Differential and Integral Equations, Vol. 9, N. 3, 465-479, (1996).

NON-EXISTENCE OF POSITIVE SOLUTIONS

OF SEMILINEAR ELLIPTIC SYSTEMS IN RN

ENZO MITIDIERI
Dipartimento di Scienze Matematiche
Università degli Studi di Trieste
Piazzale Europa 1, 34100 Trieste, Italy

March 31 1992
1. Introduction

In this paper we study non-existence of positive classical solutions of elliptic systems

of the form
−∆u = f (u, v), −∆v = g(u, v) in RN (1.1)

where N ≥ 3 and f , g are given non-negative functions such that f (0, 0) = g(0, 0) = 0.
Recently the existence and non-existence of positive solutions for the scalar equation

−∆u = f (x, u) in RN , (1.2)

has been studied by a number of authors, see for example [2],[4],[5],[9],[10], [13],[14],[16]
and the references therein. In particular in Gidas-Spruck [5] it is proved among other
things that if f (x, u) = |u|p−1 u then (1.2) has no positive classical solutions provided
1 <p<(N+2)/(N−2). An important feature of their result is that no a priori information
on the behavior of the solutions at infinity is given.
Liouville type theorems of this kind play an important role in the study of semilinear
elliptic and parabolic equations as is shown in [6],[7],[8]. Hence it is a natural question,
in view of the possible applications, to investigate when similar results continue to hold
for systems of equations. Some interesting non-existence results have for example been
obtained for systems of semilinear elliptic equations of potential type by Noussair and
Swanson [17].

AMS Subject Classification : 35j60 , 35j70.

1
 The paper is organized as follows. In Section 2 we prove some Liouville theorems of
Gidas-Spruck type for systems of the form (1.1) and for related equations. In Section 3
we consider non-existence of positive radially symmetric solutions for a non-autonomous
system and for a related non-variational problem (see Theorem 3.3). Finally in Section 4
we discuss some results for the Dirichlet problem in a ball.

2. Non-existence results

Let u ∈ C 2 (RN ),(N ≥ 3) and consider its average on a sphere of radius r = |x|
centered at the origin, that is
Z
−1
T (u)(x) = u (x) = |SN −1 |
#
u(rω) dω,
SN −1

where ω denotes the variable on the unit sphere SN −1 . It is well known (see [23], Propo-
sition 9.2) that
(∆u)# = ∆u# . (2.1)
#
We shall call T (or ) the spherical mean operator.
For later purposes note that if f : R+ → R+ is a given convex function and
u ∈ C 2 (RN ), then a simple application of the Jensen inequality shows that

f (u)# (x) ≥ f (u# (x)), x ∈ RN . (2.2)

We also note that if u is nonnegative and superharmonic, then by (2.1) u# is also super-
harmonic and hence u# is non-increasing with respect to r.

Theorem 2.1. Suppose that f, g: [0, ∞) × [0, ∞) → [0, ∞) are continuous and such that

f (0, 0) = g(0, 0) = 0

f (x, y) ≥ a11 xk + a12 y p , (x, y) ∈ [0, ∞), (2.3)

g(x, y) ≥ a21 xq + a22 y s , (x, y) ∈ [0, ∞), (2.4)

where

(i) a11 , a12 , a21 , a22 > 0, (ii) k, s, p, q > 1.

If ½ ¾
N −2 1 p+1 1 q+1
≤ max , , , (2.5)
2 k − 1 pq − 1 s − 1 pq − 1

2
then the problem
−∆u = f (u, v) , −∆v = g(u, v) in RN (2.6)

has no positive solutions of class C 2 (RN ).

The following result will play an important role in the rest of the paper. For its proof, see
[1].

Lemma 2.1. If Ψ ∈ C 2m (RN \ {0}), N > 2m, is positive and radial and if

(−∆)k Ψ ≥ 0 in RN , for every 0 ≤ k ≤ m,

then for every r ∈ (0, ∞) we have (r = |x|),

rΨ0 (r) + (N − 2m)Ψ(r) ≥ 0. (2.7)

Remark 2.1. Note that (2.7) is equivalent to (rN −2m Ψ(r))0 ≥ 0. Hence

Ψ(r) ≥ Cr2m−N , C > 0, r > r0 . (2.8)

Proof of Theorem 2.1. Applying the spherical mean operator to (2.6) and using (2.2)
and (2.3) − (2.4) we obtain for r ∈ (0, ∞)

−∆u# ≥ a11 (u# )k + a12 (v # )p

−∆v # ≥ a21 (u# )q + a22 (v # )s

(u# )0 (0) = (v # )0 (0) = 0;

hence
−(rN −1 (u# (r))0 ≥ rN −1 {a11 u# (r)k + a12 v # (r)p }, (2.9)

−(rN −1 (v # (r))0 ≥ rN −1 {a21 u# (r)q + a22 v # (r)s }. (2.10)

Integrating (2.9) − (2.10) on (0, r) and taking into account that u# , v # are non-increasing,
we obtain
−ru# (r)0 ≥ Cr2 {a11 u# (r)k + a12 v # (r)p },

−rv # (r)0 ≥ Cr2 {a21 u# (r)q + a22 v # (r)s }.

