Professional Documents
Culture Documents
James SERRIN
School of Mathematics - University of Minnesota - Minneapolis (Minnesota)
Abstract
We study the asymptotic behavior of radially symmetric ground state solutions of a quasilinear
elliptic equation involving the m-Laplacian. The case of two vanishing parameters is considered: we
show that these parameters have opposite effects on the asymptotic behavior. Moreover the results
highlight a surprising phenomenon: different asymptotic behaviors are obtained according to whether
n > m2 or n m2 , where n is the dimension of the underlying space.
Introduction
Let m u = div(|u|m2 u) denote the degenerate m-Laplace operator and consider the quasilinear
elliptic equation
m u = um1 + up1
in IRn ,
(Pp )
where n > m > 1, m < p < m , 0 and
m =
nm
.
nm
By the results in [Ci, GST] (see also [AP1, BL] for earlier results in the case m = 2) we know that (Pp )
admits a ground state for all p, in the given ranges. Here, by a ground state we mean a C 1 (IRn ) positive
distribution solution of (Pp ), which tends to zero as |x| . Since in this paper we only deal with
radial solutions of (Pp ), from now on by a ground state we shall mean precisely a radial ground state. It
is known [PS, ST] moreover that radial ground states of (Pp ) are unique.
Equation (Pp ) is of particular interest because of the choice of the power m 1 for the lower order
term: if m = 2 (i.e. m = ) this is just the linear case, while for any m > 1 the lower order term has
Supported by the Italian MURST project Metodi Variazionali ed Equazioni Differenziali non Lineari.
the same homogeneity as the differential operator m , a fact which allows the use of rescaling methods.
Moreover, this case is precisely the borderline between compact support and positive ground states, see
[FLS, Section 1.3].
It is our purpose to study the behavior of (radial) ground states of (Pp ) as p m , 0. As far
as we are aware, the asymptotic behavior of solutions of (Pp ) has been studied previously only for the
vanishing parameter = m p and only in the case of bounded domains, see [AP2, GP, Ga, KP, R1, R2]
and references therein.
Consider first the case when = 0. Then (Pp ) becomes
m u = up1
in IRn ,
(Pp0 )
which by [NS2, Theorem 5] admits no ground states (recall p < m ). It is of interest therefore to study
the behavior of the ground states u of (Pp ) as 0 and p is fixed: in Theorem 1 below we prove in
this case that u 0 uniformly on IRn and moreover estimate the rate of convergence. As a side result,
the arguments used in the proof of Theorem 1 allow us to show that the corresponding ground states u
converge to a Dirac measure concentrated at x = 0 when , see Theorem 9 in Section 4 below.
Next, let p = m and > 0; then (Pp ) becomes
1
m u = um1 + um
in IRn ,
)
(Pm
which by the results in [NS1] again admits no ground states. Thus we next study the behavior of ground
states u of (Pp ) as = m p 0 with > 0 fixed. We prove in Theorem 2 that u then converges to a
Dirac measure concentrated at the origin, namely, u(0) and u(x) 0 for all x 6= 0, while also, at
the same time, u converges strongly to 0 in any Lebesgue space Lq (IRn ) with m 1 q < m . Our study
also reveals a striking and unexpected phenomenon: the asymptotic behavior is different in the two cases
n m2 and n > m2 ; for instance, in the case m = 2 (i.e. m = ) there is a difference of behavior
between the space dimensions n = 3, 4 and n 5. More precisely, if n > m2 we show that u(0) blows
2
up asymptotically like (nm)/m while if n m2 it blows up at a stronger rate. This phenomenon
is closely related with the Lm summability of functions which achieve the best constant in the Sobolev
embedding D1,m Lm , see [Ta] and (1) below for the explicit form of these functions.
Finally, let both p = m and = 0; then equation (Pp ) reads
1
m u = um
in IRn ,
0 )
(Pm
Ud (x) = d 1 + D d nm |x|) m1
i nm
m
(d > 0),
(1)
where D = Dm,n = (m 1)/(n m)n1/(m1) . Since the effects of vanishing m p and are in some
sense opposite, it is reasonable to conjecture that there exists a continuous function h, with h(0) = 0,
such that if = h(), p = m , then ground states u of (Pp ) converge neither to a Dirac measure nor
to 0! In Theorem 4 below we prove the surprising fact that when n > m2 this equilibrium occurs exactly
2
when and are linearly related, h() Const . Moreover in this case the corresponding ground states
0 ), namely a function of the
u then converge uniformly to a suitably concentrated ground state of (Pm
family (1), with the parameter d = Ud (0) representing a measure of concentration and depending on
the limiting value of the ratio h()/.
