinder, and of the multiplicity n − 1,
if ϕ = cylinder, and 0 is not an eigenvalue, if ϕ = sphere; in the first case, the eigenfunctions are
hσj ϕ, ν(ϕ)i, j = 1, . . . , 21 n(n − 1), where σj are generators of the Lie algebra of SO(n + 1).
Before we proceed to the proof of this result, we mention the following important property of Va (ϕ): the
equation H(ϕ) − ahϕ, ν(ϕ)i = 0 breaks the scaling and translational symmetry. Indeed, using the relations
H(λϕ) = λ−1 H(ϕ), ν(λϕ) = ν(ϕ), ∀λ ∈ R+ ,
H(ϕ + h) = H(ϕ), ν(ϕ + h) = ν(ϕ), ∀h ∈ R
(4.3)
n+1
H(gϕ) = H(ϕ), ν(gϕ) = gν(ϕ), ∀g ∈ O(n + 1),
,
(4.4)
(4.5)
using that g ∈ O(n + 1) are isometries in Rn+1 and using the notation Ha (ϕ) := H(ϕ) − ahϕ, ν(ϕ)i, we obtain
Hλ−2 a (λϕ) = λ−1 Ha (ϕ), ∀λ ∈ R+ ,
Ha (ϕ + h) + ahh, ν(ϕ)i = Ha (ϕ), ∀h ∈ R
Ha (gϕ) = Ha (ϕ), ∀g ∈ O(n + 1).
(4.6)
n+1
,
(4.7)
(4.8)
MCF Lectures, October 26, 2014
17
Proof of Theorem 16. If an immersion ϕ satisfies the soliton equation H(ϕ) = ahϕ, ν(ϕ)i, then by (4.6), we
have Hλ−2 a (λϕ) = 0 for any λ > 0. Differentiating this equation w.r.to λ at λ = 1, we obtain dHa (ϕ)ϕ =
−2ahϕ, ν(ϕ)i.
Now, choosing ξ to be equal to the tangential projection, ϕT , of ϕ, and subtracting the equation
(4.1) from the last equation, we find dHa (ϕ)hϕ, ν(ϕ)iν(ϕ) = −ahϕ, ν(ϕ)i. Since by the definition (4.2),
dHa (ϕ)f ν(ϕ) = dN Ha (ϕ)f and Hess Va (ϕ) = dHa (ϕ), this proves the first statement.
To prove the second statement, we observe that the soliton equation implies, by (4.7), that Ha (ϕ + sh) +
ashh, ν(ϕ)i = 0 and any constant vector field h. Differentiating this equation w.r.to s at s = 0, we obtain
dHa (ϕ)h = −ahh, ν(ϕ)i. Now, choosing ξ to be equal to the tangential projection, hT , of h, and subtracting
the equation (4.1) from the last equation, we find dHa (ϕ)hh, ν(ϕ)iν(ϕ) = −ahh, ν(ϕ)i, which together with
(4.2) gives the second statement.
Finally, to prove the third statement, we differentiate the equation Ha (g(s)ϕ) = 0, where g(s) is a oneparameter subgroup of O(n + 1), w.r.to s at s = 0, to obtain dHa (ϕ)σϕ = 0, where σ denotes the generator
of g(s). Now, choosing ξ in (4.1) to be equal to the tangential projection, (σϕ)T , of σϕ, and subtracting the
equation (4.1) from the last equation, we find dHa (ϕ)hσϕ, ν(ϕ)iν(ϕ) = 0, which together with (4.2) gives the
third statement. Finally, an explicit and elementary computation performed below shows that the normal
hessian, HessN
sph Va (ϕ) on a sphere has no eigenvalue 0.
In what follows a > 0.
Corollary 17. The multiplicity of the eigenvalue 0 (zero multiplicity meaning that 0 is not an eigenvalue)
differentiates between spheres, cylinders and other surfaces. In particular, a self-similar surface, ϕ, with
a > 0 satisfying HessN Va (ϕ) > 0 on the subspace span{hϕ, ν(ϕ)i, ν j (ϕ), j = 1, . . . , n + 1}⊥ , is a sphere.
Corollary 18. There are no smooth, embedded self-similar (a > 0) surfaces in Rn+1 , satisfying HessN Va (ϕ) ≥
0 on the subspace span{hϕ,q
ν(ϕ)i, ν j (ϕ), j = 1, . . . , n + 1}⊥ , which are close to S k × Rn−k , where S k is the
round k−sphere of radius
k
a
(for k = 1, . . . , n).
Theorem 16 implies that if ϕ is not spherically symmetric, then 0 is an eigenvalue of HessN Va (ϕ) of
multiplicity at least n + 1. This gives the second part of the first corollary, while the second one follows
from the fact that if there is a sequence of non-spherical self-similar surfaces approaching a sphere, then by
Theorem 16 and a standard spectral result, the normal hessian on the sphere would have the eigenvalue 0
which contradicts the third statement.
Remark 1. a) Strictly speaking, if the self-similar surface is not compact, then hϕ, ν(ϕ)i and ν j (ϕ), j =
1, . . . , n + 1, generalized eigenfunctions of HessN Va (ϕ). In the second case, the Schnol-Simon theorem (see
Appendix D.1 or [13]) implies that the points −2a and −a belong to the essential spectrum of HessN Va (ϕ).
pa
b) We show below that the normal hessian, HessN
sph Va (ϕ), on the sphere of the radius
n , given in
2a
(4.22), has no other eigenvalues below n , besides −2a and −a. A similar statement, but with n replaced by
n − 1, we have for the cylinder.
We call the eigenfunction hϕ, ν(ϕ)i, ν j (ϕ) and hσj ϕ, ν(ϕ)i, j = 1, . . . , n+1, the scaling, translational and
rotational modes. They originate from the normal projections of scaling, translation and rotation variations.
MCF Lectures, October 26, 2014
4.1
18
Spectra of hessians for spheres and cylinders
It is easy to show (for a systematic derivation see Subsection 4.5) that the normal hessians on the n−sphere
n−k
n
n
n
n
k
and (n,
q k)−cylinder, SR and the standard n−sphere S = S1 and the n−cylinder CR = SR × R of radius
R=
n−k
a
in Rn+1 , , are given by
a
HessN
sph Va (ϕ) = − ∆Sn − 2a,
n
(4.9)
on L2 (Sn ), where ∆Sk is the Laplace-Beltrami operator on the standard n−sphere Sk , and
HessN
cyl Va (ϕ) = −∆y − ay · ∇y −
a
∆ n−k − 2a,
n−k S
(4.10)
acting on L2 (C n ), where the standard round cylinder, C n = Sn−k × Rk , is naturally embedded in Rn+1 as
2
C n = {(ω, y) : |ω| = 1} ⊂ Rn+1 .
pa
To describe the spectra of the normal hessians on the n−sphere and (n, k)−cylinder, of the radii
n
q
a
and n−1
, we observe that it is a standard fact that the operator −∆ = −∆Sn is a self-adjoint operator on
L2 (Sn ). Its spectrum is well known (see [30]): it consists of the eigenvalues l(l + n − 1), l = 0, 1, . . . , of the
multiplicities
n+l
n+l−2
m` =
−
.
n
n
Moreover, the eigenfunctions corresponding to the eigenvalue l(l + n − 1) are the restrictions to the sphere
Sn of harmonic polynomials on Rn+1 of degree l and denoted by Ylm (the spherical harmonics),
− ∆Ylm = l(l + n − 1)Ylm ,
l = 0, 1, 2, 3, . . . , m = 1, 2, . . . , ml .
(4.11)
In particular, the first eigenvalue 0 has the only eigenfunction 1 and the second eigenvalue n has the eigenfunctions ω 1 , · · · , ω n+1 .
a
:= HessN
Consequently, the operator Lsph
a
sph Va (ϕ) = − n (∆Sn + 2n) is self-adjoint and its spectrum
a
consists of the eigenvalues n (l(l + n − 1) − 2n) = a(l − 2) + na l(l − 1), l = 0, 1, . . . , of the multiplicities
m` . In particular, the first n + 2 eigenvectors of Lsph
(those with l = 0, 1) correspond to the non-positive
a
eigenvalues,
0
0
j
j
Lsph
Lsph
(4.12)
a ω = −2aω ,
a ω = −aω , j = 1, . . . , n + 1,
and are due to the scaling (l = 0) and translation (l = 1) symmetries.
We proceed to the cylindrical hessian Lcyl
:= HessN
a
cyl Va (ϕ), given in (4.24). The variables in this
a
operator separate and we can analyze the operators −∆y − ay · ∇ and − n−k
(∆Sn−k + 2(n − k)) separately.
a
The operator − n−k (∆Sn−k + 2n) was already analyzed above. The operator −∆y − ay · ∇ is the Ornstein Uhlenbeck generator, which can be unitarily mapped by the gauge transformation
a
v(y, w) → v(y, w)e− 4 |y|
2
into the the harmonic oscillator Hamiltonian Hharm := −∆y + 14 a2 |y|2 − ka. Hence the linear operator
2
a
−∆y − ay · ∇ is self-adjoint on the Hilbert space L2 (R, e− 2 |y| dy). Since, as was already mentioned, the
MCF Lectures, October 26, 2014
19
a
operator − n−k
(∆Sn−k + 2(n − k)) is self-adjoint on the Hilbert space L2 (C n ), we conclude that the linear
a
2
operator Lcyl
on the Hilbert space
L2 (Rk × Sn−k , e− 2 |y| dydw). Moreover, the spectrum of
a is self-adjoint
n P
o
k
−∆y − ay · ∇ is a 1 si : si = 0, 1, 2, 3, . . . , with the normalized eigenvectors denoted by φs,a (y), s =
(s1 , . . . , sk ),
(−∆y − ay · ∇)φs,a = a
k
X
si φs,a ,
si = 0, 1, 2, 3, . . . .
(4.13)
1
a
a
(∆Sn−k +2(n−k)) is n−k
(l(l+n−k−1)−2(n−k)) =
Using that we have shown that the spectrum of − n−k
P
k
a
1 si , si = 0, 1, 2, 3, . . ., we conclude the spectrum
n−k l(l + n − k − 1) − 2a, l = 0, 1, . . . , and denoting |s| =
of the linear operator Lcyl
,
for
k
=
1,
is
a
a
spec(Lcyl
)
=
`(`
+
n
−
2)
:
|s|
=
0,
1,
2,
3,
.
.
.
;
`
=
0,
1,
2,
.
.
.
(|s|
−
2)a
+
,
(4.14)
a
n−1
with the normalized eigenvectors given by φs,l,m,a (y, w) := φs,a (y)Ylm (w). This equation shows that the
non-positive eigenvalues of the operator Lcyl
a , for k = 1 (so that |s| = s), are
1
a 4
) ((s, l) = (0, 0)), due
• the eigenvalue −2a of the multiplicity 1 with the eigenfunction φ0,0,0,a (y) = ( 2π
to scaling of the transverse sphere;
1
a 4 m
• the eigenvalue −a of the multiplicity n with the eigenfunctions φ0,1,m,a (y) = ( 2π
) w , m = 1, . . . , n
((s, l) = (1, 1)), due to transverse translations;
a 14 √
• the eigenvalue 0 of the multiplicity n with the eigenfunctions φ1,1,m,a (y) = ( 2π
) aywm , m = 1, . . . , n
((s, l) = (0, 1)), due to rotation of the cylinder;
a 41 √
• the eigenvalue −a of the multiplicity 1 with the eigenfunction φ1,0,0,a (y) = ( 2π
) ay ((s, l) = (1, 0));
1
a 4
• the eigenvalue 0 of the multiplicity 1 with the eigenfunction φ2,0,0,a (y) = ( 2π
) (1−ay 2 ) ((s, l) = (2, 0)).
The last two eigenvalues are not of the broken symmetry origin and are not covered by Theorem 16. They
indicate instability of the cylindrical collapse.
4.2
Connection of the spectrum and geometry of self-similar surfaces
For a 6= 0, the soliton equation, hϕ, ν(ϕ)i = a−1 H(ϕ), and Theorem 16 imply that the mean curvature H is
an eigenfunction of HessN Va (ϕ) with the eigenvalue −2a. This fact has important consequences, which are
used in the next result in order to connect the spectrum to the geometry of S.
Theorem 19. (Colding - Minicozzi) For a self-similar surface with a > 0, HessN Va (ϕ) ≥ −2a iff either
H(x) > 0 ∀x ∈ S, or H(x) < 0 ∀x ∈ S.
To prove this theorem we will use the Perron-Frobenius theory (see Appendix D.2) and its extension as
given in [27]. We begin with
MCF Lectures, October 26, 2014
20
Definition 1. We say that a linear operator on L2 (S) has a positivity improving property iff either A, or
e−A , or (A + µ)−1 , for some µ ∈ R, takes non-negative functions into positive ones.
Proposition 20. The normal hessian, HessN Va (ϕ), has a positivity improving property.
Proof. By standard elliptic/parabolic theory, e∆ and (−∆ + µ)−1 , for any µ > 0, has strictly positive integra
kernel and therefore is positivity improving. To lift this result to HessN Va (ϕ) := −∆ − |W |2 − a1 + ϕ · ∇
we proceed exactly as in [27].
Since, the operator HessN Va (ϕ) is bounded from below and has a positivity improving property, it
satisfies the assumptions of the Perron-Frobenius theory (see Appendix D.2) and its extension in [27] to the
case when the positive solution in question is not an eigenfunction. The latter theory, Proposition 1 and
Remark 1 imply the statement of Theorem 19.
Corollary 21. Let ϕ be a self-similar surface. We have
(a) For a < 0 (ϕ is an expander), H(ϕ) changes the sign.
(b) For a = 0, inf HessN Va (ϕ) < 0.
(c) For a > 0, if ϕ is an entire graph over Rn and inf HessN Va (ϕ) ≥ −2a, then ϕ is a hyperplane.
Indeed, (a) follows directly from Theorems 16 and 19, while (b) follows from the fact that for a = 0, 0 is
an eigenvalue of the multiplicity ≥ n + 2 and therefore, by the Perron-Frobenius theory, it is not the lowest
eigenvalue of HessN Va (ϕ). (c) is based on the fact that entire graphs over Rn cannot have strictly positive
mean curvature. (check, references)
Theorems 19 and 10 imply
Corollary 22 (Colding - Minicozzi). The only smooth, complete, embedded self-similar (a > 0) surfaces
in Rn+1 of polynomial
growth, satisfying HessN Va (ϕ) ≥ −2a, are S k × Rn−k , where S k is the round
q
k−sphere of radius
4.3
k
a.
Linearized stability of self-similar surfaces
Given a self-similar surface ϕ, we consider for example the manifold of surfaces obtained from ϕ by symmetry
transformations,
Mself−sim := {λgϕ + z : (λ, z, g) ∈ R+ × Rn+1 × SO(n + 1)}.
By the spectral theorem above, it has unstable and central manifolds corresponding to the eigenvalues
−2a, −a and 0. Hence, we can expect only the dynamical stability in the transverse direction.
Definition 2 (Linearized stability of self-similar surfaces). We say that a self-similar surface φ, with a > 0, is
linearly stable (for the lack of a better term) iff HessN Va (ϕ) > 0 on the subspace span{hϕ, ν(ϕ)i, ν i (ϕ), i =
⊥
⊥
1, . . . , n+1, hσj ϕ, ν(ϕ)i, j = 1, . . . , 21 n(n−1)} (i.e. on span{scaling, translational, rotational modes} ).
MCF Lectures, October 26, 2014
21
(I.e. the only unstable motions allowed are scaling, translations and rotations.)
A weaker notion of stability was introduced in [3]:
Definition 3 (F −stability of self-similar surfaces of [3]). We say that a self-similar surface φ, with a > 0,
is F −stable iff the normal hessian satisfies HessN Va (ϕ) ≥ 0 on the subspace span{hϕ, ν(ϕ)i, ν j (ϕ), j =
1, . . . , n + 1}⊥ .
Remark 2. 1) The F −stability, at least in the compact case, says that the HessN Va (ϕ) has the smallest
possible negative subspace, i.e ϕ has the smallest possible Morse index.
2) The reason the F −stability works in the non-compact case is that, due to the separation of variables
for the cylinder = (compact surface) ×Rk , the orthogonality to the negative eigenfunctions of the compact
factor removes the entire branches of the essential spectrum. This might not work for warped cylinders.
3) For minimal surfaces (strict) stability implies the the surface is a (strict) minimizer of the volume
functional V (ψ). As is already suggested by the discussion above, this is not so for self-similar surfaces, they
are saddle points possibly satisfying some min-max principle.
4) Remark 1(c) shows that the spheres and cylinders are F −stable. However, we show in Section 6 that
cylinders are dynamically unstable.
N
cyl
:= HessN
By the description of the spectra of the normal hessians of Lsph
sph Va (ϕ) and La := Hesscyl Va (ϕ)
a
in the previous subsection, we conclude that the spherical collapse is linearly stable while the cylindrical one
is not.
We will show in Sections 5 and 6 that indeed the spherical collapse is (nonlinearly) stable while the
cylindrical one is not. We will also show that the last two eigenvalues of Lcyl
a in the list above are due to
translations of the point of the neckpinch on the axis of the cylinder and due to shape instability, respectively.
4.4
Explicit expression for the normal hessians
Now, we compute explicit expression for the normal hessian, HessN Va (ϕ).
Theorem 23. The normal hessian, HessN Va (ϕ) at a self-similar ϕ (i.e. H(ϕ) = aϕ · ν) is given by
HessN Va (ϕ) = −∆ − |W |2 − a(1 − ϕ · ∇).
(4.15)
Proof. Recall that by the definition
R (1.8) and the formula (2.8), we have gradh Va (ϕ) = (H − aϕ · ν)ν
in the Riemann metric h(ξ, η) := S λ ξηρ. For normal variations, η = f ν, this gives gradh Vanorm (ϕ) =
(H(ϕ) − aϕ · ν). We have to compute d gradh Vanorm (ϕ) = d(H − aϕ · ν)ν w.r.to normal variations. We claim
that
dN Hf = (−∆ − |W |2 )f,
N
d ϕ · νf = f − ϕ · ∇f.
We begin with some simple calculations. We have
(4.16)
(4.17)
MCF Lectures, October 26, 2014
22
Lemma 24 (Simons, Schoen-Simon-Yau, Hamilton, Huisken, Sesum). We have the following identities
(a) dν(ϕ)η = −g ij hν, ∂i ηi∂uj ϕ = −hν, ∇ηi.
(b) dbij (ϕ)η = −hν, (∂i ∂j + Γkij ∂k )ηi.
(c) dg ij η = g im g jn dgmn η, dgmn η = h∂m ϕ, ∂n ηi + h∂n ϕ, ∂m ηi.
Proof. Denote ϕj := ∂uj ϕ and ηj := ∂uj η. For (a), use that hν(ϕs ), ϕsi i = 0 to obtain h∂s ν(ϕs ), ϕsi i =
−hν(ϕs ), ∂s ϕsi i, which implies hdν(ϕ)η, ϕi i = −hν(ϕ), ηi i and therefore dν(ϕ)η = g ij hdν(ϕ)η, ϕi iϕj =
−g ij hν(ϕ), ηi iϕj , which gives the desired statement.
