Вы находитесь на странице: 1из 18

Calc. Var.

(2006) 26(3): 313330


DOI 10.1007/s00526-006-0007-3

Calculus of Variations

Marcello Lucia

A blowing-up branch of solutions for a mean


field equation

Received: 2 December 2004 / Revised: 14 September 2005 / Accepted: 10 October 2005 /


Published online: 15 March 2006
c Springer-Verlag 2006


Abstract We consider the equation



 u
1
e
,
u =  u
||
e

u H01 ().

If  is of class C 2 , we show that this problem has a non-trivial solution u for


each (8, ). The value depends on the domain and is bounded from
below by 2 j02 , where j0 is the first zero of the Bessel function of the first kind of
order zero ( 2 j02 > 8). Moreover, the family of solution u blows-up as
8.
Keywords Mean field equations Moser-Trudinger inequality Mountain pass
theorem Faber-Krahn inequality
1 Introduction
This paper deals with the problem :
 u

e
1
u =  u
,
||
e

u H01 (),

(1.1)

where  is a bounded open set of IR2 . Clearly u 0 is a solution of (1.1) and we


are here interested in finding non-trivial solution for this problem.
Equations involving exponential nonlinearities are motivated by several problems arising in geometry and physics. Already in 1853, in connection with the
M. Lucia (B)
Dept. Math., National Center for Theoretical Sciences, Kuang Fu Rd, Hsinchu, Taiwan
E-mail: lmarcello@math.cts.nthu.edu.tw

314

M. Lucia

geometrical problem of finding a metric on  having a prescribed constant Gauss


curvature, Liouville [17] studied the semilinear equation:
u = eu .

(1.2)

Beside geometrical interpretations, Eq. (1.2) is also related to several fields of


physics. For example, it arises in some problems of combustion [4]. Also, in statistical mechanics, Caglioti, et al. [6] and Kiessling [14] proved that the Gibbs
measures defined by a system of N -particles having logarithmic interactions in a
domain  of R2 is described (when N ) by the non-local equation
eu
,
u
e

u = 

u H01 ().

(1.3)

They also proved that (1.3) has a solution for each < 8, while uniqueness of
solutions on simply connected domains was derived by Suzuki for each < 8
in [25] and by Chang et al. for = 8 in [7].
The PDE in (1.1) has been studied with several boundary conditions. In [24],
in connection with the asymptotical behavior of some solutions derived in the
framework of the Chern-Simons theory (see [26]), Struwe and Tarantello proved
existence of periodic solutions. In biology, PDE of the type (1.1) with Neumann
boundary conditions are connected to some models of chemotaxis (see for example [27]). If we consider homogenous Dirichlet boundary conditions, Problem (1.1) has also some very interesting applications in the theory of plasma or
point-vortex gas. For example, in the study of the thermal equilibrium of a twodimensional plasma confined by a magnetic field in a cylinder, Smith and ONeil
[23] analyzed the following equation:
eu
,
u
e

u = 

u H01 (),

(1.4)

where R, 0 are constants which are related to the inverse temperature


and the rotation frequency of the plasma. In [23] the problem is discussed on a
ball and their numerical simulations indicate that (1.4) has a unique solution if
< 8, and may have multiple solutions when > 8.
For the Problem (1.4), existence and uniqueness of solution on any bounded
domain are easy to prove when 0. The question becomes more subtle when
this parameter is positive. In [20], we noted that Problem (1.4) has always a solution when < 8. Moreover, we also noted that the method used already by
Suzuki in [25] shows that this solution is unique when the domain is simply connected. On non-simply connected domain, we were able to handle the question of

uniqueness when = ||


. In this case, we are reduced to the Problem (1.1) for
which u 0 is a trivial solution. The main result of [20], shows that u 0 is
actually the unique solution as far as 8 (for any bounded domain).
The goal of this paper is to construct non-trivial solution beyond this value
8. We will do this by using a variational method. Indeed, Problem (1.1) is the
Euler-Lagrange equation of the functional :





1
1

2
u
I (, u) =
|u| log
e +
u, u H01 (). (1.5)
2 
|| 
|| 

A blowing-up branch of solutions for a mean field equation

315

As a consequence of the Moser-Trudinger inequality (see [22]), this functional is


well-defined and of class C . A straightforward calculation shows moreover that
its critical points are weak solution of Problem (1.1).
To state the main result of this paper, we introduce the space of H 1 ()functions of mean zero which are constant on the boundary :



Hc1 () := H 1 () :
= 0, H01 () for some R , (1.6)


(where is a non-prescribed constant). In this space let us consider the least eigenvalue of the Laplacian :

() := inf


|| :
2

Hc1 (),

=1 .
2

(1.7)

