Advanced Nonlinear Studies 3 (2003), 513-541

Existence of Radial Solutions for Quasilinear

Elliptic Equations with Singular Nonlinearities

Beatrice Acciaio, Patrizia Pucci
Dipartimento di Matematica e Informatica
Universit`a degli Studi di Perugia
Via Vanvitelli 1, 06123 Perugia, Italy
e-mail: beatrice.acciaio@stat.unipg.it, pucci@dipmat.unipg.it
Received 17 September 2003
Communicated by Antonio Ambrosetti
Abstract
We prove the existence of radial solutions of the quasilinear elliptic equation
div(A(|Du|)Du) + f(u) = 0 in R
n
, n > 1, where f is either negative or positive
for small u > 0, possibly singular at u = 0, and grows subcritically for large u.
Our proofs use only elementary arguments based on a variational identity. No
dierentiability assumptions are made on f.
1991 Mathematics Subject Classication. 35J60, 35J70.
Key words. Quasilinear singular elliptic equations
1 Introduction
Recently, for the pLaplacian equation when p > 1, under general conditions for the
nonlinearity f, besides other properties Tang in [18] proved existence of crossing
radial solutions for f positive near u = 0, while Gazzola, Serrin and Tang in [7]
proved existence of radial ground states in R
n
for f negative near u = 0. Montefusco
and Pucci in [9] extended the results of [7] to the general quasilinear case, considered
here. The main purpose of this paper is to extend the existence results of [7] and
[18] to general quasilinear elliptic problems, using a unied proof, and also to extend

This research was partially supported by the Italian MIUR project titled Metodi Variazionali
ed Equazioni Dierenziali Non Lineari.
513
514 B. Acciaio, P. Pucci
them and those of [9], introducing a new subcritical condition on f at innity that,
in canonical cases, interesting in applications, is more general than the subcritical
condition used in [7], [18] and [9]. For instance, in the case 1 < p < n, the two
typical examples covered in our paper, but not in that of [7] and [18], are given by
f(u) = u
p

1
+ log u and f(u) = u
p

1
+ u
q1
, with 1 < q < p

, and for u R
+
,
where p

= np/(n p).
In particular, we are interested in nding sucient conditions for existence of
radial ground states of the quasilinear elliptic equation
divA([Du[)Du + f(u) = 0 in R
n
, n > 1, (1.1)
when f < 0 near u = 0. By a ground state we mean a nonnegative nontrivial
solution of (1.1) which tends to zero at innity. Moreover, with the same technique,
we are also able to prove the existence of a radial crossing solution of (1.1) in its
maximal continuation interval where u > 0 and u

< 0, when f > 0 near u = 0,

already established in [18] for the pLaplacian equation.
In addition to the ground state problem, when f < 0 near u = 0, we can
also consider existence of nontrivial radial solutions of the homogeneous Dirichlet
Neumann free boundary problem
divA([Du[)Du + f(u) = 0 in B(0, R) R
n
,
u > 0 in B(0, R), u =
u

= 0 on B(0, R).
(1.2)
For t > 0 we set (t) = tA(t) and assume that
(1) is of class C
1
(R
+
), R
+
:= (0, ),
(2)

(t) > 0 for t > 0, and (t) 0 as t 0,

(3) There exists a positive number p, with 1 < p n, such that t
1p
(t) is a
nondecreasing function on R
+
. Note that (3) is equivalent to

(t)
p 1
t
(t), t > 0. (1.3)
As in [7], [18], [9], and more generally in natural existence settings, we are
concerned with subcritical nonlinearities f since 1 < p n. Specic hypotheses
are given in the statements of the main Theorems 5.15.4 below. Throughout the
paper f is assumed of type
(f1) f C(R
+
) L
1
[0, 1].
Condition (f1) was rst used in [16] for the uniqueness problem, but without
suitable attention to the diculties involving this type of singularity at u = 0. Only
recently in [6], a careful denition of solution for (1.1) under a singularity of type
(f1) was given, in order to avoid the undened nature of f(u) at u = 0. We shall
adopt that denition.
Quasilinear Elliptic Equations 515
Moreover, putting F(u) =
_
u
0
f(v)dv, in the main existence theorems we assume
also that either
(f2) there exists b > 0 such that F(u) < 0 for 0 < u < b, F(b) = 0 and f(b) > 0,
as in [7] and [9], or
(f3) there exists c > 0, possibly innite, such that f(u) > 0 for 0 < u < c,
as in [18].
The behavior of f near 0 is of crucial importance for the existence results given
in Sections 5. We shall identify two mutually exclusive situations:
Regular case: f can be extended by continuity at u = 0, with f(0) = 0;
Singular case: f cannot be extended by continuity at u = 0, with f(0) = 0.
For a complete discussion on the extemsive background and literature concerning
related previous results, including as well the classical scalar eld equation and the
regular case with f(0) = 0, we refer the reader to [5] and [7] and the references
therein.
The paper is organized as follows: in Section 2 some preliminary qualitative
properties for solutions of (1.1) and (1.2) are given, including a necessary and su-
cient condition in order that a solution of (1.1) has compact support. In Section 3 we
present and summarize the main properties of solutions of the corresponding initial
value problem. Section 4 is devoted to proving the existence of crossing solutions
and also to showing the connections between the subcritical growth condition
(Q1) The function Q(v) = pnF(v) (n p)vf(v), v R
+
, is locally bounded near
v = 0 and there exist > d and (0, 1) such that Q(v) 0 for all v
and
limsup
v
Q(
1
v)
_
v
p1
f(
2
v)
_
n/p
= for all
1
,
2
[, 1],
used in [18] and [9], and the new condition
(Q) The function Q(v) = pnF(v) (n p)vf(v), v R
+
, is locally bounded near
v = 0 and there exist > d and (0, 1) such that Q(v) 0 for all v
and
limsup
v
Q(
1
v)[v
p+1
f(
2
v)]
n/p
= for every
1
,
2
[, 1],
introduced in this paper, where in both conditions d = b under (f2) and d = 0 under
(f3). Section 4 ends with some remarks and examples proving the independence of
the two dierent growth hypotheses (Q) and (Q1), and the main fact that, under
(f1) and liminf
u
f(u) = k
0
> 0, condition (Q1) is stronger than (Q), see Proposi-
tion 4.3. Finally, in Section 5, we give some existence results, also using the further
assumption
(4) k = inf
t>0
(t)
t
p1
> 0.
516 B. Acciaio, P. Pucci
In particular: Theorem 5.3 is the analogue of Theorem 4.1 of [9] and Theorem 5.4
extends Theorem 5.1 of [18]; while Theorems 5.15.2 give the same results under
the new condition (Q) and without assuming (4).
2 General theory and behavior of solutions
We consider the quasilinear singular elliptic problem
divA([Du[)Du + f(u) = 0 in R
n
, n > 1,
u 0, u , 0 in R
n
,
(2.4)
under the following main structural assumptions. For t > 0 we set (t) = tA(t), as
mentioned in the Introduction, and assume the validity of (1)(3) throughout
the paper. Note that can be extended by continuity at t = 0 by (1) and (2),
that is C(R
+
0
) C
1
(R
+
), with (0) = 0. We introduce the function
H(t) = t(t)
_
t
0
(s)ds =
_
t
0
s

(s)ds, t > 0. (2.5)

Therefore, H is strictly increasing on R
+
0
, with H(0) = 0. Moreover, by (3), that
is (1.3), it is also evident that
p 1
p
B(t) H(t) < B(t) for all t > 0, (2.6)
where
B(t) = t(t), t 0, (2.7)
is strictly increasing on R
+
0
, with B(0) = 0, by (2), and moreover t
p
B is non
decreasing on R
+
by (3). Clearly (t) as t by (3) and in turn both
H(t) and H(t)
_
(t) as t by (2.6).
Throughout the paper we also assume that f satises (f1), so that it is clear that
F(u) =
_
u
0
f(v)dv is well dened and is of class C(R
+
0
) C
1
(R
+
), with F(0) = 0.
Since (2.4) is possibly singular when u = 0, it is necessary to carefully dene
the meaning to be assigned to solutions of (2.4), and in analogy with [16] and [6],
we introduce the following:
Denition A semiregular nonnegative (weak) radial solution u of (2.4) is a non
trivial nonnegative radial function of class C
1
(R
n
0), which is a classical distri-
bution solution of (2.4) in the open (support) set = x R
n
0 : u(x) > 0,
and is bounded near x = 0.
Of course nonnegative semiregular radial solutions are then of class C
1
(R
+
),
bounded near r = 0, and, we shall see, they satisfy
[r
n1
A([u