3
On the other hand, it follows from (2.7) with m = 1 that

(N − 2)u# (r) ≥ −ru# (r)0 ≥ Cr2 {a11 u# (r)k + a12 v # (r)p },

(N − 2)v # (r) ≥ −rv # (r)0 ≥ Cr2 {a21 u# (r)q + a22 v # (r)s }.

Solving these inequalities, we obtain for r > 0,

u# (r) ≤ Cr− max{2/(k−1),2(p+1)/pq−1} , v # (r) ≤ Cr− max{2/(s−1),2(q+1)/pq−1} . (2.11)

Combining (2.11) with (2.8), it follows that

Cr2−N ≤ u# (r) ≤ Cr− max{2/k−1,2(p+1)/pq−1} ,

Cr2−N ≤ v # (r) ≤ Cr− max{2/s−1,2(q+1)/pq−1} .

Thus if ½ ¾
N −2 1 p+1 1 q+1
< max , , ,
2 k − 1 pq − 1 s − 1 pq − 1
we obtain a contradiction for large r. This concludes the proof in the case of strict inequality
in (2.5).
Next, consider the case
½ ¾
N −2 1 p+1 1 q+1
= max , , , .
2 k − 1 pq − 1 s − 1 pq − 1
N −2
Suppose that 2 = p+1
pq−1 , the other cases being similar. It follows from (2.7) and (2.9) −
(2.10) that for r > 0,

−(r(u# (r))0 + (N − 2)u# (r))0 ≥ Cr{a11 u# (r)k + a12 v # (r)p } ≥ Cr1+2p u# (r)pq . (2.12)

Integrating (2.12) on (r, t) and taking into account (2.7) and Remark 2.1, we obtain
Zt
# 0 N −2 pq
r(u ) (r) + (N − 2)u (r) ≥ C(u (r)r
# #
) ξ 1+2p+2pq(2−N ) dξ
r

1 h i
= C(u# (r)rN −2 )pq t2+2p+(2−N )pq − r2+2p+(2−N )pq . (2.13)
2 + 2p + (2 − N )pq

Proceeding to the limit t → ∞ on the right hand side of (2.13) (note that 2+2p+(2−N )pq <
0) and multiplying the resulting inequality by rN −3 , we obtain

(u# (r)rN −2 )0 ≤ Cr−1 (u# (r)rN −2 )pq , r > r0 > 0. (2.14)

4
Finally, integrating (2.14) on (r, t) we get
1 £ # ¤ t
(u (t)tN −2 )1−pq − (u# (r)rN −2 )1−pq ≥ C log( ). (2.15)
1 − pq r

Passing to the limit t → ∞ in (2.15) leads to a contradiction. This concludes the proof.

If in Theorem 2.1 we suppose that a11 = a22 = 0 and a12 , a21 > 0, then by the same
method we obtain, for r > 0,

u# (r) ≤ C r−2(p+1)/(pq−1) , v # (r) ≤ C r−2(q+1)/(pq−1) .

Corollary 2.1. Suppose that 1 < p ≤ q, N ≥ 3,and

N −2
(pq − 1) ≤ max {p + 1, q + 1} .
2
Then the problem
−∆u ≥ v p , −∆v ≥ uq in RN

has no positive solutions of class C 2 (RN ).

Similarly, if 1 = p < q we obtain

u# (r) ≤ C r−4/(q−1) , v # (r) ≤ C r−2(q+1)/(q−1)

which gives

N
Corollary 2.2. Suppose that N > 4 and 1 < q ≤ .Then the problem
N −4

∆2 u ≥ u q , ∆u ≤ 0 in RN (2.16)

has no positive solutions of class C 4 (RN ).

In order to prove Corollary 2.2 it is sufficient to write (2.16) in the form

−∆u ≥ v , −∆v ≥ uq

and to proceed as in the proof of Theorem 2.1.

The generalization of Theorem 2.1 to (even) polyharmonic equations is contained in

the following

5
Proposition 2.1. Assume that m ≥ 2, N > 2m, and

N
1<q≤ .
N − 4m

Then for any s = 1, 2, . . . , 2m − 1, the problem

∆2m u ≥ uq , (−1)s ∆s u ≥ 0 in RN

has no positive solutions of class C 2m (RN ).