If there is no risk of confusion, we will write  instead of (). By using Schwarz


symmetrization and calculating the value of  for a ball and the disjoint union of
two balls, we will first prove that || 2 j02 , with j0 the first zero of the
Bessel function J0 (for the properties of J0 , we refer to [28]). Let us emphasize
that since the functions in Hc1 () have not a prescribed value on the boundary,
such bound from below does not follow immediately from the classical FaberKrahn inequality (see [3] or [8]) available for the first or second eigenvalue of
the Laplacian with homogeneous Dirichlet boundary condition. As a byproduct of
this bound on , we see that the interval (8, ||) is not empty and therefore
the following Theorem applies for any bounded open set :
Theorem 1.1 Let  R2 and set E := (8, ||). Then,
(1) There exists D E dense in E such that Problem (1.1) admits a nontrivial
solution (, u ) for each D;
(2) If furthermore  is of class C 2 , non-trivial solution exist for each E.
By setting S := {(, u) E H01 () : (, u) solves (1.1), u 0}, we also
have
lim inf

(,u)S ,8

u L 2 = .

(1.8)

This paper is organized in the following way. In Sect. 2, we state the two main
ingredients that will be used to derive our existence result. The first is a deformation Lemma that has been introduced in [19]. The second tool is a compactness
result for sequences of solutions to (1.1). In Sect. 3, we prove a Faber-Krahn type
inequality for the value  and show that || 2 j02 > 8. In Sect. 4, we show
that in the range (8, ||) the functional exhibits a Mountain Pass geometry.
This fact together with our Deformation Lemma allow to conclude in Sect. 4 the
existence of a non-trivial critical point for almost all (8, ||). Using then
some a priori estimates derived from a blow-up analysis, we conclude the proof
of Theorem 1.1.

316

M. Lucia

2 Deformation Lemma and a priori estimates


The main problem that we have to deal with the functional I (, ) is the difficulty
to verify the so-called Palais-Smale condition. Actually, in [18] we show that
when = 8 N (N N), the Palais-Smale condition is not satisfied for the
functional




1
2
u
I (, u) = 1
|u| log
e , u H01 ().
(2.1)
2 
|| 
These same arguments show that the functional I (, ) defined by (1.5) admits
unbounded Palais-Smale sequence when = 8 N . In [24] and [10], this problem
is overcome by using a trick due to Struwe. They exploit some monotonicity of
the min-max value c() to construct bounded Palais-Smale sequence for almost
all in some range of the parameter. Here, we use the approach presented in [18]
and then in [19], where we proved a deformation Lemma for the functional (2.1).
Instead of using the classical flow defined by the gradient of the functional I (, ),
we introduced a new flow to overcome the problem of verifying the Palais-Smale
condition.
We first introduce a definition:
Definition 2.1 Given two sets A B H01 (), we say that A is a deformation
retract of B if there exists a map : [0, 1] H01 () H01 () satisfying
(a) is continuous;
(b) (t, u 0 ) = u 0 for all (t, u 0 ) [0, 1] A;
(c) for t = 1, (1, ) maps B on A.
The deformation lemma we will need can be stated as follows :
Proposition 2.2 Let (0, ) and 0 < a < b. Then,
(1) either there exists a sequence (n , u n ) solution of Problem (1.1) satisfying:
a I (n , u n ) b,

n (0, ],

n ;

(2) or, {u : I (, u) a} is a deformation retract of {u : I (, u) b}.


Proof Let us set


J:

H01 ()

R,

1
u
log
||

and
I : H01 () R,

1
2




||




u,




|u|2 J (u).

Then, up to some minor modifications, we can follow the proof of the deformation
Lemma given in [18] (a more general version is given in [19]).


Above deformation lemma will allow to construct solution for a dense set of
value of the parameter belonging to some interval of R. For C 2 domain, existence
in the full set will be a consequence of the following a priori estimate.

A blowing-up branch of solutions for a mean field equation

317

Proposition 2.3 Let  R2 be of class C 2 . Assume (n , u n ) is a sequence of


solution to Problem (1.1) with:
n

n .

and

Then, there exists C > 0 such that


 un 
e
sup  u C
n

e

n N.

(2.2)

Proof We first consider the a priori estimates near the boundary. Let h n be the
solution of
n
h n =
(2.3)
, h n H01 ().
||
By setting
vn := u n + h n

and

n h n (x)
e
,
un
e

K n (x) := 

we rewrite Problem (1.1) as follows:


vn = K n evn ,

vn H01 ().