[)u

)]

+ r
n1
f(u) = 0 in J = r > 0 : u(r) > 0. (2.8)
Quasilinear Elliptic Equations 517
As in [6], it will rst be shown that the denition of nonnegative semiregular
radial solution is compatible with that of classical solution in the regular case,
namely when f is continuous in R
+
0
with f(0) = 0.
Proposition 2.1 Let u = u([x[) be a nonnegative semiregular radial solution of
(2.4), where f C(R
+
0
), with f(0) = 0. Then u is a classical C
1
solution of (2.8)
in R
+
.
Proof. If J = R
+
there is nothing to prove. Otherwise let J

be any component of
J = r > 0 : u(r) > 0, let r, r

and r
0
< r
1
be the endpoints of J

. Then
using standard distribution arguments, we get
r
n1
A([u

(r)[)u

(r) (r

)
n1
A([u

(r

)[)u

(r

) =
_
r
r

s
n1
f(u(s))ds. (2.9)
Let r

r
1
and observe that necessarily u(r
1
) = u

(r
1
) = 0 by denition of a
semiregular nonnegative solution. This gives
r
n1
A([u

(r)[)u

(r) =
_
r
r
1
s
n1
f(u(s))ds. (2.10)
Moreover, when r
0
> 0 we may also let r r
0
in (2.10) and thus obtain
_
r
1
r
0
s
n1
f(u(s))ds = 0. (2.11)
Now let r be a xed point of R
+
J. Since f(0) = 0, it is easy to see that
_
I(r
1
, r)
s
n1
f(u(s))ds =

J
i
I(r
1
, r)
_
J
i
s
n1
f(u(s))ds = 0, (2.12)
where I(r
1
, r) is the interval with endpoints r
1
, r; the sum is taken over all the
components J
i
of J contained in I(r
1
, r); and (2.11) is used at the second step.
Finally from (2.10) and (2.12) one gets
r
n1
A([u

(r)[)u

(r) =
_
r
r
s
n1
f(u(s))ds. (2.13)
Here r can be any point in J and also by the computation of (2.12) it is clear that
(2.13) is also correct if r R
+
J. That is (2.13) holds for all r R
+
. On the
other hand, both sides of (2.13) are continuously dierentiable, so u satises (2.8)
in R
+
, which was to be proved.
In analogy with [16] and [6] we give
Proposition 2.2 Let u = u([x[) be a nonnegative semiregular radial solution of
(2.4), with
liminf
r0
+
u(r) = > 0. (2.14)
518 B. Acciaio, P. Pucci
Then A([u

[)u

C
1
(J) and u is a solution in J of (2.8). Moreover,
lim
r0
+
u

(r) = 0 and lim

r0
+
u(r) = . (2.15)
Proof. Let u be a nonnegative semiregular radial distribution solution of (2.4),
so that u is in particular of class C
1
(R
n
0). It is clear that u solves (2.8) in the
sense of distributions in J. By (f1) we have f u C(J). Hence, as in (2.9),
r
n1
A([u

(r)[)u

(r) r
n1
0
A([u

(r
0
)[)u

(r
0
) =
_
r
r
0
s
n1
f(u(s))ds, (2.16)
for any interval [r
0
, r] J. Thus A([u

[)u

is actually of class C
1
(J), and (2.8) holds
in J.
To obtain (2.15), choose > 0 so small that f u L

[0, ]; this can be done

since u is positive and bounded near 0 by denition and (2.14), and also by (f1).
Then, letting r
0
0
+
in (2.16), we see that
lim
r
0
0
+
r
n1
0
A([u

(r
0
)[)u

(r
0
) = , (2.17)
where is a nite number. We claim that = 0. Assume for contradiction that
,= 0. Then by (2.17) and (3)
[u

(r
0
)[ =
1
_
[[r
(n1)
0
_
[1 + o(1)]
_
[[
1/(p1)

1
(1)
_
r
(n1)/(p1)
0
[1 + o(1)]
as r
0
0
+
, so u cannot be bounded near 0, since 1 < p n, which is the required
contradiction. Hence = 0 in (2.17) and the claim is proved.
Now, letting r
0
0
+
in (2.16), we obtain
r
n1
A([u

(r)[)u

(r) =
_
r
0
s
n1
f(u(s))ds, 0 < r < ,
that is, ([u

(r)[) Cr, where C > 0 is an appropriate constant, since f u

L

[0, ] by (2.14) and (f1). Hence (2.15) immediately follows.

Remark If also satises condition (4), given in the Introduction, then it is
easily seen that (2.15) holds in the stronger form
u

(r) = O
_
r
1/(p1)
_
and u(r) = O
_
r
p/(p1)
_
as r 0
+
. (2.18)
Let u be a xed semiregular nonnegative radial solution of either (2.4) or
(1.2). In order to unify the notation we shall dene R = for (2.4), and clearly
R < is the positive radius given already in (1.2). By Proposition 2.2 then u is a
nonnegative classical solution of (2.8) in J (0, R) such that
u C
1
[0, R), A([u

[)u

C
1
(J),
u(0) = > 0, u

(0) = 0, u 0.
(2.19)
It is useful to dene w(r) = A([u

(r)[)u

(r).
Quasilinear Elliptic Equations 519
Lemma 2.3 The function w is of class C
1
[0, R) and is a solution of
(r
n1
w)

+ r
n1
f(u) = 0 on J. (2.20)
Moreover, denoting by r
0
the rst zero of u in (0, R), if any, or otherwise r
0
= R,
we have
w(r) =
1
r
n1
_
r
0
s
n1
f(u(s))ds, 0 < r < r
0
, (2.21)
w(0) = 0, w

(0) =
f()
n
. (2.22)
Finally, putting (r) = [u

(r)[, there holds

lim
r0
+
B((r))
r
= 0 and
B()
r
C[0, R). (2.23)
Proof. Of course w(0) = 0, since u

(0) = 0 by (2.19); the function w is a solution of

(2.20) by (2.8). Integrating over [0, r], as long as r < r
0
, we get
r
n1
w(r) =
_
r
0
s
n1
f(u(s))ds,
so (2.21) holds. From (2.19) we have
w

(r) = f(u(r)) +
n 1
r
n
_
r
0
s
n1
f(u(s))ds,
and by LHopitals rule (2.22), as well as the C
1
regularity of w at r = 0, follows at
once.
Finally,
B((r))
r
= u

(r)
w(r)
r
,
and (2.23) follows at once by (2.22) and (2.19).
Corollary 2.4 If u

(r) ,= 0 at some r, with 0 < r < r

0
, then u

exists at this point

and satises (2.8) in the form

()u

n 1
r
() + f(u) = 0, = [u

[. (2.24)
Proof. By (2.21) of Lemma 2.3 and the fact that u

(r) ,= 0 we have by (2)

[u

(r)[ =
1
_

_
r
0
_
s
r
_
n1
f(u(s))ds

_
,
since () = by (3). Now the integral is not zero, so that the function on the
right hand side is dierentiable at r by (1). Hence u

exists at r and from (2.8)

we get exactly (2.24), since [u

(r)[ = (r) > 0.