2
ϕ
To prove (b), we use bij = −h ∂u∂i ∂u
, νi (see Lemma 38 of Appendix A.2), relation (a) and the equation
j
2
2
2
2
ϕ
ϕ
dϕη = η, to find dbij (ϕ)η = −h ∂u∂i ∂uj dϕη, νi−h ∂u∂i ∂u
, dνηi = −h ∂u∂i ∂uj η, νi+g mn hν, ηm ih ∂u∂i ∂u
, ϕn i. Next,
j
j
using the Gauss formula
∂2ϕ
= bij ν + Γkij ϕk
(4.18)
∂ui ∂uj
2
(see e.g. [25]), hν, ϕn i = 0 and hϕk , ϕn i = gkn gives dbij (ϕ)η = −h ∂u∂i ∂uj η, νi − g mn gkn Γkij hν, ∂m ηi, which
implies the statement (b).
The first statement in (c) follows by differentiating the relation gg −1 = 1, where g = (gij ), which gives
d(g ) = g −1 dgg −1 , and the second, from the definition gij = hϕi , ϕj i and the relation dϕη = η.
−1
Using the definition H = g ij bij , we find dHη = bij dg ij + g ij dbij . We start with the second term. Notice
that the definition ∆f = div grad f = ∇i ∇i f and relations to
div V = ∇i V i =
∂V i
i
i
ij ∂f
+ Γm
,
mi V , (grad(f )) = g
∂ui
∂uj
(4.19)
∂
m
ij ∂f
ij
k
imply ∆f = ( ∂u
i + Γmi )g
∂uj . Now, take η = f ν. Then the last relation gives g (∂i ∂j + Γij ∂k )η =
k
ν∆f + f (∂i ∂j + Γij ∂k )ν + 2∂i f ∂i ν. Using this relation, the statement (c) of Lemma 24 and the relation
hν, ∂j νi = 0, we obtain
dbij η = hν, (∂i ∂j + Γkij ∂k )(f ν)i = ∆f + f hν, ∂i ∂j νi.
Furthermore, using hν, ∂j νi = 0, again, and using hξ, ηi = g mn hξ, ϕm ihϕn , ηi, we find hν, ∂i ∂j νi = −h∂i ν, ∂j νi =
−g mn h∂i ν, ϕm ihϕn , ∂j νi. Next, the relation h∂k ν, ϕl i = −hν, ∂k ϕl i = bkl gives hν, ∂i ∂j νi = −g mn bim bjn .
Collecting the above equalities, we arrive at
dbij (f ν) = ∆f − g mn bim bjn f,
(4.20)
and therefore g ij dbij (f ν) = ∆f − g ij g mn bim bjn f = ∆f − |W |2 f .
Next, for the first term in dH = bij dg ij + g ij dbij , we have, by Lemma 24 (c), dg ij = g im g jn dgmn . Now,
using Lemma 24 (c) again and using the relation hν, ∂j νi = 0, we find, for η = f ν,
dgmn (f ν) = hϕm , ∂n (f ν)i + hϕn , ∂m (f ν)i = f [hϕm , ∂n νi + hϕn , ∂m νi] = 2f bmn .
MCF Lectures, October 26, 2014
23
Combining the last two relations gives bij dg ij (f ν) = 2f g im g jn bmn bij = 2|W |2 f . Collecting the relations
above gives (4.16).
Finally, using Lemma 24 and the equation dϕη = η, we find
dhϕ, νiη = hη, νi − g ij hν, ∂i ηihϕ, ϕj i.
(4.21)
Now, taking η = f ν and using ∂i (f ν) = ∂i f ν + f ∂i ν, hν, νi = 1 and hν, ∂j νi = 0, we arrive at (4.17).
4.5
Hessians for spheres and cylinders
We finish this section with illustrating the above result by computing the normal hessians on the n−sphere
and (n, k)−cylinder.
pn
n
n+1
of radius R =
Explicit expressions. 1) For the n−sphere SR
, we have bij = R−1 gij =
a in R
stand
stand
n
and the standard n−sphere Sn = S1n , respectively, which
Rgij
, where gij and gij
are the metrics on SR
2
−2
gives |W | = nR = a. Moreover, since ν(ϕ) = ϕ, we have that ϕ · ∇ = 0 and therefore
a
HessN
sph Va (ϕ) = − ∆Sn − 2a,
n
(4.22)
on L2 (Sn ), where ∆Sk is the Laplace-Beltrami operator on the standard n−sphere Sk .
q
n−k
n
n+1
stand
= SR
× Rk of radius R = n−k
2) For the n−cylinder CR
, we have bij = R−1 gij = Rgij
,
a in R
n−1
stand
for i, j = 1, . . . , n − k, and bαβ = δαβ , α, β = n − k + 1, . . . , n. Here gij and gij
are the metrics on SR
and Sn−1 = S1n−1 , respectively. This gives
|W |2 = (n − k)R−2 = a.
(4.23)
Moreover, letting the standard round cylinder C n = S n−k × Rk be naturally embedded as C n = {(ω, x) :
2
|ω| = 1} ⊂ Rn+1 , we see that the immersion ϕ is given by ϕ(ω, y) = (χ(ω, y), y) for some χ(ω, y), where
y = y(x) := λ−1 (t)x. This gives ϕ · ∇ = ω · ∇Sn−k + y · ∇y = y · ∇y , which together with the previous relation,
implies
HessN
cyl Va (ϕ) = −∆y − ay · ∇y −
a
∆ n−k − 2a,
n−k S
(4.24)
acting on L2 (C n ).
4.6
Minimal and self-similar submanifolds
For a self-similar hypersurface S in a manifold M , with immersion ϕ (satisfying H(ϕ) = aϕ · ν), the normal
hessian, HessN Va (ϕ) is given by
HessN Va (ϕ) = −∆ − |W |2 − a(1 − ϕ · ∇) − RicM (ν, ν),
where RicM (ν, ν) is the Ricci curvature of M .
(4.25)
MCF Lectures, October 26, 2014
5
24
Stability of elementary solutions I. Spherical collapse
The next two results indicate some possible scenarios for the long-time behavior of surface under mean
curvature flow.
Theorem 25.
• (Ecker-Huisken) The hyperplanes are asymptotically stable.
• (Kong-Sigal) The spherical collapse is asymptotically stable.
• (Gang - Sigal) The cylindrical collapse is unstable.
We prove the second statement which we formulate precisely. Recall that we are considering the mean
curvature flow, starting with a hypersurface S0 in Rn+1 , which is given by an immersion, ψ0 : Sˆ → Rn+1 ),
of some fixed n−dimensional hypersurface Sˆ ⊂ Rn+1 (i.e. ψ(·, t) : Sˆ → Rn+1 ). We look for solutions of the
mean curvature flow,
∂ψ
= −H(ψ)ν(ψ)
∂t
(5.1)
ψ|t=0 = ψ0 ,
as immersions, ψ(·, t) : Sˆ → S, of Sˆ giving the hypersurfaces S(t). (Recall that H(x) and ν(x) are mean
curvature and the outward unit normal vector, at x ∈ S(t), respectively.)
Theorem 26. Let a surface S0 , defined by an immersion x0 ∈ H s (Sn ), for some s > n2 + 1, be close to
Sn , in the sense that kψ0 − 1kH s 1. Then there exist t∗ < ∞ and z∗ ∈ Rn+1 , s.t. (5.1) has the unique
solution, S(t), t < t∗ , and this solution contracts to the point z∗ , as t∗ → ∞. Moreover, S(t) is defined, up
to a (time-dependent) reparametrization, by an immersion ψ(·, t) ∈ H s (Sn ), with the same s, of the form
ψ(ω, t) = z(t) + λ(t)ρ(ω, t)ω,
(5.2)
where ρ, λ(t), z(t) and a(t) satisfy
with s := t∗ − t, κ1 :=
1
2
+
1
2a∗ (1
−
√
2a∗ s + O(sκ1 ),
˙
a(t) = −λ(t)λ(t)
= a∗ + O(sκ2 ),
(5.3)
z(t) = z∗ + O(sκ3 )
r
1
n
kρ(·, t) −
kH s . s 2n ,
a(t)
(5.5)
λ(t) =
1
2n ),
κ2 :=
1
2a∗ (1
−
1
2n )
and κ3 :=
(5.4)
(5.6)
1
2a∗ (n
+
1
2
−
1
2n ).
Moreover, |z∗ | 1.
Note that our condition on the initial surface does not impose any restrictions of the mean curvature of
this surface.
The form of expression (5.2) above is a reflection of a large class of symmetries of the mean curvature
flow:
• (5.1) is invariant under rigid motions of the surface, i.e. x 7→ Rx + a, where R ∈ O(n + 1), a ∈ Rn+1
and x = x(u, t) is a parametrization of S(t), is a symmetry of (5.1).
• (5.1) is invariant under the scaling x 7→ λx and t 7→ λ−2 t for any λ > 0.
MCF Lectures, October 26, 2014
25
We utilize these symmetries in an essential way by defining the centres, z(t), of the closed surfaces S(t) and
using the rescaling of the equation (5.1) by a parameter λ(t). This leads to a transformed MCF depending
explicitly on these parameters and determining their behaviour. Then we define Lyapunov-type functionals
and derive a series of differential inequalities for them from which we obtain our main results. We omit some
technical details for which we refer to [24, 1].
Notation. The relation f . g for positive functions f and g signifies that there is a numerical constant
C, s.t. f ≤ Cg. In this section we denote the inner product in Rn+1 by the dot: x · y.
5.1
Rescaled equation
˜ ) = λ−1 (t)(S(t) − z(t)),
Instead of the surface S(t), it is convenient to consider the new, rescaled surface S(τ
Rt
where λ(t) and z(t) are some differentiable functions to be determined later, and τ = 0 λ−2 (s)ds. The new
surface is described by ϕ, which is, say, an immersion of some fixed n−dimensional hypersurface Ω ⊂ Rn+1 ,
˜ ), i.e. ϕ(·, τ ) : U → S(t)). Thus the new collapse
i.e. ϕ(·, τ ) : Ω → Rn+1 , (or a local parametrization of S(τ
variables are given by
Z
t
ϕ(ω, τ ) = λ−1 (t)(ψ(ω, t) − z(t)) and τ =
λ−2 (s)ds.
(5.7)
0
R t −2
∂z
(s)ds.
Let λ˙ = ∂λ
∂t and ∂τ be the τ -derivative of z(t(τ )), where t(τ ) is the inverse function of τ (t) = 0 λ
∂ψ
∂ϕ ∂τ
∂z
−2 ∂z
−1 ∂ϕ
−1
˙
˙
Using that ∂t = ∂t + λϕ + λ ∂τ ∂t = λ ∂τ + λϕ + λ ∂τ and H(λϕ) = λ H(ϕ), we obtain from (5.1)
the equation for ϕ, λ and z:
∂z
∂ϕ
˙
· ν(ϕ) = −H(ϕ) + (aϕ − λ−1 ) · ν(ϕ) and a = −λλ.
∂τ
∂τ
(5.8)
Thus S(t) is given by the datum (ϕ, λ, z), satisfying
p Eq. (5.8). Note that the equation (5.8) has static
solutions (a = a positive constant, z = 0, ϕ(ω) = na ω, ω ∈ Sn ).
5.2
The graph represetation
In the next proposition ∆ is the Laplace-Beltrami operator in is the standard metric on Sn , ∇i ρ =
2
∂ ρ
∂ui uj
∂ρ
− Γkij ∂u
k,
Γkij
1 kn ∂gin
2 g ( ∂uj
∂gjn
∂ui
∂ρ
∂ui
(in a
∂gij
∂un ).
n
local parametrization x = x(u)) and (Hess ρ)ij =
where
=
+
−
Here
and in what follows the summation over the repeated indices is assumed. (Hess is the hessian on S and, if ∇
∂v
the Riemann connection on Sn , which acts on vector fields as (∇i v)j = ∂uji − Γkij vk , then (Hess)ij = ∇i ∇j .)
For more details see Appendix A. In this section we present the following
˜ ) = λ−1 (t)(S(t) − z(t)), for 0 ≤ t < T , for some T > 0, and τ , given in (5.7), and
Proposition 27. Let S(τ
˜ ) is defined by an immersion
for λ(t) and z(t) differentiable functions on [0, T ), and assume S(τ
ϕ(ω, τ ) = ρ(ω, τ )ω
(5.9)
˜ ) satisfies (5.8) if
of Sn for some functions ρ(·, τ ) : Sn → R+ , differentiable in their arguments. Then S(τ
and only if ρ and z satisfy the equation
∂ρ
= G(ρ) + aρ − λ−1 zτ · ω + λ−1 z˜τ · ρ−1 ∇ρ,
∂τ
(5.10)
MCF Lectures, October 26, 2014
where z˜τ k =
∂xi i
z ,
∂uk τ
zτ :=
∂z
∂τ
26
and (check the sign of the last term on the r.h.s.)
G(ρ) =
1
n ∇ρ · Hess(ρ)∇ρ − ρ|∇ρ|2
∆ρ
−
−
.
ρ2
ρ
ρ2 (ρ2 + |∇ρ|2 )
(5.11)
Proof. By a reparametrization (see Proposition 1), the equation (5.8) is equivalent to the equation
∂ϕ
∂z
· ν(ϕ) = −H(ϕ) + (aϕ − λ−1 ) · ν(ϕ).
∂τ
∂τ
(5.12)
∂ρ
−1/2
For the graph ϕ(ω, τ ) = ρ(ω, τ )ω, we have ∂ϕ
ρ and
∂τ = ∂τ ω. By results of Appendix A.8, we have ν ·ω = p
−1/2 2
2
2
hν, ϕi = p
ρ , where p := ρ + |∇ρ| . These relations, together with the equation (5.12), the expression
∂z
· ν(ϕ) = zτ · ω − +˜
zτ · ρ−1 ∇ρ, give the
(A.34) for the mean curvature H(ϕ) and a similar computation ∂τ
equation (5.10).
We consider (5.10) as an equation for the unknowns ρ(t), z(t) and a(τ ), with λ(t) determined from the
˙
equation λ(t)λ(t)
= −a(τ (t)) (see (5.8)). Additional equations come from imposing the subsidiary conditions
r
n
n
hρ(·, τ )iS =
,
(5.13)
a(τ )
R
where by hf iSn we denote the average, hf iSn := |S1n | Sn f , of f over Sn , fixing the average radius of the
rescaled surface, and
Z
ρ(ω, τ )ω j = 0, j = 1, . . . , n + 1,
(5.14)
Sn
fixing the center ofR the rescaled surface. To see the latter we observe that the latter conditions are equivalent
to the conditions Sn ((ψ − z) · ω)ω j = 0, j = 1, . . . , n + 1, which can be said to determine the center of the
surface.
5.3
Reparametrization of solutions
Our goal is to decompose a solution, ρ(ω, τ ), of the equation (5.10) into a leading part which is a sphere of
some radius τ -dependent and small fluctuation around this part. This would give a convenient reparametrization of our solution.
We decompose ρ(ω, τ ) into the sphere, which (a) is closest to it and (b) is a stationary solution to (5.10),
and a remainder. By (a)
q the remainder (fluctuation) is orthogonal to this sphere and by (b), the radius of
n
the sphere is equal to a(τ
) , where a(τ ) is the parameter-function appearing in (5.10). In other words, we
write
r
ρ(·, τ ) =
n
+ φ(ω, τ ),
a(τ )
with φ ⊥ 1 in L2 (Sn ).
(5.15)
Introduce the notation ω 0 = 1. Then the relations (5.15), (5.13) and (5.14) give
φ(·, τ ) ⊥ ω j , j = 0, . . . , n + 1, in L2 (Sn ).
(5.16)
MCF Lectures, October 26, 2014
5.4
27
Lyapunov-Schmidt decomposition
(changed φ to ξ) Let ρ solve (5.10) and assume it can be written as ρ(ω, τ ) = ρa(τ ) +ξ(ω, τ ), with ρa =
and ξ ⊥ ω j , j = 0, . . . , n + 1. Plugging this into equation (5.10), we obtain the equation
∂ξ
= −La ξ + Nλ,z (ξ) + F,
∂τ
pn
a
(5.17)
where La = −dGa (ρa ), with Ga (ρ) := G(ρ) + aρ, ρ = ρa + ξ, N (ξ) = G(ρa + ξ) − G(ρa ) − dG(ρa )ξ,
∂xi i
∂a
∂z
and z˜τ k = ∂u
F = −∂τ ρa − λ−1 zτ · ω and, recall, zτ = ∂τ
k zτ . Let aτ = ∂τ . We compute
√
a
n −3/2
La = (−∆ − 2n), F =
a
aτ − λ−1 zτ · ω,
(5.18)
n
2
Nλ,z (ξ) := N (ξ) + λ−1 z˜τ · ρ−1 ∇ξ,
(5.19)
N (ξ) = −
nξ 2
∇ξ · Hess(ξ)∇ξ − ρ|∇ξ|2
(ρa + ρ)ξ∆ξ
−
−
.
ρ2 ρ2a
ρρ2a
ρ2 (ρ2 + |∇ξ|2 )
(5.20)
Now, we project (5.17) onto span{ω j , j = 0, . . . , n + 1}. In the rest of the section, all the norms and
inner products are in the sense of L2 (Sn ). By ξ⊥ω j , j = 0, . . . , n + 1, we have
√
n −3/2
a
aτ |Sn | = hNλ,z (ξ), 1i ,
(5.21)
2
− cλ−1 zτj = Nλ,z (ξ), ω j ,
(5.22)
R
1
n
j 2
where j = 1, . . . , n + 1, and c := Sn (ω ) = n+1 |S |. Indeed, this equation follows from
(i) ∂τ ξ, ω j = − ξ, ∂τ ω j = 0, La ξ, ω j = ξ, La ω j = 0, j = 0, . . . , n + 1;
D√
E √
(ii) hF, 1i = 2n a−3/2 aτ , 1 = 2n a−3/2 aτ |Sn |; F, ω j = −λ−1 zτ · ω, ω j = −cλ−1 zτj ,
j = 1, . . . , n + 1.
Projecting (5.17) onto the orthogonal complement of span{ω j , j = 0, . . . , n + 1}, we find
∂ξ
⊥
= −La ξ + Nλ,z
(ξ),
∂τ
(5.23)
⊥
˙
where Nλ,z
(ξ) is the corresponding projection of Nλ,z (ξ). We use the equations (5.21) - (5.23) and λ(t)λ(t)
=
−a(τ (t)) in order to determine a(τ ), z(t), ξ(t) and λ(t).
Remark 3. The operator La = −dGa (ρa ) is the hessian Hess Va (ρa ) = d grad Va (ρa ) of the modified volume
functional Va (ρ) at the sphere ρa .
5.5
Linearized operator
p
In this section we discuss the operator La := −∂Ga ( na ) = na (−∆ − 2n) acting on L2 (Sn ), which is the
p
linearization of the map −Ga (ρ) = −G(ρ) − aρ at ρa = na . We have already encountered this operator in
MCF Lectures, October 26, 2014
28
Section 2, as the normal hessian of modified volume functional around a sphere. Its spectrum was described
in that section. Here we mention only the following relevant for us fact: The only non-positive eigenvalues
of La are
La ω 0 = −2aω 0 , La ω j = −aω j , j = 1, . . . , n + 1
(5.24)
and the next eigenvalue is
2a
n
and consequently, on their orthogonal complement we have
hξ, La ξi ≥
2a
kξk2 if ξ ⊥ ω j , j = 0, . . . , n + 1.
n
(5.25)
This coercivity estimate will play an important role in our analysis. This is the reason why we need the
conditions (5.16).
