On a phase field problem driven by interface area and

interface curvature ∗
Xiaofeng Ren †‡ Juncheng Wei §
Department of Mathematics and Statistics Department of Mathematics
The George Washington University Chinese University of Hong Kong
Washington, DC 20052, USA Hong Kong, PRC

Abstract
A two component system driven by both interface area and interface curvature is studied
with a new phase field model. We show that if the curvature impact in the system is strong
enough, there exist bubble profiles. A bubble profile describes a pattern of an inner core of one
component surround by an outer membrane of the other component. It is a radial solution to a
fourth order nonlinear PDE. We show the existence of such profiles in all dimensions, although
the profile is unstable if the dimension is greater than two.

1 Introduction
The bending energy plays a central role in the study of vesicle membranes formed by certain am-
phiphilic molecules [4, 5]. In the isotropic case it may be expressed as a surface integral [13, 22]
Z
Eb = {a1 + a2 (κ − c0 )2 + a3 G} ds. (1.1)
Γ

Here Γ is a closed surface in R3 representing a vesicle membrane, κ is the mean curvature of the
surface, and G is the Gauss curvature of the surface. The constant a1 represents the surface tension
caused by the interaction effects between the vesicle material and the ambient fluid; a2 is the bending
rigidity and a3 is the stretching rigidity, both of which are determined by the interaction properties
of the amphiphilic molecules. The last constant c0 is the spontaneous curvature describing an
asymmetric effect.
In (1.1) the integral of the first term a1 leads to the area of the surface Γ. The integral of the
third term a3 G gives a topological invariant due to the Gauss-Bonnet Theorem. We may therefore
ignore the third quantity. The most interesting part in (1.1) is the second term. In the case c0 = 0,
it is equation to a2 times Z
κ2 ds. (1.2)
Γ
∗ Abbreviated title: Infterface area and interface curvature
† Corresponding author. Phone: 1 435 797-0755; Fax: 1 435 797-1822; E-mail: ren@math.usu.edu
‡ Supported in part by NSF grant DMS-0509725, DMS-0754066.
§ Supported in part by an Earmarked Grant of RGC of Hong Kong.

1
The integral (1.2) is known as the Willmore functional [25].
The functional (1.1) may be studied by a diffusive interface method [9, 10]. To explain this ap-
proach, let us first recall the Allen-Cahn equation [1] that is often used to study the phase separation
phenomenon in condensed matter physics. It is a second order nonlinear parabolic equation,
ut (x, t) = ǫ2 ∆u(x, t) − f (u(x, t)), x ∈ D ⊂ Rn , t > 0, (1.3)
with the Neumann boundary condition on ∂D. The parameter ǫ is positive and small. The nonlinear
function f is a balanced cubic function, such as f (u) = u(u − 1/2)(u − 1). It can be viewed as the
negative gradient flow, in L2 space, of the free energy functional
ǫ2
Z
Iac (u) = [ |∇u|2 + F (u)] dx, (1.4)
D 2
Ru
where F (u) = 0 f (u) du is the anti-derivative of f . If f (u) = u(u − 1/2)(u − 1), then F (u) =
1 2 2
4 u (1 − u) . A steady state u = u(x) of (1.3), i.e. a critical point of (1.4), is a solution of

∂u(x)
−ǫ2 ∆u(x) + f (u(x)) = 0, if x ∈ D; = 0, if x ∈ ∂D (1.5)
∂ν
where ν is the outward normal vector to ∂D.
The free energy (1.4) models a two component system whose conformation solely depends on the
area of the interfaces separating the two components. If u(x) is close to 0, then the first component
occupies x; if u(x) is close to 1, then the second component occupies x. The interfaces separating
the two components are the regions where u(x) is somewhat greater than 0 and less than 1. Given
a configuration u(x) with x in an interface region, we may roughly interpret −ǫ2 ∆u + f (u) as the
mean curvature of the interface at x. The equation (1.5) then states that at an equilibrium state,
the mean curvature of the interface must be everywhere equal to 0.
In such an interface area driven system, it is difficult for the two components to co-exist. Casten
and Holland [3] (and Matano [15] independently) showed that when D is bounded and convex, any
non-constant solution of (1.5) must be unstable. More recently in the study of polymer blends (see
Tang and Freed [24]) an additional molecular weight dependent curvature term is found to contribute
to the free energy. In this case one observes two immiscible homopolymers, one forming an outer
membrane and the other constituting an inner core.
This morphology pattern may be explained phenomenologically through a very simple model
motivated by (1.1). Let us consider the situation in two dimensions. Suppose that the two compo-
nents are separated by a closed curve Γ in R2 . We propose that the free energy Ic of the system is
given by Z Z
2
Ic (Γ) = κ ds + γ ds, (1.6)
Γ Γ
where s is the length element, κ is the curvature and γ > 0 is a parameter. If we assume that Γ is
a circle of radius ρ, then the curvature is everywhere ρ1 and (1.6) becomes
2π
+ 2πγρ, (1.7)
ρ
A stable configuration is obtained by minimizing (1.7) with respect to ρ. One finds that
1
ρ= √ . (1.8)
γ

2
 In this paper we study a more sophisticated phase field version of (1.6). As in the Allen-Cahn
approach we let u be the phase field variable of a two component system. Again u(x) ≈ 0 means
that x is taken by one component; u(x) ≈ 1 means that x is taken by the other component. The
free energy of the system is now
1 1
Z Z
2
I(u) = |∆u − f (u)| dx + γ [ |∇u|2 + F (u)] dx. (1.9)
2 D D 2

Here ∆u − f (u) plays the role of curvature and 12 |∇u|2 + F (u) plays the role of length element. The
constant 1/2 in front of the first integral is put there merely for simplicity.
We will study (1.9) in the general case of n dimensions, i.e. D ⊂ Rn with n being a positive
integer. Although in (1.6) we have assumed that γ is positive, here we allow γ to be negative if
n ≥ 3. In this paper we are only interested in the situation where (1.9) is sufficiently different from
(1.4), so we assume that |γ| is small.
The functional (1.9) is a phase field version of (1.6) in n dimensions. As (1.6) is generalized to a
problem in Rn , Γ becomes an n − 1 dimensional hyper-surface, κ the mean curvature of the surface
and ds the surface element. In the n = 2 and n ≥ 4 cases we are guided by the simple problem
(1.6). However when n = 1 or n = 3, (1.6) does not yield much useful information. We must rely
on (1.9) only and carry out some careful analysis.
The Euler-Lagrange equation of (1.9) is a fourth order partial differential equation

∆(∆u − f (u)) − f ′ (u)(∆u − f (u)) − γ(∆u − f (u)) = 0. (1.10)

If D has a boundary, then we have the Neumann boundary conditions

∂u ∂(∆u − f (u))
= = 0 on ∂D. (1.11)
∂ν ∂ν
If we introduce a new variable v = ∆u − f (u), then (1.10) may be written as a system

∆u − f (u) − v = 0, ∆v − f ′ (u)v − γv = 0 in D. (1.12)

If D has a boundary, then u and v both should satisfy the Neumann boundary condition there.
In this paper we study the outer membrane/inner core pattern mentioned earlier using (1.9).
More specifically we seek radially symmetric solutions of (1.10). The domain D is the entire space
Rn . We require that the solutions u = u(|x|) = u(r) satisfy the conditions

u(0) > 0, lim u(r) = 0. (1.13)

r→∞

We often call such a solution a bubble profile. Recall that γ is either positive or negative, but |γ| is
sufficiently small. This means that the curvature term in the free energy (1.9) is significant and the
problem is very different from the Allen-Cahn problem (1.4). Note that the Allen-Cahn problem
does not have a bubble profile solution.
Our main results are the existence of bubble profiles in 1-dimension and 2-dimensions if γ is
positive and sufficiently small, and the existence of bubble profiles in n = 3 and n ≥ 4 dimensions
if γ is negative and sufficiently close to 0.
These results are proved by the so-called localized energy method which is a combination of
the Liapunov-Schmidt reduction argument and variational techniques. Let ρ be the location of the

3
interface of bubble profile u in the sense that u(ρ) = 1/2. Near ρ, u has a rather particular shape.
This shape is mostly described by a function H given in (2.1). When r is much less than ρ, u(r) is
close to 1; when r is much larger than ρ, u(r) is close to 0. Much of our paper is devoted to locating
ρ. We will see that as γ → 0, ρ → ∞. The construction is divided into two steps: in the first step,
we fix ρ large and solve a nonlinear problem with an orthogonal condition. In the next step, we
locate ρ by finding a critical point for a reduced energy function involving ρ only. For the localized
energy method used in other problems, see [2, 6, 7, 11, 12,R 21, 20].
R There is a well know relationship between Iac and Γ
ds, i.e. Ic without the curvature part:
Γ
ds is the Gamma-limit of Iac to as ǫ → 0. See De Giorgi [8], Modica and Mortola [17], Modica
[16], Kohn and Sternberg [14], etc, for this theory. We do not know R if a Gamma-convergence type
theory between I and Ic is available. The curvature part of Ic , i.e. Γ κ2 ds, the Willmore functional
[25] (also see Simon [23]). There are some partial results regarding the convergence of
Z
[−ǫ2 ∆u + f (u)]2 dx (1.14)
D

to the Willmore functional (see Moser [18]).