520 B. Acciaio, P. Pucci
A natural energy function associated to semiregular nonnegative radial solu-
tions u of (2.4) is given by
E(r) = H((r)) + F(u(r)), = [u

[. (2.25)
Lemma 2.5 The energy function E is of class C
1
(J), with
E

(r) =
n 1
r
B((r)) in J. (2.26)
Proof. Obviously, by (f1) and (2.19)
dF(u(r))
dr
= f(u(r))u

(r),
this formula being valid only when u(r) > 0, namely in J. Moreover, by (2.5) and
(2),
H() =
_

0
td(t) =
_
()
0

1
()d
and, since ((r)) = [sgn u

(r)]w(r), we get
dH((r))
dr
= (r)[sgn u

(r)]w

(r) = u

(r)w

(r).
Therefore, by (2.20), on J
E

(r) = u

(r)
_
w

(r) + f(u(r))
_
= u

(r)
n 1
r
w(r)
and (2.26) follows at once.
Now let u be either a xed semiregular nonnegative radial ground states of
the problem (2.4), namely a semiregular nonnegative radial solution of (2.4) in
the above sense such that
lim
|x|
u(x) = 0, (2.27)
or a xed semiregular nonnegative radial solution of the corresponding homoge-
neous DirichletNeumann free boundary problem (1.2).
With the respective end conditions at R = in (2.4), (2.27) and at R <
in (1.2), the problems (2.4), (2.27) and (1.2) can be unied again into the single
statement (2.8) and (2.19) in J (0, R).
Theorem 2.6 If u(t
0
) = 0 for some t
0
> 0, then u 0 on [t
0
, R).
Proof. Since u 0, clearly u

(t
0
) = 0. Hence E(t
0
) = 0 by (2.25), (2.5) and
(f1). Assume for contradiction that there is t
1
, with t
0
< t
1
R such that again
u(t
1
) = 0 and u(r) > 0 in (t
0
, t
1
). Then (t
0
, t
1
) J and E

0 in (t
0
, t
1
) by (2.26).
Quasilinear Elliptic Equations 521
Clearly E(t
1
) = 0 when t
1
< R, and we claim that E(r) 0 as r t
1
= R.
Indeed, if R < , the claim is obvious by the boundary conditions of (1.2); while
if R = , since u(r) 0 as r by (2.27), then F(u(r)) 0 by (f1), and
so E(r) decreases to a nite nonnegative limit as r by (2.25) and (2.5).
Consequently, u

(r) limit = 0 as r , since u(r) 0 as r , and H is

strictly increasing and positive by (2.5). Therefore E(r) 0 as r , as claimed.
Hence E E

0 in (t
0
, t
1
) and so B() 0 on [t
0
, t
1
) by (2.26), namely u

0
on [t
0
, t
1
) by (2.7). This contradiction completes the proof.
By Theorem 2.6 it follows that any solution of (2.19) and (2.8), with the given
end conditions (2.27) and those in (1.2), has as its (open) support set J exactly
an initial interval (0, r
0
), with r
0
R. In turn, one deduces that actually E
C
1
[0, R), and that (2.26) holds in the entire maximal interval [0, R). Therefore for
any 0 s
0
< r < R we have
E(r) E(s
0
) =
_
r
s
0
n 1
s
B((s))ds. (2.28)
Clearly E(0) = F() by (2.25) and (2.19). Thus, letting s
0
0
+
in (2.28), we
obtain
E(r) = F() (n 1)
_
r
0
B((s))
s
ds, 0 r < R. (2.29)
As in the proof of Theorem 2.6, if R = and u(r) 0 as r , then u

(r) 0
and E(r) 0 as r . Hence, by (2.27) or (1.2), the nonnegative continuous
function B((s))/s is also integrable on [0, R), R , with
_
R
0
n 1
s
B((s))ds = F(). (2.30)
In summary, a semiregular nonnegative radial ground state of (2.4), (2.27),
or a semiregular nonnegative radial solution of the corresponding free boundary
problem (1.2), has the property that
u(0) = > 0, u

(0) = 0, u(R) = u

(R) = 0, (2.31)
where respectively R = or R < . Furthermore, by (2.29) and (2.30)
E(r) =
_
R
r
n 1
s
B((s))ds 0, (2.32)
and clearly also
E(0) = F() =
_
R
0
n 1
s
B((s))ds > 0. (2.33)
Lemma 2.7 If s
0
0 is a critical point of u, with u(s
0
) > 0, then either u(r)
u(s
0
) for r > s
0
and f(u(s
0
)) 0, or u(r) u(s
0
) for r > s
0
and f(u(s
0
)) 0.
522 B. Acciaio, P. Pucci
Proof. Let s
0
0 be a critical point of u. Assume for contradiction that there are
two points t
1
, t
2
> s
0
such that u(t
1
) > u(s
0
) and u(t
2
) < u(s
0
). Then, there is r in
the interval, with endpoints t
1
and t
2
, such that u(r) = u(s
0
) and u is not constant
on [s
0
, r]. Now by (2.28) and (2.25)
H((r)) +
_
r
s
0
n 1
s
B((s))ds = 0
and both terms are nonnegative by (2), (2.5) and (2.7). Thus in particular
B() 0 on [s
0
, r], so u

0 on [s
0
, r] by (2.7), which is impossible. Hence we have
two cases: either u(r) u(s
0
) for r > s
0
, or u(r) u(s
0
) for r > s
0
.
In the rst case, since u

(s
0
) = 0, then w(s
0
) = 0 by (2), and by (2.20) and
the regularity of w established in Lemma 2.3 at r = s
0
, we have
f(u(s
0
)) = w

(s
0
) 0.
Indeed, otherwise w

(s
0
) > 0, and so there is t
3
> s
0
such that w

(r) > 0 on [s
0
, t
3
];
in turn w(r) > w(s
0
) = 0 and u

(r) > 0 on [s
0
, t
3
], which gives u(r) > u(s
0
) on
[s
0
, t
3
], which contradicts (2.31).
In the same way, in the second case, it follows that f(u(s
0
)) 0, completing
the proof of the lemma.
It is convenient to introduce the following additional condition on f.
(f4) There exists > 0, possibly innite, such that F(u) 0 whenever both f(u)
= 0 and u (0, ).
Clearly if (f4) holds for some > 0, then it continues to hold for all

(0, ).
Consequently there exists a maximal , possibly innite, for which (f4) is valid.
Without loss of generality we can assume that in (f4) is maximal.
One can see also from (f4) that if F(u
0
) 0 at some point u
0
, 0 < u
0
< ,
then F(u) 0 for all u [0, u
0
]. Conversely, it is evident that if F(u) 0 on some
interval [0, u
0
], then the maximal u
0
.
Furthermore if either (f2) or (f3) holds, where (f2) and (f3) are the main
qualitative conditions of f given in the Introduction, then (f4) is satised, with
= supv > d : f(u) > 0 for u (d, v), where
d =
_
b, if (f2) holds,
0, if (f3) holds.
(2.34)
In this case > d 0 and = if f(u) > 0 for all u > d, while f() = 0 if
< .
Proposition 2.8 Let u be a semiregular nonnegative radial ground state of (2.4),
(2.27), or a semiregular nonnegative radial solution of the corresponding free
boundary problem (1.2), so that (2.31) holds. Then r = 0 is a maximum of u
and u

0 on [0, R); furthermore f() 0 and F() > 0.

If (f4) holds, then also
Quasilinear Elliptic Equations 523
(i) f() > 0, when 0 < < ,
(ii) u

(r) < 0 when r > 0 and 0 < u(r) < .

Proof. By Lemma 2.7 and the condition u

(0) = 0 one sees that u(r) u(0) for

r > 0, since otherwise u(r) u(0) = > 0 for all r > 0, and so u cannot approach
0 as r R

, contradicting (2.31). The fact that f() 0 similarly follows from

Lemma 2.7, and clearly F() > 0 by (2.33).
Next assume for contradiction that u

(s
0
) > 0 for some s
0
> 0. Since u(s
0
)
u(0), as shown above, there is a minimum s in (0, s
0
), with u(s) < u(s
0
), and so
u(r) u(s) for r > s by Lemma 2.7. If u(s) > 0, then u(r) cannot approach 0 as
r R

, contradicting (2.31). Therefore u(s) = 0 with s > 0, and by Theorem

2.6 we get u 0 on [s, R): thus u

(s
0
) = 0, which is again a contradiction. Hence
u

(r) 0 on [0, R).