N
Proof. First we consider the case 1 < q < N −4m . Introducing the new variables
(u0 , u1 , . . . , u2m−1 ) with u0 = u, it follows that (u0 , u1 , . . . , u2m−1 ) is a positive C 2 (RN )
solution of the equivalent system

−∆u0 = u1 , −∆u1 = u2 ,..., −∆u2m−1 ≥ uq in RN . (2.17)

Applying the spherical mean operator to (2.17) and using (2.2) there results
¡ # ¢q
−∆u# #
0 = u1 , −∆u# #
1 = u2 , . . . − ∆u#
2m−1 ≥ u in RN . (2.18)

Writing (2.18) in radial coordinates and taking into account that u#

i (r), i = 0, 1, . . . , 2m−1,
is non-increasing, we obtain for r > 0

u# 2 #
i (r) ≥ Cr ui+1 (r) , i = 0, 1, 2, . . . , 2m − 1.

It follows that for r > 0,

u#
0 (r) ≥ Cr
2(2m−1) #
u2m−1 (r)

and
−∆u#
2m−1 ≥ Cr
2q(2m−1) #
(u2m−1 (r))q . (2.19)

Hence
−(2(2m−1)+2)/(q−1)
u#
2m−1 (r) ≤ Cr . (2.20)

Combining (2.20) with the fact that u#

2m−1 (r) ≥ C0 r
2−N
, C0 > 0, we obtain a contradiction
for large r.
N
Next, we consider the case q = N −4m . Using (2.19) it follows that

0 0
−(r(u# #
2m−1 (r)) + (N − 2)u2m−1 (r)) ≥ Cr
1+2q(2m−1) #
u2m−1 (r)q . (2.21)

6
Integrating (2.21) on (r, t) and proceeding as in (2.13),we get

Zt
0
r(u# #
2m−1 (r)) + (N − 2)u2m−1 (r) ≥ C ξ 1+2q(2m−1) u#
2m−1 (ξ) dξ ≥
q

Zt
N −2 q
≥ C(u#
2m−1 (r)r ) ξ 1+2q(2m−1)+q(2−N ) dξ,
r

and by letting t → ∞,
r(u# #
2m−1 (r) + (N − 2)u2m−1 (r) ≥

C
≥ (u# (r)rN −2 )q r2+q(2−N )+2q(2m−1) .
q(N − 2) − 2 + 2q(1 − 2m) 2m−1
Thus
N −2 0
(u#
2m−1 (r)r ) ≥

C
≥ (u# (r)rN −2 )q rq(2−N )+2q(2m−1)+N −1 .
q(N − 2) − 2 + 2q(1 − 2m) 2m−1
Finally by observing that q(2 − N ) + 2q(2m − 1) + N − 1 = −1, the demonstration is
completed as in the proof of Theorem 2.1.

Remark 2.2. By employing the same methods, we can obtain non-existence results for
systems of the form
−∆ui = fi (x, u1 , u2 , . . . , uk ) in RN ,

where fi : RN × [0, ∞)k → [0, ∞), (i = 1, 2, . . . , k), are continuous functions which satisfy

X
k
pij −1
fi (x, u1 , u2 , . . . , uk ) ≥ aij |x|αij |uj | uj ,
j=1

with aij , αij > 0 and pij > 1 for i, j = 1, 2, . . . , k.

If k ≥ 3, then the analogue of condition (2.5) becomes very involved. In order to

illustrate this fact, consider the following example.

Suppose that (u, v, w) ∈ C 2 (RN ) is a positive solution of

−∆u = |x|a v p , −∆v = |x|b wk , −∆w = |x|c uq in RN . (2.22)

7
where a, b, c > 0 and p, q, k > 1, and N ≥ 3. By applying the spherical mean operator to
(2.22), we obtain

p k q
−∆u# ≥ |x|a (v # ) , −∆v # ≥ |x|b (w# ) , −∆w# ≥ |x|c (u# ) in RN .

Then as before it is seen that, for r > 0,

u# (r) ≤ C r−(2+a+p(2+b)+kp(2+c))/(pqk−1) ,

v # (r) ≤ C r−(2+b+k(2+c)+qk(2+a))/(pqk−1) ,

w# (r) ≤ C r−(2+c+q(2+a)+pq(2+b))/(pqk−1) .

Hence , if
(N − 2)(pqk − 1) ≤ max {A, B, C} ,

where

A = 2 + a + p(2 + b) + kp(2 + c), B = 2 + b + k(2 + c) + qk(2 + a),

C = 2 + c + q(2 + a) + pq(2 + b),

then u# = v # = w# = 0. We conjecture that the optimal condition for non-existence of

positive solutions of class C 2 (RN ) of (2.22) is

N (pqk − 1) < min {(q + 1)A, (p + 1)B, (k + 1)C} . (2.23)

This conjecture is supported by considerations related to the integral identities proved in

[20] (see also [25],[12]) and by the fact that for the special case a = b = c = 0 , k = q = 1
and (u, v, w) radial, the condition (2.23) is optimal. For other results concerning systems
of potential type of the form (2.6), see [17].

3. Problems with variational structure

In case the problem under consideration possesses more structure (for example, La-
grangian structure) we can obtain much more precise information on the non-existence
problem. The following result, which is a generalization of Theorem (3.2) of [12], is an
example in this direction.