Our paper is organized as follows. In Sections 2 and 3 we show the existence of the two dimen-
sional bubble profile when γ is positive and small. In section 4 we show the existence of the bubble
profile in the n ≥ 4 case; the n = 1 case is studied in Section 5 and the n = 3 case in Section 5.
Certain steps in these proofs are similar so we omit some details in the latter cases. We include a
section to discuss the stability of the solutions. A technical estimate is proved in the appendix.
The function F in (1.9) is assumed to be smooth. F (0) = F (1) = 0 and F (u) > 0 if u 6= 0 and
u 6= 1. We also assume that F ′′ (0) = F ′′ (1). As |u| → ∞, F (u) → ∞.
We often encounter quantities that depend on ρ, the radius of a bubble. For instance we have
a family of approximate solutions w that depend on both r, the radial coordinate, and ρ. We often
write w(r; ρ) to emphasize w’s dependence on r and ρ. However when the dependence on ρ is less
important in a piece of argument, we will write w(r) instead.
We use L2r (Rn ) and Hr4 (Rn ) to denote the subspaces of radial functions in L2 (Rn ) and H 4 (Rn )
respectively. The inner product in L2 (Rn ) is denoted by h, i. We write φ ⊥ ψ if hφ, ψi = 0. We use
k · k∞ to denote the L∞ -norm and k · k2 the L2 -norm of a function.

2 n=2
In two dimensions, the phase field problem (1.9) is consistent with the simple model (1.6).

Theorem 2.1 When γ is positive and sufficiently small, there exists a bubble profile. The radius of
the bubble is √12γ + o(γ −1/2 ).

We let H = H(y) be the solution of the following ODE on (−∞, ∞).

H ′′ − f (H) = 0, y ∈ (−∞, ∞), lim H(y) = 1, lim H(y) = 0, H(0) = 1/2. (2.1)
y→−∞ y→∞

We define a positive constant τ by Z

τ= (H ′ (y))2 dy. (2.2)
R

4
Because H has a first integral 12 (H ′ )2 − F (H) = 0 by (2.1), we also have
Z 1 p
τ= 2F (s) ds. (2.3)
0

The derivative of H, H ′ (y), decays to 0 exponentially fast as |y| → ∞. More precisely we have a > 0
and k < 0 such that
H ′ (y) = ke−a|y| + O(e−2a|y| ). (2.4)
Necessarily
a2 = f ′ (0) = f ′ (1). (2.5)
In the special case f (H) = H(H − 1/2)(H − 1)
1 x 1 1
H(y) = [tanh(− √ ) + 1], and a = √ , k = − √ . (2.6)
2 2 2 2 2
For each
1 2
ρ∈( √ ,√ ) (2.7)
2 2γ 2γ
we construct an approximate solution w to (1.10) of the form

w(r; ρ) = H(r − ρ) + β(r; ρ) (2.8)

where
β(r; ρ) = c1,ρ e−ar + c2,ρ re−ar . (2.9)
The constants c1,ρ and c2,ρ are so chosen that w′ (0) = w′′′ (0) = 0. More explicitly

H ′ (−ρ) 1 H ′′′ (−ρ)

c1,ρ = + (H ′ (−ρ) − ), (2.10)
a 2a a2
1 ′ H ′′′ (−ρ)
c2,ρ = (H (−ρ) − ). (2.11)
2 a2
Note that
H ′ (−ρ)
c1,ρ = + O(e−2aρ ), c2,ρ = O(e−2aρ ). (2.12)
a
In this paper a weighted L∞ -norm is of particular importance. Define

kζk∗ = sup |ζ(r)|eµ|r−ρ| (2.13)

r≥0

where µ is a small positive number. This µ is independent of γ and ρ. How small µ should be will
become clear later.
We denote the left side of (1.10) by S(u), i.e.

S(u) = (∆ − f ′ (u) − γ)[∆u − f (u)] (2.14)

where ∆ − f ′ (u) − γ is viewed as an operator. The equation (1.10) becomes S(u) = 0.

Lemma 2.2 kS(w)k∗ = O(γ).

5
 Proof. We start with an estimate of ∆w − f (w). Calculations show that

H ′ + β′
∆w − f (w) = β ′′ + − f (w) + f (H).
r
We consider two cases of r: r ∈ (0, θρ) and r ∈ (θρ, ∞) where θ ∈ (0, 1). In the first case

β ′′ − f (w) + f (H) = O(e−aρ ),

and
H ′ + β′ H ′ (r − ρ) − H ′ (−ρ) β ′ (r) − β ′ (0)
= +
r r r
−(1−θ)aρ −aρ
= O(e ) + O(e )

by the mean value theorem, the decay rates of H ′′ (y) on y ∈ (−∞, −(1 − θ)ρ) and the fact that
kβk∞ = O(e−aρ ). Hence
∆w − f (w) = O(e−(1−θ)aρ ). (2.15)
Consequently
(∆w(r) − f (w(r)))eµ|r−ρ| = O(e(−(1−θ)a+µ)ρ ). (2.16)
In the second case

(β ′′ − f (w) + f (H))eµ|r−ρ| = O(e−aρ (r + 1)e−ar eµ|r−ρ| )

= O(e−aρ (r + 1)e−ar eµ(r+ρ) ) = O(e−(a−µ)ρ )

and
H ′ (r − ρ) + β ′ (r) µ|r−ρ| 1
e = O( ).
r ρ
Therefore we deduce that
1
k∆w − f (w)k∗ = O( ) = O(γ 0.5 ). (2.17)
ρ
If we write S(w) as ∆z − f ′ (w)z − γz with z = ∆w − f (w), then (2.17) implies that

kγzk∗ = O(γ 1.5 ). (2.18)

z′
The term ∆z is further broken into z ′′ and r for which

H ′ + β ′ ′′
z ′′ = β (4) − (f (w) − f (H))′′ + ( )
r
and
z′ 1 H ′ + β ′ ′ 1 ′′′
= ( ) + (β − (f (H + β) − f (H))′ ).
r r r r
We again consider two cases of r. If r ∈ (0, θρ), then

H ′ + β ′ ′′ µ|r−ρ|
z ′′ (r)eµ|r−ρ| = ( ) e + O(e−(a−µ)ρ ).
r

6
Since β ′′′ − (f (H + β) − f (H))′ is 0 at r = 0, according to the mean value theorem,
1 ′′′
(β − (f (H + β) − f (H))′ ) = O(e−aρ ).
r
Therefore we have, on (0, θρ),
H ′ + β ′ ′′ 1 H ′ + β ′ ′ µ|r−ρ|
[∆z − ( ) − ( ) ]e = O(e−(a−µ)ρ ).
r r r
Note that
H ′ + β ′ ′′ 1 H ′ + β ′ ′ H ′′′ + β ′′′ H ′ + β ′ − rH ′′ − rβ ′′
( ) + ( ) = +
r r r r r3
Together with (2.16) we find
H ′′′ + β ′′′ H ′ + β ′ − rH ′′ − rβ ′′ µ|r−ρ|
[∆z − f ′ (w)z − − ]e = O(e(−(1−θ)a+µ)ρ ). (2.19)
r r3
Since H ′′′ + β ′′′ is 0 at r = 0, the mean value theorem implies
H ′′′ + β ′′′ µ|r−ρ|
e = O(e(−(1−θ)a+µ)ρ ).
r
Using Taylor expansions shows that
H ′ + β ′ − rH ′′ − rβ ′′
| | ≤ C supr∈(0,θρ) (|H (4) (r − ρ)| + |β (4) (r)|),
r3
which implies
H ′ + β ′ − rH ′′ − rβ ′′ µ|r−ρ|
e = O(e(−(1−θ)a+µ)ρ ).
r3
Therefore, for r ∈ (0, θρ),
(∆z − f ′ (w)z)eµ|r−ρ| = O(e(−(1−θ)a+µ)ρ ).
When r > θρ, similar argument shows that
H ′′′ + β ′′′ H ′ + β ′ − rH ′′ − rβ ′′ H ′ + β ′ µ|r−ρ|
[∆z − f ′ (w)z − − − f ′
(w) ]e = O(e−(a−µ)ρ ).
r r3 r
In this case
H ′ + β ′ − rH ′′ − rβ ′′ µ|r−ρ| 1
e = O( 2 ),
r3 ρ
and
H ′′′ + β ′′′ H ′ + β ′ µ|r−ρ| (f ′ (H) − f ′ (w))H ′ + β ′′′ − f ′ (w)β ′ µ|r−ρ| e−(a−µ)ρ
[ − f ′ (w) ]e = e = O( ).
r r r ρ
Therefore for r > θρ,
1
(∆z − f ′ (w)z)eµ|r−ρ| = O( ).
ρ2
Combining the two cases of r we find
1
k∆z − f ′ (w)zk∗ = O( ) = O(γ),
ρ2
which implies, by (2.18), that kS(w)k∗ = O(γ).