To show (i) it is enough to observe that if f() = 0 and 0 < < , then
F() 0 by assumption (f4). This is impossible by (2.33), proving (i).
To obtain (ii), assume now for contradiction that there is a point s
0
> 0 such
that u

(s
0
) = 0 and 0 < u(s
0
) < . Since u

(r) 0 for r 0, then both u(r) u(s

0
)
for 0 r < s
0
and u(r) u(s
0
) for s
0
< r < R. Of course, w(s
0
) = 0 by (2). We
claim that also w

(s
0
) = 0. Indeed, if w

(s
0
) > 0, then w would be strictly increasing
at s
0
, namely u

would change sign at s

0
, which is impossible since u

(r) 0 on
[0, R). Analogously, the case w

(s
0
) < 0 also cannot occur.
Since w is a C
1
[0, R) solution of (2.20) by Lemma 2.3, we get f(u(s
0
)) = 0.
Also 0 < u(s
0
) < , so that F(u(s
0
)) 0 by (f4). Hence by (2.32) and (2.25), with
r = s
0
,
0
_
R
s
0
n 1
s
B((s))ds = E(s
0
) = F(u(s
0
)) 0,
which implies u

0 on [s
0
, R) by (2) and (2.7). Thus u(r) u(s
0
) > 0 for
s
0
r < R, again contradicting (2.31). This completes the proof of (ii).
The next result gives a necessary and a sucient condition for a semiregular
nonnegative radial ground state to have compact support.
Theorem 2.9 Let u be a semiregular nonnegative radial ground state of (2.4),
(2.27), with u(0) = > 0.
(i) If F(u) 0 for all values 0 < u < , for some > 0, and
_
0
+
du
H
1
([F(u)[)
< , (2.35)
then u has compact support.
(ii) Conversely, assume there exists > 0 and a nondecreasing function G :
[0, ) R, with G(0) = 0, such that [F(u)[ G(u) for all u [0, ). If u has
compact support, then
_
0
+
du
H
1
(G(u))
< . (2.36)
524 B. Acciaio, P. Pucci
Proof. Let u be a semiregular nonnegative radial ground state as in the theorem,
so R = .
(i) Suppose (2.35) holds. We denote by r

> 0 any point such that 0 u(r) <

on (r

, ). Condition (f4) clearly holds with , as noted above. Thus by

Proposition 2.8 (ii) we have u

(r) < 0 for all r (r

, ) for which u(r) > 0. Hence

by Theorem 2.6, either u 0 for all r suciently large, or u > 0 and u

< 0 on
(r

, ). In the rst case we are done. Otherwise, denoting by r = r(u) the inverse
function on (r

, ), by (2.32) and (2.25), we get on (r

, )
H((r)) = F(u(r)) +
_

r
n 1
s
B((s))ds > F(u(r)),
or u

(r) < H
1
([F(u(r))[) on (r

, ) by the assumption that F(u) 0 on [0, )

and the fact that H() = by (3). That is, writing r = r(u) and putting
= u(r

), we have
1
r

(u)
< H
1
([F(u)[) for u u((r

, )) = (0, ),
since u(r) 0 as r . By integration over (u(r), ), r > r

,
_

u(r)
du
H
1
([F(u)[)
>
_
u(r

)
u(r)
r

(u)du = r r

.
Hence, letting r , there results
_

0
du
H
1
([F(u)[)
= .
This contradicts (2.35) and completes the proof of part (i) of the theorem.
(ii) Let u have compact support. Then by Theorem 2.6 and the rst part of
Proposition 2.8 there is r
0
> 0 such that u

(r) 0 and 0 < u(r) u(0) =

on (0, r
0
), while u 0 on [r
0
, ). Let r

(0, r
0
) be some xed point such that
0 < u(r) < on (r

, r
0
). By (2.28) and (2.25) for 0 < r

< r < r
0
we have E(r
0
) = 0
and
H((r)) = F(u(r)) +
_
r
0
r
n 1
s
B((s))ds G(u(r)) + c
1
_
r
0
r
H((s))ds,
by assumption, (3) and (2.6), with c
1
= p

(n 1)/r

. Applying Gronwalls in-

equality to e
c
1
(r

r)
_
r
0
r
H((s))ds, we obtain
H((r)) G(u(r)) + c
1
_
r
0
r
G(u(s))e
c
1
(sr)
ds.
Now G(u(r)) is nonincreasing on (r

, r
0
), since G is nondecreasing by assumption
and u is nonincreasing on (r

, r
0
). Hence
H((r)) CG(u(r)), with C = e
c
1
(r
0
r

)
> 1.
Quasilinear Elliptic Equations 525
Therefore, since u

(r) 0,
u

(r) H
1
(CG(u(r))) on (r

, r
0
).
We can now apply Lemma 1.3.3 of [5] since the main assumption (ii) in [5] is clearly
satised. Indeed, here p
_
t
0
(s)ds B(t) in the entire R
+
0
by (2.7), (2.6) and (2.5).
Consequently by Lemma 1.3.3 of [5] there is a constant D > 0 such that
H
1
(Ct) DH
1
(t) for all 0 = G(0) < t < G(u

),
where u

= u(r

) . Hence
u

(r) DH
1
(G(u(r))) on (r

, r
0
).
Integrating on [s
0
, r], with r

< s
0
< r < r
0
, we get
_
u(s
0
)
u(r)
du
H
1
(G(u))
=
_
r
s
0
u

(s)
H
1
(G(u(s)))
ds D(r s
0
).
Letting r r

0
, this yields
_
u(s
0
)
0
du
H
1
(G(u))
D(r
0
s
0
),
that is (2.36) holds. This completes the proof of part (ii) of the theorem.
As an immediate consequence of Theorem 2.9 we obtain
Corollary 2.10 Let u be a semiregular nonnegative radial ground state of (2.4),
(2.27), with u(0) = > 0, and assume f 0 on (0, ), for some > 0, and f , 0.
(i) Then u(r) > 0 for every r > 0 if and only if
_
0
+
du
H
1
([F(u)[)
= . (2.37)
(ii) In particular, u(r) > 0 for every r > 0 if
_
0
+
du
[F(u)[
1/p
= . (2.38)
(iii) Furthermore, if satises also
(4) k = inf
t>0
(t)
t
p1
> 0,
then u(r) > 0 for every r > 0 if and only if
_
0
+
du
[F(u)[
1/p
= .
526 B. Acciaio, P. Pucci
Proof. Statement (i) is a direct consequence of Theorem 2.9, with G = [F[. By
(3), (2.7) and (2.6) it is clear that H(t) B(t) B(1)t
p
for all 0 t 1, with
B(1) > 0. Hence t H
1
(B(1)t
p
) since H() = , and so H
1
(s) cs
1/p
, where
c = [B(1)]
1/p
, for 0 s B(1). Consequently the validity of (2.38) implies (2.37),
and the proof of (ii) is complete.
Furthermore, if (4) holds, by (4) and (2.6) again, we also have H(t) (p
1)kt
p
/p for all t 0. Consequently H
1
(s) Cs
1/p
for every s 0, where
C = [(p1)k/p]
1/p
, and H
1
(s) cs
1/p
for 0 s B(1), as shown above in part
(ii). Hence (2.38) is now equivalent to (2.37), and this proves (iii) by (i).
Remarks To see that (4) is not automatic under the assumption (1)(3), note
that (t) = t
p1
(e
t
1) veries (1)(3) but not (4). Of course (t) = t
p1
e
t
satises (1)-(4), with k = 1.
Assume also (4). If f is singular at u = 0, with [f(u)[ u
q
as u 0
+
,
then by (f1) we must have q (1, 0). Hence [F(u)[
1/p
u
(q+1)/p
as u 0
+
,
so
_
0
+
[F(u)[
1/p
du < and every semiregular nonnegative radial ground state
of (2.4), with u(0) = > 0, is compactly supported on R
+
by Corollary 2.10
(iii). While, if f is regular at u = 0 and q > 0, then
_
0
+
[F(u)[
1/p
du < when
0 < q < p 1, and every semiregular nonnegative radial ground state of (2.4),
with u(0) = > 0, is compactly supported on R
+
by Corollary 2.10 (iii).
If again [f(u)[ u
q
as u 0
+
, with q > 1 by (f1), then by Corollary 2.10 (iii)
and Proposition 2.8 it follows that u(r) > 0, u

(r) < 0 for all r R

+
if and only if
q p 1. In this case, since p > 1, this means that f is continuous at u = 0, with
f(0) = 0, namely we are in the regular case.
For a more general wider discussion on the validity of the strong maximum and
compact support principles for solutions, radial or not, of quasilinear singular elliptic
inequalities, as well as on applications of these principles to variational problems
on manifolds and to existence of radial dead cores, one is referred to [17]. See also
[14].
3 Properties of solutions of the corresponding ini-
tial value problem
Semiregular nonnegative radial solutions u of (2.4), (2.27), or of (1.2), are also
solutions of the initial value problem
_
[r
n1
((r))]

r
n1
f(u(r)) = 0, r > 0,
u(0) = > 0, u

(0) = 0,
(3.39)
by (2.31) and the rst part of Proposition 2.8.
From now on, besides (1)(3), we assume that either (f2) or (f3) holds,
see the Introduction, so that, as noted above, (f4) is also satised, with given by
(2.34). Moreover, from now on we also assume that
(f5) f Lip
loc
(0, ).
Quasilinear Elliptic Equations 527
In this section we shall consider only solution of (3.39) under the further restric-
tion
d < < , (3.40)
where d and are given in (2.34).
Lemma 3.1 Problem (3.39) and (3.40) has a unique classical solution u in a neigh-
borhood of the origin. Moreover, u

(r) < 0 for r small and positive. The solution is

unique as long as it exists and remains in (0, ).
Proof. Local existence and uniqueness of solution of the initial value problem (3.39)
and (3.40) follows from Proposition A4 of the Appendix of [5], since (3) is equiv-
alent to (1.3), and in turn for 0 < t < p 1

(t) (t) > 0, (3.41)

which is the main condition of Proposition A4 of [5] in the special case = 1.
Moreover by Proposition 2.8 (i) and (ii) we have f() > 0 and u

(r) is negative for

small r.
By Proposition 2.8, any semiregular nonnegative radial solution u of (2.4),
(2.27), or of (1.2), satisfying (3.40), remains a classical solution of (3.39) as long as
it remains in (0, ]. Moreover, by Lemma 3.1 the unique local solution of (3.39) and
(3.40) has the property that u

< 0 and u < in some interval to the right of zero.