Let us heuristically describe the phenomena highlighted by our results. When p m with fixed,
the mass of the ground state u of (Pp ) tends to concentrate near the point x = 0, that is, all other points
of the graph are attracted to this point: in order to let the other points fit near x = 0 the maximum
level u(0) is forced to blow up. When 0 with p fixed, the ground state spreads, since now x = 0
behaves as a repulsive point, forcing the maximum level to blow down in order not to break the graph.
When both = m p and tend to 0 at the equilibrium velocity = h(), the point x = 0 in neither
attractive nor repulsive: in this case, a further striking fact is that the exponential decay of the solution
u of (Pp ) at infinity reverts to a polynomial decay.
The outline of the paper is as follows. In the next section we state our main results, Theorems 1 5.
Then in Section 3 we present background material on radial ground states, including an estimate for the
asymptotic decay as r of ground states of (Pp ), see Theorem 8. This estimate, along with Theorems
6 and 7 in Section 3, seems to be new and may be useful in other contexts. These results allow us to give
a simple proof of Theorem 5 while the proofs of Theorems 1 4 are given in subsequent sections.
Main results
The existence and uniqueness of radial ground states for equation (Pp ) is well known [GST, ST]. We
state this formally as
Proposition 1 For all n > m > 1, m < p < m and > 0 the equation (Pp ) admits a unique radial
ground state u = u(r), r = |x|. Moreover u0 (r) < 0 for r > 0.
We start the asymptotic analysis of (Pp ) by maintaining p fixed and letting 0. An important role
will be played by the rescaled problem ( = 1)
m v = v m1 + v p1
in IRn .
(Qp )
By Proposition 1 there exists a unique (radial) ground state v of (Qp ), so that the constant
= v(0)
(2)
pm 1
1
m1
|x| m1 + o m1 |x| m1
as 0.
(3)
From Theorem 1 we can also obtain a result which, while slightly beyond the scope of the paper, is
nevertheless worth noting. It states that the unique solution of (Pp ) for fixed p < m tends to a Dirac
measure as , see Theorem 9 in Section 4.
We now maintain > 0 fixed and let p m . In order to state our main asymptotic result for this
case, it is convenient to introduce the beta function B(, ) defined by
B(a, b) =
ta1
dt,
(1 + t)a+b
a, b > 0.
Then we put
m,n = n
m
nm
and
m,n
"
m1
nm
= n
n
m
m1
n(m1) nm2
, m
m
n
B n(m1)
,
m
m
2 B
m1 #n/m
(nm)/m2
for n > m2
n(m 1) n
B
,
m
m
(n = measure S n1 ).
lim
"
(nm)/m2
u(0) =
m,n
if n > m2
(4)
if n m2 .
(5)
0 IRn
um1 = 0,
lim
0 IRn
um = m,n .
(6)
Theorem 2 gives a complete description of the asymptotic behavior of u when n > m2 ; it leaves open
only the exact behavior when n m2 . This latter question is considered in more detail in Section 5.2.
The results given there, while not as precise as in the case n > m2 , nevertheless provide significant insight
into the behavior of u(0) as 0 beyond that described in the second case of (4), see Lemma 8.
Condition (6) shows that, as 0, not only does u approach a Dirac measure (u(0) and
u(|x|) 0 for |x| 6= 0), but also that the Lm norm of u approaches a non-zero finite limit. It is a
remarkable fact, also, that the limit relation (6) is independent of the value of . (It may be worthwhile
to note as well that if the exponent m 1 in the first equation of (6) is replaced by a general value q then
the limit remains 0 for m 1 < q < m but becomes if q > m ).
4
Remark. The constants in Theorem 2 in the important case m = 2, n > 4 are given by
2,n
(n2)/4
B n2 , n4
4n
2
=
n n
2
(n 2) B 2 , 2
2,n =
n
n n
[n(n 2)]n/2 B
,
,
2
2 2
lim u(0)|u(x)|m1
and
lim
0 IRn
nm
m1
m1
Cm,n |x|1n
|u|m = m,n .
(7)
(8)
Finally, we may accurately describe the behavior of the ground states of (Pp ) when = m p and
approach zero simultaneously.
Theorem 4 For > 0 and m < p < m , let u be the unique ground state of (Pp ). Then for all d > 0
there exists a positive continuous function () = (, d) such that
2
(i) () (d/m,n )m /(nm) as 0 (when n > m2 ), and () 0 as 0 (when n m2 ).
(ii) If = (), p = m , then u(0) = d.
Moreover
u Ud
as = m p 0
uniformly on IRn , where Ud is the function defined in (1).
If , 0 without respecting the equilibrium behavior Const (in the case n > m2 ), the central
height u(0) of the ground state may either converge to zero or diverge to infinity. We note finally that
as soon as the asymptotic behavior of u(0) as p m is more accurately determined in the case n m2
of (4), one also gets a more precise statement of (i): of course, the equilibrium behavior will no longer
be Const .
To conclude the section, we supply two global estimates for u(0), supplementing the asymptotic
conditions (3) and (4).