7
 1
Lemma 2.3 I(w) = 2πτ ( + γρ) + O(γ 1.5 ) where τ is given in (2.2).
2ρ
Proof. It is easy to see that
1
H ′ + β′
Z Z
|∆w − f (w)|2 dx = 2π [β ′′ + − f (w) + f (H)]2 r dr
R2 r
Z0 ∞
1 ′ 1
= 2π (H (y))2 dy + O( 3 )
−ρ ρ ρ
2πτ
= + O(γ 1.5 );
ρ
Z ∞
1 1
Z
[ |∇w|2 + F (w)] dx = 2π [ (H ′ (y))2 + F (y)] rdr + O(e−2aρ )
R2 2 −∞ 2
= 2πτ ρ + O(e−2aρ ).

The lemma follows.

In the next section we will show that there is a particular ρ, called ργ , and a small function
φ(·; ργ ) such that S(w(·; ργ ) + φ(·; ργ )) = 0.

3 Reduction to one dimension

Around w the linearized operator of S is Lρ given by

Lρ φ = (∆ − f ′ (w))2 φ − (∆w − f (w))f ′′ (w)φ − γ(∆ − f ′ (w))φ. (3.1)

Here (∆ − f ′ (w))2 and ∆ − f ′ (w) are linear operators.

We define an approximate kernel

h(r; ρ) = H ′ (r − ρ) + b1,ρ e−ar + b2,ρ re−ar (3.2)

where b1,ρ = O(e−aρ ) and b2,ρ = O(e−aρ ) are constants so chosen that h′ (0) = h′′′ (0) = 0.
Let πρ be the projection operator to the subspace perpendicular to h:

< g, h >
πρ g = g − h. (3.3)
khk22

We view
1 2
M = {w(·; ρ) : √ < ρ < √ } (3.4)
2 2γ 2γ
as a one-dimensional submanifold in Hr4 (R2 ). At each w(·; ρ) we define an approximate normal
subspace
Fρ = {φ ∈ Hr4 (R2 ) : φ ⊥ h(·; ρ)}. (3.5)
In each Fρ we look for a φ(·; ρ) so that

πρ S(w(·; ρ) + φ(·; ρ)) = 0. (3.6)

8
We write the last equation as
πρ (S(w) + Lρ φ + Nρ φ) = 0
where the higher order, nonlinear operator Nρ is given by

Nρ φ = −(∆ − f ′ (w) − γ)(f (w + φ) − f (w) − f ′ (w)φ)

−(f ′ (w + φ) − f ′ (w))(∆φ − (f (w + φ) − f (w)))
−(∆w − f (w))(f ′ (w + φ) − f ′ (w) − f ′′ (w)φ). (3.7)

We would like to turn the last equation to the following fixed point form

φ = −(πρ Lρ )−1 (πρ S(w) + πρ Nρ (φ)). (3.8)

To this end we need to specify the function space in which the fixed point argument is made and
also establish the fact that πρ Lρ is invertible.
First we note that Lρ can be defined as an operator from Hr4 (R2 ) to L2r (R2 ).

Lemma 3.1 The operator πρ Lρ from {φ ∈ Hr4 (R2 ) : φ ⊥ h(·; ρ)} to {η ∈ L2r (R2 ) : η ⊥ h(·; ρ)}
satisfies the Fredholm Alternative. In particular the operator is onto if it is one-to-one.

Proof. Let q = f ′ (0) − f ′ (w) and z = ∆w − f (w), we note that

Lρ φ = (∆ − f ′ (w))2 φ − zf ′′ (w)φ − γ(∆ − f ′ (w))φ

= (∆ − f ′ (0))2 φ − γ(∆ − f ′ (0))φ
+(∆ − f ′ (0))qφ + q(∆ − f ′ (w))φ − γqφ − zf ′′ (w)φ
:= Qφ + Pρ φ

and
hLρ φ, hi
πρ Lρ φ = Qφ + Pρ φ − h (3.9)
khk22
with

Qφ = (∆ − f ′ (0))2 φ − γ(∆ − f ′ (0))φ (3.10)

Pρ φ = (∆ − f ′ (0))qφ + q(∆ − f ′ (w))φ − γqφ − zf ′′ (w)φ. (3.11)

Note that Q is a well-behaved operator, an isometry indeed, from Hr4 (R2 ) to L2r (R2 ). Let us denote
{φ ∈ L2r (R2 ) : φ ⊥ h} by {h}⊥ . If we are given an equation πρ Lρ φ = η with η ∈ L2r (R2 ) ∩ {h}⊥ and
with φ expected in Hr4 (R2 ) ∩ {h}⊥ , we apply the operator πρ Q−1 to both sides to find
hLρ φ, hi −1
φ + πρ [Q−1 Pρ φ − Q h] = πρ Q−1 η. (3.12)
khk22

However we must show that the operator πρ Q−1 used to make this transformation is one-to-one
and onto from L2r (R2 ) ∩ {h}⊥ to Hr4 (R2 ) ∩ {h}⊥ . To show that the operator is one-to-one, we let
πρ Q−1 g = 0 for some g ∈ L2r (R2 ) ∩ {h}⊥ . There exists c ∈ R such that Q−1 g = ch, i.e. cQh = g.
Multiply by h and integrate to find
Z Z
0=c hQh = c [|(∆ − f ′ (0))h|2 + γ(|∇h|2 + f ′ (0)h2 )].
R2 R2

9
Hence c = 0, and consequently g = 0. To show that πρ Q−1 is onto, we must be able to solve
πρ Q−1 g = ξ for any ξ ∈ Hr4 (R2 ) ∩ {h}⊥ , i.e. we look for c ∈ R and g ∈ L2r (R2 ) ∩ {h}⊥ such that

Q−1 g + ch = ξ, i.e. g + cQh = Qξ.

Multiply the last equation by h and integrate to find

R
2 hQξ dx
c = RR .
R 2 hQh dx

We then find g by setting g = Qξ − cQh. This g is necessarily perpendicular to h.

For the Fredholm Alternative to hold in (3.12), the operator from Hr4 (R2 ) to itself on the left of
(3.12) side should be of the form Identity + Compact. To see the compactness we note that both q
and z in Pρ decay to 0 at ∞. Hence by the Sobolev imbedding theory Q−1 Pρ is a compact operator.
hLρ φ,hi
Also a rank one operator like φ → khk 2 h is compact.
2
More importantly we prove the following estimate for πρ Lρ under the weighted L∞ -norm.

Lemma 3.2 There exists C > 0 independent of γ and ρ such that if πρ Lρ φ = g, φ ⊥ h and
g ∈ C(R), then
kφk∗ + k∆φk∗ ≤ Ckgk∗ .

Before we prove this lemma, we need a technical estimate. This estimate was used by Ni and
Wei [19]. They stated a version on a bounded ball. We include a proof for our entire space situation
in the appendix.

Lemma 3.3 Let φ ∈ C 2 [0, ∞) satisfy φ′ (0) = 0, limr→∞ φ(r) = 0, and

φ′
|φ′′ + − f ′ (0)φ| ≤ c0 e−µ|r−ρ| .
r
Then
2ec0 −µ|r−ρ|
|φ(r)| ≤ e[|φ(ρ)| + ]e .
f ′ (0)

Proof of Lemma 3.2. Let πρ Lρ φ = g with φ ⊥ h. Then there exists d1,ρ ∈ R such that

Lρ φ = g + d1,ρ h.