We claim that the solution can be continued either for all r > 0 with u(r) > 0 and
u

(r) < 0, or else reaches a rst point r

where either u(r

) = 0 and u

(r

) 0, or
u(r

) > 0 and u

(r

) = 0. To prove this, rst note that by (2.28), with s

0
= 0, and
(2.25), since F > 0 and increasing on (d, ) by either (f2) or (f3), then
H([u

(r)[) F() F(u(r)) F() + max

v[0,d]
[F(v)[ = M

R
+
,
by (2.33). Therefore, since H() = by (3), we also obtain that [u

(r)[
H
1
(M

) as long as u exists and 0 < u(r) . This shows the claim.

In what follows we assume that the solution u of (3.39) and (3.40) is continued
exactly until a rst point r

is reached where either u(r

) = 0 and u

(r

) 0, or
u(r

) > 0 and u

(r

) = 0. If no such a point occurs, we put r

= .
We denote by J

the maximal domain of continuation of any semiregular non

negative radial solution u of (2.4), (2.27), or of (1.2), satisfying (3.40), where it
is positive, namely J

= (0, r

), r

nite or not, is the maximal open interval of

continuation under the restriction
u > 0, < u

< 0 in J

. (3.42)
As mentioned earlier, the main purpose of this paper is to extend to the general
problem (2.4), (2.27) and to the corresponding free boundary problem (1.2), the
recent existence results given by Gazzola, Serrin and Tang in [7], under (Q1) and
528 B. Acciaio, P. Pucci
(f2), and by Tang in [18], under (Q1) and (f3), for the pLaplacian case, namely
when (t) = t
p1
, p > 1, in (1)(3), to the general setting of [9] on , under the
slightly more general growth condition (Q), when the natural assumption (f1) holds,
and using a unied technique. We shall also completely generalize the existence
results of [9].
Lemma 3.2 Let u be a solution of (3.39) and (3.40), and let J

= (0, r

) be the
corresponding maximal interval of denition in the sense of (3.42).
(i) The limit

= lim
rr

u(r) (3.43)
belongs to [0, b) if (f2) holds, and

= 0 if (f3) is satised. Moreover if

> 0, then u

(r

) = 0, with r

possibly innite.
(ii) Let > d. If > , then there exists a unique value r = r() J

such that
u(r) = .
(iii) If r

= , then lim
r
u

(r) = 0.
Proof.
(i) Clearly the limit in (3.43) exists and is nonnegative, since u is decreasing
and positive in J

by (3.42), in particular

[0, ) by (3.40).
First let (f2) hold. By contradiction, suppose

[b, ). Then b

<
u(r) < in J

, and this implies [r

n1
((r))]

> 0 by (3.39) and (3.40), that is

r
n1
((r)) is strictly increasing on J

.
If r

is nite and u(r

) =

b > 0, then u

(r

) = 0 by (3.42), and in turn

r
n1
((r)) tends to 0 as r r

. On the other hand, r

n1
((r)) is 0 at r = 0.
This is impossible since r
n1
((r)) is strictly increasing on J

.
If r

= , then J

= R
+
and E C
1
(R
+
) is nonincreasing in R
+
by Lemma
2.5. Therefore by (2.25) also H((r)) admits nite limit at , and in turn by
(2.5) also (r) admits limit as r . Consequently u

(r) 0 as r . Now,
rewriting the equation in (3.39) in the equivalent form
[((r))]

+
n 1
r
((r)) f(u(r)) = 0, r J

= R
+
,
we obtain that lim
r
[((r))]

= f(

) > 0, since

[b, ). This is impossible

since ((r)) > 0 on R
+
and lim
r
((r)) = 0 by (2).
Now let (f3) hold. If we assume by contradiction that

> 0, then, omitting b,

we can repeat the above proof word by word and reach the desired conclusion.
If

> 0 and r

< , then lim

rr

(r) = u

(r

) = 0 again by (3.42). If

> 0
and r

= , then by Lemma 2.5 and (2.25), as shown above, lim

r
u

(r) = 0. This
case can occur only when (f2) holds.
Quasilinear Elliptic Equations 529
(ii) The claim follows easily, by denition of J

, since u is strictly decreasing.

(iii) Let r

= , then the claim follows word by word as in the proof of (i),

since H((r)) approaches a nite limit as r by (2.25).
Lemma 3.3 Let u be a solution of (3.39) with maximal interval J

and assume
(3.40). Then for any t
0
J

and > 0, there exists > 0 such that if v is a

solution of (3.39) with [u(0) v(0)[ < , then v = v(r) is dened on [0, t
0
] and
sup
[0,t
0
]
[u(r) v(r)[ +[u

(r) v

(r)[ < .
Proof. The proof of Lemma 2.3 of [9] can be repeated since it used only the fact
that f Lip
loc
(0, ) together with (f1).
Lemma 3.4 (NiPucciSerrin) Let u be a solution of (3.39). Set
Q(v) = pnF(v) (n p)vf(v), v R
+
,
and
P(r) = (n p)r
n1
u(r)u

(r)A([u

(r)[) + pr
n
E(r), 0 < r < r

. (3.44)
Then
P(r)
_
r
0
Q(u(s))s
n1
ds. (3.45)
This follows at once by direct calculation from (2.8), (2.6) and (2.26). See the
proof of Lemma 2.3 of [9].
4 Preparatory existence results
To establish existence of semiregular nonnegative radial solutions of (2.4), (2.27)
or of (1.2) under (f2), or existence of semiregular radial crossing solutions of (2.4)
under (f3), using a unied approach, we follow the main ideas used in [5], [7], [18]
and [9]. In particular, we shall rst prove existence theorems for the corresponding
initial value problem (3.39), under condition (3.40).
Using the same notation of the previous section, we introduce the set
I

= > d : r

< ,

= 0, u

(r

) < 0.
Hence if I

,= and I

, the corresponding solution u

of (3.39) is a semi
regular radial crossing solution of (2.4).
Lemma 4.1 Let u be a semiregular radial solution of (3.39), under the restriction
(3.40), in its maximal continuation interval J

. If / I

, then
r <
Cu(r)[F + F(u(r))]
F(u(r))B
1
(p

[F + F(u(r))])
for every r J

, (4.46)
530 B. Acciaio, P. Pucci
where p

is the Holder conjugate of p,

C = (n 1)p

, F = max
v[0,d]
F

(v), (4.47)
B
1
is the inverse function of B given in (2.7), and F

= maxF, 0.
Proof. Assume by contradiction that there is r, with 0 < r < r

such that
r
CU[F + F(U)]
F(U)B
1
(p

[F + F(U)])
, (4.48)
where U = u(r). Put
M = sup
[r,r

)
(r) = (R
1
),
where R
1
[r, r

) by the continuity of u

on J

, (3.42), Lemma 3.2 (i) and (iii),

since / I

. By (2.5), (2.25) and (2.28) at s

0
= r and r = r

, we have
F(U) < E(r) = E(r

) + (n 1)
_
r

r
B((s))
s
ds (n 1)
(M)
r
_
r

r
u

(s)ds
=
(n 1)(M)U
r
,
by (2.32), (2.7), (3.42) and (2). Hence
F(U) <
(n 1)(M)U
r
. (4.49)
Similarly, by (2.25) and (2.28) at s
0
= R
1
and r = r