(2.4)

Since  is of class C 2 , we have h n W 2,2 () [Thm 8.12, [12]]. From Sobolev
embedding [Thm. 7.26, [12]], we derive h n C 1 () and also that
|h n | + |h n | C0 |n |,

(2.5)

where C0 := C0 (). From (2.5) and since K n > 0, we get


| K n |
= |h n | C0 |n |.
Kn

(2.6)

Since n is bounded, the right hand-side of (2.6) is uniformly bounded and we


can therefore apply the arguments of [Prop.4, [21]] to Eq. (2.4) together with the
[Thm. 2.1, [11]] (note that vn > 0). In this way, we deduce the existence of a
  of  (depending on  and the uniform bound on | K | )
neighbourhood 
K
in which
vn L (
(2.7)
) C n N.
 v
Using (2.5), the a priori estimate (2.7) together with  e n || (since vn 0),
 the following a priori estimate
we get on the set 
eh n (x) evn (x)
eu n (x)


=
C
un
h n evn
e
e

 n N.
x 

(2.8)

We are left with the problem of obtaining interior a priori estimates. To do this,
we consider again the function h n defined by (2.3) and we set


un
e
(2.9)
wn := u n + h n log
and Vn (x) := n eh n .


318

M. Lucia

Hence, if (n , u n ) is a solution of Problem (1.1), the sequence wn satisfies :

wn
on ,
n = Vn (x)e
w


Vn (x)ewn = n .


Assume that sup {wn } . We will show that this implies = 8 N .


From (2.8) and (2.5), we see that
sup{wn } C



n N

and

sup {wn } + (as n ).


\

Therefore, since 0 < C1 Vn C2 , we may use the main result of Brezis (i = 1, . . . , m) such
Merle [5] and find m disjoint balls Bi = B(ai , r )  \ 
that
sup{wn } + and sup{wn } ,
Bi

 m B(ai , r )). Since Vn is furthermore relatively


where we have set K :=  \ (
i=1
2
0
compact in Cloc (), the results of Li-Shafrir [16] give
m 

i=1

Therefore, we derive


Vn ewn =


m B
i=1
i

Vn ewn 8 N .
Bi

Vn ewn +

Vn ewn +
K





Vn ewn ,

that is

|).
n = 8 N + o(1) + O(|
|). Since |
| is arbitrarily small, we
By taking n , we get = 8 N + O(|

actually obtain = 8 N .
We can now say more by using the refined blow-up analysis that has been done
in [9]. Indeed, let h n be defined by (2.3) and vn := u n + h n . Then,
vn = n

eh n evn
,
eh n evn

vn H01 (),

n 8 N .

Since sup wn (wn defined in (2.9)), we have sup


Eq. (2.10) we derive that sup vn . Since
 log(eh n ) = h n =

e

vn
vn

e

(2.10)

and using

n
> 0,
||

the results of [9] show that n 8 N > 0. In other words, blow-up can hap the
pen only for n approaching 8 N from the right. Since we assume n ,
proposition is proved.


We finally point out the following result which will be used in the last section.

A blowing-up branch of solutions for a mean field equation

319

Proposition 2.4 Assume (n , u n ) R H01 () is a sequence of solutions of (1.1)


satisfying :

eu n
(i)
|u n |2 C or (ii)  u C.
n

e
Then, up to a subsequence, (n , u n ) ( , u)
R H01 () and ( , u)
is a
solution of Problem (1.1).
Proof If (i) holds, then as a consequence of the Moser-Trudinger equation (see
[22] or [2]), we get
u n
u

in H01 ()

and

eu n
eu


un
u
e
e

Using (2.11) with the fact that




  un
e
1

,
u n = n

un
||


e

in L 2 ().

H01 (),

(2.11)

(2.12)

we derive the conclusion. If (ii) holds, by using u n as a test function in (2.12) we


derive immediately that
|u n | C
2

which shows that






 |u n |

|u n | C

1/2
|u n |

C. Hence the conclusion follows from (i).




3 A Faber-Krahn inequality
In order to prove in the following section that the functional exhibits some mountain pass geometry, we need to estimate the range of the parameter where the
quadratic form D 2 I(,0) is positive definite. By calculating explicitly D 2 I(,0) (see
Prop. 4.1), we are led to consider the value  defined by (1.7) and to find a bound
from below for . The goal of this section is to show :
|| 2 j02 ,

(3.1)

where here and in the sequel jn denotes the first zero of the Bessel function Jn
whose definition and properties can be found in [28]. In particular, from (3.1), we
see that the interval (8, ||) is never empty.
The difficulty here is that we deal with function in the space Hc1 () (defined
by (1.6)) for which the boundary value is not prescribed. If we were working
in the space H01 () then we would be able to apply directly results obtained
already by Krahn [15]. More precisely, given any R2 , let us denote by
1 () 2 () . . . the eigenvalues of the Laplacian in H01 (). The FaberKrahn inequality states that :
1 ()|| j02 ,

(3.2)

320

M. Lucia

(see for example [3] or [8]). By applying this inequality on the nodal domains of
the second eigenfunction, Krahn [15] also gave a bound from below of the second
eigenvalue :
2 ()|| 2 j02 .
(3.3)
Here, though we may not apply directly the results (3.2) and (3.3), we are nevertheless able to derive a similar conclusion for the value ().
We start by noting that from classical results of variational theory the minimization problem associated with (1.7) admits always a solution. More precisely,
there exists Hc1 () satisfying


||2 with
2 = 1.
(3.4)
() =


We check then easily that ((), ) solves the eigenvalue problem :


 = () ,

Hc1 (),

0.