If the lemma does not hold, then we may assume that kgk∗ = o(1) and kφk∗ + k∆φk∗ = 1. Let
ψ = ∆φ − f ′ (w)φ. Clearly kψk∗ = O(1) and ψ satisfies

∆ψ − f ′ (w)ψ − zf ′′ (w)φ − γψ = d1,ρ h + g

where z = ∆w − f (w).
If we multiply the last equation by h and integrate over R2 , then integration by parts shows that
Z
o(ρ) = d1,ρ (ρ h2 dr + O(1)).
R

10
This implies that d1,ρ = o(1). It follows that

k∆ψ − f ′ (w)ψk∗ ≤ kzf ′′ (w)k∞ kφk∗ + γkψk∗ + d1,ρ khk∗ + kgk∗ = o(1).
ψ
Now we prove that kψk∗ = o(1). Assume this is not true. Then we consider ψ̃ = kψk∗ , which
satisfies kψ̃k∗ = 1 and
k∆ψ̃ − f ′ (w)ψ̃k∗ = o(1) (3.13)
following the last estimate. Simple elliptic regularity argument shows that ψ̃(· − ρ) converges in
2
Cloc (R) to a function Ψ̃ as γ → 0. It follows that kΨ̃k∗ ≤ 1 and Ψ̃ satisfies Ψ̃′′ − f ′ (H)Ψ̃ = 0.
Therefore Ψ̃ = d2 H ′ for some d2 ∈ R. This implies that

hψ̃, hi = d2 (2πρτ ) + o(ρ). (3.14)

On the other hand we multiply

1
(∆φ − f ′ (w)φ) = ψ̃
kψk∗
by h and integrate over R2 to find that

O(1) = hψ̃, hi.

2
Combined with (3.14) we deduce that d2 = 0 and ψ̃(· − ρ) → 0 in Cloc (R). We now return to (3.13)
and find
k∆ψ̃ − f ′ (0)ψ̃k∗ ≤ k∆ψ̃ − f ′ (w)ψ̃k∗ + k(f ′ (w) − f ′ (0))ψ̃k∗ = o(1). (3.15)
Since ψ̃(ρ) = o(1), Lemma 3.3 implies that kψ̃k∗ = o(1). A contradiction to kψ̃k∗ = 1.
Finally we consider φ in the equation

∆φ − f ′ (w)φ = ψ

with kψk∗ = o(1). Again elliptic regularity argument shows that φ(· − ρ) → d3 H ′ in Cloc
2
(R) for
2
some d3 ∈ R. Our assumption φ ⊥ h implies that d3 = 0. Hence φ(· − ρ) → 0 in Cloc (R). As before

k∆φ − f ′ (0)φk∗ ≤ kψk + k(f ′ (w) − f ′ (0))φk∗ = o(1).

Lemma 3.3 again implies that kφk∗ = o(1). Moreover

k∆φk∗ ≤ k∆φ − f ′ (w)φk∗ + kf ′ (w)φk∗ ≤ kψk∗ + kf ′ (w)k∞ kφk∗ = o(1).

We have now reached a contradiction to kφk∗ + k∆φk∗ = 1.

A consequence of Lemma 3.2 is the following.

Lemma 3.4 πρ Lρ : Hr4 (R2 ) ∩ {h}⊥ → L2r (R2 ) ∩ {h}⊥ is a one-to-one and onto map. Moreover
(πρ Lρ )−1 is also an operator from {g ∈ C[0, ∞) : kgk∗ < ∞, g ⊥ h} to {φ ∈ C 2 [0, ∞) : φ′ (0) =
0, kφk∗ + k∆φk∗ < ∞, φ ⊥ h}, whose norm is bounded by a constant independent of γ and ρ.

11
 Proof. Lemma 3.2 shows that πρ Lρ is one-to-one. Hence it is onto by the Fredholm Alternative,
Lemma 3.1. Since every continuous function with finite k · k∗ -norm is in L2r (R2 ), one can apply
(πρ Lρ )−1 to such a function. Lemma 3.2 yields a bound of (πρ Lρ )−1 in this setting.
Now we define the proper space on which the fixed point argument is done. Let

Zρ = {φ ∈ C 2 [0, ∞) : φ′ (0) = 0, kφk∗ < ∞, k∆φk∗ < ∞, φ ⊥ h(·; ρ)}. (3.16)

In Zρ we define a norm
kφkZ = kφk∗ + k∆φk∗ . (3.17)
We write the right side of (3.8) as Tρ φ. Based on Lemmas 3.2 and 3.4 we know that Tρ is
well-defined on Zρ . We show that Tρ is a contraction map with a fixed point.

Lemma 3.5 There exists φ(·; ρ) so that kφ(·; ρ)kZ = O(γ) and πρ S(w + φ) = 0.

Proof. Define a closed ball Bρ in Zρ :

Bρ = {φ ∈ Zρ : kφkZ ≤ d1 γ}. (3.18)

Here d1 is a positive constant independent of γ and ρ to be fixed soon. For each φ ∈ Bρ , we have,
by Lemmas 2.2 and 3.2,

kTρ φkZ ≤ C(kS(w)k∗ + kNρ φk∗ ) ≤ C(γ + C1 kφk2Z ) ≤ C(γ + C1 d21 γ 2 ).

The last quantity is less than d1 γ when γ is small, if we choose d1 to be sufficiently large. This
shows that Tρ maps Bρ into itself. Next we take φ1 and φ2 from Bρ and consider

kTρ φ1 − Tρ φ2 kZ ≤ CkNρ φ1 − Nρ φ2 k∗ ≤ Ckφ1 − φ2 kZ (kφ1 kZ + kφ2 kZ ) ≤ Cd1 γkφ1 − φ2 kZ .

Hence Tρ is a contraction map when γ is small. This yields a unique solution φ of πρ S(w + φ) = 0
in Bρ .
To find a particular ρ = ργ so that S(w(·; ργ ) + φ(·; ργ )) = 0, we consider the free energy of
w(·; ρ) + φ(·; ρ): I(w(·; ρ) + φ(·; ρ)).
1
Lemma 3.6 1. I(w(·; ρ) + φ(·; ρ)) = 2πτ ( + γρ) + O(γ 1.5 ).
2ρ
1
2. There exists ργ = √ + o(γ −1/2 ) such that S(w(·; ργ ) + φ(·; ργ )) = 0.
2γ
Proof. Expanding I shows that
1
Z Z Z
I(w + φ) = I(w) + S(w)φ dx + φLρ φ dx + O( [|φ|3 + |∆φ|3 ] dx)
2 2 R2 R2
ZR Z ∞
1
Z
= I(w) + S(w)φ dx + φLρ φ dx + O(kφk3Z e−3µ|r−ρ| r dr)
R2 2 R2 0
1
Z Z
3
= I(w) + S(w)φ dx + φLρ φ dx + O(kφkZ ρ)
2 2 R2
ZR
1
Z
= I(w) + S(w)φ dx + φLρ φ dx + O(γ 2.5 ) (3.19)
R2 2 R2

12
by Lemma 3.5. From the equation πρ S(w + φ) = 0 we find, since φ ⊥ h,
Z Z Z Z
0 = S(w + φ)φ dx = S(w)φ dx + φLρ φ dx + φNρ φ dx
R2 R2 R2 R2
Z Z Z ∞
= S(w)φ dx + φLρ φ dx + O(kφk∗ kNρ φk∗ e−2µ|r−ρ| r dr)
R2 R2 0
Z Z
= S(w)φ dx + φLρ φ dx + O(γ 2.5 ).
R2 R2

We can rewrite (3.19) as

1
Z
I(w + φ) = I(w) + S(w)φ dx + O(γ 2.5 ). (3.20)
2 R2

Using Lemmas 2.2 and 3.5 we deduce

Z Z
S(w)φ dx = O(kS(w)k∗ kφk∗ e−2µ|r−ρ| rdr) = O(kS(w)k∗ kφk∗ ρ) = O(γ 1.5 ).
R2 R2