, we get
H(M) = F(

) F(u(R
1
)) + (n 1)
_
r

R
1
B((s))
s
ds
F (n 1)
(M)
R
1
_
r

R
1
u

(s)ds
F +
(n 1)(M)U
R
1
,
since u(R
1
) U being R
1
r, E(r

) = F(

) 0 if (f2) holds and is zero if (f3)

is satised, since u

(r

) = 0 by Lemma 3.2 (i), by assumption / I

, and nally
u(r)

U since

0. Hence
H(M) F +
(n 1)(M)U
r
. (4.50)
By (4.47)(4.49) and (2.7) it follows at once that
(M) >
r
n 1
F(U)
U

_
n 1
r
U
F(U)
p

[F + F(U)]
_
,
Quasilinear Elliptic Equations 531
namely by (2)
M
p

>
n 1
r
U
F(U)
[F + F(U)]. (4.51)
Now from (4.50), (2.6), (4.51) and (4.49) we nally have
F H(M) (n 1)
(M)
r
U
M
p

(M) (n 1)
(M)
r
U = (M)
_
M
p

n 1
r
U
_
>
r
n 1
F(U)
U
n 1
r
U
_
F + F(U)
F(U)
1
_
= F,
which is the required contradiction.
Theorem 4.2 (Existence) Let (f1) and either (f2) or (f3) hold. If
1
= , we
also assume that:
(Q) The function Q is locally bounded near v = 0 and there exist > d and
(0, 1) such that Q(v) 0 for all v and
limsup
v
Q(
1
v)[v
p+1
f(
2
v)]
n/p
= for every
1
,
2
[, 1].
Then I

,= .
Proof. We discuss separately the two cases < and = .
Let < . Problem (3.39) and (3.40) admits a unique classical solution u,
which depends continuously on by Lemmas 3.1 and 3.3. Hence u(r, ) as
uniformly on every bounded interval of R
+
0
. Since (d, ), for suciently
close to , by Lemma 3.2 (ii), there is a unique value t

such that u(t

, ) = ,
where = (+d)/2 (d, ). We claim that t

is not bounded above as .

For, otherwise, 0 < t

t < as , for an appropriate constant t, and all

the corresponding solutions u(r, ) would be as r t, which contradicts the
fact that u(r, ) converges uniformly on bounded sets to as .
Now, putting u(t

, ) = in the righthand side of (4.46), we nd a nite

number which is independent of . This is impossible by the previous argument.
Consequently there is (d, ) such that the corresponding solution u(r, ),
with u(t

, ) = , has the property that

t

>
C [F + F( )]
F( )B
1
(p

[F + F( )])
,
namely I

by Lemma 4.1, or u is a crossing solution by the denition of I

,
and the proof is complete when < .
Now let = . Assume for contradiction that u(r, ) is a global solution for
all (d, ), namely J

= R
+
for all (d, ).
We take and as stated in (Q), with > , and, without loss of generality,
we take suciently close to 1 such that:
>

and 0 <
(1 )
r

=
u(0) u(r

)
r

1, (4.52)
532 B. Acciaio, P. Pucci
where r

is the unique point r such that u(r

) = by Lemma 3.2 (ii). This is

clearly possible since by (2.19)
lim
r0
+
u(r) u(0)
r
= u

(0) = 0 and lim

1
r

= 0.
Integrating (3.39) on [0, r], with r (0, r

), we obtain
r
n1
((r)) =
_
r
0
s
n1
f(u(s))ds.
Hence, putting
f(
2
) = max
[,]
f(u) > 0,
2
[, 1],
we have
r
n1
((r)) f(
2
)
r
n
n
.
Since () = by (3), then by (2)
(r)
1
_
f(
2
)r
n
_
.
Integration from 0 to r

yields
(1 )
_
r

1
_
f(
2
)r
n
_
dr r

1
_
f(
2
)r

n
_
,
by (2). Thus
f(
2
)
n

1
r

_
(1 )
r

_
=
1
r

__
(1 )
r

_
p/(p1)
_
r

(1 )
_
1/(p1)
_

1
r

(1 )

__
(1 )
r

_
p/(p1)
_
,
by (3) and (4.52). This shows that

1
_
(1 )f(
2
)
n
_

_
(1 )
r

_
p/(p1)
, (4.53)
namely
r

(1 )
_

1
_
(1 )f(
2
)
n
__
(p1)/p
.
Now, we take < 1 suciently close to 1 so that
(1 )f(
2
)
n
1.
Quasilinear Elliptic Equations 533
Hence
_
(1 )f(
2
)
n
_
1/(p1)

1
_
(1 )f(
2
)
n
_

1
(1),
since the function
1/(p1)

1
(), R
+
0
, is nondecreasing by (3). There-
fore

1
(1)
_
n
(1 )f(
2
)
_
1/(p1)

1
_
(1 )f(
2
)
n
_

_
(1 )
r

_
p/(p1)
,
by (4.53), or
r

(1 )
(p+1)/p
[
1
(1)]
(p1)/p
n
1/p

(p+1)/p
[f(
2
)]
1/p
,
namely
r

(p+1)/p
[f(
2
)]
1/p
, C

=
(1 )
(p+1)/p
[
1
(1)]
(p1)/p
n
1/p
. (4.54)
Let r

be the unique value of r such that

u(r

) = . (4.55)
Hence r

> r

, since u(r

) = > = u(r

) by (4.52), and u

< 0 on J

= R
+
.
Furthermore, since , I

, by (4.46)
r

< C

:=
C[F + F()]
F()B
1
(p

[F + F()])
, (4.56)
where C is given in (4.47) and B in (2.7).
Let
Q

= inf
0<v
Q(v) > (4.57)
by (Q). Moreover there is a suitable constant
1
[, 1] such that
Q(
1
) = min
v
Q(v) 0 (4.58)
by (Q) since > . Consequently, by construction,
Q(u(r))
_
_
_
Q(
1
), if 0 < r < r

0, if r

r r

[Q

[, if r > r

.
(4.59)
By Lemma 3.4, that is (3.44) and (3.45), with r > r

,
pr
n
E(r) P(r)
__
r

0
+
_
r

+
_
r
r

_
s
n1
Q(u(s))ds. (4.60)
534 B. Acciaio, P. Pucci
Hence, using (4.59), we get
pr
n
E(r)
Q(
1
)
n
C
n

[
p+1
f(
2
)]
n/p

[Q

[r
n
n
,
by (4.54). By (2.25), (4.56) and the fact that F is positive and increasing on (d, ),
and so on (d, ],
pnH((r)) >
_
C

+ 1
_
n
Q(
1
)[
p+1
f(
2
)]
n/p
[Q

[ npF(),
for r

< r < r

+1. By (Q) we can take suciently large so that H((r)) H(),

that is (r) by (2.5). Consequently
u(r

+ 1) = u(r

) +
_
r

+1
r

(s)ds = 0,
which contradicts the fact that u > 0 on J

= R
+
, and completes the proof.
We now recall the main growth condition, denoted here by (Q1), for existence
in the principal case = , used in [7], [9] and [18]:
(Q1) The function Q is locally bounded near v = 0 and there exist > d and
(0, 1) such that Q(v) 0 for v and
limsup
v
Q(
1
v)
_
v
p1
f(
2
v)
_
n/p
= for all
1
and
2
in [, 1].
Proposition 4.3 Let = . Assume that (f1) holds and also that
(F1) liminf
u
f(u) = k
0
> 0,
with k
0
possibly innity, is satised. Then
(Q1) = (Q).
Proof. By (F1), for u suciently large
f(u) > 0 and uf(
2
u) 1/ for all
2
[, 1].
Consequently by (Q1)
Q(
1
u)u
(p+1)n/p
[f(
2
u)]
n/p
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
,
and the implication follows at once.
Theorem 4.2 generalizes Lemma 3.5 of [9], since in general (f1), (F1), (Q)
and either (f2) or (f3) do not imply the validity of (Q1), as shown by the following
examples. For brevity we dene the functions f under consideration in the examples
Quasilinear Elliptic Equations 535
only for large u, and without further mention we assume that the functions f satisfy
(f1) and either (f2) or (f3), and the corresponding functions Q(v) = pnF(v)
(n p)vf(v), v > 0, are locally bounded near v = 0. Finally, in the next examples,
1 < p < n and p