The fact that the n + 2 non-positive eigenvectors of La are related to change of the radius and position
of the euclidean sphere comes from the scaling and translational symmetries of (5.1). Indeed, let α = (r, z),
where r := na , and ρα be the real function on Sn , whose graph is the sphere Sα in Rn+1 of radius R, centered
at z ∈ Rn+1 , i.e. graph(ρ) := {ρ(ω)ω : ω ∈ S n } = Sα . The function ρα satisfies the equation |ρα (x)ˆ
x −z| = r,
or ρα (x)2 + |z|2 − 2ρα (x)z · x
ˆ = r2 , and therefore it is given by
p
ρα (ˆ
x) = z · x
ˆ + r2 − |z|2 + (z · x
ˆ )2 ,
where, recall, x
ˆ=
x
|x| .
By the symmetries, the equation Ga (ρα ) = 0 holds for any α. Differentiating it w.r.to α, we find
∂Ga (ρα )∂α ρα + ∂α Ga (ρα ) = 0. If we introduce the operator Lα := −∂Ga (ρα ), then the latter equations
become Lα ∂r ρα = −2a∂r ρα , Lα ∂z ρα = 0, i. e. ∂α ρα are eigenfunctions of the operator Lα .
Since
Lα = La + O(|z|) and
∂r ρα (x) = 1 + O(|z|), ∂zj ρα (x) = x
ˆj + O(|z|),
(5.26)
which gives - in the leading order - the n + 2 non-positive eigenvectors of La , these equations imply (5.24).
The n + 2 non-positive eigenvectors of La span the unstable - central subspace of the tangent space of
the fixed point manifold, {Sr,z |r ∈ R+ , z ∈ Rn+1 } (manifold of spheres), for the rescaled MCF. We use
the parameters of size, a, of thep
sphere (1 parameter) and its translations z (n + 1 parameters) to make the
fluctuation ξ(ω, τ ) := ρ(ω, τ ) − na orthogonal to the unstable-central modes, so that to be able to control
it.
5.6
Lyapunov functional
We introducep
the operator L = −∆ − 2n, acting on L2 (Sn ), which is related to the linearized operator
La := −∂Ga ( na ) = na (−∆ − 2n), introduced above, as La := na L. Using the spectral information about
obtained above we see that on functions ξ obeying (5.17) and ξ⊥ω j , j = 0, . . . , n + 1, tit is bounded below
as hL0 ξ, ξi ≥ kξk2 , we derive in this section some differential inequalities for certain Sobolev norms of such
a ξ. These inequalities allow
us to prove a priori estimates for these Sobolev norms. For k ≥ 1, we define
the functional Λk (ξ) = 21 ξ, Lk ξ .
Using the coercivity estimate (5.25) and standard elliptic estimates, we obtain (see Proposition 8 of [1],
with R2 = n/a)
MCF Lectures, October 26, 2014
29
Proposition 28. There exist constants c > 0 and C > 0 such that
ckξk2H k ≤ Λk (ξ) ≤ Ckξk2H k .
Proposition 29. Let k >
n
2
+ 1. If φ satisfies (5.17), then there exists a constant C > 0 such that
∂τ Λk (ξ) ≤ −
k+1
2a
a
Λk (ξ) − [
− C(Λk (ξ)1/2 + Λk (ξ)k )]kL 2 ξk2 .
n
2n
(5.27)
Proof. We have, using (5.17), or (5.23),
1
a
ξ, Lk ξ = ∂τ ξ, Lk ξ = − Lξ, Lk ξ + Nλ,z (ξ), Lk ξ ,
(5.28)
2
n
D k
E
k+1
k
where, recall, we use the notation ρ = ρa + ξ. We write Lξ, Lk ξ = 21 kL 2 ξk2 + 21 L 2 ξ, LL 2 ξ and use
D k
E D k
E
k
k
the coercivity estimate (5.25) for the second term, L 2 ξ, LL 2 ξ ≥ L 2 ξ, L 2 ξ = Λk (ξ), to obtain
∂τ
1 k+1
Lξ, Lk ξ ≥ kL 2 ξk2 + 2Λk (ξ).
2
(5.29)
Now, we begin estimating the next term
E
D k−1
k+1
| Nλ,z (ξ), Lk ξ | = | La 2 Nλ,z (ξ), L 2 ξ |
≤ kL
k−1
2
Nλ,z (ξ)kkL
k+1
2
(5.30)
ξk.
Recall the definition (5.19) of Nλ,z (ξ). To estimate the contribution of the first term on the r.h.s. in (5.19),
we use the inequality, which is a special case of Lemma 12 of Appendix D of [1],
kL
k−1
2
1/2
N (ξ)k . (Λk (ξ) + Λkk (ξ))kL
k+1
2
ξk.
(5.31)
(The operator La of this paper
is identified with the first (main) part of the operator LR0 := − R12 (∆ + n) +
R
1
n
2
R2 Av, where Av ξ := |Γ| Γ ξ, of [1], once we set R = n/a.) (To prove estimates of the type of (5.31), one
uses the Leibnitz rule for fractional derivatives and fractional integration by parts in the Besov spaces, in
k−1
order to distribute the derivatives in L 2 N (ξ) evenly among various factors, see [1].)
To estimate the contribution of the second term on the r.h.s. in (5.19), we use Eq. (5.22) to obtain
λ−1 zτ . N (ξ) + λ−1 z˜τ · ρ−1 ∇ξ, ω . kN (ξ)kL1 + λ−1 |zτ |k∇ξkL1 .
Next, using (5.20), where, recall, ρ = ρa + ξ and ρa =
pn
a,
(5.32)
and assuming that |ξ| ≤ 21 ρa , we have that
kN (ξ)kL1 . (k∇ξk2L4 + kξkH 1 )kξkH 2 .
(5.33)
This together with (5.32) gives, provided that kξkH 1 1,
|zτ | . λ(k∇ξk2L4 + kξkH 1 )kξkH 2 .
(5.34)
MCF Lectures, October 26, 2014
30
The last two equations, together with Proposition 28, imply that
kL
k−1
2
1/2
(λ−1 z˜τ · ρ−1 ∇ξ)k ≤ C(Λk (ξ) + Λkk (ξ))kL
k+1
2
ξk.
Collecting the estimates (5.30), (5.31) and (5.35), we find
k+1
1/2
| Nλ,z (ξ), Lk ξ | ≤ C(Λk (ξ) + Λkk (ξ))kL 2 ξk2 .
(5.35)
(5.36)
Relations (5.28), (5.29) and (5.36) yield (5.27).
Remark 4. (5.27) is an approximate monotonicity formula for the Lyapunov functional Λk (ξ). In general,
the functional Λ(ξ) = na Λ1 (ξ) = 21 hξ, La ξi is the leading term in the expansion of the modified volume
functional Va (ϕ + ξ) at the self-similar surface ϕ. What is the relation between the monotonicity formula
(2.9) and the inequality (5.27)?
5.7
Proof of Theorem 26
We begin with reparametrizing the initial condition. First, we write the immersion ψ0 (ω) as ψ0 (ω) =
z0 + ϕ0 (ω), where z0 ∈ Rn+1 is s.t.
Z
(ψ0 (ω) − z0 ) · ω)ω j = 0, j = 1, . . . , n + 1.
(5.37)
Sn
so that
R
Sn
(ϕ0 (ω) · ω)ω j = 0.
Next, reparametrizing
Sn , if necessary, we can assume that ϕ0 (ω) is of the form ϕ0 (ω) = ρ0 (ω)ω, which
R
j
implies that Sn ρ0 (ω)ω = 0, j = 1, . . . , n + 1. Given ρ0 we define a0 by an0 = hρ0 iSn , where, recall,
R
hf iSn := |S1n | Sn f , the average of f over Sn , so that
r
n
ρ0 (ω) =
+ ξ0 (ω), with ξ0 ⊥ 1 in L2 (Sn ).
(5.38)
a0
The last two statements imply that ξ0 ⊥ ω j , j = 0, . . . , n + 1, where, recall, ω 0 = 1.
1
Since the initial condition, ψ0 (ω), is sufficiently close to the identity, ξ0 (ω) and a0 satisfy Λk (ξ0 ) 2 +
1
1
Λk (ξ0 )k ≤ 10C
, Λk (ξ0 ) 1 and |a0 − n| ≤ 10
(see (5.38)), where the constant C is the same as in
Proposition 29.
˙
We consider the equations (5.21) - (5.23) and λ(t)λ(t)
= −a(τ (t)) for a(τ ), z(t), ξ(t) and λ(t). The
˙
initial value problem λ(t)λ(t)
= −a(τ (t)) for λ(t) and λ(t0 ) = λ0 , where τ = τ (t) given by (5.7), can be
solved explicitly to obtain λ(t) = λ(a, λ0 )(t), where λ(a, λ0 )(t) is the positive function given by
Z t
λ(a, λ0 )(t) := (λ20 − 2
a(τ (s))ds)1/2 .
(5.39)
t0
To solve for a(τ ), z(t) and ξ, we use the spaces At0 ,δ , Zt0 ,δ = C 1 (It0 ,δ , Rn+1 ) and H s (Sn ), s > n/2 + 1,
respectively. Here, t0 ≥ 0, δ > 0, It0 ,δ := [t0 , t0 + δ], and
1
1
At0 ,δ := C 1 (It0 ,δ , [n − , n + ]).
2
2
(5.40)
MCF Lectures, October 26, 2014
31
In what follows we deal only with the unknowns a(τ ), z(t) and ξ(t) and consider λ(t) as given by (5.39).
Now, a standard parabolic theory gives the local existence result for the equations (5.21) - (5.23).
Now, let
a
}.
4nC
By continuity, T > 0. Assume T < ∞. Then ∀τ ≤ T we have by Proposition 29 that ∂τ Λk (ξ) ≤ − 2a
n Λk (ξ).
1
We integrate this equation to obtain Λk (ξ) ≤ Λk (ξ0 )e− 2 τ ≤ Λk (ξ0 ). This implies
1
1
a
.
Λk (ξ(T )) 2 + Λk (ξ(T ))k ≤ Λk (ξ0 ) 2 + Λk (ξ0 )k ≤
8nC
1
T = sup{τ > 0 : Λk (ξ(τ )) 2 + Λk (ξ(τ ))k ≤
1
a
Therefore proceeding as above we see that there exists δ > 0 such that Λk (ξ(τ )) 2 + Λk (ξ(τ ))k ≤ 4nC
, for
τ ≤ T + δ, which contradicts
the
assumption
that
T
<
∞
is
the
maximal
existence
time.
So
T
=
∞
and
Rτ
Rτ
2
2
Λk (ξ) ≤ Λk (ξ0 )e− n 0 a(s)ds . By Proposition 28 we know that kξk2H k . Λk (ξ0 )e− n 0 a(s)ds .
Now, we use Eq. (5.21) to obtain
−3/2 a
aτ . N (ξ) + λ−1 z˜τ · ρ−1 ∇ξ, 1 . kN (ξ)kL1 + λ−1 |zτ |k∇ξkL1
(5.41)
|a−3/2 aτ | . (k∇ξk2L4 + kξkH 1 )kξkH 2 .
(5.42)
This together with (5.33) gives
This implies |a(τ ) − n| ≤
1
4,
and
1
1
|a(τ )− 2 − a(0)− 2 | ≤
1
2
Z
τ
3
|a(s)− 2 aτ (s)|ds .
0
Z τ
1
≤ Λk (ξ0 )
e− 2 s 1.
Z
τ
Λk (ξ)ds
0
(5.43)
0
Next, by (5.34),
Z
τ
Z
τ
|z(τ ) − z(0)| ≤
|zτ (s)|ds .
λ(s)Λk (ξ)ds
0
0
Z τ
1
≤ Λk (ξ0 )
e−(n+ 2 )s 1.
(5.44)
0
Rt
Rt
Observe that λ20 − λ(t)2 = 2 0 a(τ (s))ds. Let t∗ be the zero of the function λ20 − 2 0 a(τ (s))ds. Since
|a(τ ) − n| ≤ 21 , we have t∗ < ∞ and λ(t) → 0 as t → t∗ . Similarly to (5.43), we know that
Z τ2
1
|a(τ2 )−1/2 − a(τ1 )−1/2 | .
e− 2 s ds → 0,
τ1
1
as τ1 , τ2 → ∞. Hence there exists a∗ > 0, such that |a(τ ) − a∗ | . e−(1− 2n )τ . Similar arguments show
Rt
1
1
that there exists z∗ ∈ Rn+1 such that |z(τ ) − z∗ | . e−(n+ 2 − 2n )τ . Then λ2 = λ20 − 2 0 a(τ (s))ds =
Rt
R t ds
Rt
1
ds
2 t ∗ a(τ (s))ds = 2a∗ (t∗ − t) + o(t∗ − t). The latter relation implies that τ = 0 λ(s)
2 = 2a
0 (t∗ −s)(1+o(1))
∗
p
1
1
1
1
1
1
−(1− 2n )τ
and
= O((t∗ −t) 2a∗ (1− 2n ) ). So λ(t) = 2a∗ (t∗ − t)+O((t∗ −t) 2 + 2a∗ (1− 2n ) ), ρ(ω, τ (t)) =
q therefore e
1
n
2n .
The latter inequality with k = s, together with
a(τ (t)) + ξ(ω, τ (t)), and kξ(ω, τ (t))kH k . (t∗ − t)
estimates on a, z and λ obtained above, together with (5.7) and (5.9), proves Theorem 26. MCF Lectures, October 26, 2014
6
32
Stability of elementary solutions II. Neckpinching
We consider boundariless hypersurfaces, {Mt | t ≥ 0}, in Rn+2 , given by immersions X(·, t) : Sn ×R → Rn+1 ,
evolving by the mean curvature flow Then X satisfy the evolution equation:
∂X
= −H(X)ν(X),
∂t
(6.1)
where ν(X) and H(X) are the outward unit normal vector and mean curvature at X ∈ Mt , respectively.
We are interested in evolution of surfaces starting with those, close to cylinders. Our goal is to show
that
(a) for initial surfaces arbitrary close to the cylinder, the MCF becomes singular in a finite time by
collapsing the radius at one point along the axis - the neckpinch;
(b) while the point and time of the neckpinch depend on the initial conditions, the profile of the neckpinch
is universal;
(c) describe the asymptotics of the parameters involved.
The initial surface could be arbitrary close to a cylinder, which shows that the cylindrical collapse to
the axis is unstable.
There are two new elements in the analysis of the neckpinch as compared to the collapse to a point,
which make the problem much more difficult:
(i) cylinders have a larger group of transformations compared to spheres: besides of shifts there are
rotations of the axis (tilting),
(ii) the linearized map on the cylinder acts on functions on the cylinder - a non-compact domain - which
make it much more difficult to analyse.
In the rest of this section we present the latest result on the neckpinch for a fixed cylinder axis, explain
its proof and discuss possible extension to the case when the axis is allowed to move (tilt).
6.1
Existing Results
Write points in Rn+2 as (x, w) = (x, ω 1 , . . . , ω n+1 ). (We label components of points of Rn+2 as (x0 , x1 , . . . , xn+1 ).)
2
The round cylinder R × Sn is naturally embedded as C n+1 = {(x, w) : |w| = 1} ⊂ Rn+2 . We say a surface
given by immersion X is a (normal) graph over the cylinder C n+1 , if there a function u : C n+1 → R+ (called
the graph function) s.t. X = Xu , where
Xu : (x, w) 7→ (x, u(x, w)w).
(6.2)
Assume the solution, S(t), of the MCF with an initial condition M0 , given by a graph X0 = Xu0 over
the cylinder C n+1 with a graph function u0 (x, ω), is given by a graph X0 = Xu0 over the cylinder C n+1 with a
MCF Lectures, October 26, 2014
33
graph function u(x, ω, t). We say S(t) undergoes a neckpinching at time t∗ and at a point x∗ if inf u(·, t) > 0
for t < t∗ and inf u(·, t) → 0 as t → t∗ and inf u(·, t) → 0 at the single point x∗ .
All existing results on the neckpinch, except for [9, 10], consider axi-symmetric, compact or periodic
surfaces, i.e. initial graph functions u0 (x, ω), independent of ω and defined on a finite interval, say [a, b], for
some −∞ < a < b < ∞, with u0 (x) > 0 for a < x < b and either u0 (a) = u0 (b) = 0 or ∂x u0 (a) = ∂x u0 (b) = 0
(the Dirichlet or Neumann boundary conditions). [9] considers a much more difficult case of axi-symmetric
surfaces, with are initial conditions, u0 (x), defined on the entire axis R and satisfying u0 (x) > 0 ∀x ∈ R
and lim inf |x|→∞ u0 (x, ω) > 0, while [10] treats non-axi-symmetric surfaces along the entire axis R, but with
some symmetries fixing the cylinder axis.
For simplicity, we present the result of [10] and then discuss a general (unpublished) approach to general,
non-axi-symmetric surfaces, without fixing the cylinder axis. In what follows, we use the notation hxi :=
1
(1 + x2 ) 2 .
Initial conditions. We consider initial surfaces, S0 , given by immersions Xu0 , which are graphs over the
cylinder C n+1 , with the graph functions u0 , which are positive, have, modulo small perturbations, global
minima at the origin, are slowly varying near the origin and are even w.r. to the origin.
11
More precisely, we assume that (a) u0 (x, ω) satisfies for (k, m) = (3, 0), ( 10
, 0), (1, 2) and (2, 1) the
estimates
k+m+1
2 1
2
0x
2k
≤
Cε
,
ku0 − ( 2n+ε
)
k,m
0
2ς0
(6.3)
√
u0 (x, ω) ≥ n,
weighted bounds on derivatives to the order 5.
symmetries: u0 (x, ω) = u0 (−x, ω), either u0 (x, ω) = u0 (x), or n = 2 and u0 (x, θ) = u0 (x, θ + π).
The last condition fixes the the point of the neckpinch and the axis of the cylinder. Here we used the norms
kφkk,m := hxi−k ∂xm φL∞ .
Neckpinching profile.
Theorem 30 ([9, 10]). Under initial conditions described above and for ε0 sufficiently small, we have
(i) there exists a finite time t∗ such that inf u(·, t) > 0 for t < t∗ and limt→t∗ u(·, t) > 0, for x 6= 0 and
= 0, for x = 0;
(ii) there exist C 1 functions ζ(y, ω, t), λ(t), c(t) and b(t) such that
s
2n + b(t)y 2
u(x, ω, t) = λ(t)[
+ ζ(y, ω, t)]
c(t)
with y := λ−1 (t)x and the remainder ζ(y, ω, t) satisfying, for some constant c,
X
khyi−m ∂yn ζ(y, ω, t)k∞ ≤ cb2 (t).
m+n=3,n≤2
(6.4)
(6.5)
MCF Lectures, October 26, 2014
34
Dynamics of scaling parameter
Theorem 31 ([9, 10]). (iii) the parameters λ(t), b(t) and c(t) satisfy the estimates
Here s := t∗ − t and λ0 = √
s := t∗ − t.
1
λ(t)
= s 2 (1 + o(1));
b(t)
=
− lnn|s| (1 + O( | ln1|s| |3/4 ));
c(t)
=
1+
1
ε
2ς0 + n0
1
ln |s| (1
(6.6)
+ O( ln1|s| )).
, with ς0 , ε0 > 0 depending on the initial datum and o(1) is in
(iv) if u0 ∂x2 u0 ≥ −1 then there exists a function u∗ (x) > 0 such that u(x, t) ≥ u∗ (x) for R\{0} and t ≤ t∗ .