We have now turned (3.20) to

I(w + φ) = I(w) + O(γ 1.5 ). (3.21)
The first part of the lemma then follows from Lemma 2.3.
If we minimize I(w(·; ρ) + φ(·; ρ)) with respect to ρ, then based on the first part of this lemma
we know that I(w(·; ρ) + φ(·; ρ)) is minimized at some ργ and
1
ργ = √ + o(γ −1/2 ).
2γ
We now show that if ρ = ργ , S(w(·; ργ ) + φ(·; ρ)) = 0. Since πρ S(w + φ) = 0, there exists cρ such
that S(w + φ) = cρ h. We differentiate I(w + φ) with respect to ρ to find
dI(w(·; ρ) + φ(·; ρ)) ∂w ∂φ ∂w ∂φ
Z Z Z
= S(w + φ)( + ) dx = cρ [ h dx + h dx].
dρ R2 ∂ρ ∂ρ R2 ∂ρ R2 ∂ρ

Here ∞
∂w ∂(H(r − ρ) + β)
Z Z
h dx = 2π h rdr = 2π(τ ρ + o(ρ));
R2 ∂ρ 0 ∂ρ
φ ⊥ h implies that
∂φ ∂h
Z Z
h dx = − φ dx.
R 2 ∂ρ R2 ∂ρ
Then ∞
∂φ ∂h −µ|r−ρ|
Z Z
| h dx| ≤ kφk∗ 2π | |e rdr = kφk∗ O(ρ) = o(ρ).
R 2 ∂ρ 0 ∂ρ
Therefore
dI(w + φ)
= cρ (2πτ ρ + o(ρ)).
dρ
If ρ is equal to ργ , then
0 = cργ (2πτ ργ + o(ργ )),
i.e. cργ = 0 and S(w(·; ργ ) + φ(·; ργ )) = 0.
The last lemma completes the proof of Theorem 2.1.

13
4 n≥4
If we consider (1.6) for n ≥ 4, then with Γ being a n − 1 dimensional sphere

Ic (Γ) = ωn−1 ρn−3 + ωn−1 γρn−1 (4.1)

where ωn−1 is the area of n − 1 dimensional unit sphere. It is clear that if γ > 0, the right side is
increasing in ρ. Only if γ < 0, there exists a critical point, but this critical point is a maximum. In
the phase field model, we have the similar phenomenon.

Theorem 4.1 When γqis negative and sufficiently close to 0, there exists a bubble profile. The
radius of the bubble is (n−1)(n−3)
−2γ + o(γ −1/2 ).

Proof. The proof of this theorem is almost identical to the proof of Theorem 2.1. The main
difference occurs in the last step:

(n − 1)2 ρn−3
I(w + φ) = ωn−1 τ ( + γρn−1 ) + o(γ (3−n)/2 ) (4.2)
2
where ωn−1 is the area of the n − 1 dimensional unit sphere. If γ < 0, the above quantity has a
maximum at s
(n − 1)(n − 3)
ργ = + o(γ −1/2 ).
−2γ
The detail of the proof is left to the reader.

5 n=1
When n = 1, the phase field problem is far more complex than (1.6). A zero dimensional sphere is
just the union of two points in R, and ρ is half the distance between the two points. This sphere
has no curvature. Hence
Ic (Γ) = 2γ, (5.1)
a constant independent of ρ. No conclusion can be drawn (5.1). But for the phase field problem, we
have the following result.

Theorem 5.1 When γ is positive and sufficiently small, there exists a bubble profile. The radius of
1
the bubble is 2a log γ1 + o(log γ1 ).

Let
1 1 1 1
ρ∈( log , log ). (5.2)
4a γ a γ
For each ρ satisfying (5.2) we define an approximate solution w. Compared to the n = 2 and n ≥ 4
cases, the construction of w is more complex. We let

H ′ (−ρ)
α(x; ρ) = c0,ρ e−ax , where c0,ρ = , so that H ′ (−ρ) + α′ (0) = 0. (5.3)
a

14
Now we define a function g(y; ρ) on (−∞, ∞) which is the solution of

g ′′ − f ′ (H)g + α(y + ρ)(f ′ (0) − f ′ (H)) = dρ H ′ , g(0) = −α(ρ). (5.4)

In (5.4) the constant dρ is chosen so that

Z Z
dρ (H ′ )2 dy = α(y + ρ)(f ′ (0) − f ′ (H(y)))H ′ (y) dy. (5.5)
R R

This ensures that (5.4) is solvable. The condition g(0) = −α(ρ) gives a unique solution. We calculate
the right side of (5.5):
Z Z
′ ′ ′
α(y + ρ)(f (0) − f (H(y)))H (y) dy = α(ρ) e−ay (f ′ (0)H ′ − H ′′′ ) dy
R R
y=∞
= α(ρ)[−e−ay H ′′ (y) − ae−ay H ′ (y)]|y=−∞ = α(ρ) lim [e−ay H ′′ (y) + ae−ay H ′ (y)]
y→−∞
2 −2aρ
= 2akα(ρ) = 2k e + o(e−2aρ ).

Therefore
2k 2 e−2aρ
dρ = + o(e−2aρ ). (5.6)
τ
We include g in the construction of w. One last term is β which is given as

f ′′ (0)c20,ρ −2ax
β(x; ρ) = c1,ρ e−ax + c2,ρ xe−ax + e . (5.7)
6a2
It is a solution of
f ′′ (0)α2
(D2 − f ′ (0))2 β = (D2 − f ′ (0)) . (5.8)
2
The constants c1,ρ and c2,ρ are chosen so that

β ′ (0) = −H ′ (−ρ) − α′ (0) − g ′ (−ρ) = −g ′ (−ρ), β ′′′ (0) = −H ′′′ (−ρ) − α′′′ (0) − g ′′′ (−ρ). (5.9)

Here
c1,ρ = O(e−2aρ ), c2,ρ = O(e−2aρ ). (5.10)
Now we set
w(x; ρ) = H(x − ρ) + α(x; ρ) + g(x − ρ; ρ) + β(x; ρ). (5.11)
′ ′′′
Our choice of β ensures that w (0) = w (0) = 0. Note that this β is different from the one (2.9)
used in the n = 2 case.
We again need the weighted L∞ norm:

kζk∗ = sup |ζ(x)|eµ|x−ρ| . (5.12)

x∈R

Lemma 5.2 There exists δ > 0 independent of γ and ρ such that kS(w)k∗ = O(γ 1+δ ).

15
 Proof. We start with w′′ − f (w). Note that

w′′ − f (w) = H ′′ + α′′ + g ′′ + β ′′ − f (H + α + g + β)

= α′′ + g ′′ + β ′′ + f (H) − f (H + α + g + β)
= α′′ + g ′′ + β ′′ − f ′ (H)(α + g + β)
−(f (H + α + g + β) − f (H) − f ′ (H)(α + g + β))
= dρ H ′ + β ′′ − f ′ (H)β − (f (H + α + g + β) − f (H) − f ′ (H)(α + g + β)).

At this point we consider two cases of x: x ∈ (0, θρ) and x ∈ (θρ, ∞) where θ ∈ (0, 1). In the
first case we write
f ′′ (0) 2
w′′ − f (w) = dρ H ′ + β ′′ − f ′ (0)β − α + N1 (x; ρ) (5.13)
2
with
f ′′ (0) 2
N1 (x; ρ) = (f ′ (0) − f ′ (H))β − (f (H + α + g + β) − f (H) − f ′ (H)(α + g + β) − α ). (5.14)
2
Note that
sup{|N1 (x; ρ)| + |N1′ (x, ρ)| + |N1′′ (x, ρ)| : x ∈ (0, θρ)} = O(e−3θaρ ).
Hence

sup{(|N1 (x; ρ)| + |N1′ (x, ρ)| + |N1′′ (x, ρ)|)eµ|x−ρ| : x ∈ (0, θρ)} = O(e−(3θa−µ)ρ ). (5.15)
′′
From the equation that β satisfies we see that β ′′ − f ′ (0)β − f 2(0) α2 = c3,ρ e−ax . Hence (5.13)
becomes
w′′ − f (w) = dρ H ′ + c3,ρ e−ax + N1 (r; ρ).
To estimate the size of c3,ρ note that the derivative of w′′ − f (w) at x = 0 is 0, by our construction
of w. Therefore
0 = dρ H ′′ (−ρ) − ac3,ρ + O(e−3θaρ ).
This shows, with (5.6), that c3,ρ = O(e−3θaρ ) and

f ′′ (0) 2
β ′′ − f ′ (0)β − α = O(e−3θaρ ).
2
Now we can write (5.13) as
w′′ − f (w) = dρ H ′ + M (x; ρ) (5.16)
with M satisfying

sup{(|M (x; ρ)| + |M ′ (x, ρ)| + |M ′′ (x, ρ)|)eµ|x−ρ| : x ∈ (0, θρ)} = O(e−(3θa−µ)ρ ). (5.17)

For the second case, x > θρ, since α(x) = O(e−aρ )e−ax and β(x) = O(e−2aρ )xe−ax ,

w′′ − f (w) = dρ H ′ + β ′′ − f ′ (H)β − (f (H + α + g + β) − f (H) − f ′ (H)(α + g + β))

= dρ H ′ + M (r; ρ)

16
with

sup{(|M (x; ρ)| + |M ′ (x, ρ)| + |M ′′ (x, ρ)|)eµ|x−ρ| : x ∈ (θρ, ∞)} = O(e−(2a+δ1 )ρ ) (5.18)

for some δ1 > 0.