(3.5)

In the case of a ball B, the first eigenvalue (B) of Problem (3.5) can be calculated
explicitly and we have :
Proposition 3.1 Let B(0, R) be a ball of R2 . Then,
 2
j1
(B) = 2 (B) =
,
R

(3.6)

where 2 (B) is the second eigenvalue of the Dirichlet-Laplacian on B and j1 is


the first zero of the Bessel function J1 . In particular,
(B)|B| = j12 > 8.

(3.7)

Proof To simplify the notation we will write  instead of (B). We first study
Problem (3.5) in the space of radial functions. If the eigenfunction solving (3.5)
is radial, it has to solve the ODE Problem:

(r  )

= (r ), r (0, R),

 Rr
(3.8)

(r )r dr = 0,
 (0) = 0, 0.
0

By integrating the left and right hand-side of above ODE on (0, R), we see that
Problem (3.8) can be reformulated as

(r  )

= (r ),
r (0, R),
(3.9)
 r

(0) = (R) = 0, 0.
Therefore, since  is the least eigenvalue of (3.5) it has to coincide with the
first positive eigenvalue of the Neumann-Laplacian restricted to the class of ra
dial function. Denoting by j  the first zero of J0 and using well-known properties
of the Bessel functions, we deduce that:
  2
  
j
j
(r ) = C J0
.
r
(C = 0),  =
R
R

A blowing-up branch of solutions for a mean field equation

321

Since on the one hand j  coincides with the first zero j1 of J1 and on the other
hand ( Rj1 )2 = 2 (B), we actually have :

(r ) = C J0

j1
r
R


and

=

j1
R

2
= 2 (B).

This concludes the case when is radial. Assume now that is not radial. Differentiate now Eq. (3.5) with respect to the angle variable . We get :




=
on B,
= 0 on B.

Since is not radial we have


0 and therefore  coincides with an eigenvalue of the Laplacian defined on the space H01 (B). By noting that
must change
sign and since  is the least eigenvalue of (3.5) we obtain as in the radial case
 = 2 (B).


We turn now our attention to open sets which are a union of two disjoint balls.
Proposition 3.2 Let B1 = B(a1 , R1 ), B2 = B(a2 , R2 ) be two disjoint balls with
R2 R1 . Then,
(B1 B2 )|B1 B2 | 2 j02 .
Proof To simplify the notation, we will write  instead of (B1 B2 ). Consider
Hc1 (B1 B2 ) satisfying (3.4) and we remind that (, ) solves (3.5). We
first deal with the radial case and then consider non-radial function. More
precisely, consider the subset of functions Hc1 (B1 B2 ) satisfying :
B1 R,
B2 R,

x
(x a1 ) is radial ,
x
(x a2 ) is radial ,

(3.10)
(3.11)

and define
1
(B1 B2 ) := { Hc1 (B1 B2 ) : satisfies (3.10) and (3.11)}.
Hc,rad
1
(B1 B2 ) and 0 on B2 .
Case 1: Hc,rad
In this case we have

 =  on B1

= 0,
( a1 ) radial .

B1

Hence, we are reduced to the first part of the proof of Prop. 3.1 and we deduce
 = ( Rj11 )2 . Therefore,
|B1 B2 | = j12

R12 + R22
R12

j12 .

(3.12)

322

M. Lucia

1
Case 2: Hc,rad
(B1 B2 ), 0 on B1 and 0 on B2 .
We see in this case that is equal to

x B(a1 , R1 ),
(x a1 ) = J0 ( |x a1 |)

(x a2 ) = A J0 ( |x a2 |) x B(a2 , R2 ) (A = 0),

where (A, ) will be subjected to the conditions


J0 ( R1 ) = A J0 ( R2 ),
 R2
 R1

J0 ( r )r dr +
A J0 ( r )r dr = 0.

(3.13)

(3.14)

From the system (3.14), we deduce that  must satisfy




J0 ( R2 )

R1

J0 (t)t dt + J0 ( R1 )

R2

J0 (t)t dt = 0.