6.2
Rescaled surface
Let Mt denote a smooth family of smooth surfaces, given by immersions X(·, t) : R × Sn → Rn+1 and
evolving by the mean curvature flow, (6.1). It is convenient to rescale it as follows
Y (y, ω, τ ) = λ−1 (t)X(x, ω, t),
(6.7)
where y = y(x, t) and τ = τ (t), with
y(x, t) := λ−1 (t)[x − x0 (t)]
Z
and
τ (t) :=
t
λ−2 (s) ds,
(6.8)
0
respectively, where x0 (t) marks the center of the neck. If the point of the neckpinch is fixed by a symmetry
of the initial condition, then we set x0 (t) = 0. We derive the equation for Y . For most of the derivation, we
suppress time dependence.
−1
H(Y ) and ν(X) = ν(Y ) (see (4.3) - (4.5)),
Let λ˙ = ∂λ
∂t . Using X = λY and the relations H(X) = λ
∂y ∂Y
∂y
∂Y ∂Y
−2 ∂Y
−1
˙
˙
˙ −1 , we obtain from
= λY + λ ∂t , ∂t = λ ∂τ + ∂t ∂y and ∂t = −λλ y, and the definition a := −λλ
(6.1) the equation for Y , λ, z and g:
∂X
∂t
∂Y
= −H(Y )ν(Y ) + aY − ay∂y Y.
∂τ
The equation (6.9) has the static solutions: a = a positive constant, Y (x, w) = (x,
6.3
(6.9)
pn
a w).
Linearized map
Consider
p n the equation (6.9). This equation has the static solutions Y (y, w) = Xρ (y, w) = (y, ρw), with
ρ =
a . They describe the cylinders of the radii ρ, shifted by z and rotated by g. We would lie to
investigate the linearized stability of these solutions. To this end, we linearize the r.h.s. of (6.9) around
these static solutions, considering only variations in the normal direction, to obtain (see (4.24))
La := −∂y2 − ay∂y +
a
(∆Sn + 2n)
n
MCF Lectures, October 26, 2014
35
where ∆Sn is the Laplace-Beltrami operator on Sn . We have already encountered this operator in Section 2, as
the normal hessian of modified volume functional around a cylinder. (Recall that the operator − na (∆Sn +2n),
as the normal hessian of modified volume functional around a sphere. It played an important role in our
study of spherical collapse.)
For a fixed a, properties of the operator La were described in Section 2: it is self-adjoint on the Hilbert
a 2
space L2 (R × Sn , e− 2 y dydw) and its spectrum consists of the eigenvalues [k − 2 + n1 l(l + n − 1)2 ]a, k =
0, 1, 2, 3, . . . ; ` = 0, 1, 2, . . . , of the multiplicities ml , with the normalized eigenvectors given by
φk,l,m,a (y, w) := φk,a (y)Ylm (w),
(6.10)
where, recall, φk,a (y) are the normalized eigenvectors of the linear operator −∂y2 − ay∂y corresponding
to the eigenvalues ka : k = 0, 1, 2, 3, . . . , and Ylm are the eigenfunctions of −∆Sn corresponding to the
eigenvalue l(l + n − 1) (the spherical harmonics). This implies that the only non-positive eigenvalues of La
are −2a, −a, 0, corresponding to the eigenvectors, labeled by (k, l) = (0, 0), (0, 1), (1, 0), (1, 1), (2, 0), with
multiplicities 1, n + 2; n + 2 (so that the total multiplicity of the non-positive eigenvalues is 2n + 5). The
first three eigenfunctions of the operator −∂y2 − ay∂y , corresponding to the eigenvalues, −2a, −a, 0, are
φ0,a (y) = (
a 1√
a 1
a 1
) 4 , φ1,a (y) = ( ) 4 ay, φ2,a (y) = ( ) 4 (1 − ay 2 ),
2π
2π
2π
(6.11)
and first n + 2 eigenfunctions of the operator −∆Sn , corresponding to the first eigenvalue 0, and the second
eigenvalue n are 1 and w1 , · · · , wn+1 , respectively. Hence, La has
1
a 4
) ((k, l) = (0, 0)), due
• the eigenvalue −2a of the multiplicity 1 with the eigenfunction φ0,0,0,a (y) = ( 2π
to scaling of the transverse sphere;
1
a 4 m
• the eigenvalue −a of the multiplicity n + 1 with the eigenfunctions φ0,1,m,a (y) = ( 2π
) w , m =
1, . . . , n + 1 ((k, l) = (1, 1)), due to transverse translations;
a 14 √
• the eigenvalue 0 of the multiplicity n + 1 with the eigenfunctions φ1,1,m,a (y) = ( 2π
) aywm , m =
1, . . . , n + 1 ((k, l) = (0, 1)), due to rotation of the cylinder;
a 14 √
) ay ((k, l) = (1, 0)),
• the eigenvalue −a of the multiplicity 1 with the eigenfunction φ1,0,0,a (y) = ( 2π
due to translations of the point of the neckpinch on the axis of the cylinder;
1
a 4
• the eigenvalue 0 of the multiplicity 1 with the eigenfunction φ2,0,0,a (y) = ( 2π
) (1−ay 2 ) ((k, l) = (2, 0)),
due to shape instability.
The modes φ1,a and φ2,a are due to the location of the neckpinch (the shift in the x direction) and
its shape (see [9] and below). To see how they originate, we replacepthe cylindrical, static (i.e. y and
n
τ -independent) solution Y (y, w) = XVa (y, w) = (y, ρa w), with Va =
a , to (6.9) with a constant, by a
modulated cylinder solution
r
2n + by 2
Yab (y, w) = XVab (y, w) = (y, Vab (y)w), with Vab (y) =
,
(6.12)
2a
which is derived later and which incorporates some essential features of the neckpinch. They are approximate
static solutions to (6.9) with a constant, −H(Yab ) + aYab · ν(Yab ) ≈ 0. Differentiating it w.r.to a, x0 (see
MCF Lectures, October 26, 2014
36
(6.8)) and b and using that b is sufficiently small and therefore by 2 can be neglected in a bounded domain,
we arrive at vectors proportional to φ0,a , φ1,a and φ2,a .
Recall from Section 6.8 that the modes φk,l,m,a , (k, l, m) = (0, 1, 1), . . . , (0, 1, n + 1), are coming from the
transverse shifts, and φk,l,m,a , (k, l, m) = (1, 1, 1), . . . , (1, 1, n+1), (i.e. φ1,a (y)wm ' ywm , m = 1, . . . , , n+1),
from the rotations of the cylinder. Hence fixing the cylinder axis, eliminates these modes.
6.4
Equations for the graph function
Assume that for each t in the the existence interval, the transformed immersion Y (y, w, τ ), given in (6.7), is
a graph over the cylinder Y (y, w, τ ) = (y, v(y, w, τ )w). where y and τ are blowup variables defined by (6.8),
∂f
with x0 = 0. Then, similarly to the derivation of the equation (5.10) and using ∂f
∂t = ∂τ ∂t τ, for any
−2
−2 ˙
−2
˙ we see that
function f of the variables τ , ∂t τ = λ and ∂t y = −λ λxx = λ ayx, where, recall, a = −λλ,
(6.9) implies
∂τ v = G(v) − ay∂y v + av,
(6.13)
where the map G(v) given by
G(v) := v −2 ∆v + ∂y2 v − nv −1 −
−
v −4 h∇v, Hess v ∇vi − v −2 |∇v|2
1 + (∂y v)2 + v −2 |∇v|2
2v −2 ∂y vh∇v, ∇∂y vi + (∂y v)2 ∂y2 v
,
1 + (∂y v)2 + v −2 |∇v|2
(6.14)
with ∇ and Hess denoting the Levi-Civita covariant differentiation and hessian on a round Sn of radius one,
with the components ∇i and ∇i ∇j in some local coordinates, respectively.
Initial conditions for v. If λ0 = 1, then the initial conditions for u given in Theorem 30 implies that
there exists a constant δ such that the initial condition v0 (y) is even and satisfy for (k, m) = (3, 0), ( 11
10 , 0),
(1, 2) and (2, 1) the estimates
2
k+m+1
2
1
0y
kv0 (y, ω) − ( 2n+ε
) 2 kk,m ≤ Cε0
1− 1 ε
n 0
v0 (y, ω) ≥
√
,
(6.15)
n,
weighted bounds on derivatives to the order 5,
symmetries: v0 (y, ω) = v0 (−y, ω), v0 (y, θ) = v0 (y, θ + π).
6.5
Reparametrization of the solution space
Adiabatic solutions. The equation (6.13) has the following cylindrical, static (i.e. y and τ -independent)
solution
r
n
Va :=
and a is constant ⇐⇒ ucyl (t).
(6.16)
a
MCF Lectures, October 26, 2014
37
In the original variables t and x,p
this family of solutions corresponds to the cylinder (homogeneous) solution
X(t) = (x, R(t)ω), with R(t) = R02 − 2nt.
A larger family of approximate solutions is obtained by assuming that v is slowly varying and ignoring
∂τ v and ∂y2 v in the equation for v to obtain the equation
ayvy − av +
n
=0
v
(adiabatic or slowly varying approximation). This equation has the general solution
r
2n + by 2
Vab (y) :=
,
(6.17)
2a
pn
p
2
b
2π 14
√
with b ∈ R. Note also that for by 2 small, Vab (y) ≈ na (1 + by
4n ) =
a + 4a an ( a ) [φ0,a (y) + φ2,a (y)], i.e.
pn
the static solution a is modified by vectors along its unstable and central subspaces.
In what follows we take b ≥ 0 so that Vab is smooth. Note that Va0 = Va .
The rescaled equation (6.13) has two unknowns v and a. Hence we need an extra restriction specifying
a or λ.
Reparametrization of solutions. We want to parametrize a solution to (6.13) by a point on the manifold
Mas , formed by the family of almost solutions (6.17) of equation (6.13),
Mas := {Vab | a, b ∈ R+ , b ≤ },
equippped with the Riemannian metric
0
hη, η i :=
Z
0
ηη e−
ay 2
2
dydw,
(6.18)
and the fluctuation almost orthogonal to this manifold (large slow moving and small fast moving parts of
the solution):
v(w, y, τ ) = Va(τ )b(τ ) (y) + φ(w, y, τ ),
(6.19)
with
φ(·, τ ) ⊥ φka(τ ) in L2 (R × Sn , e−
a(τ ) 2
2 y
dydw), k = 0, 1, 2.
(6.20)
2
(Provided b is sufficiently small, the vectors φ0a = 1 and φ2a
√ = (1 − ay ) are almost tangent vectors to the
manifold, Mas . Note that φ is already orthogonal to φ1a = ay, since by our initial conditions the solutions
are even in y. For technical reasons, it is more convenient to require the fluctuation to be almost orthogonal
to the manifold Mas .)
As in the spherical collapse case, we consider (6.13) as an equation for a and φ and the equation (6.20),
with k = 0, leads to the additional equation on the parameter a. Moreover, the parameters x0 (see (6.8))
and b are determined from the equations (6.20), k = 1, 2.
MCF Lectures, October 26, 2014
6.6
38
Lyapunov-Schmidt splitting (effective equations)
In what follows we consider only surfaces of revolution. The decomposition (6.19) becomes
v(y, τ ) = Va(τ ),b(τ ) (y) + φ(y, τ ),
(6.21)
Substitute (6.21) into (MCF) to obtain
∂τ φ = −Lab φ + Fab + Nab (φ)
(6.22)
where Lab is the linear operator given by
Lab := −∂y2 + ay∂y − 2a +
aby 2
2+
by 2
n
and the functions F (a, b) and N (a, b, φ) are defined as
Fab :=
1 2n + by 2 1
1
b3 y 4
y2
],
(
) 2 [Γ1 + Γ2
+
−
2
2a
2n + by 2
n (2n + by 2 )2
with
∂τ a
2a
(6.23)
+ 2a − 1 + nb ,
Γ1
:=
Γ2
:= −∂τ b − b(2a − 1 + nb ) −
Nab (ξ)
2a
2
:= − nv 2n+by
2ξ −
b2
n,
(6.24)
(∂y v)2 ∂y2 v
1+(∂y v)2 .
Here we ordered the terms in Fab according to the leading power in y 2 .
Remember that
φ(·, τ ) ⊥ 1, a(τ )y 2 − 1 in L2 (R, e−
a(τ ) 2
2 y
dy).
(6.25)
2
Project the above equation on 1, a(τ )y − 1 =⇒ the equations for the parameters a, b.
Estimating φ Let U (τ, σ) be the propagator generated by −Lab . By Duhamel principle we rewrite the
differential equation for φ(y, τ ) as
Z τ
φ(τ ) = U (τ, 0)φ(0) +
U (τ, σ)(F + N )(σ)dσ.
(6.26)
0
The key problem in estimating φ(y, τ ) is to estimate the propagator U (τ, σ). We have to use (6.25).
Proposition 32. There exist constants c, δ > 0 such that if b(0) ≤ δ, then for any g ⊥ 1, a(τ )y 2 −
a(τ ) 2
1 in L2 (R, e− 2 y dy) and τ ≥ σ ≥ 0, we have
khzi−3 U(τ, σ)gk∞ . e−c(τ −σ) khzi−3 gk∞ .
MCF Lectures, October 26, 2014
39
Estimating the linear propagator. I. Recall U (τ, σ) is the propagator generated by −Lab . We write
Lab = L0 + V, where
1
L0 := −∂y2 + ay∂y − 2a, V (y, τ ) ≥ 0 and |∂y V (y, τ )| . b 2 (τ ).
(6.27)
(L0 is the Ornstein-Uhlenbeck generator related to the harmonic oscillator Hamiltonian.)
Denote the integral kernel of U(τ, σ) by K(x, y). We have the representation (see Appendix E)
K(x, y) = K0 (x, y)heV i(x, y),
(6.28)
where K0 (x, y) is the integral kernel of the operator e−(τ −σ)L0 and
Z R
τ
heV i(x, y) = e σ V (ω(s)+ω0 (s),s)ds dµ(ω).
(6.29)
Here dµ(ω) is a harmonic oscillator (Ornstein-Uhlenbeck) probability measure on the continuous paths
ω : [σ, τ ] → R with the boundary condition ω(σ) = ω(τ ) = 0 and
(−∂s2 + a2 )ω0 = 0 with ω(σ) = y and ω(τ ) = x.
Estimating the linear propagator. II.
(6.30)
By a standard formula we have
1
K0 (x, y) = 4π(1 − e−2ar )− 2
√
ae2ar e
−a
(x−e−ar y)2
2(1−e−2ar )
,
where r := τ − σ. To estimate U (x, y) we use that, by the explicit formula for K0 (x, y) given above,
|∂yk K0 (x, y)| .
e−akr
(|x| + |y| + 1)k K0 (x, y),
(1 − e−2ar )k
and by an elementary estimate
1
|∂y heV i(x, y)| ≤ b 2 r.
(6.31)
Estimate for ear ≤ β −1/32 (τ ) (r := τ − σ) and then iterate using the semi-group property.
Now, (6.28) implies Equation ( 6.31) by the following lemma.
Lemma 33. Assume in addition that the function V (y, τ ) satisfies the estimates
1
V ≤ 0 and |∂y V (y, τ )| . β − 2 (τ )
(6.32)
where β(τ ) is a positive function. Then
Z R
τ
1
|∂y e σ V (ω0 (s)+ω(s),s)ds dµ(ω)| . |τ − σ| sup β 2 (τ )
σ≤s≤τ
Proof. By Fubini’s theorem
Z R
Z
Z
τ
V (ω0 (s)+ω(s),s)ds
σ
∂y e
dµ(ω) = ∂y [
0
τ
V (ω0 (s) + ω(s), s)ds]e
Rτ
σ
V (ω0 (s)+ω(s),s)ds
dµ(ω)
MCF Lectures, October 26, 2014
40
Equation ( 6.32) implies
Z τ
Rτ
1
|∂y
V (ω0 (s) + ω(s), s)ds| ≤ |τ − σ| sup β 2 (τ )|, and e σ V (ω0 (s)+ω(s),s)ds ≤ 1.
σ≤s≤τ
σ
Thus
Z
|∂y
e
Rτ
σ
V (ω0 (s)+ω(s),s)ds
1
2
dµ(ω)| . |τ − σ| sup β (τ )|
Z
1
dµ(ω) = |τ − σ| sup β 2 (τ )|
σ≤s≤τ
σ≤s≤τ
to complete the proof.
Derivation of Proposition 32 from (6.28) and (6.31).
Bootstrap Let (a, b, φ) be the neck parametrization of the rescaled solution v(y, τ ). To control the function
φ(y, τ ), we use the estimating functions
Mk,m (T ) := max b−
k+m+1
2
τ ≤T
with (k, m) = (3, 0), ( 11
10 , 0), (2, 1), (1, 2). Let |M | :=
(τ )khyi−k ∂ym φ(·, τ )k∞ ,
P
i,j
M := (Mi,j ), (i, j) = (3, 0), (
(6.33)
Mi,j and
11
, 0), (1, 2), (2, 1).
10
(6.34)
Using Proposition 32 and a priori estimates obtained with help of a maximum principle (see Appendix
??), we find
Proposition 34. Assume that for τ ∈ [0, T ] and
1
|M (τ )| ≤ b− 4 (τ ), v(y, τ ) ≥
1√
2n, and ∂ym v(·, τ ) ∈ L∞ , m = 0, 1, 2.
4
Then there exists a nondecreasing polynomial P (M ) s.t. on the same time interval,
1
Mk,m (τ ) ≤ Mk,m (0) + b 2 (0)P (M (τ )),
(6.35)
Corollary 35. Assume |M (0)| 1. On any interval [0, T ],
1
|M (τ )| ≤ b− 4 (τ ) =⇒ |M (τ )| . 1.
The analysis presented above goes through also in the non-radially symmetric case, with the cylinder
axis fixed, but with one caveat. The feeder bounds have to be extended to the non-radially symmetric case
and new bounds on θ- and mixed derivatives, ∂yn ∂θm v(y, θ, τ ), have to be obtained. This requires additional
tools (differential inequalities for Lyapunov-type functionals.
MCF Lectures, October 26, 2014
6.7
41
General case. Transformed immersions
Let Mt denote a smooth family of smooth hypersurfaces, given by immersions X(·, t) and evolving by the
mean curvature flow, (6.1). Instead of the surface Mt , it is convenient to consider the new, rescaled surface
˜ τ = λ−1 (t)g(t)−1 (Mt − z(t)), where λ(t) and z(t) are some differentiable functions and g(t) ∈ SO(n + 2),
M
Rt
to be determined later, and τ = τ (t) := 0 λ−2 (s)ds. If we look for X as a graph over a cylinder, then we
have to rescale also the variable x along the cylinder axis, y = λ−1 (t)(x − x0 (t)) (see (6.8)). The new surface
is described by Y , which is an immersion of the fixed cylinder C n+1 , i.e. Y (·, τ ) : C n+1 → Rn+2 , given by
Y (y, w, τ ) = λ−1 (t)g −1 (t)(X(x, w, t) − z(t)).
Rt
where y = y(x, t) and τ = τ (t) are given by y = λ−1 (t)(x − x0 (t)) and τ = 0 λ−2 (s)ds.
(6.36)
We derive the equation for Y . For most of the derivation, we suppress the time dependence. Let λ˙ = ∂λ
∂t .
Using that X = z + λgY and the relations (4.3) - (4.5) , which imply H(X) = λ−1 H(Y ) and ν(X) = gν(Y ),
and using
∂X
˙
= z˙ + λgY
+ λgY
˙ + λg Y˙ ,
∂t
we obtain from (6.1) the equation for Y , λ, z and g:
∂Y
= −H(Y )ν(Y ) + (a − y∂y )Y − g −1 gY
˙ − λ−1 g −1 z.
˙
∂τ
(6.37)
The equation (6.37)
has the static solutions (a = a positive constant, z =constant, g =constant, x0 =constant,
p
Y (x, w) = (x, na w)).