Theorem 5 Let u be a ground state of (Pp ). Then
1/(pm)
u(0) >
mp
mn p(n m)
u(0) <
p n m(n 1)
m n p(n 1)
(9)
1/(pm)
(10)
The proof of this result is given in next section. By setting = 1 in Theorem 5 we obtain related
estimates for the parameter = v(0) in Theorem 1. Also from Theorem 2 we have the following
asymptotic formula for , with = m p 0,
2
if n > m2 ;
(m, n, p)
1.6
1.8
2.11< <57.67
1.2
1.9
3.89< <37.61
1.1
2.4
1.6
5.36< <10.72
1.2
1.4
9.1< <525.22
In this section we consider the ground state problem for the general equation
m u = f (u)
in IRn ,
(11)
where the function f is assumed only to be continuous on [0, ) and to obey the condition
f (0) = 0,
f (u) < 0
for u near 0.
(12)
A radial ground state u = u(r), r = |x|, of (11) is in fact a C 1 solution of the ordinary differential
equation
n 1 0 m2 0
(|u0 |m2 u0 )0 +
|u |
u + f (u) = 0,
r > 0,
r
(13)
0
u(0) = > 0, u (0) = 0
for some initial value > 0. For our purposes the dimension n may in fact be considered as any real
number greater than m.
Put
Z u
F (u) =
f (s) ds
(14)
0
(15)
r
n
as r 0,
F () = (n 1)
as r ,
|u0 (r)|m
dr
r
and
E(r) > 0
r 0,
E(r) 0
as r .
In the next result we recall a Pohozaev-type identity due to Ni and Serrin [NS1].
Proposition 3 Let u = u(r) be a radial ground state of (13), and put
Q(r) = nmF (u) (n m)uf (u).
(16)
Then the functions rn1 Q(r) and rn1 F (u(r)) are in L1 (0, ), and moreover
Z
Q(r)rn1 dr = 0.
(17)
Remark. In other terms, the result of Proposition 3 says that the functions Q(|x|) and F (u(|x|)) are in
R
L1 (IRn ) and that IRn Q(|x|) dx = 0.
For completeness we give a proof of Proposition 3. By direct calculation, using (13), one finds that
P (r) =
Q(t)tn1 dt,
r > 0,
where
P (r) = (n m)rn1 u(r)u0 (r)|u0 (r)|m2 + mrn E(r).
Since E =
m1 0 m
m |u |
m1 0
|u (r)|m
m
for all sufficiently large r. Using Proposition 2 then gives rn1 |u0 |m1 Const. and
rn |F (u(r))|, rn E(r) Const. r(nm)/(m1)
(18)
r 0
Q(t)tn1 dt = 0.
But from (18) we get rn1 |F (u(r))| L1 (0, ), while also uf (u) < 0 for all sufficiently large r. Thus the
previous equation together with the definition of Q(r) shows in fact that rn1 Q(r) is in L1 (0, ) and
that (17) holds. This completes the proof.
2
Proposition 3 has the following important consequence.
7
f or 0 < s < .
(19)
Then > .
Proof. Suppose for contradiction that . Then since u0 < 0 for r > 0, it follows that u(r) < for all
r > 0. In turn, by the hypothesis (19) we have Q(r) = nmF (u) (n m)uf (u) < 0 for all r > 0, which
contradicts Proposition 3.
2
An upper bound for u(0) can also be obtained in some circumstances, as in the following
Theorem 7 Suppose f 0 (s) 0 whenever f (s) > 0 and that there exists > 0 such that
nF (s) (n 1)sf (s) 0
f or s .
(20)
Then < .
Proof. We assert that the function r
7
(r) = r1 |u0 (r)|m1 is decreasing on (0, ). By direct
calculation, using (13),
r0 (r) = f (u) n(r).
If f (u) 0 then 0 < 0. On the other hand, for all r such that f (u) > 0, we have (f (u) n(r))0 =
f 0 (u)u0 n0 (r) n0 (r), by hypothesis. Consequently
(r0 )0 n0 .
By integration this gives rn+1 0 (r) r1n+1 0 (r1 ) on any interval (r1 , r) where f (u) > 0. The assertion
now follows by an easy argument, once one notes notes that rn+1 0 (r) 0 as r 0.
Now by Proposition 2 and the assertion, we have
F () = (n 1)
|u0 (r)|m
dr = (n 1)
r
Since by Proposition 2 we also have (0) = f ()/n, this gives nF () (n 1)f () < 0. The conclusion
now follows from the main hypothesis (20).