(3.15)

We check easily that (3.15) is equivalent to

R1 J0 ( R2 )J0 ( R1 ) + R2 J0 ( R1 )J0 ( R2 ) = 0,


equation which can be written as

1 J0 ( R1 )
1 J0 ( R2 )
+
= 0.

R2 J0 ( R2 )


R1 J0 ( R1 )

(3.16)

Hence, we are led to study the zeros of the function


(0, j1 ) (0, j1 ) R,

(x, y)

1 J0 (x) 1 J0 (y)
+
.
x J0 (x)
y J0 (y)

(3.17)

In Prop. .2 of the appendix, we give a proof that any (x, y) zero of (3.17) has the
property x 2 + y 2 2 j02 . By applying this result to (3.16), we obtain :
R12 + R22 2 j02 .
Therefore,



|B1 B2 | =  R12 + R22 2 j02 .

(3.18)

1
Case 3: Hc,rad
(B1 B2 ).
In this case up to translation we may assume that B1 is centered at the origin and
that is not radial on B1 . By considering the derivative of with respect to the
angular variable , we get :




=
,

= 0 on B1 ,


0 on B1 .

A blowing-up branch of solutions for a mean field equation

323

Hence,  has to be an eigenvalue of Laplacian in H01 (B1 ). We then derive easily


that:
 2


j1
|B1 B2 | =
R12 + R22 j12 .
(3.19)
R1
By comparing (3.12), (3.18) and (3.19), we conclude the proof of the theorem.


By using Schwarz symmetrization together with Propositions 3.1 and 3.2, we
are able to handle the case of a general open bounded set :
Proposition 3.3 Let  R2 . Then, || 2 j02 .
Proof Let be defined by (3.4). By setting c := | , we consider
+ := {x  : (x) c}

and

 := {x  : (x) < c}.

Choose two disjoint balls B + , B with


|B + | := |+ |

and

|B | := | |.

Consider on these two balls, the Schwarz symmetrization


+ : B+ R

and

: B R

of the functions |+ and ()| (for the definition and property of the
Schwarz symmetrization, we refer to [13]). Let us finally set :
 +
(x)
x B +;

: B B R, x
(3.20)

(x) x B .
The crucial point is that the sets {|+ > t} (respectively {| > t}) do not
touch the boundary of + (respectively  ). Hence, the same arguments used to
derive the classical Faber-Krahn inequality give:



2
2
2
 ||
+|| +  ||
() =
=
2
2
2
+ + 
 

+ 2
2
+ | | + B | |

B
+ 2
2
+ | | + B | |
 B
2
+
| |
= B B
2
B + B | |
(B + B ).

(3.21)

In the case | | = 0, we have |B | = 0 and Hc1 (B + ). Hence, relation (3.21) together with Prop. 3.1 give
()|| (B + )|| = (B + )|B + | = j12 .

(3.22)

We derive the same conclusion if |+ | = 0. If |+ | = 0 and | | = 0, from


relation (3.21) and Prop. 3.2 we derive :
()|| (B + B )|| = (B + B )|B + B | 2 j02 .
The estimates (3.22) and (3.23) give the conclusion since 2 j02 < j12 .

(3.23)



324

M. Lucia

To conclude this section, we emphasize that the value of || may be very
large. More precisely, given M > 0, we can always find a domain  that has the
property ()|| M. This can be justified easily by considering rectangles.
Proposition 3.4 Let R = (0, a) (0, b) with a b 1. Then,

(R)|R| = 2


1
4
+
ab.
a2
b2

(3.24)

Proof We will set  := (R). Let be defined by (3.4) and differentiate (3.5)
with respect to the variable y. We see that the function :=
y satisfies the
following eigenvalue problem with mixed boundary conditions :

 = ,

(0, y) = (a, y) = 0

(x, 0) = (x, b) = 0
n
n

y (0, b),

(3.25)

x (0, a).

We check easily that


y 0. Therefore,  is an eigenvalue of the Problem (3.25).
Hence, we necessarily have





m
n
= Cm,n sin
x cos
y
m = 1, 2, 3, . . .
y
a
b

n = 0, 1, 2 . . .

(3.26)
must change sign.
Due to the boundary condition satisfied by , we see that
Hence, we cannot have n = 0 in (3.26). Therefore, has to be of the form :

 

m
n
= Cm,n sin
x sin
y + f (y) m, n = 1, 2, 3, . . .
a
b
Using the boundary conditions, we deduce that f 0, which gives :

= Cm,n sin




m
n
x sin
y
m, n = 1, 2, 3, . . .
a
b

(3.27)

If (m, n) = (1, 1), we check easily that the function (3.27) cannot satisfy our
initial problem (3.5). For (m, n) = (2, 1), the function (3.27) is of mean zero and
satisfies then the Problem (3.5). Hence, we deduce


  
2

= Cm,n sin
x sin
y
a
b


and

=

2
a

2

 2

+
.
b




A blowing-up branch of solutions for a mean field equation

325

4 Existence result
The following proposition shows that the functional (1.5) has a mountain pass
geometry for each (8, ||), where  is defined by (1.7).
Proposition 4.1 Let (8, ||). Then, there exist , > 0 such that
I (, u)
Moreover, there exists u 0

H01 ()

(4.1)

such that

u 0 L 2 >
Proof We check easily that

u = .

and

I (, u 0 ) < 0.