=
np
n p
.
Examples
1. Let
f(u) = u
p

1
+
1
u
, for u u
0
, with u
0
> 0.
Clearly (F1) holds with k
0
= . Here
Q(u) = c
1
+ np log u, u u
0
,
where c
1
is an appropriate constant. Hence for u and > d suciently large,
certainly Q(u) > 0. Now, as u
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
= [c
1
+ np log (
1
u)]
u
n(p1)/p
_
(
2
u)
p

1
+
1

2
u
_
n/p
0,
so that clearly (Q1) fails, while
Q(
1
u)[u
p+1
f(
2
u)]
n/p
= [c
1
+np log (
1
u)]u
n(p+1)/p
_
(
2
u)
p

1
+
1

2
u
_
n/p
,
namely (Q) holds.
2. Dene
f(u) = u
p

1
u
q
for u u
0
, with q > 1, (4.61)
where u
0
> 1 when the minus sign is considered and (f3) holds, since = in
the examples, while u
0
> 0 in all the other cases. Again (F1) holds with k
0
= .
Hence, for c
0
= n p + np/(q 1)=const. > 0 and u u
0
,
Q(u) =
c
0
u
q1
+ c
1
,
where the appropriate constant c
1
is now assumed to be positive.
1
As u
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
=
_
c
0
(
1
u)
q1
+ c
1
_
u
n(p1)/p
_
(
2
u)
p

1
(
2
u)
q
_
n/p
0,
1
To have c
1
> 0, it is enough to dene f(u) = log u + 2 for u (0, 1], when in (4.61) the
plus sign is considered, u
0
= 1 and (f2) holds; while f(u) = u
s
+ 1 for u (0, 1], s > 1,
when in (4.61) the plus sign is considered, u
0
= 1 but now (f3) holds. Analogously, when in
(4.61) the minus sign is taken, to get c
1
> 0, we can dene f(u) = u
p

1
u
s
+ c(1 u) for
u (0, 1], with s (0, 1) and the positive constant c > 2/(1 s) + 2/(q 1), when (f2) holds;
while f(u) = u
p

1
0
u
q
0
+ 2q(u
0
u)/(q 1) for u [0, u
0
], u
0
> 1, when (f3) holds.
536 B. Acciaio, P. Pucci
that is (Q1) fails, while
Q(
1
u)[u
p+1
f(
2
u)]
n/p
=
_
c
0
(
1
u)
q1
+c
1
_
u
n(p+1)/p
_
(
2
u)
p

1
(
2
u)
q
_
n/p
,
that is property (Q) holds.
3. Take u
0
> 0 if (f2) holds and u
0
0 if (f3) is satised, then dene
f(u) = u
p

1
+ u
q1
, 1 < q < p

.
Again (F1) holds with k
0
= and, for u u
0
,
Q(u) = c
0
u
q
+ c
1
, with c
0
=
np
q
(n p) = const. > 0,
and c
1
is an appropriate constant. As u ,
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
= [c
0
(
1
u)
q
+ c
1
]
u
n(p1)/p
[(
2
u)
p

1
+ (
2
u)
q1
]
n/p
0,
namely (Q1) fails, while
Q(
1
u)[u
p+1
f(
2
u)]
n/p
= [c
0
(
1
u)
q
+c
1
]u
n(p+1)/p
[(
2
u)
p

1
+(
2
u)
q1
]
n/p
,
that is (Q) holds.
4. Now let p = n and
f(u) = pu
p1
e
u
p
for u u
0
,
with u
0
> 0 if (f2) holds and with u
0
0 when (f3) is satised. Clearly (F1)
is satised with k
0
= and Q(u) = p
2
_
e
u
p
+ c
1
_
, where c
1
is an appropriate
constant. Hence, as u , for
1
=
2
,
Q(
1
u)
u
p1
f(
2
u)
=
p[e
(
1
u)
p
+ c
1
]

p1
2
e
(
2
u)
p
p
1p
1
,
namely (Q1) fails, while for all
1
,
2
[, 1], with (0, 1),
Q(
1
u)[u
p+1
f(
2
u)] =
p1
2
p
3
[e
(
1
u)
p
+ c
1
]u
2p
e
(
2
u)
p
,
that is (Q) holds.
Proposition 4.4 If = , and f satises (f1) and
(F2) lim
u
uf(u) = k
1
0, with k
1
nite,
then the reverse implication of Proposition 4.3 holds, namely
(Q) = (Q1).
Quasilinear Elliptic Equations 537
Proof. Indeed, for u suciently large
Q(
1
u)u
n
[uf(
2
u)]
n/p
c
0
Q(
1
u)u
n
[uf(
2
u)]
n/p
,
where c
0
is a positive constant, and the implication follows at once.
Clearly (F2) implies that f(u) 0 as u , a situation which is not so
interesting in applications and in any case the existence problem could be treated
for such nonlinearities using much simpler techniques.
In general (f1), (F2), (Q1) and either (f2) or (f3), do not imply (Q), as shown
by the following examples. As before, for brevity the functions f under consideration
in the next examples are dened only for large u and assumed to satisfy (f1) and
either (f2) or (f3), without further mention. Again the corresponding functions
Q(v) = pnF(v) (n p)vf(v), v > 0, are supposed to be locally bounded near
at v = 0. Finally, in the next examples 1 < p n, with p

= if p = n and
p

=
np
n p
if 1 < p < n.
Example 5. Let
f(u) = u
q1
for u u
0
, u
0
> 0, q p.
Clearly (F2) holds with k
1
= 0 and Q(u) = c
1
c
2
u
q
for u u
0
, where the appro-
priate constant c
1
is now assumed to be positive
2
and c
2
= npnp/q =const.> 0.
Hence
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
=
2
n(1q)/p
[c
1
c
2
(
1
u)
q
] u
n(pq)/p
,
that is (Q1) holds, while
Q(
1
u)[u
p+1
f(
2
u)]
n/p
=
2
n(q1)/p
[c
1
c
2
(
1
u)
q
] u
n(p+q)/p
,
where = c
1

2
n(p+1)/p
when q = p and = 0 when q < p, so that (Q) fails.
6. Indeed, if
f(u) = u
q1
for u large, with q > p,
then (Q) and (Q1) are valid if and only if q < p

.
7. Let
f(u) = e
u
for u u
0
,
with u
0
> 0 under (f2) and with u
0
0 under (f3). Condition (F2) is satised
with k
1
= 0, and for u u
0
Q(u) = c
1
[np +(n p)u]e
u
, where the appropriate
2
To have c
1
> 0, it is enough to dene f(u) = log u + 1 for u (0, 1], when u
0
= 1 and (f2)
holds; while f(u) = u
s
for u (0, 1], s > 1, when u
0
= 1 but now (f3) holds.
538 B. Acciaio, P. Pucci
constant c
1
> 0 is again assumed to be positive.
3
Q(
1
u)
_
u
p1
f(
2
u)
_
n/p
= c
1
[np + (n p)
1
u]e

1
u
u
n(p1)/p
e
n
2
u/p
,
that is (Q1) holds, while
Q(
1
u)[u
p+1
f(
2
u)]
n/p
= c
1
[np + (n p)
1
u]e

1
u
u
n(p+1)/p
e
n
2
u/p
0,
namely (Q) fails.
Therefore Theorem 4.2 extends the result of Theorem 5.1 of [18] also to the
nonlinearity satisfying (Q).
8. Dene
f(u) = u
r1
u
q1
, 1 < q < r < p

,
for u u
0
, where u
0
> 0 when the plus sign is considered and (f2) holds, and
u
0
> 1 when the minus sign is considered and (f3) is satised, while in the other
cases u
0
0. Clearly (F1) holds with k
0
= and Q(u) = c
1
+
r
u
r

q
u
q
for
u u
0
, where c
1
> 0 is an appropriate constant and

r
=
pn
r
(n p) > 0 and
q
=
pn
q
(n p) > 0.
Then (Q1) is satised, and so also the weaker condition (Q), since (F1) holds.
9. Let
f(u) = u
q1
log u for u large, with 0 q < p

.
Then for u > 0 suciently large
Q(u) =
_
_
_
u
q
_
log u
_
np
q
(n p)
_

np
q
2
_
+ c
0
, 0 < q < p

,
np
2
log
2
u (n p) log u + c
1
, q = 0,
Clearly Q(u) 0 for u and an appropriate > d, since here q < p

, as
requested in the main assumptions. Moreover, after some calculation, one sees that
both (Q) and (Q1) hold.
Indeed, it is worth noting that the existence Theorem 4.2, as well as those in [7]
and [18] for the pLaplacian case, and in [9] for general divergence equations, can
be applied in particular when
f(u) = u
q1
for u large, p < q < p