2
Using Theorems 6 and 7 it is now easy to obtain the
Proof of Theorem 5. Equation (Pp ) can be written in the form (11), or (13), with
f (s) = sm1 + sp1 ,
Q(r) = mum +
mp
mn p(n m)
mn p(n m) p
u .
p
1/(pm)
Moreover
nF (s) (n 1)sf (s) =
n m(n 1) m n p(n 1) p
s +
s .
m
p
p n m(n 1)
m n p(n 1)
1/(pm)
We conclude the section by showing that radial ground states u = u(r) of (P ) have exponential decay
as r approaches infinity. This is well-known in the case m = 2, see [BL, Theorem 1 (iv)]: we give here a
different proof in the general case m > 1.
Theorem 8 Suppose that there exist constants , , > 0 such that f satisfies the inequality
sm1 f (s) sm1
f or 0 < s < .
(21)
Then there exist constants 0 , 1 , 2 > 0, (depending on m, n, , ) such that, for r suitably large,
u(r) 0 er
|u0 (r)| 1 er
|u00 (r)| 2 er .
(22)
Remark. For general nonlinearities f in (13), one usually expects polynomial decay at infinity, see
[NS1, Lemma 5.1] [ST, Proposition 2.2]. Nevertheless, Theorem 8 is not entirely unexpected, since the
nonlinearity (21) has borderline behavior which separates compact support and positive ground states,
see [FLS, Section 1.3].
Proof of Theorem 8. Obviously u = u(r) satisfies (13). Let R 0 be such that u(r) when r R.
Since u 0 as r , it is clear that such a value R exists. By Proposition 2 and the right hand
inequality of (21) we thus obtain
m1 0
u0 (r)
>
u(r)
m1
1/m
r R.
(23)
integrating this inequality on the interval [R, r] yields the first part of the result, with
0 = eR ,
= (/(m 1))1/m .
(24)
(25)
Since f (u) < 0 for u near 0, it follows that rn1 |u0 (r)|m1 is ultimately decreasing, clearly to a nonnegative limit as r (this is the first result of Proposition 2). By the exponential decay proved
9
above, the limit must be 0. Therefore we can integrate (25) on [r, ) for r R to obtain, with the help
of (21),
r
n1
m1
|u (r)|
n1
f (u(t)) dt <
n1 m1
(t)dt
m1
0
r R.
n1 0
|u (r)|m1 f (u).
r
From the right hand inequality of (21) we get f (u) 0 for r R, which shows that u00 (r) > 0 for all
r R. Further, from the left hand inequality,
n1
um1 (r)
|u0 (r)| +
.
(m 1)R
m 1 |u0 (r)|m2
m1
(m2)/m
u(r) 2 er
r R.
Remarks. The first estimate of (22) requires only the right hand inequality of (21) for its validity.
It almost goes without saying that the function f (u) = um1 + up1 satisfies (21) for suitable , .
Proof of Theorem 1
Let u = u(r) be a ground state of (Pp ). Define v = v(r) by means of the rescaling
v(r) =
1/(pm)
r
1/m
(26)
so that v is the unique ground state of the rescaled equation (Qp ). By the definition (2) and by (26) one
has u(0) = 1/(pm) .
Next, from (Qp ) we find, as in (25),
0
m1
|v (r)|
=
=
=
10
p1 m1
n
!1/(m1)
rm/(m1) + o(rm/(m1) )
as r 0.
This, together with (26), yields (3) and completes the proof of Theorem 1.
(27)
2
1/m |x|
1/m |x|
for all fixed x 6= 0 and sufficiently large . Finally, taking = , with small, we get
u(x) 0 1/(pm) e
1/m |x|
1/m |x|
= o(e
1/m |x|
as . The conclusion now follows at once, since clearly by appropriate choice of and we can
assume that is any number less than 1/(m 1)1/m .
2
Proof of Theorem 2
The argument is delicate, covering a number of pages. For the proof of (4) we need to distinguish the
two cases n > m2 and m < n m2 ; this is done in Sections 5.1 and 5.2 below. The proof of (5) and (6)
is given in Section 5.3.
We shall prove (4) first, for the case = 1, and then obtain the general estimate by means of the
rescaling (26).
Thus we assume that u = u(r) satisfies (13) with f (s) = sm1 + sp1 , namely
(|u0 |m2 u0 )0 +
n 1 0 m2 0
|u |
u um1 + up1 = 0
r
(28)
with u(0) = . From the estimate (9) in Theorem 5 we have always > 1 (since p > m) and, more
precisely,
1/(pm)
mp
>
(n m)
11
where
p = m .
Hence
>
m2 1
nm
!1/(pm)
pm K
(29)
(|w0 |m2 w0 )0 +
n1
0 m2 w 0
r |w |
(30)
wm1 + wp1 = 0
(31)
w(0) = 1, w0 (0) = 0
where = (pm) . Note that < 1 since > 1, and also, by (29), 0 as 0. Now define the
modified nonlinearity
f (s) = sm1 + sp1
and the corresponding functions (see (14) and (16))
F (s) =
m 1 p
s + s ,
m
p
(n m) p
w (r).
p
(32)
(nm)/m
(33)
r > 0.