Suppose that N ≥ 3, and p, q > 1 and let a, b: [0, ∞) → [0, ∞) be functions such that

8
 (i) a, b ∈ C [0, ∞) ∩ C 1 [0, ∞) and a(r), b(r) > 0 for r > 0.

(ii) (a(r)rγ )0 , (b(r)rγ )0 ≥ 0, r > 0, with

2(p + 1)(q + 1) − N (pq − 1)

γ=
p+q+2

(iii) limr→∞ a(r)rγ = limr→∞ b(r)rγ = ∞.

Theorem 3.1. Suppose that assumptions (i)–(iii) are satisfied. Then the problem

−∆u = a(|x|)v p , −∆v = b(|x|)uq in RN , (3.1)

has no positive radial solutions of class C 2 (RN ).

Proof. First, we consider the asymptotic behavior of (u, v). By writing (3.1) in radial
coordinates, we obtain for r > 0,

−(ru0 (r) + (N − 2)u(r))0 = ra(r)v p , −(rv 0 (r) + (N − 2)v(r))0 = rb(r)uq .

Using Remark 2.1 and (ii), it follows that for 0 < s ≤ t,

Zt
0 N −2 p
su (s) + (N − 2)u(s) ≥ a(s)s (v(s)s
γ
) ξ 1+p(2−N )−γ dξ
s

≥ a(s)sγ v p (s)s(N −2)p+1

1 h i
× t1+p(2−N )−γ − s1+p(2−N )−γ , (3.2)1
1 + p(2 − N ) − γ
and
Zt
0 N −2 q
sv (s) + (N − 2)v(s) ≥ b(s)s (u(s)s γ
) ξ 1+q(2−N )−γ
s

≥ b(s)sγ uq (s)s(N −2)q+1

1 h i
× t1+q(2−N )−γ − s1+q(2−N )−γ . (3.2)2
1 + q(2 − N ) − γ

Since u0 , v 0 < 0 and 1+p(2−N )−γ ,1+q(2−N )−γ < 0, by letting t → ∞ in (3.2)1 −(3.2)2
we see that for r ≥ r0 > 0 we have

u(r) ≥ Cr2 a(r)v p (r) , v(r) ≥ Cr2 b(r)uq (r).

9
Hence
u(r) ≤ C r−2(p+1)/(pq−1) a(r)−1/(pq−1) b(r)−p/(pq−1) ,

v(r) ≤ C r−2(q+1)/(pq−1) a(r)−q/(pq−1) b(r)−1/(pq−1) . (3.3)

An application of the identity (2.5) in [12] (see also [21],[20],[22],[25]), gives

ZR ZR
0
− v (r)v (r)a(r)r dr −
p N
uq (r)u0 (r)b(r)rN dr =
0 0

ZR
= (N − 2) u0 (r)v 0 (r)rN −1 dr + RN u0 (r)v 0 (r). (3.4)
0

We rewrite the left hand side of (3.4) as follows gives

ZR ZR
0
− v (r)v (r)a(r)r dr −
p N
uq (r)u0 (r)b(r)rN dr
0 0

ZR ZR
1 γ 0 N −γ N −γ
= v p+1
(r)(a(r)r ) r dr + v p+1 (r)a(r)rN −1 dr
p+1 p+1
0 0

ZR ZR
1 γ 0 N −γ N −γ
+ u q+1
(r)(b(r)r ) r dr + uq+1 (r)b(r)rN −1 dr
q+1 q+1
0 0

1 1
− up+1 (R)b(R)RN − v p+1 (R)a(R)RN . (3.5)
q+1 p+1
Multiplying (3.1)1 by v , and (3.1)2 by u, and integrating by parts, we get

ZR ZR
N −1 0 0 0 N −1
R u (R)v(R) + u (r)v (r)r dr = v p+1 (r)a(r)rN −1 dr (3.6)
0 0

ZR ZR
N −1 0 0 0 N −1
R u(R)v (R) + u (r)v (r)r dr = uq+1 (r)b(r)rN −1 dr . (3.7)
0 0

For any θ ∈ R, it follows from (3.6) and (3.7) that

ZR
u0 (r)v 0 (r)rN −1 dr = θRN −1 u0 (R)v(R) + (1 − θ)RN −1 u(R)v 0 (R)
0

10
 ZR ZR
+θ v p+1 (r)a(r)rN −1 dr + (1 − θ) uq+1 (r)b(r)rN −1 dr.
0 0

On the other hand, from (3.4) and (3.5) we obtain

ZR ZR
1 1
v p+1 (r)(a(r)rγ )0 rN −γ dr + uq+1 (r)(b(r)rγ )0 rN −γ dr
p+1 q+1
0 0

ZR
N −γ
= (θ(N − 2) − ) v p+1 (r)a(r)rN −1 dr
p+1
0

ZR
N −γ
+((1 − θ)(N − 2) − ) uq+1 (r)b(r)rN −1 dr
q+1
0

1 1
+ up+1 (R)b(R)RN + v p+1 (R)a(R)RN
q+1 p+1

+RN u0 (R)v 0 (R) + θ(N − 2)RN −1 u0 (R)v(R) + (1 − θ)(N − 2)RN −1 u(R)v 0 (R). (3.8)