(4.2)


 


1
2
2
| |
)
(
D I(0) (, ) =
|| 
|| 


 
 2



1
2


=
| |

||  
||  




1
| |2 .
|| 


Therefore, when < ||, the point u 0 is a strict local minimum of the
functional (1.5). The first assertion of the Proposition follows.
Let us now prove the second statement. Given (a, )  (0, ), consider
the function :
82
(x) = log
x IR2 ,
(4.3)
(1 + 2 |x a|2 )2
which is solution of the problem

u
2
u = e on IR ,
eu < .
(4.4)
IR2

Fix then a ball B := B(a, R)  and consider the projection on H01 (B) of
defined by
 = , H01 (B).
2

8
We see easily that = log( (1+
2 R 2 )2 ), that is

82
(x) = (x) log
(1 + 2 R 2 )2


x B.

By using (4.5), a straightforward calculation shows then that



||2 = 16 log 2 + O(1),
B



log
e = log(2 ) + O(1),
B

= O(1).
B

(4.5)

(4.6)
(4.7)
(4.8)

326

M. Lucia

Define now the function



(x) :=

(x)
0

if x B,
if x  \ B.

(4.9)

Using (4.6), (4.7) and (4.8) we deduce that


I (, 
) = (8 ) log 2 + O(1).

(4.10)

Hence, on the one hand L 2 (by 4.6) and on the other hand for each
> 8, we have I (, ) as (by 4.10). This proves the second
assertion of the proposition.


We are now able to prove that Problem (1.1) has a non-trivial solution when
the parameter E := (8, ||)
Proof of Theorem 1.1 The proof is divided in three parts. Based on the Deformation Lemma given in Sect. 2, we first prove existence of non-trivial critical points
for a dense set of value of E. In a second part, by assuming  to be of class
C 2 , existence is obtained for the full set E by using the a priori estimates given in
Prop. 2.3 (following the idea of [24]). Finally we prove that the set of solutions is
unbounded in H01 when tends to 8.
1) Existence for a dense set.
Given (8, ||), choose , > 0 and u 0 H01 () given by Prop. 4.1. Let
us consider the set of continuous curve in H01 () connecting u 0 to u 0 :
 := { C([0, 1], H01 ()) : (0) = 0, (1) = u 0 },
and consider also the minmax value :
c := inf max I (, (t)).
 t[0,1]

(4.11)

If the Palais-Smale condition would hold, the arguments of [1] would show that
this minmax value c is a critical value for the functional I (, ). Here, without this
compactness condition, we prove a slightly different result. We first note that since
u 0 L 2 and > 0, we have
c > 0.

(4.12)

Assume there is no sequence (n , u n ) such that:



I (n , u n ) = 0, n [0, ),
c  < I (n , u n ) < c + ,

n ,
c
with  = .
2

(4.13)

Then, by Prop. 2.2, I c is a deformation retract of I c+ through a continuous


map : [0, 1] H01 () H01 () where
I c+ := {u : I (, u) < c + }

and

I c := {u : I (, u) < c }.

Since u 0 , 0 I c , for each  and t [0, 1], we note that the deformed
curve
t : [0, 1] H01 (), s (t, (s)),

A blowing-up branch of solutions for a mean field equation

327

is still in the set . Consider now any 0  having the property



0 (s) I c+
s [0, 1],
for some s [0, 1],
I (0 (s)) (c , c + )

(4.14)

(such curve exists by the definition of the minmax value c). Then, we have
inf max I ( (s)) max I ((t, 0 (s)))

 s[0,1]

s[0,1]

t [0, 1];

but at t = 1, the right hand-side is less than c  due to (4.14) and the fact that
is a deformation retract. A contradiction with the definition of c. Hence, the
property (4.13) must hold.
2) Existence on (8, ||) when  is of class C 2 .
Let (8, ||) Consider a sequence (n , u n ) satisfying (4.13). Using
Prop. 2.3 and Prop. 2.4, we deduce that (up to a subsequence) (n , u n ) converges
u)
strongly in R H01 () to a (,
solution of Problem (1.1). Moreover, from (4.13)
we also derive
u)
0 < c  < I (,
< c + .
As a consequence, we deduce that u 0.
3) Proof of Property (1.8).
If (1.8) does not hold, then there exists a sequence (n , u n ) S such that
n 8

and

u n 2L 2 C.