Note that we do not fix λ(t) and x0 (t), z(t) and g(t) but instead consider them as free parameters to be
determined from the evolution of v in equation (6.37).
Remark 5. Another way to write (6.36) is Y = (Tλ Tz Tg X) ◦ Sλ , where Tλ , Tz and Tg are the scaling,
translation and rotation maps defined by Tλ X := λ−1 X, Tz X := X − z, z ∈ Rn+2 , and Tg X := g −1 X,
Rt
g ∈ SO(n + 2), and acting pointwise and Sλ−1 : (x, ω, t) → (y = λ−1 (t)x, w, τ = 0 λ−2 (s)ds).
6.8
Translational and rotational zero modes
(do not probably need) Recall that the maps X = Xρzg , with z, g and ρ constant, describe the cylinders
of the radii ρ, shifted by z and rotated by g. They are static solutions of (6.37) with a constants, i.e. solve
−H(Xzgρ ) + aXzgρ · ν(Xρzg ) = 0,
(6.38)
pn
provided ρ = a . Differentiating this equation w.r. to a, z i and g ib , b = n + 2, we arrive at the scaling,
translational and rotational modes. We expect that these are exactly the modes mentioned in Section 2 and
Subsection 6.3.
6.9
Collapse center and axis
In this section, extending [24], we introduce a notion of the ’centre’ and ’axis’ of a surface, close to the round
cylinder C n+1 , and show that such centre and axis exist. We define the cylinder axis, A, by a unit vector
MCF Lectures, October 26, 2014
42
α ∈ Sn+1 so that A := Rα. Moreover, we identify the axis vector α with an element g ∈ SO(n+2)/SO(n+1)
s.t. gen+2 = α.
Our eventual goal will be to show that the centers z(t) and axis vectors α(t) of the solutions Mt to the
MCF converge to the collapse point, z∗ , and axis vector, α∗ . For a boundaryless surface S, given by an
immersion X : C n+1 → Rn+2 , we define the center, z, and the axis vector, α (or g s.t. gen+2 = α), by the
relations
Z
(P ⊥ g −1 (X(x, w) − z) · w)xk wi = 0, i = 1, . . . , n + 1, k = 0, 1,
(6.39)
C n+1
where P ⊥ is the orthogonal projection to the subspace {(0, w)} ⊂ Rn+2 . Here and in what follows, the
integrals over the set C n+1 (or R × Sn ), which do not specify the measure, are taken w.r.to the measure
a 2
a 2
e− 2 x dwdx or e− 2 y dwdy. These are 2n + 2 equations for the 2n + 2 unknowns z ∈ Ran Pg⊥ and g ∈
SO(n+2)/SO(n+1), where SO(n+1) is the group of rotations of the subspace {(w, 0)} ⊂ Rn+2 . Note that the
rotated, shifted and dilated (transformed) cylinders, Cλzg , defined by the immersions Xλzg : C n+1 → Rn+2 ,
given by
Xλzg (x, w) := z + g(x, λw),
(6.40)
satisfy (6.39), which justifies our interpretation of its solutions as the centre and axis vector of a surface.
a
2
Denote by H 1 (C n+1 , Rn+2 ) the Sobolev space with the measure e− 2 x dwdx. We introduce the following
Definition 4. A surface M, given by an immersion X : C n+1 → Rn+2 , is said to be H 1 −close to a
transformed cylinder Cλzg , iff Y := λ−1 g −1 (X − z) ∈ H 1 (C n+1 , Rn+2 ) close, in the H 1 (C n+1 , Rn+2 )-norm,
to the identity 1 : C n+1 → C n+1 .
Proposition 36. Assume a surface M, given by an immersion X : C n+1 → Rn+2 , is H 1 −close to a
+
¯
¯ ∈ Rn+1 and g¯ ∈ SO(n + 2). Then there exists g ∈ SO(n +
transformed cylinder Cλ¯
¯ zg
¯ , for some λ ∈ R , z
⊥
2)/SO(n + 1) and z ∈ Ran Pg such that (6.39) holds.
Proof. By replacing X by X new , if necessary, we may assume that z¯ = 0, g¯ = 1 and λ = 1. The relations
(6.39) are equivalent to the equation F (X, z, g) = 0, where
F (X, z, g) = (Fik (X, z, g), i = 1, . . . , n + 1, k = 0, 1),
with Fik (X, z, g) equal to the l.h.s. of (6.39). Clearly F is a C 1 map from H 1 (C n+1 , Rn+2 ) × Rn+1 × SO(n +
2)/SO(n + 1) to R2n+2 . We notice that F (1, 0, 1) = 0. We solve the equation F (X, z, g) = 0 near (1, 0, 1),
using the implicit function theorem. To this end we have to calculate the derivatives ofRF w.r.to z and g at
X = 1, z = 0, g = 1. The derivatives with respect to z i are easy ∂zi Fjk |z=0,g=1 = − C n+1 wi wj xk , which
gives
∂zi Fjk |z=0,g=1 = −
1
|Sn |δij δk,0 ,
n+1
(6.41)
for i, j = 1, . . . , n + 1, k = 0, 1. To calculate the derivatives of F w.r.to g, we write g ∈ SO(n + 2)/SO(n + 1)
as a product of rotations g ab in two-dimensional planes, i.e. involving only variables xa and xb , with a = 0
and b = 1, . . . , n + 1, and denote by ∂gab the derivative w.r.to the angle of the corresponding rotation.Using
MCF Lectures, October 26, 2014
43
that ∂gab g −1 |z=0,g=1 = −`ab , the generator of the rotation in the ab−plane and using that 1(x, w) = (x, w),
we calculate
Z
(`ab (w, x) · (0, w))ω j xk .
(6.42)
∂gab Fjk |X=1,z=0,g=1 = −
C n+1
P
We see that, if a, b ∈ {1, . . . , n + 1}, then `ab (x, w) · (0, w) = −wb wa + wa wb + i6=a,b (wi )2 and therefore
R
(` (x, w) · (0, w))ω j = R0. Furthermore, if a = 0 and b = 1, . . . , n + 1, then `ab (x, w) · (0, w) = xwb +
Sn ab
P
i 2
j
i6=a,b (w ) and therefore Sn (`ab (x, w) · (0, w))ω = δb,j x. This gives, for a = 0,
Z
∂gai Fjk |X=1,z=0,g=1 = cn
a
2
x2 e− 2 x dxδij δk,1 ,
(6.43)
R
for i, j = 1, . . . , n + 1, k = 0, 1. Hence dF is invertible and we can apply implicit function theorem to show
that for any X close to 1, there exists z and g, close to 0 and 1, respectively, such that F (X, z, g) = 0.
˜ τ , i.e. the immersions (6.36), are graphs over the round
Assume that the transformed surfaces, M
cylinder C n+1 determined by the functions v : C n+1 × [0, T ) → R+ , z(t) ∈ Rn+2 , g(t) ∈ SO(n + 2), as
Y (y, ω, τ ) = (y, v(y, w, τ )w), or
X(x, w, t) = z(t) + λ(t)g(t)(y, v(y, w, τ )w).
(6.44)
¯
Assume also there are functions z¯(t) ∈ Rn+2 , g¯(t) ∈ SO(n + 2)/SO(n + 1) and λ(t)
∈ R+ , s.t. X(·, t) is
in
the
sense
of
the
definition
(4).
Then
Proposition
36 implies that
H 1 −close to a transformed cylinder Cλ¯
¯ zg
¯
there exists z(t) ∈ Rn+2 and g(t) ∈ SO(n + 2)/SO(n + 1), s.t. (6.39) holds. If X is a (λ, z, g)−graph over
C n+1 , i.e. it is of the form (6.44), then v satisfy
Z
v(y, w, τ )y k wj = 0, j = 1, . . . , n + 2, k = 0, 1.
(6.45)
C n+1
To apply Proposition 36 to the immersion X(·, t) : C n+1 → Rn+2 , solving (6.1), we pick z¯(t) to be a
piecewise constant function constructed iteratively, starting with z¯(t) = 0 and α
¯ (t) = e1 for 0 ≤ t ≤ δ
for δ sufficiently small (this works due to our assumption on the initial conditions), and z¯(t) = z(δ) and
α
¯ (t) = α
¯ (δ) for δ ≤ t ≤ δ + δ 0 and so forth. This gives z(t) ∈ Rn+1 and g(t) ∈ O(n + 2), s.t. (6.45) holds.
6.10
Equations for the graph function
Assume that for each t in the the existence interval, the transformed immersion Y (y, w, τ ), given in (6.36),
is a graph over the cylinder
Y (y, w, τ ) := λ−1 (t)g(t)−1 (X(x, w, t) − z(t)) = (y, v(y, w, τ )w),
(6.46)
where y and τ are blowup variables defined by (6.8). Then, similarly to the derivation of the equation
∂f
−2
(5.10) - (5.11) or (6.13), and using ∂f
and
∂t = ∂τ ∂t τ, for any function f of the variables τ , ∂t τ = λ
˙ − x0 ) − λ−1 x˙ 0 = λ−2 (ay − λ−1
∂t y = −λ−2 λ(x
∂x0
),
∂τ
MCF Lectures, October 26, 2014
44
˙ we see that (6.37) implies (cf. (6.13))
where, recall, a = −λλ,
∂τ v = G(v) − a(y∂y − 1)v − λ−1
∂g
∂z
∂x0
∂y v − g −1 v − λ−1 g −1
∂τ
∂τ
∂τ
(6.47)
Rt
∂z
be the τ -derivative of z(t(τ )), etc, with t(τ ) the inverse function of τ (t) = 0 λ−2 (s)ds, and the
where ∂τ
map G(v) given by (5.11). For v have the orthogonal decomposition (6.19) - (6.20). This together with
a(τ ) 2
(6.45) implies that φ(·, τ ) satisfy (again in L2 (R × Sn , e− 2 y dydw))
φ(·, τ ) ⊥ 1, y, y 2 , y k wj , j = 1, . . . , n + 1, k = 0, 1.
(6.48)
Remark 6. Other approaches: 1) Write a point, p, on the hypersurface Mt as say p = q + xω + u(x, ν)ν,
where q gives the shift of the cylinder axis, ω is the unit vector in the direction of the cylinder axis, ω · q = 0,
ν is the outward unit normal to the cylinder surface. To find the equation for u, we compute p˙ · νMt =
P4
˙ for some γi and for ϕ1 = cos θ, ϕ1 = cos θ, ϕ2 = sin θ, ϕ3 = x cos θ, ϕ4 = x sin θ, while
i=1 γi ϕi + γ0 u,
v(y, θ, τ ) is now decomposed as v(y, θ, τ ) = Va(τ ),b(τ ) + φ(y, θ, τ ), with φ ⊥ {1, y, y 2 }. One can also consider
a moving frame (e1 , e2 , e3 ), with e3 = ω, so that ν = e1 cos θ + e2 sin θ.
2) Write v(y, θ, τ ) = V˜a(τ ),b(τ ) +φ(y, θ, τ ), where V˜a(τ ),b(τ ) = Va(τ ),b(τ ) +β0 (τ )y+β1 (τ ) cos θ+β2 (τ ) sin θ+
β3 (τ )y cos θ + β4 (τ )y sin θ and φ ⊥ {1, y, y 2 , cos θ, sin θ, y cos θ, y sin θ}. In other words, one ”dresses” up the
main, finite dimensional term absorbing into it all the neutral-unstable modes (singular, secular behaviour).
7
Volume-preserving mean curvature flow and isoperimetric problem (to do)
8
Submanifolds (to do)
9
Ricci flow (to do)
10
Evolution of graphs (to do)
11
Local existence (to do)
A
Elements of Theory of Surfaces
In this appendix we sketch some results from the theory of surfaces in Rn+1 . For detailed expositions see
[25, 2, 5].
MCF Lectures, October 26, 2014
45
By a hypersurface S in Rn+1 , we often mean an n dimensional surface element, which is an equivalence
class of parameterized surface elements, i.e. maps ψ : U → Rn+1 , where U ⊂ Rn is open, s.t. dψ(u) is oneto-one (i.e. of the maximal rank) for every u ∈ U . Such maps are called (local) immersions. Here dψ(u) is
the linear map from Rn to Rn+1 , defined as dψ(u)ξ = ∂s |s=0 ψ(σs ), where σs is a curve in U so that σs=0 = u
n+1
s
, ψ 0 : U 0 → Rn+1 ,
and ∂σ
∂s = ξ. The equivalence relation is given as follows: two immersions, ψ : U → R
0
0
are said to be equivalent, if there is a diffeomorphism, ϕ : U → U , s.t. ψ = ψ ◦ ϕ.
We call ψ a parametrization of S. We are interested in properties of S, which are independent of a
particular parametrization, i.e. depend only on the equivalence class of parametrizations.
The tangent space, Tx S, to S at x is defined as the vector space of velocities at x of all curves in S
s
passing through x, i.e. Tx S := {η ∈ Rn+1 : ∃γs : [−, ] → S, s. t. γs=0 = x and ∂γ
∂s = η}. Let γs be as in
−1
the definition and σs := ψ (γs ). Then σs is a curve in U through the point u = ψ −1 (x) and η = dψ(u)ξ,
n
s
where ξ := ∂σ
∂s . Thus, we have Tx S = dψ(u)(R ) for x = ψ(u).
The inner product on Tx S is defined in the usual way: g(v, w) = hv, wi , ∀v, w ∈ Tx S. Here h·, ·i is the
inner product in Rn+1 . The form g(v, w) is called the metric on S or the first fundamental form of S.
Often, a surface S will be given either as a level set of some function ϕ : Rn+1 → R, i.e. S = ϕ−1 (0) =
{x ∈ Rn+1 : ϕ(x) = 0}, or as a graph of some function f : U → R. In the latter case, it is a zero level set of
the function ϕ(x) = xn+1 − f (u), x = (u, xn+1 ), and it has the parametrization ψ(u) := (u, f (u)).
A.1
Mean curvature
Let ν(x) be the unit outward normal vector to S at x. The map ν : S → S n given by ν : x → ν(x) is
called the Gauss map and the map Wx := dν(x) : Tx S → Tν(x) S n is called the Weingarten map or the shape
s
operator. Here dν(x)η = ∂s |s=0 ν(γs ), where γs is a curve on S so that γs=0 = x and ∂γ
∂s = η. We have the
following theorem.
Theorem 37. Ran(Wx ) ⊂ Tx S and Wx is self-adjoint, i.e. g(Wx ξ, η) = g(ξ, Wx η).
Proof. The statement Ran(Wx ) ⊂ Tx S is trivial, since as a little contemplation shows, Tν(x) S n = Tx S (both
subspaces are defined by the same normal vector ν(x)). It also follows from the elementary, but useful,
computation
1
hWx ξ, ν(x)i = h∂s |s=0 ν(γs ), ν(x)i = ∂s |s=0 hν(γs ), ν(γs )i = 0,
2
where γs is as above. This implies that Wx ξ ∈ Tx S.
Let ϕ = ϕst be a parametrization of a two-dimensional surface in S such that ϕst |s=t=0 = x, ∂s |s=t=0 ϕst =
ξ and ∂t |s=t=0 ϕst = η. Then
hξ, Wx ηi = h∂s |s=t=0 ϕ, ∂t |s=t=0 ν(ϕ)i = ∂t |s=t=0 h∂s ϕ, ν(ϕ)i − h∂t ∂s |s=t=0 ϕ, ν(x)i .
D 2
E
∂ ϕ
Since h∂s ϕ, ν(ϕ)i = 0, this implies hξ, Wx ηi = − ∂s∂t
(0, 0), ν(x) . Similarly we have
hWx ξ, ηi = −
which implies that hξ, Wx ηi = hWx ξ, ηi .
∂2ϕ
(0, 0), ν(x) ,
∂s∂t
MCF Lectures, October 26, 2014
46
The form hv, Wx wi is called the second fundamental form of S.
The principal curvatures are all the eigenvalues of the Weingarten map Wx . The mean curvature H(x)
is the trace of Wx , which is the sum of all principal curvatures,
H(x) = Tr Wx .
(A.1)
The Gaussian curvature is the determinant of Wx , which is the product of the principal curvatures.
If S is a level set of some function ϕ : Rn+1 → R, i.e. S = ϕ−1 (0) = {x ∈ Rn+1 : ϕ(x) = 0},
∇ϕ(x)
then ν(x) = |∇ϕ(x)|
. Note that ν(x) is defined on the entire Rn+1 and therefore we can introduce the
T
j
matrix Ax = (∂xi ν (x)). The associated linear map has the following properties: a) ATx ν = 0 and b)
Wx = Ax |Tx S . Indeed, since 0 = ∂xi ν j ν j = 2ν j ∂xi ν j , we have ATx ν = 0. Furthermore, let γs be a
curve through a point x, with the initial velocity ξ, i.e. γs (0) = x, and with ∂s γs |s=0 = ξ. We have
∂γ i
dν(x)ξ = ∂s |s=0 ν(γs ) = ∂xi ν(x) ∂ss |s=0 = ∂xi νξ i = Ax ξ. This implies that Wx = Ax |Tx S . The properties a)
and b) yield that Tr(Wx ) = Tr(Ax ) = ∂xi ν i (x) = div ν(x), which gives
∇ϕ(x)
H(x) = div ν(x) = div
.
(A.2)
|∇ϕ(x)|
If S is a graph of some function f : U → R, then it is a zero level set of the function ϕ(x) = xn+1 −
√ (u),1)
f (u), x = (u, xn+1 ), and therefore by the formula above ν(x)|x=ψ(u) = (−∇f
and
2
1+|∇f |
H(x) = − div
A.2
∇f
p
1 + |∇f |2
.
(A.3)
Matrix representation of the first and second fundamental forms
We denote by T S the collection of all tangent planes, the tangent bundle. A vector field V is a map
∂ψ
V : S → T S assigning to every x ∈ S a tangent vector Vx ∈ Tx S. Given an immersion ψ : U → S, { ∂u
i (u)}
∂ψ
is a basis in Tx S, where x = ψ(u). Varying u ∈ U , we obtain the basis, { ∂ui }, for vectors fields on S (if
∂
the immersion ψ is fixed, one writes this basis as { ∂u
i }). This allows to write vector fields in the coordinate
∂ψ
form as v = v i ∂ui . Plug the coordinate form of v and w into g(v, w) to obtain g(v, w) = gij v i wj , where
∂ψ ∂ψ
gij :=
,
.
∂ui ∂uj
∂ψ
The matrix (tensor) {gij } determines the first fundamental form of S in the basis { ∂u
i }. If S is a graph of
∂f ∂f
some function f : U → R, i.e. S can be parameterized as ψ(u) = (u, f (u)). Then gij (u) = δij + ∂u
i ∂uj .
Notation. {g ij } denotes the inverse matrix. The summation is understood over repeated indices. The
indices are raised and lowered by applying g ik or gik as in bij = g ik bkj . Unless we are dealing with the
Euclidian metric, one of the indices are upper and the other lower. (For the Euclidian metric, δij , the
position of the indices is immaterial.) The quadratic form hWx v, wi , where v, w ∈ Tx S, is called the second
fundamental form of S. We have
MCF Lectures, October 26, 2014
47
Lemma 38. For a parametrization ψ of S, consider the matrix elements of Wx in the basis
bij := h
∂ψ
∂ui ,
∂ψ
∂ψ
, Wx , j i.
i
∂u
∂u
(A.4)
Then
bij = −h
∂2ψ
, νi
∂ui ∂uj
and
H = g ij bji .