N −γ
Now choosing θ = (N −2)(q+1) yields

N −γ N −γ
θ(N − 2) − = (1 − θ)(N − 2) − = 0.
p+1 q+1

Hence the right hand side of (3.8) reduces to

1 1
up+1 (R)b(R)RN + v p+1 (R)a(R)RN + u0 (R)v 0 (R)RN
q+1 p+1

+RN u0 (R)v 0 (R) + θ(N − 2)RN −1 u0 (R)v(R) + (1 − θ)(N − 2)RN −1 u(R)v 0 (R).

Since u0 (r), v 0 (r) < 0, ru0 (r) + (N − 2)u(r) and rv 0 (r) + (N − 2)v(r) ≥ 0, and 0 < θ < 1
there results

RN u0 (R)v 0 (R) + θ(N − 2)RN −1 u0 (R)v(R) + (1 − θ)(N − 2)RN −1 u(R)v 0 (R)

= θRN −1 u0 (R){Rv 0 (R) + (N − 2)v(R))} + (1 − θ){Ru0 (R) + (N − 2)u(R)} ≤ 0.

11
Finally by using (3.3) we obtain

uq+1 (R)b(R)RN ≤

≤ CR{N (pq−1)−2(p+1)(q+1)/(pq−1)} a(R)−(q+1)/pq−1 b(R)−(p+1)/pq−1 (3.9)

v p+1 (R)b(R)RN ≤

≤ CR{N (pq−1)−2(p+1)(q+1)/(pq−1)} a(R)−(p+1)/pq−1 b(R)−(q+1)/pq−1 . (3.10)

Hence, by (iii) the right hand side of (3.9) − (3.10) converges to 0 as R → ∞. Thus (3.8)
implies

Z∞ Z∞
1 γ 0 N −γ 1 0
v p+1 (r) (a(r)r ) r dr + uq+1 (r) (b(r)rγ ) rN −γ dr ≤ 0.
p+1 q+1
0 0

The positivity of (u, v) and (ii) now leads to a contradiction. This concludes the proof.

In the special case when

a(|x|) = |x|α , b(|x|) = |x|β

we obtain

Corollary 3.1. Let N ≥ 3, and p, q > 1. If

N (pq − 1) − 2(p + 1)(q + 1)

< min (α, β),
p+q+2

then the problem

−∆u = |x|α v p , −∆v = |x|β uq in RN ,

has no positive radial solutions of class C 2 (RN ).

It is possible to generalize Theorem 3.1 to some classes of higher order equations. A

natural result in this direction is contained in the following

Theorem 3.2. Let N > 2m, p > 1, and assume that

(i) a, b ∈ C [0, ∞) ∩ C 1 [0, ∞) and a(r), b(r) > 0 for r > 0.

12
 (N +2m)−p(N −2m)
(ii) (a(r)rδ )0 ≥ 0 for r > 0, with δ = 2

(iii) limr→∞ a(r)rδ = ∞.

Then the problem

m k
(−∆) u = a(|x|)up , (−∆) u ≥ 0 ,0 ≤ k ≤ m − 1 in RN

has no positive radial solutions of class C 2m (RN ).

The proof is similar to that of Theorem 3.1, making use as well of the following identity
( see [12]; here we recall an equivalent ”even” version of the identity, the corresponding
”odd” case can be obtained as in [12]): If u ∈ C 4m (BR ) (m ≥ 1), then
Z
∆2m u (x, Du) dx =
BR

Z X
2m−1 Z
4m − N 2m N − 2(2m − 2k) ∂
=( ) ∆ u u dx + ( ) (∆2m−1−k u) ∆k u ds
2 2 ∂n
BR k=0 ∂BR

X
m−1 Z
∂ ∂
+ { (∆k u) (x, D(∆2m−k−1 u) + (∆2m−k−1 u) (x, D(∆k u))
∂n ∂n
k=0 ∂B
R

−(D(∆2m−k−1 u), D(∆2m−k−1 u)) (x, n) }ds

X
m−1 Z Z
1
− k
∆ u∆ 2m−k
u ds − (∆m u)2 (x, n) ds
2
k=1 ∂B ∂BR
R

For the sake of brevity, the proof of Theorem 3.2 will be omitted.

The next result deals with a system which is not necessarily variational.

Theorem 3.3. Suppose that N ≥ 3 and that p,q,k,s satisfy

(i) 0 ≤ k ≤ 1 , 0≤s≤1 p, q > 1

n o
N (p+1)(q+1)+k(q+1)−sk (p+1)(q+1)+k(q+1)+s(p+1)
(ii) 2 < min pq−(1−s)(1−k) , pq+k(q+1)+s(p+1)−1 .