;
f(u) = u
q1
log u for u large, 0 q < p

,
3
To get c
1
> 0, it is enough to dene f(u) = log u+log 2+e
1/2
for u (0, 1/2], when u
0
= 1/2
and (f2) holds; while it is enough to take u
0
= 0 when (f3) holds.
Quasilinear Elliptic Equations 539
since both (Q) and (Q1) hold.
Now, assuming the further condition (4), given in the Introduction, we obtain
another criterion, which extends recent results of [7] and [18], established for the
pLaplacian case, and also Theorem 4.1 of [9].
Theorem 4.5 Suppose that (4), (Q1), (f1) and either (f2) or (f3) hold. Then
I

,= .
Proof. Obviously the only new case to be proved is when = . Indeed, we can
proceed essentially as in the proof of Theorem 4.2. By (Q1) there is > d such that
Q(u) 0 for u . Choose > /, where is the constant specied in (Q1), and
let u

be the corresponding solution of (3.39), with maximal domain J

= (0, r

)
in the sense of (3.42). We denote by r

and r

respectively the unique values of r

in which u

reaches and , by Lemma 3.2 (ii). Clearly r

< r

.
Hence, putting
f(
2
) = max
[,]
f(u) > 0,
2
[, 1],
and integrating (3.39) on [0, r], for any r (0, r

), we have
kr
n1

p1
(r) r
n1
((r)) =
_
r
0
s
n1
f(u(s))s
f(
2
)
n
r
n
,
by (4). Thus
(r)
_
f(
2
)
kn
_
1/(p1)
r
1/(p1)
.
Integration from 0 to r

yields
(1 )
p 1
p
_
f(
2
)
kn
_
1/(p1)
r
p/(p1)

,
and therefore
r

_

p1
f(
2
)
_
1/p
, where C

=
_
kn
_
(1 )p
p 1
_
p1
_
1/p
, (4.62)
which is the exact analogue of (4.54). Continuing the proof as that of Theorem 4.2,
we obtain (4.55)(4.60). Now, by (4.60), (2.25) and (4.62),
pnH((r)) >
_
C

+ 1
_
n
Q(
1
)
_

p1
f(
2
)
_
n/p
[Q

[ npF(),
for r

< r < r

+1, where C

, Q

and Q(
1
) are given in (4.56), (4.57) and (4.58),
respectively. Thus by (Q1) we can take suciently large so that H((r)) H()
and conclude the proof exactly as in Theorem 4.2.
540 B. Acciaio, P. Pucci
5 Main existence results
Here we turn to the main problems (2.4), (2.27) and (1.2) under the main conditions
(1)(3).
Theorem 5.1 Assume (f1), (f2) and (Q). Then problem (2.4) admits a positive
semiregular radial ground state if (2.37) holds, while a compactly supported semi
regular nonnegative nontrivial ground state if (2.37) fails. In this second case the
ground state u is also a solution of the free boundary problem (1.2).
Proof. Put
I
+
= b :

> 0.
Following the proofs of Lemmas 3.2 and 3.3 of [9], word by word, we prove that I
+
is not empty and open. Similarly, using the proof of Lemma 3.6 of [9], we can also
show that I

is also open, and by the above Theorem 4.2 is not empty. Therefore
there should be a number / I
+
. I

, with b < < , since b I

+
by Lemma
3.2 of [9], such that (3.39) admits a positive solution u

, with u

(0) = , in its
maximal continuation open interval J

= (0, r

). Since / I
+
, then

= 0, while
since / I

, then either r

= or r

< , and in both cases u

(r

) = 0 by
Lemma 3.2 (i) and (iii). In the rst case u

is a semiregular positive radial ground

state of (2.4), and the latter u

is a semiregular positive radial solution of (1.2),

with R = r

, or a compactly supported radial ground state of (2.4), accordingly to

Corollary 2.10 (i).
Theorem 5.2 Assume (f1), (f3) and (Q). Then problem
div(A([Du[)Du) + f(u) = 0 in R
n
, n > 1,
u , 0 in R
n
,
(5.63)
admits a semiregular crossing radial solution u in its maximal continuation open
interval J

= (0, r

), with u(0) = (0, ), u

(0) = 0, r

< , u(r

) = 0 and
u

(r

) < 0.
Proof. By Theorem 4.2 there is I

and the conclusion follows at once.

Theorem 5.3 Assume (4), (f1), (f2) and (Q1). Then problem (2.4) admits a
positive semiregular radial ground state if (2.38) holds, while a compactly supported
semiregular nonnegative nontrivial ground state if (2.38) fails. In this second
case the ground state u is also a solution of the free boundary problem (1.2).
Proof. We proceed, word for word, as in the proof of Theorem 5.1, where now I

,=
by Theorem 4.5 and by Corollary 2.10 (iii).
Theorem 5.4 Assume (4), (f1), (f3) and (Q1). Then problem (5.63) admits a
semiregular crossing radial solution u in its maximal continuation open interval
J

= (0, r

), with u(0) = (0, ), u

(0) = 0, r

< , u(r

) = 0 and u

(r

) < 0.
Proof. This is an immediate consequence of Theorem 4.5.
Quasilinear Elliptic Equations 541
References
[1] F.V. Atkinson and L.A. Peletier, Ground states of u = f(u) and the Emden-Fowler
equation, Arch. Rational Mech. Anal. 93 (1986), 103-127.
[2] H. Berestycki and P.L. Lions, Non-linear scalar eld equations, L Existence of a
ground state, Arch. Rational Mech. Anal. 82 (1983), 313-345.
[3] S. Coleman, V. Glazer and A. Martin, Actions minima among solutions to a class of
Euclidean scalar eld equations, Comm. Math. Phys. 58 (1978), 211-221.
[4] L. Erbe and M. Tang, Uniqueness theorems for positive radial solutions of quasilinear
elliptic equations in a ball, J. Dierential Equations 138 (1997), 351-379.
[5] B. Franchi, E. Lanconelli and J. Serrin, Existence and uniqueness of nonnegative
solutions of quasilinear equations in R
n
, Advances in Math. 118 (1996), 177-243.
[6] M. Garca-Huidobro, R. Manasevich, P. Pucci and J. Serrin, Qualitative properties of
ground states for singular elliptic equations with weights, submitted for publication,
(pages 27).
[7] F. Gazzola, J. Serrin and M. Tang, Existence of ground states and free boundary
problems for quasilinear elliptic operators, Adv. Dierential Equations 5 (2000), 1-30.
[8] N. Kawano, E. Yanagida and S. Yotsutani, Structure theorems for positive radial
solutions to u +K(|x|)u
p
= 0 in R
n
, Funkcial. Ekvac. 36 (1993), 557-579.
[9] E. Montefusco and P. Pucci, Existence of radial ground states for quasilinear elliptic
equations, Adv. Dierential Equations 6 (2001), 959-986.
[10] W.M. Ni and J. Serrin, Nonexistence theorems for quasilinear partial dierential
equations, Rend. Circ. Mat. Palermo, Suppl. 8 (1985), 171-185.
[11] W.M. Ni and J. Serrin, Existence and nonexistence theorems for ground states of
quasilinear partial dierential equations. The anomalous case, Atti Accad. Naz. Lincei
Rend. Cl. Sci. Fis. Mat. Natur. 77 (1986), 231-257.
[12] L.A. Peletier and J. Serrin, Uniqueness of positive solutions of semilinear equations
in R
n
, Arch. Rational Mech. Anal. 81 (1983), 181-197.
[13] L.A. Peletier and J. Serrin, Uniqueness of nonnegative solutions of semilinear equa-
tions in R
n
, J. Dierential Equations 61 (1986), 380-397.
[14] P. Pucci, M. Rigoli and J. Serrin, The compact support principle for singular elliptic
inequalities on complete manifolds, in preparation.
[15] P. Pucci and J. Serrin, A general variation identity, Indiana Univ. Math. J. 35(1986),
681-703.
[16] P. Pucci and J. Serrin, Uniqueness of ground states for quasilinear elliptic operators,
Indiana Univ. Math. J. 47 (1998), 501-528.
[17] P. Pucci and J. Serrin, The strong maximum principle revisited, to appear in J. Dif-
ferential Equations, (pages 65).
[18] M. Tang, Existence and uniqueness of fast decay entire solutions of quasilinear elliptic
equations, J. Dierential Equations 164 (1999), 155-179.