Combining (5.17) and (5.18) we have

w′′ − f (w) = dρ H ′ + M

with
kM k∗ + kM ′′ k∗ = O(e−(2a+δ1 )ρ ) (5.19)
by choosing θ to be sufficiently close to 1 and µ sufficiently small.
Now we estimate S(w) by (5.19). In

S(w) = (D2 − f ′ (w) − γ)(dρ H ′ + M ) = dρ (f ′ (H) − f ′ (w))H ′ − γdρ H ′ + (D2 − f ′ (w) − γ)M,

clearly kγdρ H ′ k∗ = O(γ 2 ) and k(D2 − f ′ (w) − γ)M k∗ = O(γ 1+δ ) for some δ > 0. As for dρ (f ′ (H) −
f ′ (w))H ′ , note that

kdρ (f ′ (H) − f ′ (w))H ′ k∗ ≤ dρ kf ′ (H) − f ′ (w)k∞ kH ′ k∗ = dρ O(e−aρ ) = O(e−3aρ ).

Hence kS(w)k∗ = O(γ 1+δ ).

4k 4 e−4aρ 2k 2 e−2aρ
Lemma 5.3 I(w) = + γ(2τ − ) + o(γ 2 ).
τ a
Proof. . Using (5.18) and (5.6) we find
1
Z
(w′′ − f (w))2 dx
2 R
Z ∞
4k 4 e−4aρ
= d2ρ (H ′ (x − ρ))2 dx + o(e−4aρ ) = d2ρ τ + o(e−4aρ ) = + o(γ 2 );
0 τ
1 ′2
Z
[ |w | + F (w)] dx
R 2
Z ∞ Z ∞
′ 2
= [|w | + 2F (w)] dx = [|H ′ (x − ρ) + α′ (x)|2 + 2F (H + α)] dx + o(e−2aρ )
0 0
Z −ρ Z ∞
′ 2
= 2τ − 2 H (y) dy + [2H ′ α′ + 2f (H)α + (α′ )2 + f ′ (H)α2 ] dx + o(e−2aρ )
−∞ 0
Z −ρ Z ∞ Z ∞
= 2τ − 2 H ′ (y)2 dy + 2H ′ (x − ρ)α(x)|x=0
x=∞
+ (α′ )2 dx + f ′ (H)α2 dx + o(e−2aρ )
−∞ 0 0
Z −ρ Z ∞
= 2τ − 2 k 2 e2ay dy − 2H ′ (−ρ)α(0) + 2a2 α2 dx + o(e−2aρ )
−∞ 0
k 2 e−2aρ 2k 2 e−2aρ k 2 e−2aρ 2k 2 e−2aρ
= 2τ − − + + o(e−2aρ ) = 2τ − + o(γ).
a a a a
This proves the lemma.

17
 The rest of the proof is analogous to that of Theorem 2.1. Define h as in (3.2). For each ρ we find
φ(·; ρ) ⊥ h(·; ρ) so that πρ S(w + φ) = 0. The Contraction Mapping Principle used in the argument
also shows, with the help of Lemma 5.2, that

kφ(·; ρ)k∗ = O(γ 1+δ ), kφ′′ (·; ρ)k∗ = O(γ 1+δ ). (5.20)

We then expand I(w + φ) as follows.

1
Z Z
I(w + φ) = I(w) + S(w)φ dx + φLρ φ dx + O(γ 3+3δ )
R 2 R
1
Z
= I(w) + S(w)φ dx + O(γ 3+3δ )
2 R
Z ∞
= I(w) + O(kS(w)k∗ kφk∗ ) e−2µ|x−ρ| dx + O(γ 3+3δ )
0
2+2δ
= I(w) + O(γ ).

Finally we minimize I(w(·; ρ)+φ(·; ρ)) with respect to ρ. Lemma 5.3 shows that I(w+φ) is minimized
at some
1 γτ 1
ργ = − log + o(log ). (5.21)
2a 4ak 2 γ
This completes the proof of Theorem 5.1.

6 n=3
1 1
κ2 ds = 2
R
When n = 3, for a sphere κ = ρ and Γ ρ2 4πρ = 4π. Hence

Ic (Γ) = 4π + 4πγρ2 , (6.1)

which has no critical point for positive ρ. The phase field problem is again very different.
Let l(s) be the inverse function of

2k 2 e−2aρ
ρ→− . (6.2)
τρ
1
Here l : (−∞, 0) → (0, ∞). As s tends to 0, l(s) grows to ∞, but more slowly than − 2a log(−s)
does.

Theorem 6.1 When γ is negative and sufficiently close to 0, there exists a bubble profile. The
radius of the bubble is l(γ) + o(l(γ)).

We let
l(γ)
ρ∈( , 2l(γ)). (6.3)
2
We define a family of approximate solutions

w(r) = H(r − ρ) + β(r; ρ) (6.4)

where β is the same as the one (2.9) used in the n = 2 case.

18
Lemma 6.2 There exists δ > 0 independent of γ and ρ so that kS(w)k∗ = O(γ (1+δ)/2 ).

Proof. We start with

2H ′ + 2β ′
∆w − f (w) = β ′′ + − f (w) + f (H).
r
As in (2.17),
1 1
k∆w − f (w)k∗ = O( ) = O( ). (6.5)
ρ l(γ)
Let z = ∆w − f (w). In S(w) = (∆ − f ′ (w))z − γz we first note from the above estimate that
γ
kγzk∗ = O( ).
l(γ)

It suffices to estimate ∆z − f ′ (w)z. In the n = 3 case, ∆ = 1r D2 r for radial functions. Hence we

deduce that
1 2
∆z − f ′ (w)z = (D − f ′ (w))(rz)
r
1 2
= (D − f ′ (w))[(rβ)′′ − r(f (w) − f (H)) + 2H ′ ]
r
1
= {(D2 − f ′ (w))[(rβ)′′ − r(f (w) − f (H))] − 2(f ′ (w) − f ′ (H))H ′ }.
r
When r ∈ (θρ, ∞) where θ ∈ (0, 1) is independent of ρ and γ, we find
C −aρ −ar
|∆z − f ′ (w)z| ≤ [e re + e−aρ re−ar e−a|r−ρ| ];
ρ
hence for small µ
|(∆z − f ′ (w)z)|eµ|r−ρ| = O(e−(a+δ1 )ρ ), r ∈ (θ, ρ), (6.6)
for some δ1 > 0 where δ1 is independent of ρ and γ.
When r ∈ (0, θρ) we write

∆z − f ′ (w)z
1
= {(D2 − f ′ (w))[(rβ)′′ − f ′ (0)rβ + f ′ (0)rβ − r(f (w) − f (H))] − 2(f ′ (w) − f ′ (H))H ′ }
r
1
= {(D2 − f ′ (0))2 (rβ) + (f ′ (0) − f ′ (w))(D2 − f ′ (0))(rβ)
r
+(D2 − f ′ (w))(f ′ (0)rβ − r(f (w) − f (H)) − 2(f ′ (w) − f ′ (H))H ′ }.

The largest term appears to be (D2 − f ′ (0))2 (rβ). However

(D2 − f ′ (0))2 (rβ) = r(D2 − f ′ (0))2 β + 4(D2 − f ′ (0))β ′ = 4(D2 − a2 )β ′ = 8a2 c2,ρ e−ar ,

which is small. By choosing θ close to 1 and µ small we find δ1 > 0 such that

|∆z − f (w)z|eµ|r−ρ| = O(e−(a+δ1 )ρ ), r ∈ (0, θρ). (6.7)

19
From (6.6) and (6.7) we deduce that

k∆z − f ′ (w)zk∗ = O(e−(a+δ1 )ρ ) = O(γ (1+δ)/2 )

for some δ > 0 independent of ρ and γ, and consequently by (6.5) kS(w)k∗ = O(γ (1+δ)/2 ).