Then the problem

−∆u = uk v p , −∆v = uq v s in RN (3.11)

13
has no positive radial solutions of class C 2 (RN ).

Proof. As in the proof of Theorem 3.1, we obtain for r > 0

u(r) ≤ Cr−2(p+1−s)/(pq−(1−s)(1−k)) , v(r) ≤ Cr−2(q+1−k)/(pq−(1−s)(1−k)) . (3.12)

Multiplying (3.11)1 by v, and (3.11)2 by u, and integrating by parts on (0, R), we obtain

ZR ZR
−u0 (R)v(R)RN −1 + u0 (r)v 0 (r)rN −1 dr = uk (r)v p+1 (r)rN −1 dr, (3.13)
0 0

ZR ZR
−u(R)v 0 (R)RN −1 + u0 (r)v 0 (r)rN −1 dr = uq+1 (r)v s (r)rN −1 dr. (3.14)
0 0

We claim that
lim u0 (R)v(R)RN −1 = lim u(R)v 0 (R)RN −1 = 0. (3.15)
R→∞ R→∞

Indeed, by (2.7)

|u0 (R)v(R)RN −1 | |u(R)v 0 (R)RN −1 | ≤ (N − 2)u(R)v(R) (3.16)

and (3.15) then follows from (3.12) and the assumption (ii).

Using (ii) again, together with (3.12), we get

Z∞ Z∞
uk (r)v p+1 (r)rN −1 dr , uq+1 (r)v s (r)rN −1 dr < ∞.
0 0

Hence by (3.16) and (3.13) − (3.14),

Z∞ Z∞ Z∞
0 0 N −1 N −1
u (r)v (r)r dr = k
u (r)v p+1
(r)r dr = uq+1 (r)v s (r)rN −1 dr. (3.17)
0 0 0

Proceeding as in the proof of Theorem 3.1 there results

ZR ZR
0
− u (r)v (r)v (r)r dr −
k p N
uq (r)v s (r)u0 (r)rN dr
0 0

14
 ZR
= (N − 2) u0 (r)v 0 (r)rN −1 dr + RN u0 (R)v 0 (R) + RN −1 u0 (R)v(R) + RN −1 u(R)v 0 (R)
0

ZR
≤ (N − 2) u0 (r)v 0 (r)rN −1 dr. (3.18)
0

Integrating the left hand side of (3.18) by parts on (0, R) yields

ZR
1 N
− v p+1 (R)uk (R)RN + uk (r)v p+1 (r)rN −1 dr
p+1 p+1
0

ZR
k 1
+ uk−1 (r)u0 (r)v p+1 (r)rN dr − uq+1 (R)v s (R)RN
p+1 q+1
0

ZR ZR
N N −1 s
+ u q+1 s
(r)v (r)r dr + uq+1 (r)v 0 (r)v s−1 (r)rN dr
q+1 q+1
0 0

ZR
≤ (N − 2) u0 (r)v 0 (r)rN −1 dr. (3.19)
0

On the other hand, by (2.7),

 
 k ZR s
ZR 
(2 − N ) uk (r)v p+1 (r)rN −1 dr + uq+1 (r)v s (r)rN −1 dr
p + 1 q+1 
0 0

ZR ZR
k s
≤ uk−1 (r)v p+1 (r)u0 (r)rN dr + uq+1 (r)v s−1 (r)v 0 (r)rN dr.
p+1 q+1
0 0

Hence from (3.18) − (3.19),

1 1
− v p+1 (R)uk (R)RN − uq+1 (R)v s (R)RN
p+1 q+1

ZR ZR
N + k(2 − N ) N −1 N + s(2 − N )
+ k
u (r)v p+1
(r)r dr + uq+1 (r)v s (r)rN −1 dr
p+1 q+1
0 0

ZR
≤ (N − 2) u0 (r)v 0 (r)rN −1 dr. (3.20)
0

15
Passing to the limit R → ∞ in (3.20) and using (3.12) and (3.17), we finally obtain

½ ¾ Z∞
k s 1 1
0≤ (N − 2)(1 + + ) − N( + ) uk (r)v p+1 (r)rN −1 dr,
p+1 q+1 p+1 q+1
0

which contradicts (ii) and concludes the proof.

For related results concerning the parabolic problem associated with (3.11), see [3],[11].

4. Some remarks on positive solutions of problems in bounded domains

In this section we shall establish some non-existence results for positive solutions of
semilinear systems of the form

−∆u = f (u, v) , −∆v = g(u, v) in Ω (4.1)

u=v=0 on ∂Ω, (4.2)

where Ω ⊂ RN , N ≥ 3, is a smooth bounded domain and f , g are given smooth functions.