(34)
Proof. We make use of the function H introduced in Lemma 2.1 in [KP]: here however it will be applied
without a previous Emden-Fowler inversion. Thus set
H(r) = (m 1)rn |w0 (r)|m (n m)rn1 w(r)|w0 (r)|m1 +
nm n
r w(r)f (w(r)).
n
rn 2 m1
m w
(r) (n m)wp1 (r) w0 (r).
n
12
w(R) =
!1/(pm)
(0, 1);
see (29) and recall from Proposition 2 that w0 < 0 and w < 1 for r > 0. Hence it is easy to see that H is
strictly increasing on [0, R] and strictly decreasing on [R, ). Moreover, H(0) = 0 and limr H(r) = 0
by Theorem 8. Consequently
H(r) > 0
r > 0.
(35)
Consider the function
(r) =
|w0 (r)|m1
r wn(m1)/(nm) (r)
(r)
wn(m1)/(nm) (r)
where (r) = |w0 (r)|m1 /r (see the proof of Theorem 7). By using (31) again we find that
0 (r) =
n
1
H(r).
n m rn+1 wm(n1)/(nm) (r)
From (35) it follows that is strictly increasing on [0, ). Therefore, by Proposition 2 we have
(r) > lim (t) =
t0
hence
|w0 (r)|
>
wn/(nm) (r)
1
n
f (1)
1
=
;
n
n
1/(m1)
r1/(m1)
r > 0.
The conclusion (34) follows upon integration, and the proof is complete.
For later use we observe that the function z = z(r) defined in (33) satisfies the equation
n 1 0 m2 0
|z |
z + (1 )z m 1 = 0.
r
(|z 0 |m2 z 0 )0 +
Now let
C1 = C1 () =
nm
m2
/(pm)
(36)
(37)
Then by differential calculus (recalling that p = m and = (pm) ) we find without difficulty that
1
f (s) C1 sm
s > 0
and
lim C1 = 1.
(38)
r > 0.
(39)
(40)
Integrating on [0, r], and taking into account (38) and Lemma 1, yields
rn1 |w0 (r)|m1 =
tn1 z m
(t)dt =
C1
rn1 |z 0 (r)|m1 ,
1
the last equality being obtained by a similar integration of (36) on [0, r]. Therefore,
|w0 (r)| < C2 /(m1) |z 0 (r)|
where
C2 =
C1
1
1
m1
r > 0,
(41)
r > 0,
that is, C2 /(m1) 1 > 0 since z(r) < 1 for r > 0 by (34) and the fact that < 1. This completes the
proof.
2
The following technical lemmas will be crucial in the sequel. To simplify their presentation, we shall
think of the functions w = w(r) and z = z(r), given in (30) and (33), to be defined over the space
IRn instead of on r 0; that is, w = w(|x|) and z = z(|x|). In particular, w then satisfies the partial
differential equation
m w = f (w) = wm1 + wp1 ,
= (pm)
(42)
We observe also that w(|x|) decays exponentially as |x| , so that the integrals below are well defined.
Lemma 3 We have
c1
IRn
IRn
c2
IRn
wp ,
where = pm , p = m , and
1
c1 =
n
nm
m
2
c2 =
nm
.
m2
Proof. By Proposition 3 applied to the ground state w of (31) we get, with the help of the second part
of (32),
Z
Z
Z
Z
(n m)
nm
m
wm +
wp = 0,
that is,
wm =
wp .
p
mp
IRn
IRn
IRn
IRn
But p (m, m ), so the conclusion follows at once.
14
Lemma 4 We have
Z
(nm)/m
IRn
w (C )
IRn
|w|m (C )(nm)/m ,
where C is a Sobolev constant for the embedding of D1,m (IRn ) into Lm (IRn ).
Proof. If we multiply (42) by w and integrate by parts, we obtain
Z
IRn
|w|
IRn
w +
IRn
w <
IRn
wp .
(43)
Using (38) and the fact that C1 1 by (37), equation (42) can also be written in the form m w =
IRn
|w|
IRn
Z
IRn
|w|
m /m
(44)
by the Sobolev inequality. Solving this relation for IRn |w|m gives the second inequality of the lemma;
the first is then obtained from (43). This completes the proof.
2
R
5.1
By (33) we see that z(|x|) |x|(nm)/(m1) as |x| , so z Lm (IRn ) if and only if n > m2 . This
allows us to derive
Lemma 5 Let n > m2 . Then there exists A > 0 (depending only on m, n) such that
(nm)/m
f or all
m 1 m2
0,
,
n nm
(45)
Proof. Define z(|x|) to be the function given by (33) with the parameter fixed at the value
(m 1)(n m)
.
n2 m(m 1)
Using (9) with = 1, an easy calculation shows that for in the range stated in the lemma we have
= (pm) (0, ). Hence, for the given range of , we infer from (34) that
Z
IRn
IRn
IRn
zm c
IRn
wm c1
IRn
wp c1 (C )(nm)/m .