2k 2 e−2aρ
Lemma 6.3 I(w) = 4π[2τ − + γτ ρ2 ] + O(γ 1+δ ) for some δ > 0 independent of ρ and γ.
a
Proof. It is easy to see that
Z ∞
1 1
Z
[ |∇w|2 + F (w)] dx = 4π [ (H ′ (r − ρ) + β ′ )2 + F (H + β)] r2 dr = 4πτ ρ2 + O(ρe−aρ ). (6.8)
R 3 2 0 2

The estimate of the first part of I(w) is a bit more involved. Note that

1
Z
|∆w − f (w)|2 dx
2 R3
Z ∞
2(H ′ + β ′ )
= 2π |H ′′ + β ′′ + − f (H + β)|2 r2 dr
0 r
Z ∞
= 2π |rβ ′′ + 2β ′ + 2H ′ − rf ′ (H)β|2 dr + O(e−(2a+δ1 )ρ )
0
Z ∞
= 2π |2H ′ − 2ac1,ρ e−ar + rc1,ρ e−ar (f ′ (0) − f ′ (H))|2 dr + O(e−(2a+δ1 )ρ )
0

for some δ1 > 0, where δ1 is independent of ρ and γ. Here c1,ρ comes from the definition (2.9) of β.
We now write the last quantity as 2π(T1 + T2 + T3 + T4 + T5 + T6 ) where
∞ ∞ −ρ
2k 2 e−2aρ
Z Z Z
T1 = 4(H ′ )2 dr = 4(H ′ )2 dy = 4τ − 4(H ′ )2 dy = 4τ − + O(e−3aρ )
0 −ρ −∞ a
∞
2(H ′ (−ρ))2 2k 2 e−2aρ
Z
T2 = 4a2 c21,ρ e−2ar dr = 2c21,ρ a = = + O(e−3aρ )
a a
Z0 ∞
T3 = r2 c21,ρ e−2ar (f ′ (0) − f ′ (H))2 dr = O(e−(2a+δ1 )ρ )
0
Z ∞ Z ∞
T4 = −8aH ′ c1,ρ e−ar dr = −8ac1,ρ e−aρ H ′ (y)e−ay dy
0 −ρ
Z ∞ Z ∞
′ −ar ′ ′ −aρ
T5 = 4c1,ρ H re (f (0) − f (H)) dr = 4c1,ρ e (y + ρ)H ′ e−ay (f ′ (0) − f ′ (H)) dy
0 −ρ
Z ∞
T6 = −4ac21,ρ e−2ar r(f ′ (0) − f ′ (H)) dr = O(e−(2a+δ1 )ρ ).
0

We focus on T5 . Note that

Z ∞ Z ∞
T5 = 4c1,ρ e−aρ [ H ′ (y)y(f ′ (0) − f ′ (H))e−ay dy + H ′ (y)ρ(f ′ (0) − f ′ (H))e−ay dy]
−ρ −ρ

20
where
Z ∞
H ′ (y)y(f ′ (0) − f ′ (H))e−ay dy
−ρ
Z ∞
= (H ′ f ′ (0) − H ′′′ )ye−ay dy
−ρ
Z ∞
= [H ′ f ′ (0)ye−ay − H ′ (ye−ay )′′ ] dy − H ′′ (y)ye−ay |∞ ′
−ρ + H (y)(ye
−ay ′ ∞
) |−ρ
−ρ
Z ∞
= 2a H ′ (y)e−ay dy − H ′′ (−ρ)ρeaρ − H ′ (−ρ)eaρ − aH ′ (−ρ)ρeaρ ;
−ρ
Z ∞
H ′ (y)ρ(f ′ (0) − f ′ (H))e−ay dy]
−ρ
Z ∞
= ρ (H ′ f ′ (0) − H ′′′ )e−ay dy
−ρ
= ρ[−H ′′ (y)e−ay |∞ ′
−ρ − aH (y)e
−ay ∞
|−ρ ]
′′ aρ ′ aρ
= ρ[H (−ρ)e + aH (−ρ)e ].

Hence Z ∞
T5 = 4c1,ρ e−aρ [2a H ′ (y)e−ay dy − H ′ (−ρ)eaρ ],
−ρ

and consequently
4k 2 e−2aρ
T1 + ... + T6 = 4τ − 4c1,ρ H ′ (−ρ) + O(e−(2a+δ1 )ρ ) = 4τ − + O(e−(2a+δ1 )ρ ).
a
Therefore
1 2k 2 e−2aρ
Z
|∆w − f (w)|2 dx = 4π(2τ − ) + O(e−(2a+δ1 )ρ ). (6.9)
2 R3 a
The lemma now follows from (6.8) and (6.9).
Note that in I(w),
2k 2 e−2aρ
2τ − + γτ ρ2
a
is maximized at ρ = l(γ). For each ρ satisfying (6.3) we find φ ⊥ h so that πρ S(w + φ) = 0.
Here h is again given by (3.2). As we use the fixed point argument, Lemma 6.2 implies that
kφk∗ = O(γ (1+δ)/2 ). Then we find
1
Z
I(w + φ) = I(w) + S(w)φ dx + O(γ 1+δ )
2 R3
Z ∞
= I(w) + O(kS(w)k∗ kφk∗ ) e−2µ|r−ρ| r2 dr
0
= I(w) + O(kS(w)k∗ kφk∗ ρ2 )
= I(w) + O(γ 1+δ l2 (γ)).

Finally we maximize I(w + φ) with respect to ρ. Theorem 6.1 follows from Lemma 6.3.

21
7 Discussion
The stability of the bubble profiles constructed in this paper should depend on the dimension of
the space. But first it is obvious that by differentiating the equation of a bubble solution u =
u(x1 , x2 , ..., xn ) with respect to xj , j = 1, 2, ..., n, one obtains an eigenfunction with eigenvalue 0.
This 0 eigenvalue is a consequence of the translation invariance of our problem. We can only discuss
stability modulo translation.
In the cases n = 3 and n ≥ 4, we have obtained the solutions by maximizing I(w + φ) with
respect to ρ. This means that a solution is a maximum of I when restricted in the submanifold
{w(·; ρ) + φ(·; ρ)}. Hence the solution must be unstable; actually it must be a saddle point.
In the cases n = 1 and n = 2, our conjecture is that the solutions are stable modulo translation.
We will present a complete spectral analysis of all the bubble solutions elsewhere.

A Proof of Lemma 3.3

The proof mainly consists of comparison argument. We separate the two cases r ∈ (ρ, ∞) and
r ∈ (0, ρ).
In the case r ∈ (ρ, ∞), we consider an auxiliary function

A(y) = e−µy , y = r − ρ ∈ (0, ∞). (1.1)

In this appendix we work with the y-coordinate instead of the r-coordinate. Then

A′ µe−µy
∆A − f ′ (0)A = A′′ + − f ′ (0)A = µ2 e−µy − − f ′ (0)e−µy
y+ρ y+ρ
µ f ′ (0) −µy
= (µ2 − − f ′ (0))e−µy < − e
y+ρ 2
provided that µ is sufficiently small. Now we have φ, as a function of y as well, such that

φ′
|φ′′ + − f ′ (0)φ| ≤ c0 e−µy , lim φ(y) = 0.
y+ρ y→∞

Let
2c0
R(y) = [|φ(0)| + ]A(y) − φ(y),
f ′ (0)
and we have
2c0
R(0) = |φ(0) + | − φ(0) ≥ 0, lim R(y) = 0,
f ′ (0) y→∞

and
R′ 2c0 f ′ (0)
R′′ + − f ′ (0)R ≤ −[|φ(0)| + ′ ] A(y) + c0 e−µy ≤ 0. (1.2)
y+ρ f (0) 2
We claim R(y) ≥ 0 on (0, ∞). Otherwise there exists y∗ ∈ (0, ∞) such that R(y) ≥ R(y∗ ) for all
y ∈ (0, ∞) and R(y∗ ) < 0. But at this minimum point y∗ ,

R′′ (y∗ ) ≥ 0, R′ (y∗ ) = 0, −f ′ (0)R(y∗ ) > 0.