Using again Prop. 2.4, we see that, up to a subsequence, (n , u n ) converges


strongly in R H01 () to (8, u)
solution of (1.1). Since  is of class C 2 , stan But, from [20], we know
dard regularity results show that u C 2 () C 0 ().
2
0

that u 0 is the unique solution in C () C () of Problem (1.1) at = 8.


Therefore, we deduce that u 0, namely (8, 0) is a bifurcation point for Problem (1.1). We claim that this is impossible. Indeed, as a consequence of the MoserTrudinger inequality ([2, 22]), we derive that the mapping
F : H01 () L 2 (),

eu
,
u
e

u 

is of class C 1 with


 
1
1
D F(0) ( ) =
.

||
|| 
Using (4.15) and (n , u n L 2 ) (8, 0), we deduce easily that
u 0 solving :
verges weakly in H01 () to some 

 
8
1

u=

u

u , 
u H01 ().
||
|| 

(4.15)
un
u n L 2

con-

Therefore, recalling the definition of () in (1.7) we obtain


8
().
||

(4.16)

But (4.16) is a contradiction since the Faber-Krahn inequality proved in Proposition 3.3 says that actually ()|| > 8. This shows that (1.8) must hold.



328

M. Lucia

Acknowledgements The author is grateful to Prof. A. Bahri, Prof. M. Kiessling, Prof. Y.Y. Li
for very fruitful discussions and to Prof. C.S. Lin for bringing [9] to his attention. He also
thanks Prof. Cao Daomin for his warm hospitality and his support at the Institute of Applied
Mathematics (CAS, Beijing) where this paper was accomplished.
1 J0 (x)
x J0 (x)

Appendix A: Zeros of

1 J0 ( y)
y J0 ( y)

Let Jn be the Bessel function of the first kind of order n and denote by jn the first zero of Jn . We
note for the sequel that j1 coincides with the first zero of J0 (see [28]). The goal of this appendix
is to locate the zeros of the problem:
1 J0 (y)
1 J0 (x)
+
= 0,

x J0 (x)
y J0 (y)

(x, y) (0, j1 ) (0, j1 ).

(A.1)

With this aim, let us define the function


 : (0, j1 ) R,

1 J0 (s)
.
s J0 (s)

(A.2)

Rewrite Eq. (A.1) as follows :


(x) + (y) = 0,
and define

(x, y) (0, j1 ) (0, j1 ),

(A.3)

Z := {(x, y) (0, j1 ) (0, j1 ) : (x) + (y) = 0}.

Using the well-known properties of the Bessel functions [28], we check easily
( j0 ) = 0,
lim (s) = ,

s0

(A.4)

lim (s) = +,

s j1

(A.5)

1
(A.6)
+ s|(s)|2 > 0 s (0, j1 ).
s
From (A.4), we have ( j0 , j0 ) Z and the relation (A.6) shows that the function  is strictly
monotone. In particular, we deduce :
 (s) =

y
(x) + (y) strictly monotone x (0, j1 ),
x
(x) + (y) strictly monotone y (0, j1 ).

(A.7)
(A.8)

We now prove the following result on the location of the set Z :


Proposition .2 Let (x, y) Z . Then, x 2 + y 2 2 j02 .
Proof We will show that

Z {(x, y) : x 2 + y 2 = 2 j02 } = {( j0 , j0 )}.


This fact together with the monotonicity properties (A.7) and (A.8) will then imply that

Z {(x, y) (0, ) (0, ) : x 2 + y 2 2 j02 } = {( j0 , j0 )},


which will conclude the proposition.
To prove (A.9), let us consider the function
 : (0,


2 j0 ) R,

x (x) + 


 2
2 j0 x 2 ,

(A.9)

A blowing-up branch of solutions for a mean field equation

329

. Using (A.6), the


and note that (A.9) is equivalent to the fact that j0 is the unique zero of 
 is given by :
derivative of the function 





1
x
1
 (x) = + x2 (x) 

+ 2 j02 x 2 2 2 j02 x 2

x
2 j02 x 2
2 j02 x 2



j2 x2
(x).
+ x (x)  2 j02 x 2 
= 2 02
(A.10)
2
x(2 j0 x )
Moreover at the point j0 , we have
( j0 ) = 0,


 ( j0 ) = 0,


(A.11)

and from (A.10) a straightforward calculation shows that


 ( j0 ) =


4
< 0.
j02

(A.12)

 has some zero on the open interval ( j0 , j1 ). Consider the first


Assume now that the function 
 (x0 ) 0.
zero x0 > j0 on this interval. By the properties (A.11) and (A.12) we deduce that 
 (x0 ) < 0. A contradiction.
But on the other hand, by evaluating (A.10) at x0 , we deduce that 
 has no zero on the interval ( j0 , j1 ) and by symmetry, the same property holds on the
Hence, 
 which establishes the statement (A.9). 