15
A/,
(46)
where A (
c/c1 )C (nm)/m depends only on m, n. Finally, using the given restriction
0<
m 1 m2
n nm
(47)
m/(nm)
(45) now follows immediately, and the proof is complete.
Together with the inequality > 1, Lemma 5 implies the important conclusion
1
as 0.
(48)
Lemma 6 Let n > m2 . Then there exists K 0 > 0 (depending only on m, n) such that
= pm K 0
f or all
m 1 m2
0,
,
n nm
Proof. We have
pm =
n
m m m
nm
m
( nm )2
m
pm
( nm )2
m
It remains to show that the right side is bounded, but this follows directly from the fact that (1/s)s is
bounded ( e1/e ) on (0, ). The proof is complete.
2
Remark. A short calculation, taking into account the restriction (47), shows that in fact we can choose
2
2
K 0 = Am(nm+1)/n e(nm) /em .
We can now complete the proof of (4). Here it is convenient to revert to the original understanding
that w = w(r) and z = z(r), We first rewrite the results of Lemmas 1, 2 as
0 < z w < C3 1
for all r 0,
(49)
Q (r)r
n1
16
dr m.
Here we let 0. Clearly converges to some limit 0 [K, K 0 ], up to a subsequence; since we will
determine a unique possible value for 0 , this shows that 0 on the continuum > 0. Then by (49)
and the obvious fact that (recall 0 as 0)
(nm)/m
(50)
Both z0m rn1 and z0m rn1 are in L1 (0, ) since n > m2 , so we can let 0 and R in (50); this
gives finally
Z
Z
(n m)2 m
m
n1
z0 (r)r
dr = 0
z0 (r)rn1 dr.
nm2
0
0
By means of the change of variables s = Drm/(m1) one obtains
Z
z0m (r)rn1 dr
and
m 1 m1 n
=
D m B
m
z0m (r)rn1 dr
n(m 1) n m2
,
m
m
m 1 m1 n
n(m 1) n
=
D m B
,
.
m
m
m
(51)
(52)
Hence,
m
0 = n
nm
n(m1) nm2
, m
m
n
B n(m1)
,
m
m
2 B
(nm)/m2
(nm)/m = ( )(nm)/m 0
= m,n
as 0 (recall 1), which is just (4) for the case = 1. Since for general one has u(0) =
2
1/(pm) (nm)/m , relation (4) is proved (case n > m2 ).
5.2
The case n m2
Here z 6 Lm (IRn ) and the crucial Lemma 6 does not hold; nevertheless, we can prove the following
weaker result.
Lemma 7 Assume that n m2 . Then then there exists K 0 = K 0 (m, n) > 0 such that
2 /(n+m2 m)
17
K 0.
Proof. We argue as in the proof of Lemmas 5 and 6, with a few changes. First, from (34) we have
Z
IRn
wm
IRn
zm
IRn
zm = d <
(53)
IRn
wm c1
IRn
wp c1 (C )(nm)/m .
(54)
Next, integrating (40) over (0, ) and taking into account the exponential decay of w and w0 , as well as
(34), we get
Z
Z
Z
m1
pm
p1
pm
w
=
w
zp1 = d1 pm ,
(55)
IRn
IRn
IRn
Lp1 (IRn )
wm
Z
IRn
IRn
wm1
1 Z
IRn
wm
where = (n m)/(n + m2 m) (0, 1). Inserting (53), (54), (55) into this inequality gives, after a
little calculation,
2
2
2
m /(n+m m)(n+m )/m A1
where A1 = A1 (m, n). For < m2 /(n m)(n + m2 ) one finds in turn (compare Lemma 5)
A1
(n+m2 m)/m(m1)
(56)
As before this implies that is bounded, from which the lemma follows at once, subject of course to
the restriction on noted above.
2
From (56) it follows that 1 as 0, just as in the case n > m2 . In turn (49) holds exactly as
before, with C3 1 as 0.
For the next conclusion, we shall need a sharper form for the behavior of C3 . First, it is not difficult
to verify that the function C1 = C1 () defined in (37) satisfies C1 1 + c| log | for some constant c > 0;
we understand here and in what follows that c denotes a generic positive constant, depending only on
m and n. Moreover, by (29) we have < c, so the function C2 = C2 () defined in (41) also satisfies
C2 1 + c| log |. Finally
C3 = C2 /(m1) 1 + c| log |
(57)
for sufficiently small .
Next, let R > 0 denote the unique value of r where z(R) = | log |, where > 0 is a constant to be
determined later; note in particular that R as 0. Now, arguing from (39) and the fact that
1 < C3 < 1 + c | log |, we infer
w(r) > C3 z(r) (C3 1)
z(r)
C3 1
c
= 1
z(r) 1
z(R)
| log |
18
and
m/(m1) K1 | log |
when n = m2 .