Proof. In the proof of Theorem 37, we have shown that hξ, Wx ηi = −
D
(A.5)
E
∂2ϕ
∂s∂t s=t=0 , ν(x)
, where ϕ = ϕst
is a parametrization of a two-dimensional surface in S such that ϕst |s=t=0 = x, ∂s |s=t=0 ϕst = ξ and
∂t |s=t=0 ϕst = η. Now take ϕst = ψ(u + sei + tej ), where ei are the co-ordinate unit vectors in U . Then
∂2ψ
∂2ϕ
∂s∂t s=t=0 = ∂ui ∂uj , which proves the first equality. To show the second equality in (A.5), we write H(x) =
P
∂ψ
Tr(Wx ) =
hei , Wx ei i, where {ei } is an orthonormal basis of Tx S. Let ei = µij ψj , where ψj = ∂u
j . Let
U = (µij ). Then
hei , Wx ei i = µik µil hψk , Wx ψl i = µik µil bkl = Tr(U BU T ) = Tr(U T U B),
where B = (bji ) . We have δij = hei , ej i = µik µjl gkl = (U GU T )ij , where G = (gij ). Hence U GU T = I and
therefore G−1 = U T U . So
H = Tr(G−1 B)
(A.6)
and therefore the second equation in (A.5) follows.
∂ψ
∂ψ
By (A.4) and the general formula v = g ij h ∂u
i , vi ∂uj , we have the following useful expression
Wx
∂ψ
∂ψ
= g ij bjk k .
i
∂u
∂u
(A.7)
n
n
ˆ
by the immersion ψ(u) = Rψ(u),
of radius R in Rn+1 . We can define SR
Examples. 1) The n−sphere SR
n
n
ˆ
ˆ
where ψ(u) is the immersion for the standard n−sphere S = S1 . Then ν(ψ(u)) = ψ(u) and Lemma 38
2 ˆ
ψ
ˆ = Rh ∂ ψˆ , ∂ ψˆ i = R−1 gij = Rg stand , where gij and g stand are the metrics on S n
gives bij = −Rh ∂u∂i ∂u
, ψi
ij
ij
R
∂ui ∂uj
j
and Sn = S1n , respectively. This in particular implies H = g ij bji = nR−1 .
n−1
n
n
2) The n−cylinder CR
= SR
× R of radius R in Rn+1 . We can define CR
by the immersion φ(u, x) =
n−1
ˆ
ˆ
(Rψ(u), x), where ψ(u) is the immersion for the standard (n − 1)−sphere S
= S1n−1 . Then ν(φ(u, x)) =
2 ˆ
ˆ ∂ψ
ˆ
∂
ψ
∂
ψ
−1
stand
stand
ˆ
ˆ = Rh ,
gij = Rgij
, where gij and gij
(ψ(u),
0) and Lemma 38 gives bij = −Rh ∂ui ∂uj , ψi
∂ui ∂uj i = R
n−1
are the metrics on SR
and Sn−1 = S1n−1 , respectively, for i, j = 1, . . . , n−1, and bnn = 0. This in particular
ij
implies H = g bji = (n − 1)R−1 .
A.3
Integration.
on S. Indeed, first, we define the integrals
RThe first fundamental form allows us to define Rthe integration
R
√
√
h over an immersion patche, ψ : U → S, as ψ h := U h ◦ ψ gdn u, where g := det(gij ). ( gdn u is the
ψj
MCF Lectures, October 26, 2014
48
∂ψ
infinitesimal volume spanned byR the basis
∂uj .) Then, we extend to the entire surface by using a
R
P vectors
partition of unity, say, {χj }, as S h := j ψj χj h. It is easy to check that this expression is independent
of the choice of partition of unity, {χj }, or local immersions, ψj .
If S is locally a graph, S = graph
f , of some function f : U → R, so that one can take the immersion
p
√
ψ(u) = (u, f (u)), then gdn u = 1 + |∇f |2 dn u.
Show that V (ψ) is independent of the parametrization.
A.4
Connections
The connection, ∇, is probably a single, most important notion of Differential Geometry. It generalizes the
map
n+1
V → ∇R
V
from Rn+1 to manifolds (in our case surfaces). Here V is a vector field on Rn+1 , i.e. V : Rn+1 → Rn+1 ,
n+1
n+1
and ∇R
is the directional derivative which is defined as ∇R
: T (x) 7→ ∂s |s=0 T (γs ), where γs=0 = x,
V
V
n+1
∂γs
|
=
V
and
T
is
either
a
function
or
a
vector
field.
(We
will
not use tensors.) Note that ∇R
T =
s=0
V
∂s
P i ∂T
Rn+1
W (x) ∈
/ Tx S, i.e. “ does not belong to S
i V ∂ui . However, if W is a vector field on S, in general ∇V
”. This suggests the following definition.
If f is a function on S, we define ∇V f (x) = ∇R
V
n+1
W (x) into Tx S,
as the projection of ∇R
V
n+1
f (x). If W is a vector field on S, we define ∇V W (x)
∇V W (x) := (∇R
V
n+1
W )T (x).
(A.8)
∇V is called the covariant derivative on S ⊂ Rn+1 . So in the latter case the connection ∇ : V 7→ ∇V is a
linear map from space V(S), of vector fields on S, to the space of first order differential operators on V(S).
The covariant derivative, ∇V , has the following properties:
• ∇V is linear, i.e. ∇V (αW + βZ) = α∇V W + β∇V Z;
• ∇f V +gW = f ∇V + g∇W ;
• ∇V (f W ) = f ∇V W + W ∇V f (Leibnitz rule).
Homework: show these properties hold.
∇ is called a Levi-Civita or Riemann connection if
∇V g(W, Z) = g(∇V W, Z) + g(W, ∇V Z)
(A.9)
and is called symmetric if
∇V W − ∇W V = [V, W ].
Homework: Show that our connection is a Levi-Civita symmetric connection.
Claim: for a Levi-Civita symmetric connection, one has
hZ, ∇Y Xi =
1
{X hY, Zi + Y hZ, Xi − Z hX, Y i − h[X, Z], Y i − h[Y, Z], Xi − h[X, Y ], Zi}.
2
(A.10)
MCF Lectures, October 26, 2014
49
Thus ∇ depends only on the first fundamental form gij . To prove this relation we add the permutations of
X hY, Zi = h∇X Y, Zi + hY, ∇X Zi .
Consider a vector field V on S, i.e. a map V : S → T S. So if γs is a path in S s.t. γs |s=0 = x and
∂s γs |s=0 = V (x), then
∂
|s=0 f (γs ).
∂s
V f (x) = ∇V f (x) =
∂σ
Let σ : U → S ⊂ Rn+1 be a local parametrization of S. Then { ∂u
i } is a basis in Tσ(u) S (generally not
−1
orthonormal). We can write V (x) = ∂s γs |s=0 = ∂s σ ◦ σ ◦ γs |s=0 . By the chain rule we have V (x) =
−1
P ∂σ ∂ −1
∂σ
i
◦ γs )i |s=0 . Hence we can write V = V i ∂u
= ∂((σ ∂s◦γs )i ) |s=0 .
i , where V
i ∂ui ∂s (σ
A vector field V is defined as a map V : S → T S, but is also identified with an operator V f = ∇V f .
Let f : S → R be a function on S. Then
V f (x)
=
=
∂
◦ σ −1 ◦ γs
∂s |s=0 f ◦ σ
∂(f ◦σ) ∂((σ −1 ◦γs )i )
|s=0 ,
∂ui
∂s
which by the above relations can be rewritten as
V f (x) = V i
∂f
.
∂ui
∂
We write V = V i ∂u
i . We think of {∂ui } as a basis in T S ({(∂ui )x }, as a basis in Tx S). We think of ∂ui are
∂σ
either operators acting on functions on S or as vectors ∂ui σ. (Note that the vector ∂u
i has components δij
∂
∂σ
∂
in the basis { ∂uj } and therefore the operator associated with it is δij ∂uj = ∂ui .)
Since {∂ui } is a basis, then ∇∂ui ∂uj can be expanded in it: ∇∂ui ∂uj = Γkij ∂uk , for some coefficients Γkij ,
called the Christofel symbols. If we let X = ∂ui , Y = ∂uj and Z = ∂uk in (A.10), then we find
Γkij =
1 kl ∂glj
∂gli
∂gij
g ( i +
−
).
2
∂u
∂uj
∂ul
This can be also shown directly.
Since ∇V = ∇V i ∂ui = V i ∇∂ui and ∇∂ui W j =
∇V W
∂
j
∂ui W ,
we have
= ∇V i ∂ui (W j ∂uj )
= V i ∇∂ui (W j ∂uj )
= V i [(∇∂ui W j )∂uj + W j ∇∂ui ∂uj ]
= V i (∂ui W j ∂uj + W j Γkij ∂uk ),
which gives ∇V W = V i ∇i W k ∂uk , or (∇V W )k = V i ∇i W k , where
∇i W k ≡ ∇∂ui W k :=
∂W k
+ Γkij W j .
∂ui
MCF Lectures, October 26, 2014
A.5
50
Various differential operators
We give definitions of various differential operators used in geometry.
(1) Divergence div V = Tr ∇V , where ∇V is the map W 7→ ∇W V .
(2) Gradient. The vector field grad(f ) is defined by g(grad(f ), W ) = ∇W f ∀ vector field W , where ∇f :
V 7→ ∇V f is considered as a functional on vector fields.
(3) Laplace-Beltrami ∆ := div grad.
(4) Hessian Hess(f ) := ∇2 f . Explicitly, ∇2 f (V, W ) = (V W − ∇V W )f .
Here ∇f is viewed as (0, 1) tensor and ∇V is defined on (0, 1) tensors T : T S → R as (∇V T )(W ) =
∇V (T (W )) − T (∇V W ).
Recall the notation g = det(gij ). In local coordinates, these differential operators have the following
expressions:
1
√
div V = ∇i V i = √ ∂ui ( gV j ),
g
∂f
(grad(f ))i = ∇i f = g ij j ,
∂u
∂2f
∂f
(Hess(f ))ij = ∇i ∇j f =
− Γkij k ,
∂ui ∂uj
∂u
1
∂f
√
∆f = ∇i ∇i f = √ ∂ui ( gg ij j ).
g
∂u
Exercise:
(A.11)
(A.12)
(A.13)
(A.14)
(1) Show (A.11)- (A.14); (2) Show that grad = −(div)∗ and ∆∗ = ∆.
Proof of (A.11)- (A.14).
We have, for an orthonormal basis {ei },
div V = h∇ei V, ei i = g ij ∇∂ui V, ∂uj
= g ij ∇i V k h∂uk , ∂uj i = g ij gkj ∇i V k ,
which gives
div V = ∇i V i
To show div V =
(1.11))
√
√1 ∂ui ( gV j ),
g
we use the representation det(gij ) = eTr ln(gij ) to obtain the formula (cf.
∂g
∂gij
= gg ij k .
∂uk
∂u
√
∂gij
1 ml
√1 ∂uk g = 1 ∂uk g = 1 g ij
. On the other hand, Γm
(∂um gkl
mk = 2 g
g
2g
2
∂uk
√
1 ml
1
m
m
that Γmk = 2 g ∂uk gml and therefore √g ∂uk g = Γmk . So we have
From (A.16), we have
∂ul gmk ). It follows
div V =
(A.15)
∂V i
∂V i
1
1 ∂ √ i
√
i
+ Γm
+ √ ∂ui gV i = √
( gV ).
mi V =
i
∂u
∂ui
g
g ∂ui
(A.16)
+ ∂uk gml −
MCF Lectures, October 26, 2014
51
which gives (A.11).
If grad(f ) = (grad(f ))i ∂ui , then
which gives (A.12).
Since ∇i W j ≡ ∇∂ui W j :=
∂W j
∂ui
∂f
∂uj
= ∇∂uj f = (grad(f ))i ∂ui , ∂uj = (grad(f ))i h∂ui , ∂uj i = gij (grad(f ))i ,
+ Γjik W k , we have (Hess(f ))ji =
Furthermore, (Hess(f ))ij := ∇2 f (∂ui , ∂uj ) =
∂2f
∂ui ∂uj
∂
jk ∂f
∂ui g ∂uk
∂f
+ Γjik g k` ∂u
`.
− (∇∂ui ∂uj )f , which gives (A.13).
∂f
Finally combining the previous results, we obtain ∆f = div grad(f ) = div(g ij ∂u
j ∂ui ) =
√
∂f
√1 ∂ui ( gg ij
g
∂uj ).
Proposition 39. Let σ : U → S ⊂ Rn+1 be a local parametrization of S. Then
∆S σ = −Hν.
(A.17)
Proof. We will prove it at a point x0 ∈ S. Since (A.17) is invariant under translations and rotations
(show
T
this), we can translate and rotate S so that x0 becomes a critical point in the sense that S W = graph f
∂σ ∂σ ∂f ∂f
√(−∇f,1) 2 and
and grad(f )(x0 ) = 0. Now σ(u) = (u, f (u)) and gij = ∂u
= δij + ∂u
i , ∂uj
i ∂uj . Then ν =
1+|∇f |
ν|x=x0 = (0, 1). Note that
∆S f
∂f
= ∇i (grad(f ))i = ∇i g ij ∂u
j
∂
ij ∂f
m ij ∂f
= ∂ui (g ∂uj ) + Γmi g ∂uj
∂2f
∂g ij ∂f
m ij ∂f
= g ij ∂u
i uj + ∂ui ∂uj + Γmi g
∂uj .
2
∂ f
Since gij |x=x0 = δij and therefore g ij |x=x0 = δij , we have ∆S f |x=x0 = g ij ∂u
i uj |x=x0 =
Hence
X ∂2f
n+1
Rn+1
∆S σ|x=x0 = (0,
f |x=x0 = ν|x=x0 ∆R f |x=x0 .
2 )|x=x0 = (0, 1)∆
i
i ∂u
n+1
But H(x0 ) = − divR
Rn+1
(√ ∇
f
1+∇Rn+1 f 2
)|x=x0 = − divR
n+1
∇R
n+1
f |x=x0 = −∆R
n+1
∂2f
i ∂ui 2 |x=x0 .
P
f |x=x0 , this completes the
proof of the proposition.
A.6
Submanifolds
Recall that the Weingarten map is defined as Wx V = dν(x)V , which can be written in terms of the Euclidean
n+1
connection ∇R
as
W x V = ∇R
V
n+1
ν(x).
(A.18)
For the second fundamental form, we have consequently that for ξ, η ∈ T Rn+1 ,
Wx (ξ, η) = h∇ξR
n+1
ν(x), ηi = −hν(x), ∇R
ξ
n+1
ηi.
If S is a hypersurface in an (n + 1)− dimensional Riemannian manifold (M, s), rather than in the Euclidean
space Rn+1 , we define the Weingarten map by
W x ξ = ∇M
ξ ν(x),
(A.19)
MCF Lectures, October 26, 2014
52
where ν is the outward normal vector field on S in the metric s and ∇M is a (Levi-Civita symmetric)
connection on M , the second fundamental form by
M
Wx (ξ, η) = h∇M
ξ ν(x), ηis = −hν(x), ∇ξ ηis ,
(A.20)
where we used the relation (A.9) for Levi-Civita connections, and the mean curvature, still by (A.1).
As an example, we consider the Euclidean space Rn+1 , with the conformally flat metric σ 2 dx2 , where
σ > 0. Let σ = e−ϕ and denote ∇M by ∇σ . Then we have
n+1
∇σξ η = ∇R
ξ
η − (ξϕ)η − (ηϕ)ξ + hξ, ηiσ grad ϕ,
where ξf = ∇ξ f is the the Euclidean directional derivative (application of the vector field to a function).
Moreover, let ν and ν σ be the outward normal vector fields on S in the euclidean metric and in the metric s.
n+1
Using that hξ, νiσ = σ 2 hξ, νi = 0 for ∀ξ ∈ T S, we have ν σ = σ −1 ν and Wxσ ξ = ∇σξ ν σ (x) = ∇R
(σ −1 ν) −
ξ
n+1
(ξϕ)σ −1 ν − σ −1 (νϕ)ξ. Since ∇R
(σ −1 ν) = (ξσ −1 )ν + σ −1 ∇ξR
ξ
this gives Wxσ ξ = σ −1 Wx ξ − σ −1 (νϕ)ξ, which implies
n+1
ν, ∇R
ξ
n+1
ν = Wx ξ and ξσ −1 = σ −1 ξϕ,
Wxσ ξ = σ −1 Wx ξ − (νσ −1 )ξ = σ −1 (Wx − (νϕ))ξ.
(Here as before, νf = ∇ν f .) This gives the expression for the second fundamental form bσij = h∂xi , Wx ∂xj iσ =
σh∂xi , (Wx − (νϕ)∂xj i and therefore
bσij = e−ϕ (bij − (νϕ)gij ).
(A.21)
For the mean curvature Hσ = σ −2 gij bσij , we obtain
Hσ =
n
1
(H + 2 ∇ν σ) = eϕ (H − n∇ν ϕ).
σ
σ
(A.22)
We derive (A.22) differently using that the mean curvature arises in normal variations, η = f ν, of the
surface
volume functional V (ψ) (see (1.9)). In the conformal case, the analogue of (1.9) is dVσ (ψ)η =
R
√
√
√
√
√
H
ν
· η gσ dn u, where η = f νσ . We
use that gσ = σ n g, (d g)η
= (d g)(σ −1 f ν) = Hν · σ −1 f ν
σ
U
R
R
√
√
and νσ = σ −1 ν to compute dVσ (ψ)η = U (Hν + nσ −1 ∇σ) · η gσ dn u = U (H + nσ −1 ∇σ · ν)σ −1 f gσ dn u,
which implies Hσ = σ −1 (H + nσ −1 ∇ν σ).
Our next goal is to compute the normal hessian, HessN A, of the area functional A, for a hypersurface
S immersed in a manifold M = (Rn+1 σdx2 ), the Euclidean space Rn+1 , with the conformally flat metric
σdx2 , σ = e−2ϕ . Recall that the normal hessian, HessN A, of the area functional A, for a hypersurface
S immersed in a manifold M is given by the expression (4.25). Thus to to compute HessN A, we have to
compute the trace norm |WxM |2s , the Ricci curvature RicM (ν, ν) of M evaluated on the outward normal
vector field, ν,on S and the Laplace-Beltrami operator ∆S on S in the metric induced by the metric on M .
We begin with the trace norm |Wxσ |2σ = |WxM |2s . Since by the definition |Wxσ |2σ = σ −2 g ik g jm bσij bσkm , we
have |Wxσ |2σ = σ −1 g ik g jm (bij − ∇ν ϕgij )(bkm − ∇ν ϕgkm ), which gives
|Wxσ |2σ = e2ϕ (|Wx |2 − 2∇ν ϕH + n(∇ν ϕ)2 ).
(A.23)
Furthermore, to compute Ricσ = RicM , we use standard formulae, which give (see e.g. [19, 25])
Ricσij = (n − 1)(∇i ∇j ϕ − ∇i ϕ∇j ϕ) + (∆ϕ + (n − 1)|∇ϕ|2 )δij .
(A.24)
MCF Lectures, October 26, 2014
53
(Recall that ∇i , ∇, ∆ are the operators in the euclidean metric.)
Finally, we compute the change in the Laplace-Beltrami operator. By the equation (A.14) below, we
have
√
n
∂f
∂f
1
1
√
∆σS f = √ σ ∂ui ( g σ gσij j ) = n/2 √ ∂ui (σ 2 −1 gg ij j )
g
∂u
∂u
σ
g
n − 2 −2
n − 2 −2 j
∂f
= σ −1 ∆S +
σ (∂ui σ)g ij j = σ −1 ∆S f +
σ (∇ σ)∇j f
2
∂u
2
= e2ϕ (∆S − (n − 2)(∇j ϕ)∇j )f.