22
Therefore
R′ (y∗ )
R′′ (y∗ ) + − f ′ (0)R(y∗ ) > 0,
y∗ + ρ
a contradiction to (1.2). Hence R(y) ≥ 0, i.e.
2c0 −µ
φ(y) ≤ [|φ(0)| + ]e .
f ′ (0)
One can carry out a similar argument with
2c0
R̃(y) = [|φ(0)| + ]A(y) + φ(y)
f ′ (0)
to conclude that
2c0 −µ
−φ(y) ≤ [|φ(0)| + ]e .
f ′ (0)
Hence we obtain
2c0 −µy
|φ(y)| ≤ [|φ(0)| + ]e . (1.3)
f ′ (0)
In the case r ∈ (0, ρ), we construct A(y), with y = r − ρ ∈ (−ρ, 0) exactly as in [19]. Let χ be a
smooth cut-off function such that

χ(t) = 1 for |t| ≤ 1, χ(t) = 0 for |t| ≥ 2, 0 ≤ χ ≤ 1. (1.4)

Define the following auxiliary function

A(y) = eµy + (eµy0 − eµy )χ(µ(y + ρ)) (1.5)

where
1
y0 = −ρ + . (1.6)
µ
If y ∈ (−ρ, y0 ), then A(y) = eµy0 and
A′
A′′ + − f ′ (0)A = −f ′ (0)eµy0 ≤ −f ′ (0)eµy .
y+ρ

If y ∈ (y0 , y0 + µ1 ), then

µ 1
eµy0 ≤ eµy ≤ eeµy0 , A(y) ≥ eµy0 , ≤ ≤ µ;
2 y+ρ
hence
A′ f ′ (0) µy
A′′ + − f ′ (0)A ≤ O(µ2 )eµy − f ′ (0)eµy0 ≤ O(µ2 )eµy − e−1 f ′ (0)eµy ≤ − e
y+ρ 2e
µ
since µ is sufficiently small. If y ∈ (y0 + µ1 , 0), then A(y) = eµy and 1
y+ρ ≤ 2; hence

A′ µ2 f ′ (0) µy
A′′ + − f ′ (0)A ≤ [µ2 + − f ′ (0)]eµy ≤ − e ,
y+ρ 2 2

23
provided µ is small. Therefore for all y ∈ (−ρ, 0),

A′ f ′ (0) µy
A′′ + − f ′ (0)A ≤ − e , A′ (−ρ) = 0. (1.7)
y+ρ 2e
If we have φ, as a function of y, such that

φ′
|φ′′ + − f ′ (0)φ| ≤ c0 eµy on (−ρ, 0), φ′ (−ρ) = 0,
y+ρ
then let
2ec0
R(y) = [|φ(0)| + ]A(y) − φ(y).
f ′ (0)
Calculations show that
R′ 2ec0 f ′ (0) µy
R′′ + − f ′ (0)R ≤ −[|φ(0)| + ′ ] e + c0 eµy ≤ 0, (1.8)
y+ρ f (0) 2e

and
R(0) ≥ 0, R′ (−ρ) = 0.
We then claim that R(y) ≥ 0 for all y ∈ (−ρ, 0). Otherwise there exists y∗ ∈ [−ρ, 0) such that
R(y∗ ) < 0 and R(y) ≥ R(y∗ ) for all y ∈ (−ρ, 0). However at this y∗ , R′′ (y∗ ) ≥ 0, R′ (y∗ ) = 0, and
−f ′ (0)R(y∗ ) > 0. A contradiction to (1.8). Therefore R(y) ≥ 0, i.e.
2ec0 2ec0
φ(y) ≤ [|φ(0)| + ]A(y) ≤ e[|φ(0)| + ′ ]eµy .
f ′ (0) f (0)

In the last inequality we have used the fact that

A(y) < eeµy for all y ∈ (−ρ, 0).

To see this note that A(y) = eµy if y > y0 + µ1 . If y ∈ (y0 , y0 + µ1 ), A(y) = eµy + (eµy0 − eµy )χ(µ(y +
ρ)) ≤ eµy since y0 < y and eµy0 − eµy < 0 there. If y ∈ (−ρ, y0 ), then y0 − µ1 = −ρ < y, i.e.
1
y0 < y + µ1 , and A(y) = eµy0 < eµ(y+ µ ) = eeµy . Similarly if we consider

2ec0
R̃(y) = [|φ(0)| + ]A(y) + φ(y),
f ′ (0)

we find that
2ec0 µy
−φ(y) ≤ e[|φ(0)| + ]e .
f ′ (0)
In summary we have, on (−ρ, 0),
2ec0 µy
|φ(y)| ≤ e[|φ(0)| + ]e .
f ′ (0)

The lemma is proved.

24
References
[1] S.E. Allen and J.W. Cahn. A microscopic theroy for antiphase boundary motion and its appli-
cation to antiphase domain coarsening. Acta. Metall., 27(6):1085–1095, 1979.
[2] P.W. Bates, E.N. Dancer, and J. Shi. Multi-spike stationary solutions of the Cahn-Hilliard
equation in higher-dimension and instability. Adv. Differential Equations, 4:1–69, 1999.
[3] R.G. Casten and C.J. Holland. Instability results for reaction diffusion equations with neumann
boundary conditions. J. Differential Equations, 27(2):266–273, 1978.
[4] P. G. Ciarlet. Introduction to linear shell theory, Series in Applied Mathematics (Paris), Vol
1. Édition Scientifiques et Médicales Elsevier, Gauthier-Villars, Paris, 1998.
[5] P. G. Ciarlet. Mathematical Elasticity, III, Studies in Mathematics and its Applications, Vol
29. North-Holland, Amsterdam, 2000.
[6] E.N. Dancer and S. Yan. Multi-layer solutions for an elliptic problem. J. Diff. Equations,
194:382–405, 2003.
[7] E.N. Dancer and S. Yan. A minimization problem associated with elliptic systems of FitzHugh-
Nagumo type. Ann. Inst. H. Poincaré Anal. Non Linéaire, 21(2):237–257, 2004.
[8] E. De Giorgi. Sulla convergenza di alcune successioni d’integrali del tipo dell’area. Rend. Mat.
(6), 8:277–294, 1975.
[9] Q. Du, C. Liu, R. Ryham, and X. Wang. A phase field approach in the numerical study of the
elastic bending energy for vesicle membranes. J. Comput. Phys., 198:450–468, 2004.
[10] Q. Du, C. Liu, and X. Wang. Simulating the deformation of vesicle membranes under elastic
bending energy in three dimensions. J. Comput. Phys., 212:757–777, 2006.
[11] C. Gui and J. Wei. Multiple interior peak solutions for some singular perturbation problems.
J. Diff. Eqns., 158:1–27, 1999.
[12] C. Gui and J. Wei. On multiple mixed interior and boundary peak solutions for some singularly
perturbed neumann problems. Can. J. Math., 52:522–538, 2000.
[13] W. Helfrich. Elastic properties of lipid bilayers: theory and possible experiments. Z. Naturforsch
C, 28:693703, 1973.
[14] R. Kohn and P. Sternberg. Local minimisers and singular perturbations. Proc. Roy. Soc.
Edinburgh Sect. A, 111(1-2):69–84, 1989.
[15] H. Matano. Asymptotic behavior and stability of solutions of semilinear diffusion equations.
Publ. Res. Inst. Math. Sci., 15(2):401–454, 1979.
[16] L. Modica. The gradient theory of phase transitions and the minimal interface criterion. Arch.
Rat. Mech. Anal., 98(2):123–142, 1987.
[17] L. Modica and S. Mortola. Un esempio di Γ− -convergenza. Boll. Un. Mat. Ital. B (5), 14(1):285–
299, 1977.

25
[18] R. Moser. A higher order asymptotic problem related to phase transitions. SIAM J. Math.
Anal., 37(3):712–736, 2005.

[19] W.-M. Ni and J. Wei. On positive solutions concentrating on spheres for the Gierer-Meinhardt
system. J. Differential Equations, 221(1):158–189, 2006.

[20] X. Ren and J. Wei. Nucleation in the FitzHugh-Nagumo system: interface-spike solutions. J.
Diff. Eqns., 209(2):266–301, 2005.

[21] X. Ren and J. Wei. Droplet solutions in the diblock copolymer problem with skewed monomer
composition. Calc. Var. Partial Differential Equations, 25(3):333–359, 2006.

[22] U. Seifert, K. Berndl, and R. Lipowsky. Configurations of fluid membranes and vesicles. Phys.
Rev. A, 44:11821202, 1991.

[23] L. Simon. Existence of surfaces minimizing the Willmore functional. Comm. Anal. Geom.,
1(2):281–326, 1993.
[24] H. Tang and K.F. Freed. Free energy functional expansion for inhomogeneous polymer blends.
J. Chem. Phys., 94(2):1572–1583, 1991.

[25] T. J. Willmore. Riemannian Geometry. The Clarendon Press, Oxford University Press, New
York, 1993.

26