Proof. Assume first that n < m2 . Then for sufficiently small there holds
d1 =
IRn
IRn
wp
by (34)
c
wm
IRn
Z
c
zm
|x|<R
Z
c R
tn1
dt
1 tm(nm)/(m1)
c (m2 n)/(m1)
{R
1}
c
2
( | log |)(m n)/(nm)
by Lemma 3
by (58)
by (33)
where the last inequality is obtained by solving z(R) = | log | ( small). Rearranging with the help of
2
the relation = pm m /(nm) (and by taking into account that m2 n < m(m 1)) now yields
the first statement of the lemma.
If n = m2 , the same arguments lead to
c
dt
c
c
= log(R) | log |,
t
Lemma 8 shows at once that (4) also holds in the case m < n m2 , that is whenever n > m.
Remark. As already mentioned in the introduction, more precision in the asymptotic behavior of u(0)
is needed in the case n m2 . We conjecture that also in this case there exists a continuous increasing
function gm,n defined on [0, ) such that gm,n (0) = 0 and lim0 [gm,n ()u(0)] = 1. (gm,n (m1)/m
when n < m2 ?)
19
5.3
Dirac limits
Here we shall complete the demonstration of Theorem 2 by proving conditions (5) and (6). It will be
convenient here and in the sequel not to make the initial assumption = 1, though we continue to write
u(0) = .
From Section 5.1 we recall the basic estimate (49); with the help of (57) this can be rewritten in the
form
0 < z w < c| log |.
(59)
Here we wish to scale back to the original function u, this being accomplished by means of (26) and (30).
More specifically, in (30) it is necessary to replace u and respectively by v and ( as in (2)) because
of the initial assumption in Section 5 that = 1. The required rescaling is therefore given by
w(r) =
1
1/(pm)
r
1/m (pm)/m
1
r
u
(pm)/m
(60)
where from Theorem 1 we have 1/(pm) = . After a little calculation, (59) then leads to the basic
formula
0 < z u c| log |,
(61)
where
h
i(nm)/m
(62)
nm
m
nm
|x| m1
as 0,
To prove (6), let X = XR denote the Lebesgue space Lm over the domain {|x| < R}, and similarly
0 be the space Lm over the domain {|x| R}. By Minkowskis inequality and (61),
let X 0 = XR
kukX kz kX ku z kX c | log | k1kX .
(63)
R = m/(nm)+
where > 0 is a positive constant to be determined later. Then with the obvious change of variables
s = (pm)/m r, we find
n/m
kz km
X = n
/m+
sn1 ds
n m,n
1 + (1 )1/(m1) Dsm/(m1)
(64)
as 0, see (52) and (48) (which as shown in Section 5.2 is valid for all n > m). By the same calculation
kz km
X0 0
20
(65)
as 0, since the integration is now over the interval (/m+ , ) and the integral is convergent.
Next, one calculates that
n n/m
n 1+(nm)/m
k1kX =
R
=
n
n
in view of the definition of R. We can now determine the limit as 0 of the quantity
| log | k1kX = (n /n)(nm)/m | log |.
From Lemmas 6 and 7 it is evident that, whatever the case considered, there exists > 0 (depending
only on m, n) such that < c , provided is small. (One can check that = (n m)/m2 + 1 in fact
suffices). Hence
(nm)/m | log | c1(nm)/m | log |,
which tends to 0 as 0 if is chosen small enough. It now follows at once from (63) and (64) that
kukm
X m,n as 0.
We observe finally from the left hand inequality of (61) that
m
kukm
X 0 < kz kX 0 0
by (65). Hence
m
m
kukm
m = kukX + kukX 0 m,n ,
IRn
m1
IRn
up1 .
(66)
IRn
up1
IRn
zp1 = n 1+(nm)/m
sn1 ds
1 + (1 )1/(m1) Dsm/(m1)
(nm)(p1)/m .
Since the integral is uniformly convergent for any less than m/2(n m), we then get
Z
IRn
up1 0
as 0.
With the help of (66) this completes the proof of (6), and therefore of Theorem 2.
Proof of Theorem 3
First we prove (8). Multiplying the equation (Pp ) by u and integrating over IRn gives
Z
IRn
|u|
IRn
u +
IRn
up .
We now let 0. The first term on the right approaches 0 by (6) (and a trivial interpolation).
21
(67)
To treat the second term on the right side of (67), we slightly modify the space X from its meaning
in the previous subsection, so that now it represents the Lebesgue space Lp over the domain {|x| < R},
and similarly for the space X 0 . Then as in (64) there holds
kz kpX = n (nm)/m
/m+
sn1 ds
1 + (1 )1/(m1) Dsm/(m1)
n(nm)/m ,
the integral being convergent when < m/(n m). To evaluate the limit of the right side, note first that
on the interval 0 < s < there holds (for small )
h
i(nm)/m
< n/(m1) ,
1 + (1 )1/(m1) Dsm/(m1)
i(nm)/m
1.