(A.25)
(A.26)
(A.27)
n
In the example of the n−sphere SR
of radius R in Rn+1 considered in Subsection ??, we computed
n
stand
stand
are the metrics on SR
and Sn = S1n , respectively, which
, where gij and gij
bij = R−1 gij = Rgij
implies, in particular, that H = g ij bji = nR−1 . In this case, since ν = ω, (A.23) gives
|Wxσ |2σ = e2ϕ (nR−2 − 2nR−1 ∇ω ϕ + n(∇ω ϕ)2 ) = neϕ (R−1 − ∇ω ϕ)2 .
(A.28)
To compute Ricσ (ν σ , ν σ ), we use (A.24) and ν σ = σ −1/2 ν = σ −1/2 ω to obtain Ricσ (ν σ , ν σ ) = σ −1 Ricσ (ω, ω) =
e2ϕ [(n − 1)(∇i ∇j ϕ − ∇i ϕ∇j ϕ)ω i ω j + (∆ϕ + (n − 1)|∇ϕ|2 )].
Hence, due to (4.25), HessN A is given by
n
HessN Aσ (SR
) = −e2ϕ (R−2 ∆Sn − (n − 2)(∇j σ)∇j ) − ne2ϕ (R−1 − ∇ω ϕ)2
− e2ϕ [(n − 1) ∇i ∇j ϕω i ω j − (∇ϕ · ω)2 + |∇ϕ|2 + ∆ϕ],
(A.29)
n
HessN Aσ (SR
) = −e2ϕ R−2 (∆Sn + n) − (n − 2)(∇j σ)∇j − 2nR−1 ∇ω ϕ − (∇ω ϕ)2
+ (n − 1) ∇ω ∇ω ϕ + |∇ϕ|2 + ∆ϕ .
(A.31)
(A.30)
which gives
(A.32)
A.7
Riemann curvature tensor for hypersurfaces
A.8
Mean curvature of normal graphs over the round sphere Sn and cylinder
C n+1
Let Sˆ be a fixed convex n−dimensional hypersurface in Rn+1 . We consider hypersurfaces S given as normal
ˆ Here νˆ(ˆ
graphs θ(ˆ
x, t) = u(ˆ
x)ˆ
ν (ˆ
x), over a given hypersurface S.
x) is the outward unit normal vector to Sˆ at
ˆ
x
ˆ ∈ S. We would like to derive an expression for the mean curvature of S in terms of the graph function ρ.
To be more specific, we are interested in Sˆ being the round sphere Sn . Then a normal graph graph is
given by
θ : ω 7→ ρ(ω)ω.
(A.33)
The result is given in the next proposition.
Proposition 40. If a surface S, defined by an immersion θ : Sn → Rn+1 , is a graph (A.33) over the round
sphere Sn , with the graph function ρ : Sn → R+ , then the mean curvature, H, of S is given in terms of the
MCF Lectures, October 26, 2014
54
graph function u by the equation
H=p
−1/2
(n − ρ
−1
−3/2
∆ρ) + p
2
−1 ik j`
|∇ρ| + ρ g g ∇i ∇j ρ∇k ρ∇` ρ ,
(A.34)
where ∇ denotes covariant differentiation on Sn , with the components ∇i in some local coordinates and
p := ρ2 + |∇ρ|2 .
(A.35)
Proof. We do computations locally and therefore it is convenient to fix a point ω ∈ Sn and choose normal
coordinates at this point. Let (e1 , . . . , en ) denote an orthonormal basis of Tω Sn ⊂ Tω Rn+1 in this coordinates,
sphere
i.e. the metric on Sn at ω ∈ Sn is gij
= g sphere (ei , ej ) = δij and the Christphel symbols vanish, Γkij = 0
n
at ω ∈ S . By ∇i we denote covariant differentiation on Sn w.r.to this basis (frame). First, we compute the
metric g ij on St . To this end, we note that, for the map (A.33), one has
∇i θ = (∇i ρ)ω + uei , i = 1, . . . , n,
(A.36)
∂ψ
∂ψ
ij
where we used the relation W ∂u
i = g bjk ∂uk (see (A.7) of Appendix ??), where W is the Weigarten map,
sphere
which in our situation (i.e. bsphere
= gij
= δij and ν(ω) = ω) gives ∇i ω = ei . It follows that the
jk
Riemannian metric g induced on S has components
gij = h∇i θ, ∇j θi = ρ2 δij + ∇i ρ∇j ρ.
Recall that the symbol h·, ·i denotes the pointwise Euclidean inner product of Rn+1 . As can be easily checked
directly, the inverse matrix is
g ij = ρ−2 (δij − p−1 ∇i ρ∇j ρ).
(A.37)
It is easy to see that the following normalized vector field on S is orthogonal to all vectors (A.36) and
therefore is the outward unit normal to M is
ν = p−1/2 (ρω − ei ∇i ρ) = p−1/2 (ρω − ∇ρ).
(A.38)
Straightforward calculations show that
p1/2 ∇j ν = 2(∇j ρ)ω + ρej − ∇j ∇ρ + {· · · }ν,
where the terms in braces are easy to compute but irrelevant for what follows. One thus finds that the
matrix elements bij = h∇i θ, ∇j νi of the second fundamental form are
bij = p−1/2 (ρ2 δij + 2∇i ρ∇j ρ − ρ∇i ∇j ρ).
(A.39)
Since H = trg (h) = g ji bij , we use (A.37) and (A.39) and δji δij = n to compute:
H = p−1/2 [δji δij + 2ρ−2 δij ∇i ρ∇j ρ − ρ−2 δij ρ∇i ∇j ρ − p−1 ∇i ρ∇j ρδij − 2(ρ2 p)−1 ∇i ρ∇j ρ∇i ρ∇j ρ
+ (ρp)−1 ∇i ρ∇j ρ∇i ∇j ρ]
= p−1/2 [n + 2ρ−2 ∇i ρ∇i ρ − ρ−1 ∇i ∇i ρ − p−1 ∇i ρ∇i ρ − 2(ρ2 p)−1 |∇ρ|4 + (ρp)−1 ∇i ρ∇j ρ∇i ∇j ρ]
= p−1/2 [n + 2ρ−2 |∇ρ|2 − ρ−1 ∆ρ − p−1 |∇ρ|2 − 2(ρ2 p)−1 |∇ρ|4 + (ρp)−1 ∇i ρ∇j u∇i ∇j ρ]
= p−1/2 [n − ρ−1 ∆ρ + (ρp)−1 ∇i ρ∇j ρ∇i ∇j ρ + p−1 ρ−2 (2p|∇ρ|2 − ρ2 |∇ρ|2 − 2|∇ρ|4 )]
= p−1/2 [n − ρ−1 ∆ρ + ρ−1 p−1 ∇i ∇j ρ∇i ρ∇j ρ + p−1 |∇ρ|2 ].
which implies (A.34).
MCF Lectures, October 26, 2014
55
A different derivation is given below. (to be done)
B
Elliptic maximum principle
Consider the equation −∆f + ∇v f + b = 0 on a compact manifold, M .
Theorem 41. Assume v and b are smooth and bounded and b > 0. Then f is a constant.
Proof. By compactness of M , f should reach a maximum on M . Let x∗ be a point of the maximum. Then
∇v f (x∗ ) = 0 and ∆f (x∗ ) ≤ 0. Therefore ∆f (x∗ ) = b(x∗ ) > 0, the contradiction.
C
Some general results using the parabolic maximum principle
Theorem 42. If S0 ⊂ Bρ0 (compact and without boundary?) for some ρ0 > 0, where Bρ is the ball with
ρ2
radius ρ, center at 0, then St ⊂ B√ρ2 −2nt . In particular, (MCF) exists for time t∗ < 2n0 .
0
∂f
Proof of Theorem 42. Let f (x, t) = |x|2 + 2nt. Compute df
dt = ∂t + grad f ·
→
−
H = −Hν. We also have
k
√
∆S |x|2 = √1g ∂ui ( gg ij 2xk ∂x
∂uj )
k
√
∂xk
= 2xk ∆S xk + √2g gg ij ∂x
∂ui ∂uj
→
−
= 2x · H + 2g ij gij
→
−
= 2x · H + 2n.
So we obtain
df
dt
∂x
∂t
→
−
= 2n + 2x · H , where
= ∆S f .
d
Let f˜ = f − t for > 0, then ( dt
− ∆S )f˜ = − < 0. We claim that maxSt f˜ ≤ maxS0 f . Otherwise let t∗
˜
be the first time when maxSt f reaches a value larger than maxS0 f and f˜(x∗ ) = maxSt f˜. Then at (x∗ , t∗ ),
2
∂ f˜
df
jk ∂ f˜
dt f˜ ≤ 0, ∂ui f˜ = 0 and (∂ui ∂uj f˜) ≤ 0. Hence ∆S f˜ = (∂uj + Γiij )g jk ∂u
k = g
∂uj ∂uk ≤ 0. So dt − ∆S f ≥ 0,
contradiction! This proves our claim maxSt f˜ ≤ maxS0 f ∀ > 0. Let → 0, we have maxSt f ≤ maxS0 f ,
i.e. |x|2 ≤ ρ20 − 2nt at time t.
How the collapse take place? Some indications come from the following estimates.
Theorem 43. Let n ≥ 2 and min H > 0 on a closed surface S0 . Then under the MCF, the mean curvature
H(t) increases monotonically.
Proof. Let A be the second fundamental form of the surface. (change of notation!) The relations H =
g ij Aij , ∂t g ij = 2HAij and ∂t Aij = ∇i ∇j H − Aik Akj (see [Eckert,Regularity Theory of MCF, App B]) imply
the following differential inequality for the mean curvature
∂t H ≥ 4H +
1 3
H .
n
MCF Lectures, October 26, 2014
56
Now we prove that
H → ∞ as t → T0 ,
with T0 :=
of x,
2
( n2 Hmin
)−1 .
Indeed, let ϕ satisfy ∂t ϕ =
∂t (H − ϕ) ≥ 4(H − ϕ) +
1 3
nϕ
and ϕ|t=0 < min H|t=0 . Then, since ϕ is independent
1 3
(H − ϕ3 )
n
and
(H − ϕ)|t=0 > 0,
(C.1)
which implies
H −ϕ≥0
for
∀t ∈ [0, T ).
(C.2)
Indeed, let H = ϕ at x = x
¯, t = t¯ and H > ϕ for either x 6= x
¯, t = t¯ or t < t¯. Then H|t=t¯ has a minimum at
x=x
¯. So we obtain
Hess H(¯
x, t¯) > 0 ⇒ 4H(¯
x, t¯) = Tr Hess H(¯
x, t¯) > 0
1
⇒∂t (H − ϕ) > (H 3 − ϕ3 ) = 0 at (¯
x, t¯) ⇒ (H − ϕ)(¯
x, t)
n
grows in t
at t¯.
1
2
This implies (C.2). But ϕ(t) = Hmin (1 − n2 Hmin
t)− 2 which implies H ≥ ϕ → ∞ as t → T0 with T0 :=
2
2
−1
( n Hmin ) . Hence we have the claimed relations.
Now, by a local existence result we know that St exists on some time interval. Let [0, T ) be the maximal
time interval for the existence of the MCF. The above result implies that under the assumptions of the
theorem
2
Proposition 44. T ≤ T0 := ( n2 Hmin
)−1 < ∞.
D
D.1
Elements of spectral theory
A characterization of the essential spectrum of a Schr¨
odinger operator
In this appendix we present a result characterizing the essential spectrum of a Schr¨odinger operator in a
manner similar to the characterization of the discrete spectrum as a set of eigenvalues. We follow [?].
Theorem 45 (Schnol-Simon). Let H be a Schr¨odinger operator with a bounded potential. Then
σ(H) = closure {λ | (H − λ)ψ = 0 for ψ polynomially bounded }.
So we see that the essential spectrum also arises from solutions of the eigenvalue equation, but that
these solutions do not live in the space L2 (R3 ).
Proof. We prove only that the right hand side ⊂ σ(H), and refer the reader to [?] for a complete proof. Let
ψ be a polynomially bounded solution of (H − λ)ψ = 0. Let Cr be the box of side-length 2r centred at the
origin. Let jr be a smooth function with support contained in Cr+1 , with jr ≡ 1 on Cr , 0 ≤ jr ≤ 1, and
with supr,x,|α|≤2 |∂xα jr (x)| < ∞. Our candidate for a Weyl sequence is
wr :=
jr ψ
.
kjr ψk
MCF Lectures, October 26, 2014
57
Note that kwr k = 1. If ψ 6∈ L2 , we must have kjr ψk → ∞ as r → ∞. So for any R,
Z
Z
1
2
|wr | ≤
|ψ|2 → 0
kjr ψk2 |x|<R
|x|<R
as r → ∞. We show that
(H − λ)wr → 0.
Let F (r) =
R
Cr
|ψ|2 , which is monotonically increasing in r. We claim there is a subsequence {rn } such that
F (rn + 2)
→ 1.
F (rn − 1)
If not, then there is a > 1 and r0 > 0 such that
F (r + 3) ≥ aF (r)
for all r ≥ r0 . Thus F (r0 + 3k) ≥ ak F (r0 ) and so F (r) ≥ (const)br with b = a1/3 > 1. But the assumption
that ψ is polynomially bounded implies that F (r) ≤ (const)rN for some N , a contradiction. Now,
(H − λ)jr ψ = jr (H − λ)ψ + [−∆, jr ]ψ.
Since (H − λ)ψ = 0 and [∆, jr ] = (∆jr ) + 2∇jr · ∇, we have
(H − λ)jr ψ = (−∆jr )ψ − 2∇jr · ∇ψ.
Since |∂ α jr | is uniformly bounded,
Z
k(H − λ)jr ψk ≤ (const)
2
2
(|ψ| + |∇ψ| ) ≤ (const)
Cr+1 \Cr
So
k(H − λ)wr k ≤ C
Z
|ψ|2 .
Cr+1 Cr
F (r + 2) − F (r − 1)
F (r + 2)
≤ C(
− 1)
F (r)
F (r − 1)
and so k(H − λ)wrn k → 0. Thus {wrn } is a Weyl sequence for H and λ. D.2
Perron-Frobenius Theory
Consider a bounded operator T on the Hilbert space X = L2 (Ω).
Definition 5. An operator T is called positivity preserving/improving if and only if u ≥ 0, u 6= 0 =⇒
T u ≥ 0/T u > 0.
Note if T is positivity preserving, then T maps real functions into real functions.
Theorem 46. Let T be a bounded positive and positivity improving operator and let λ be an eigenvalue of
T with an eigenvector ϕ. Then
a) λ = kT k ⇒ λ is simple and ϕ > 0 (modulo a constant factor).
b) ϕ > 0 and kT k is an eigenvalue of T ⇒ λ is simple and λ = kT k.
MCF Lectures, October 26, 2014
58
Proof. a) Let λ = kT k, T ψ = λψ and ψ be real. Then |ψ| ± ψ ≥ 0 and therefore T (|ψ| ± ψ) > 0. The latter
inequality implies that |T ψ| ≤ T |ψ| and therefore
h|ψ|, T |ψ|i ≥ h|ψ|, |T ψ|i ≥ hψ, T ψi = λkψk2 .
Since λ = kT k = supkψk=1 hψ, T ψi, we conclude using variational calculus (see e.g. [?] or [29]) that
T |ψ| = λ|ψ|
(D.1)
i.e., |ψ| is an eigenfunction of T with the eigenvalue λ. Indeed, since λ = kT k = supkψk=1 hψ, T ψi, |ψ| is the
maximizer for this problem. Hence |ψ| satisfies the Euler-Lagrange equation T |ψ| = µ|ψ| for some µ. This
implies that µkψk2 = h|ψ|, T |ψ|i = λkψk2 and hence µ = λ. Equation (D.1) and the positivity improving
property of T imply that |ψ| > 0.
Now either ψ = ±|ψ| or |ψ| + ψ and |ψ| − ψ are nonzero. In the latter case they are eigenfunctions of T
corresponding to the eigenvalue λ : T (|ψ| ± ψ) = λ(|ψ| ± ψ). By the positivity improving property of T this
implies that |ψ| ± ψ > 0 which is impossible. Thus ψ = ±|ψ|.
If ψ1 and ψ2 are two real eigenfunctions of T with the eigenvalue λ then so is aψ1 + bψ2 for any a, b ∈ R.
By the above, either aψ1 + bψ2 > 0 or aψ1 + bψ2 < 0 ∀a, b ∈ R \ {0}, which is impossible. Thus T has a
single real eigenfunction associated with λ.
Let now ψ be a complex eigenfunction of T with the eigenvalue λ and let ψ = ψ1 + ıψ1 where ψ1 and
ψ2 are real. Then the equation T ψ = λψ becomes
T ψ1 + iT ψ2 = λψ1 + iλψ2 .
Since T ψ2 and T ψ2 and λ are real (see above) we conclude that T ψi = λψ2 , i = 1, 2, and therefore by the
above ψ2 = cψ1 for some constant c. Hence ψ = (1 + ıc)ψ1 is positive and unique modulo a constant complex
factor.
b) By a) and eigenfunction, ψ, corresponding to ν := kT k can be chosen to be positive, ψ > 0. But then
λhψ, ϕi = hψ, T ϕi = hT ψ, ϕi = νhψ, ϕi
and therefore λ = ν and ψ = cϕ.
*Question: Can the condition that kT k is an eigenvalue of T (see b) be removed?*
Now we consider the Schr¨
odinger operator H = −∆ + V (x) with a real, bounded potential V (x). The
above result allows us to obtain the following important
Theorem 47. Let H = −∆ + V (x) have an eigenvalue E0 with an eigenfunction ϕ0 (x) and let inf σ(H) be
an eigenvalue. Then
ϕ0 > 0 ⇒ E0 = inf{λ|λ ∈ σ(H)} and E0 is non-degenerate
MCF Lectures, October 26, 2014
59
and, conversely,
E0 = inf{λ|λ ∈ σ(H)} ⇒ E0 is non-degenerate and ϕ0 > 0
(modulo multiplication by a constant factor).
Proof. To simplify the exposition we assume V (x) ≤ 0 and let W (x) = −V (x) ≥ 0. For µ > sup W we have
(−∆ − W + µ)−1 = (−∆ + µ)−1
∞
X
[W (−∆ + µ)−1 ]n
(D.2)
n=0
where the series converges in norm as
kW (−∆ + µ)−1 k ≤ kW kk(−∆ + µ)−1 k ≤ kW kL∞ µ−1 < 1
by our assumption that µ > sup W = kW kL∞ . To be explicit we assume that d = 3. Then the operator
(−∆ + µ)−1 has the integral kernel
√
e− µ|x−y|
>0
4π|x − y|
while the operator W (−∆ + µ)−1 has the integral kernel
√
e− µ|x−y|
W (x)
≥ 0.
4π|x − y|
Consequently, the operator
(−∆ + µ)−1 f (x) =
1
4π
Z
√
e− µ|x−y|
f (y) dy
|x − y|
is positivity improving (f ≥ 0, f 6= 0 ⇒ (−∆ + µ)−1 f > 0) while the operator
−1
W (−∆ + µ)
1
f (x) =
4π
Z
√
e− µ|x−y|
W (x)
f (y) dy
|x − y|
is positivity preserving (f ≥ 0 ⇒ W (−∆ + µ)−1 f ≥ 0). The latter fact implies that the operators [W (−∆ +
µ)−1 ]n , n ≥ 1, are positivity preserving (prove this!) and consequently the operator
(−∆ + µ)−1 +
∞
X
[W (−∆ + µ)−1 ]n
n=1
is positivity improving (prove this!).