Very general results on the non-existence of positive solutions for semilinear elliptic
systems with Lagrangian structure can be obtained by using the variational identities
proved by Pucci and Serrin [20] and van de Vorst [25].
In this section we shall confine our discussion to systems of the form (4.1) − (4.2)
(not necessarily of Lagrangian type) and later indicate some possible generalizations. Let
Ω = BR (0) ⊂ RN , R > 0. Suppose that

(i) f, g :R2 → R are of class C 1 ,

(ii) f (0, 0) = g(0, 0) = 0 ,

(iii) fy , gx ≥ 0 for each (x, y) ∈ R+ 2 ,

(iv) there exist α ∈ R such that for every (x, y) ∈ R+ 2 we have

N {F (x, y) + G(x, y)} ≤ (N − 2) {αFy (x, y) y + (1 − α)Gx (x, y) x} ,

where
Zy Zx
F (x, y) = f (x, s) ds and G(x, y) = g(s, y) ds, (x, y) ∈ R2 .
0 0

16
Theorem 4.1. If (i)-(iv) are satisfied , then the problem

−∆u = f (u, v) , −∆v = g(u, v) in Ω (4.2)

u=v=0 on ∂Ω (4.3)

has no positive solutions of class C 2 (Ω) ∩ C 1 (Ω) .

Proof. Suppose that (u, v) is a positive solution of (4.2) − (4.3). Then (u, v) satisfies

∂F ∂G
−∆u = (u, v) , −∆v = (u, v) in Ω (4.4)
∂v ∂u

u = v = 0 on ∂Ω (4.5)

Since (4.2) − (4.3) is quasi-monotone, i.e. ,

Fyy (x, y) = fy (x, y) ≥ 0 , Gxx (x, y) = gx (x, y) ≥ 0 for every (x, y) ∈ R2+ ,

and (u, v) is a positive solution, then by applying the main result of Troy [24], we see that
(u, v) is radially symmetric and u0 (r), v 0 (r) < 0, 0 < r < R. Using the identity (2.5) of [12]
applied to (4.4) − (4.5), and the divergence theorem, we get
Z Z
∂u ∂v
(N − 2) (Du, Dv) dx + (x, n) ds
BR ∂BR ∂n ∂n
½Z Z ¾
=N F (u, v) dx + G(u, v) dx
BR BR
Z Z
∂F ∂G
+ (u, v)(x, Dv) dx + (u, v)(x, Du) dx. (4.6)
BR ∂v BR ∂u
Multiplying (4.4)1 by v and (4.4)2 by u and integrating by parts, it follows that
Z Z Z
∂F ∂G
(Du, Dv) dx = (u, v)v dx = (u, v) u dx .
BR BR ∂v BR ∂u

Hence from (4.6) we obtain

Z Z
∂F ∂G
α(N − 2) (u, v)v dx + (1 − α)(N − 2) (u, v)u dx
BR ∂v BR ∂u
½Z Z ¾ Z
∂u ∂v
−N F (u, v) dx + G(u, v) dx =− (x, n) ds
BR BR ∂BR ∂n ∂n

17
 Z Z
∂F ∂G
+ (u, v) (x, Dv) dx + (u, v) (x, Du) dx. (4.7)
BR ∂v BR ∂u

By the maximum principle the first term of the right hand side of (4.7) is negative; similarly,
the second is negative by Troy’s result and the assumption (iii) (quasi-monotonicity of the
system). This contradicts assumption (iv) and concludes the proof.

An application of Theorem 4.1 gives the following

Proposition 4.1. Assume that Ω = BR (0) ⊂ RN , R > 0. If p, q, k, s > 1 and

1 1 N −2
+ ≤ , (4.9)
p+1 q+1 N
then the problem
−∆u = uk v p , −∆v = uq v s in Ω, (4.10)

u=v=0 on ∂Ω (4.11)

has no positive solutions of class C 2 (Ω) ∩ C 1 (Ω).

The proof is obtained by a direct verification of the hypotheses (i) − (iv).

We conclude with the following generalization of Theorem 3.1 of [12].

Theorem 4.2. Let Ω ⊂ RN be a bounded, smooth domain, starshaped with respect to

0 ∈ RN . Let F ∈ C 1 (R2k ; R), k ≥ 1, and suppose that
(i) F(0) = 0
P2k
(ii) there exists (a1 , a2 , . . . , a2k ) ∈ R2k such that i=1 ai = 1
and
X
2k
∂F
N F (x) ≤ (N − 2) ai (x) xi for all x ∈ R2k .
i
∂xi

Then the problem

∂F ∂F
−∆u1 = (u1 , u2 , . . . , u2k ) , −∆u2 = (u1 , u2 , . . . , u2k ), . . .
∂u2 ∂u1
∂F
−∆u2k = (u1 , u2 , . . . , u2k ) in Ω
∂u2k−1
with
u1 = u2 = · · · u2k = 0 on ∂Ω
¡ ¢2k
has no positive solutions of class C2 (Ω) ∩ C1 (Ω) .