Hence as in (64), one obtains kz kpX m,n as 0. Also as before, kz kpX 0 0, so that finally, again
arguing as in the previous subsection,
kukpp = kukpX + kukpX 0 m,n ,
that is, IRn up m, n. The conclusion (8) follows at once from (67).
It remains to prove (7). By evaluating z 0 (r) and by using (41) and (57) we obtain
R
r1/(m1)
nm
(1 )1/(m1) D
n/m .
m1
1 + (1 )1/(m1) Drm/(m1)
(68)
1
p/m
0
r
u
.
(pm)/m
Inserting this in (68), using an obvious change of variables and then letting 0, yields
n
lim
1/(m1)
|u (r)|
nm
m1
n/m
n(nm)/m(m1) r(1n)/(m1) ,
Proof of Theorem 4
We define
1
() = (, d) =
pm
where is given by (2); here is a (well-defined) continuous function of and of course also of m, n. By
Theorem 1, when = () we have
u(0) = 1/(pm) = d,
22
(nm)/m m,n
as 0,
so that
() =
d
(nm)/m2
pm
(nm)/m2
d
m,n
!m2 /(nm)
(69)
zd (x) d 1 + D d
m
nm
|x|
m
m1
nm
m
Ud (x)
uniformly for x in IRn ; see (1) in the introduction. Together with (69) this completes the proof of (ii).
An easy consequence of the above argument is the following companion result for Theorem 4.
Corollary. Let n > m2 . In place of the condition = (), suppose that = a, where a is a positive
2
constant. Then u Ud uniformly on IRn as = p m 0, where d = a(nm)/m m,n .
Acknowledgement. The second author wishes to thank Prof. Grozdena Todorova for many valuable and
helpful conversations during the preparation of the paper.
References
[AP1] F.V. Atkinson, L.A. Peletier, Ground states of u = f (u) and the Emden-Fowler equation, Arch. Rational
Mech. Anal. 93, 1986, 103-127
[AP2] F.V. Atkinson, L.A. Peletier, Elliptic equations with nearly critical growth, J. Diff. Eq. 70, 1987, 349-365
[BL] H. Berestycki, P.L. Lions, Nonlinear scalar field equations, I, Existence of a ground state, Arch. Rational
Mech. Anal. 82, 1983, 313-345
[Ci] G. Citti, Positive solutions of quasilinear degenerate elliptic equations in IRn , Rend. Circolo Mat. Palermo 35,
1986, 364-375
[FLS] B. Franchi, E. Lanconelli, J. Serrin, Existence and uniqueness of nonnegative solutions of quasilinear equations in IRn , Advances in Math. 118, 1996, 177-243
[GP] J.P. Garca Azorero, I. Peral Alonso, On limits of solutions of elliptic problems with nearly critical exponent,
Comm. Part. Diff. Eq. 17, 1992, 2113-2126
[Ga] F. Gazzola, Critical growth quasilinear elliptic problems with shifting subcritical perturbation, to appear, Diff.
Int. Eq.
23
[GST] F. Gazzola, J. Serrin, M. Tang, Existence of ground states and free boundary problems for quasilinear elliptic
operators, Adv. Diff. Eq. 5, 2000, 1-30
[KP] M.C. Knaap, L.A. Peletier, Quasilinear elliptic equations with nearly critical growth, Comm. Part. Diff. Eq.
14, 1989, 1351-1383
[NS1] W.M. Ni, J. Serrin, Nonexistence theorems for quasilinear partial differential equations, Rend. Circolo Mat.
Palermo (Centenary Supplement), Series II, 8, 1985, 171-185
[NS2] W.M. Ni, J. Serrin, Existence and nonexistence theorems for ground states of quasilinear partial differential
equations. The anomalous case, Accad. Naz. dei Lincei, Atti dei Convegni 77, 1986, 231-257
[PS] P. Pucci, J. Serrin, Uniqueness of ground states for quasilinear elliptic operators, Indiana Univ. Math. J. 47,
1998, 501-528
[R1] O. Rey, Proof of two conjectures of H. Brezis and L.A. Peletier, Manuscripta Math. 65, 1989, 19-37
[R2] O. Rey, The role of Greens function in a nonlinear elliptic equation involving the critical Sobolev exponent,
J. Funct. Anal. 89, 1990, 1-52
[ST] J. Serrin, M. Tang, Uniqueness of ground states for quasilinear elliptic equations, to appear, Indiana Univ.
Math. J.
[Ta] G. Talenti, Best constant in Sobolev inequality, Ann. Mat. Pura Appl. 110, 1976, 353-372
24