Thus we have shown that the operator (H + µ)−1 is positivity improving. Series (D.2) shows also that
(H + µ)−1 is bounded. Since
MCF Lectures, October 26, 2014
60
hu, (H + µ)ui ≥ (− sup W + µ)kuk2 > 0,
we conclude that the operator (H + µ)−1 is positive (as an inverse of a positive operator). Finally, k(H +
µ)−1 k = sup σ((H + µ)−1 ) = (inf σ(H) + µ)−1 is an eigenvalue by the condition of the theorem. Hence the
previous theorem applies to it. Since Hϕ0 = E0 ϕ0 ⇔ (H + µ)−1 ϕ0 = (E0 + µ)−1 ϕ0 , the theorem under
verification follows. *This paragraph needs details!*.
E
Proof of the Feynmann-Kac Formula
In this appendix we present, for the reader’s convenience, a proof of the Feynman-Kac formula ( 6.28)-( 6.29)
and the estimate (6.31).
2
Let L0 := −∂y2 + α4 y 2 − α2 and L := L0 + V where V is a multiplication operator by a function V (y, τ ),
which is bounded and Lipschitz continuous in τ . Let U (τ, σ) and U0 (τ, σ) be the propagators generated
by the operators −L and −L0 , respectively. The integral kernels of these operators will be denoted by
U (τ, σ)(x, y) and U0 (τ, σ)(x, y).
Theorem 48. The integral kernel of U (τ, σ) can be represented as
Z R
τ
U (τ, σ)(x, y) = U0 (τ, σ)(x, y) e σ V (ω0 (s)+ω(s),s)ds dµ(ω)
(E.1)
where dµ(ω) is a probability measure (more precisely, a conditional harmonic oscillator, or Ornstein-Uhlenbeck,
probability measure) on the continuous paths ω : [σ, τ ] → R with ω(σ) = ω(τ ) = 0, and the pathe ω0 (·) is
given by
2ατ
e2ασ − e2αs
− e2αs
α(σ−s) e
ω0 (s) = eα(τ −s) 2ασ
x
+
e
y.
(E.2)
e
− e2ατ
e2ατ − e2ασ
Remark 7. dµ(ω) is the Gaussian measure with mean zero and covariance (−∂s2 + α2 )−1 , normalized to 1.
The path ω0 (s) solves the boundary value problem
(−∂s2 + α2 )ω0 = 0 with ω(σ) = y and ω(τ ) = x.
(E.3)
Proof of Theorem 48. We begin with the following extension of the Ornstein-Uhlenbeck process-based FeynmanKac formula to time-dependent potentials:
Z R
τ
U (τ, σ)(x, y) = U0 (τ, σ)(x, y) e σ V (ω(s),s) ds dµxy (ω).
(E.4)
where dµxy (w) is the conditional Ornstein-Uhlenbeck probability measure, which is the normalized Gaussian
measure dµxy (ω) with mean ω0 (s) and covariance (−∂s2 + α2 )−1 , on continuous paths ω : [σ, τ ] → R with
ω(σ) = y and ω(τ ) = x (see e.g. [?, ?, ?]). This formula can be proven in the same way as the one for time
independent potentials (see [?], Equation (3.2.8)), i.e. by using the Kato-Trotter formula and evaluation of
Gaussian measures on cylindrical sets. Since its proof contains a slight technical wrinkle, for the reader’s
convenience we present it below.
MCF Lectures, October 26, 2014
61
Now changing the variable of integration in (E.4) as ω = ω0 + ω
˜ , where ω
˜ (s) is a continuous
path with
R
boundary
conditions
ω
˜
(σ)
=
ω
˜
(τ
)
=
0,
using
the
translational
change
of
variables
formula
f
(ω)
dµ
xy (ω) =
R
f (ω0 + ω
˜ ) dµ(˜
ω ), which can be proven by taking f (ω) = eihω,ζi and using (E.3) (see [?], Equation (9.1.27))
and omitting the tilde over ω we arrive at (E.1).
There are at least three standard ways to prove (E.4): by using the Kato-Trotter formula, by expanding
both sides of the equation in V and comparing the resulting series term by term and by using Ito’s calculus
(see [?, ?, 28, ?]). The first two proofs are elementary but involve tedious estimates while the third proof is
based on a fair amount of stochastic calculus. For the reader’s convenience, we present the first elementary
proof of (E.4).
Lemma 49. Equation (E.4) holds.
Proof. In order to simplify our notation, in the proof that follows we assume, without losing generality,
that σ = 0. We divide the proof into two parts. First we prove that for any fixed ξ ∈ C0∞ the following
Kato-Trotter type formula holds
Y
U (τ, 0)ξ = lim
n→∞
0≤k≤n−1
k+1 k
τ, τ )e
U0 (
n
n
R
(k+1)τ
n
V (y,s)ds
kτ
n
ξ
(E.5)
in the L2 space. We start with the formula
k+1 k
τ, τ )e
U0 (
n
n
Y
U (τ, 0) −
0≤k≤n−1
=
Y
0≤k≤n−1
=
X
k+1 k
U(
τ, τ ) −
n
n
Y
U0 (
0≤j≤n j≤k≤n−1
R
(k+1)τ
n
kτ
n
k+1 k
U0 (
τ, τ )e
n
n
Y
0≤k≤n−1
R (k+1)τ
n
k+1 k
τ, τ )e
n
n
V (y,s)ds
kτ
n
V (y,s)ds
R
(k+1)τ
n
kτ
n
V (y,s)ds
j
Aj U ( τ, 0)
n
with the operator
j+1 j
Aj := U0 (
τ, τ )e
n
n
R
(j+1)τ
n
jτ
n
V (y,s)ds
− U(
j+1 j
τ, τ ).
n
n
We observe that kU0 (τ, σ)kL2 →L2 ≤ 1, and moreover by the boundness of V, the operator U (τ, σ) is
uniformly bounded in τ and σ in any compact set. Consequently
Y
k[U (τ, 0) −
0≤k≤n−1
≤ max nk
j
Y
j≤k≤n−1
k+1 k
τ, τ )e
U0 (
n
n
k+1 k
U0 (
τ, τ )e
n
n
j
. max nkAj U ( τ, 0)ξk2 .
j
n
R
R
(k+1)τ
n
kτ
n
(k+1)τ
n
kτ
n
V (y,s)ds
V (y,s)ds
]ξk2
j
Aj U ( τ, 0)ξk2
n
(E.6)
MCF Lectures, October 26, 2014
62
We show below that
j
max nkAj U ( τ, 0)ξk2 → 0,
j
n
(E.7)
as n → ∞. Equations (E.7), ( E.6), ( E.13) and imply ( E.5). This completes the first step.
In the second step we compute the integral kernel, Gn (x, y), of the operator
Gn :=
Y
0≤k≤n−1
k+1 k
U0 (
τ, τ )e
n
n
R
(k+1)τ
n
kτ
n
V (·,s)ds
in ( E.5). By the definition, Gn (x, y) can be written as
Z
Gn (x, y) =
Z
···
R
Y
U nτ (xk+1 , xk )e
(k+1)τ
n
kτ
n
V (xk ,s)ds
dx1 · · · dxn−1
(E.8)
0≤k≤n−1
with xn := x, x0 := y and Uτ (x, y) ≡ U0 (0, τ )(x, y) is the integral kernel of the operator U0 (τ, 0) = e−L0 τ .
We rewrite (E.8) as
Pn−1 R
Z
Gn (x, y) = Uτ (x, y)
where
k=0
e
Q
0≤k≤n−1
dµn (x1 , . . . , xn ) :=
Since Gn (x, y)|V =0 = Uτ (x, y) we have that
interval in R. Define a cylindrical set
(k+1)τ
n
kτ
n
V (xk ,s) ds
U nτ (xk+1 , xk )
Uτ (x, y)
R
dµn (x1 , . . . , xn ),
(E.9)
dx1 . . . dxk−1 .
dµn (x1 , . . . , xn ) = 1. Let ∆ := ∆1 × . . . × ∆n , where ∆j is an
n
P∆
:= {ω : [0, τ ] → R | ω(0) = y, ω(τ ) = x, ω(kτ /n) ∈ ∆k , 1 ≤ k ≤ n − 1}.
R
n
) = ∆ dµn (x1 , . . . , xn ). Thus, we can rewrite
By the definition of the measure dµxy (ω), we have µxy (P∆
(E.9) as
Z Pn−1 R (k+1)τ
n
V (ω( kτ
n ),s) ds
k=0
kτ
n
Gn (x, y) = Uτ (x, y) e
dµxy (ω),
(E.10)
By the dominated convergence theorem the integral on the right hand side of (E.10) converges in the sense
of distributions as n → ∞ to the integral on the right hand side of (E.4). Since the left hand side of (E.10)
converges to the left hand side of (E.4), also in the sense of distributions (which follows from the fact that
Gn converges in the operator norm on L2 to U (τ, σ)), (E.4) follows.
Now we prove (E.7). We have
j
k 1
j 1
j
max nkAj U ( τ, 0)ξk2 . n max kAj + K( τ, τ )kL2 →L2 + max nkK( τ, τ )U ( τ, 0)ξk2 ,
j
j
j
n
n n
n n
n
where the operator K(σ, δ) is defined as
Z δ
Z
K(σ, δ) :=
U0 (σ + δ, σ + s)V (σ + s, ·)U0 (σ + s, σ)ds − U0 (σ + δ, σ)
0
0
(E.11)
δ
V (σ + s, ·)ds
(E.12)
MCF Lectures, October 26, 2014
63
Now we claim that
k 1
1
kAj + K( τ, τ )kL2 →L2 . 2 ,
n n
n
(E.13)
1
sup k K(σ, δ)U (σ, 0)ξk2 → 0.
δ
(E.14)
and
0≤σ≤τ
We begin with proving the first estimate. By Duhamel’s principle we have
Z j+1
n τ
j+1 j
j+1
j+1 j
j
U(
U0 (
τ, τ ) = U0 (
τ, τ ) +
τ, s)V (y, s)U (s, τ )ds.
j
n
n
n
n
n
n
nτ
Iterating this equation on U (s, nk τ ) and using the fact that U (s, t) is uniformly bounded if s, t is on a compact
set, we obtain
Z n1 τ
j+1
j+1 j
j
1
j+1 j
kU (
U0 (
τ, τ ) − U0 (
τ, τ ) −
τ, s)V (y, s)U0 (s, τ )dskL2 →L2 . 2 .
n
n
n
n
n
n
n
0
R
On the other hand we expand e
j+1 j
kU0 (
τ, τ )e
n
n
R
(j+1)τ
n
jτ
n
(j+1)τ
n
jτ
n
V (y,s)ds
V (y,s)ds
and use the fact that V is bounded to get
j+1 j
j+1 j
τ, τ ) − U0 (
τ, τ )
− U0 (
n
n
n
n
Z
(j+1)τ
n
jτ
n
V (y, s)dskL2 →L2 .
1
.
n2
By the definition of K and Aj we complete the proof of ( E.13). A proof of (E.14) is given in the lemma
that follows.
Lemma 50. For any σ ∈ [0, τ ] and ξ ∈ C0∞ we have, as δ → 0+ ,
1
sup k K(σ, δ)U (σ, 0)ξk2 → 0.
δ
(E.15)
0≤σ≤τ
Proof. If the potential term, V , is independent of τ , then the proof is standard (see, e.g. [28]). We use the
property that the function V is Lipschitz continuous in time τ to prove ( E.15). The operator K can be
further decomposed as
K(σ, δ) = K1 (σ, δ) + K2 (σ, δ)
with
Z
K1 (σ, δ) :=
δ
U0 (σ + δ, σ + s)V (σ, ·)U0 (σ + s, σ)ds − δU0 (σ + δ, σ)V (σ, ·)
0
and
Z
K2 (σ, δ) :=
δ
Z
U0 (σ + δ, σ + s)[V (σ + s, ·) − V (σ, ·)]U0 (σ + s, σ)ds − U0 (σ + δ, σ)
0
δ
[V (σ + s, ·) − V (σ, ·)]ds.
0
Since U0 (τ, σ) are uniformly L2 -bounded and V is bounded, we have U (τ, σ) is uniformly L2 -bounded.
This together with the fact that the function V (τ, y) is Lipschitz continuous in τ implies that
Z δ
kK2 (σ, δ)kL2 →L2 . 2
sds = δ 2 .
0
MCF Lectures, October 26, 2014
64
We rewrite K1 (σ, δ) as
Z
δ
U0 (σ + δ, σ + s){V (σ, ·)[U0 (σ + s, σ) − 1] − [U0 (σ + s, σ) − 1]V (σ, ·)}ds.
K1 (σ, δ) =
0
Let ξ(σ) = U (σ, 0)ξ. We claim that for a fixed σ ∈ [0, τ ],
kK1 (σ, δ)ξ(σ)k2 = o(δ).
(E.16)
Indeed, the fact ξ0 ∈ C0∞ implies that L0 ξ(σ), L0 V (σ)ξ(σ) ∈ L2 . Consequently (see [?])
lim
s→0+
(U0 (σ + s, σ) − 1)g
→ L0 g,
s
for g = ξ(σ) or V (σ, y)ξ(σ) which implies our claim. Since the set of functions {ξ(σ)|σ ∈ [0, τ ]} ⊂ L0 L2 is
compact and k 1δ K1 (σ, δ)kL2 →L2 is uniformly bounded, we have (E.16) as δ → 0 uniformly in σ ∈ [0, τ ].
Collecting the estimates on the operators Ki , i = 1, 2, we arrive at ( E.15).
Note that on the level of finite dimensional approximations the change of variables formula can be derived
as follows. It is tedious, but not hard, to prove that
Y
Un (xk+1 , xk ) = e
−α
(x−e−ατ y)2
2(1−e−2ατ )
0≤k≤n−1
Y
Un (yk+1 , yk )
0≤k≤n−1
with yk := xk − ω0 ( nk τ ). By the definition of ω0 (s) and the relations x0 = y and xn = x we have
Gn (x, y) = Uτ (x, y)G(1)
n (x, y)
(E.17)
where
G(1)
n (x, y)
1
√
:=
4π α(1 − e−2ατ )
Z
Z
···
Y
R
Un (yk+1 , yk )e
(k+1)τ
n
kτ
n
V (yk +ω0 ( kτ
n ),s)ds
dy1 · · · dyk−1 . (E.18)
0≤k≤n−1
Since lim Gn ξ exists by ( E.15), we have lim G(1)
n ξ (in the weak limit) exists also. As shown in [?],
n→∞ Z
n→∞
Rτ
lim G(1)
e 0 V (ω0 (s)+ω(s),s)ds dµ(ω) with dµ being the (conditional) Ornstein-Uhlenbeck measure on
n =
n→∞
the set of path from 0 to 0. This completes the derivation of the change of variables formula.
Remark 8. In fact, Equations (E.5), (E.17) and (E.18) suffice to prove the estimate in Corollary 33.
References
[1] D. C. Antonopoulou, G. D. Karali and I. M. Sigal, Stability of spheres under volume preserving mean
curvature flow, Dynamics of PDE, 7 pp. 327–344, 2010.
[2] M. P. do Carmo, Riemann Geometry. Birkh¨auser 1992.
MCF Lectures, October 26, 2014
65
[3] T. H. Colding, W. P. Minicozzi II, Generic mean curvature flow I: generic singularities, Annals of Math.,
175, pp. 755–833, 2012.
[4] T. H. Colding, W. P. Minicozzi II, Minimal surfaces and mean curvature flow, arXiv:1102.1411, 2011.
[5] B. A. Dubrovin, A. T. Fomenko, S. P. Novikov, Modern Geometry, Springer, 1992.
[6] K. Ecker, Regularity Theory for Mean Curvature Flow. Birkh¨auser, 2004.
[7] K. Ecker and G. Huisken. Mean curvature evolution of entire graphs. Ann. of Math. (2), 130(3):453–471,
1989.
[8] K. Ecker and G. Huisken. Interior estimates for hypersurfaces moving by mean curvature. Invent. Math.,
105(3):547–569, 1991.
[9] Zhou Gang, I. M. Sigal, Neckpinching under the mean curvature flow, Journal of Geometric Analysis,
19(1):36–80, 2009.
[10] Zhou Gang, D. Knopf, I. M. Sigal, Neckpinch dynamics for asymmetric surfaces evolving by mean
curvature flow, arXiv 2012.
[11] Mi-Ho Giga, Yoshikazu Giga, J¨
urgen Saal, Nonlinear Partial Differential Equations Springer 2010.
[12] S. Gustafson and I.M. Sigal, The stability of magnetic vortices, Comm. Math. Phys. 210 No.2 (2000),
257-276.
[13] S. Gustafson and I.M. Sigal, Mathematical Concepts of Quantum Mechanics Springer, 3rd edition, 2011.
[14] G. Huisken. Flow by mean curvature of convex surfaces into spheres. J. Differential Geom., 20(1):237–
266, 1984.
[15] G. Huisken. Contracting convex hypersurfaces in Riemannian manifolds by their mean curvature. Invent.
Math., 84(3):463–480, 1986.
[16] G. Huisken. Asymptotic behavior for singularities of the mean curvature flow. J. Differential Geom.,
31(1):285–299, 1990.
[17] G. Huisken. Local and global behaviour of hypersurfaces moving by mean curvature. In Differential
geometry: partial differential equations on manifolds (Los Angeles, CA, 1990), volume 54 of Proc. Sympos.
Pure Math., pages 175–191. Amer. Math. Soc., Providence, RI, 1993.
[18] G. Huisken. Lecture two: singularities of the mean curvature flow. In Tsing Hua lectures on geometry
and analysis (Hsinchu, 1990–1991), pages 125–130. Int. Press, Cambridge, MA, 1997.
[19] G. Huisken and A. Polden, Geometric evolution equations for hypersurfaces.
[20] G. Huisken and C. Sinestrari. Convexity estimates of mean curvature flow and singularities for mean
convex surfaces. Acta Math., 183:45–70, 1999.
[21] G. Huisken and C. Sinestrari. Mean curvature flow singularities for mean convex surfaces. Calc. Var.
Partial Differential Equations, 8(1):1–14, 1999.
MCF Lectures, October 26, 2014
66
[22] G. Huisken and C. Sinestrari. Mean curvature flow with surgeries of two-convex hypersurfaces. Invent.
Math., 175(1):137–221, 2009.
[23] J. Jost, Geometric Analysis.
[24] W. Kong and I. M. Sigal, Stability of spherical collapse under mean curvature flow, Arxiv, 2012.
[25] W. K¨
uhnel, Differential Geometry. Curves-Surfaces-manifolds . AMS, 2006.
[26] C. Mantegazza, Lecture Notes on Mean Curvature Flow. Birkh¨auser, 2011.
[27] Yu. N. Ovchinnikov and I.M. Sigal, Ginzburg-Landau equation I. Static vortices. Part. Diff. Eqn and
their Appl. CRM Proceedings 12 (1997), 199-220.
[28] Reed and Simon, Method of Mathematical Physics, II, Academic Press.
[29] I.M. Sigal, Lecture Notes on Applied Partial Differential Equations, 2012.
[30] E. M. Stein and G. Weiss. Introduction to Fourier Analysis on Euclidean Spaces. Princeton University
Press, Princeton, N.J., 1971. Princeton Mathematical Series, No. 32.