18
The proof is similar to that of Theorem 3.1.

Acknowledgments. A great part of this work was done while I was visiting the University
of Leiden in February 1992. I wish to thank Professor L. A. Peletier and Professor C. J.
van Duijn for their kind invitation and the generous hospitality during my stay. I am very
grateful to Professor James Serrin for suggesting some improvements in the first version
of this paper. This work is partially supported by MURST 60% .

REFERENCES

[1] Ph. Clément , R. Manasevich & E. Mitidieri , Positive Solution for a Quasilinear Sys-
tem via Blow-up, Comm. Partial Diff. Equations,Vol. 18 , No.12,(1993) ,2071-2106.
N +2
[2] W. - Y. Ding & W.-M. Ni, On the Elliptic Equation ∆u + Ku N −2 = 0 and Related
Topics, Duke Math. J.,52 ,(1985), 486-506.
Local
[3] M.Behaviour
Escobedo of& Positive Solutions
H. A. Levine, of Nonlinear
Critical Blowup and Global Existence Numbers for a
Weakly Coupled System of Reaction-diffusion Equations,preprint.
[4] B. Gidas , Symmetry Properties and Isolated Singularities of Positive Solutions of Non-
linear Elliptic Equations, in ”Nonlinear Differential Equations in Engineering and Applied
Sciences”, R. L. Sternberg , ed.,Marcel Dekker, New York,(1979), 255-273.
[5] B. Gidas & J. Spruck , Global and
Elliptic Equations, Comm. Pure Appl. Math., 34,(1981), 525-598.
[6] B. Gidas & J. Spruck , A Priori Bounds for Positive Solutions of Nonlinear Elliptic
Equations, Comm. Partial Diff. Equations,6,(1981),883-901.
[7] Y. Giga, A Bound for Global Solutions of Semilinear heat Equations, Comm. Math.
Phys. 103,(1986), 414-421.
[8] Y. Giga & R Kohn, Characterizing Blowup using Similarity Variables, Indiana Univ.
Math. J . 36,(1987),1-40.
[9] T. Kusano & M. Naito, Positive Radial Solutions of Superlinear Elliptic Equations,
Hiroshima Math. J., 16,(1986),361-366.
[10] T.Kusano & M. Naito, Oscillation Theory of Entire Solutions of Second Order Super-
linear Elliptic Equations, Funckcial Ekvac., 30 ,(1987), 269-282.
[11] H. A. Levine , A Fujita Type Global Existence-global Nonexistence Theorem for a
Weakly Coupled System of Reaction-diffusion Equations, preprint.

19
[12] E. Mitidieri, A Rellich Type Identity and Applications, Comm. Partial Diff. Equa-
tions, 18 (1&2),(1993), 125-151.
[13] W.-M. Ni & J.Serrin, Non-existence Theorems for Quasilinear Partial Differential
Equations, Rend. Circ. Mat. Palermo, suppl. 8 ,(1985), 171-185.
[14] W.-M. Ni & J.Serrin, Existence and non-existence Theorems for Ground States for
Quasilinear Partial differential Equations, Atti Convegni Lincei, 77,(1985),231-257.
[15] W.-M. Ni& J.Serrin, Nonexistence Theorems for Singular Solutions of Quasilinear
Partial Differential Equations, Comm. Pure Appl. Math., 39 ,(1986), 379-399.
[16] E. Z. Noussair & C. Swanson, Positive Decaying Entire Solutions of Superlinear El-
liptic Equations, Indiana Univ. Math.J., 36,N.3,(1987),651-657.
r q&(RCN Swanson,
[17] E. Z. Noussair & C. Swanson, An L )-Theory of Subcritical Semilinear Elliptic
Problems,J.Diff. Equations, 84,(1990), 52-61.
[18] E. Z. Noussai Properties of Potential Systems in RN , J. Diff. Equa-
tions,95,(1992), 1-19.
[19] W.-M. Ni, On the Elliptic Equation ∆u + K(x)uN +2/N −2 = 0 its generalization, and
applications in geometry, Indiana Univ. Math. J., 31,(1982), 493-529.
[20] P. Pucci & J. Serrin, A General Variational Identity, Indiana Univ. Math. J.,
35,(1986), 681-703.
[21] S.- I. Pohozaev, Eigenfunctions of the equation ∆u + λf (u) = 0, Soviet Math. Dokl.,
5,(1965), 1408-1411.
[22] F. Rellich, Darstellung der Eigenwerte von ∆u+λu = 0 durch ein Randintegral, Math.
Zeit., 46,(1940), 635-636.
[23] F. Treves, Basic Linear Partial Differential Equations , Academic Press, 1975.
[24] W. C. Troy, Symmetry Properties in Systems of Semilinear Elliptic Equations, Journal
of Diff. Equations, 42, (1981),400-413.
[25] R. C. A. M. van der Vorst , Variational Identities and Applications to Differential
Systems, Arch. Rational Mech. Anal.,116,(1991) 375-398.

20