interval (0, j0 ). Therefore, j0 is the unique zero of 

References
1. Ambrosetti, A., Rabinowitz, P.H.: Dual variational methods in critical points theory and
applications. J. Funct. Anal. 14, 349381 (1973)
2. Aubin, T.: Nonlinear analysis on manifolds. Monge-Amp`ere equation. Springer-Verlag,
Berlin (1982)
3. Bandle, C.: Isoperimetric inequalities and applications. Pitman, London (1980)
4. Bervenes, J., Eberly, D.: Mathematical problems from combustion theory. Appl. Math. Sc.
Springer-Verlag, New-York 83 (1989)
5. Brezis, H., Merle, F.: Uniform estimates and blow-up behavior for solutions of u =
V (x) exp(u) in two dimensions. Comm. PDE 16, 12231253 (1991)
6. Caglioti, E., Lions, P.L., Marchioro, C., Pulvirenti, M.: A special class of stationary flows
for two-dimensional Euler equations: a statistical mechanics description. Commun. Math.
Phys. 143, 501525 (1992)
7. Chang, S.-Y.A., Chen, C.C., Lin, C.S.: Extremal functions for a mean field equation in two
dimension. Lecture on Partial Differential equations in honor of Louis Nirenbergs 75th
birthday, Chapter 4, International Press (2003)
8. Chavel, I.: Eigenvalues in Riemannian geometry. Academic Press, Orlando (1984)
9. Chen, C.C., Lin, C.S.: Sharp estimates for solutions of multi-bubbles in compact Riemann
surfaces. Comm. Pure Appl. Math. 55, 728771 (2002)
10. Ding, W., Jost, J., Li, J., Wang, G.: Existence results for mean field equations. Ann. Inst. H.
Poincare Anal. Non Lineaire 16, 653666 (1999)
11. Gidas, B., Ni, W.-M., Nirenberg, L.: Symmetry and related properties via the maximum
principle. Commun. Math. Phys. 68, 209243 (1979)
12. Gilbarg, D., Trudinger, N.S.: Elliptic partial differential equations of second order. Reprint
of the 1998 edition. Classics in Mathematics. Springer-Verlag, Berlin (2001)
13. Kawohl, B.: Rearrangements and convexity of level sets in PDE. Lecture Notes in Mathematics 1150, Springer-Verlag (1985)
14. Kiessling, M.K.-H.: Statistical mechanics of classical particles with logarithmic interactions, Comm. Pure Appl. Math. 46(1), 2756 (1993)

15. Krahn, E.: Uber


Minimaleigenschaften der Kugel in drei und mehr Dimensionen. Acta
Comm. Univ. Tartu (Dorpat) A9, 144 (1926)

330

M. Lucia

16. Li, Y.Y., Shafrir, I.: Blow-up analysis for solutions of u = V exp(u) in dimension two.
Ind. Univ. Math. J. 43, 12551270 (1994)
2 log
2a2 = 0. J. de Math. 18,
17. Liouville, J.: Sur lequation aux differences partielles ddudv
7172 (1853)
18. Lucia, M.: A Deformation Lemma for a Moser-Trudinger type functional. Nonlinear Anal.
63, 282299 (2005)
19. Lucia, M.: A Deformation Lemma with an application to a mean field equation. preprint
20. Lucia, M., Zhang, L.: A priori estimates and uniqueness for some mean field equations. J.
Differential Equations 217, 154178 (2005)
21. Ma, L., Wei, J.C.: Convergence for a Liouville equation. Comment. Math. Helv. 76(3),
506514 (2001)
22. Moser, J.: A Sharp Form of an Inequality by N. Trudinger. Indiana U. Math. J. 20, 1077
1092 (1970/71)
23. Smith, R.A., ONeil, T.M.: Nonaxisymmetric thermal equilibria of a cylindrically bounded
guiding center plasma or discrete vortex system. Phys. Fluids B 2(12), 29612975 (1990)
24. Struwe, M., Tarantello, G.: On multivortex solutions in Chern-Simons Gauge theory. Boll.
U.M.I. B(8)(1), 109121 (1998)
25. Suzuki, T.: Global analysis for a two-dimensional elliptic eigenvalue problem with the exponential nonlinearity. Ann. Inst. H. Poincare, Anal. Non lineaire 9, 367397 (1992)
26. Tarantello, G.: Multiple condensate solutions for the Chern-Simons-Higgs theory. J. Math.
Phys. 37, 37693796 (1996)
27. Wang, G., Wei, J.: Steady state solutions of a reaction-diffusion system modeling chemotaxis. Math. Nachrichten 233/234, 221236 (2002)
28. Watson, G.N.: A treatise on the theory of Bessel functions, 2nd edition. Cambridge University Press, Cambridge (1944)

Вам также может понравиться