Vol. 62, No.

3 DUKE MATHEMATICAL JOURNAL (C) April 1991

ON THE STRUCTURE OF THE CONFORMAL GAUSSIAN

CURVATURE EQUATION ON 2
KUO-SHUNG CHENG AND WEI-MING NI

1. Introduction. The purpose of this paper is to investigate the structure of the

set of all solutions of the equation

(1.1) Au + Ke 2u 0

in 2, where A 2= c2/Ox and K is a given nonpositive smooth function on 2.

We shall establish the existence of a unique maximal solution of (1.1), and in some
important cases we are able to completely classify all the solutions of (1.1). (Through-
out this paper a solution of an equation shall always mean a solution in the classical
sense.)
Equation (1.1) arises in Riemannian geometry. Let (M, 9) be a Riemannian
manifold of dimension 2 and K be a given function on M. The following question
has been raised: can we find a new metric 91 on M such that K is the Gaussian
curvature of g and gl is conformal to g (that is, gl qg for some positive smooth
function q9 on M)? If we write q9 e 2u, then this is equivalent to the problem of
solving the elliptic equation
(1.2) Aou- k + Ke 2"
0
on M where A0 and k are the Laplace-Beltrami operator and the Gaussian curvature
on M in the g-metric, respectively. In case M is compact, equation (1.2) has been
considered by many authors, and we refer the reader to the recent monograph by
Kazdan [-K] for details and references, in case M is complete and noncompact, the
first natural case seems to be M 2 and 9 is the usual Euclidean metric. In this
case, equation (1.2) reduces to equation (1.1).
The first result concerning equation (1.1) seems due to L. V. Ahl’fors [A] in 1938.
He showed that in case K (or any negative constant), equation (1.1) does not
have any solution in the entire space E2. His result was later improved and extended
by Wittich [W], Keller [Ke], Osserman [Os], Sattinger IS 1], Oleinik [O], Ni IN 1],
McOwen I-M], and Cheng and Lin [CL]. These results basically contain the
following
THEOREM A. If K < 0 in 2 and K < -Clx[ -2 near 0, then equation (1.1) does
not possess any solution on 2.

Received 2 April 1990.

First author supported in part by National Research Council, R.O.C.
Second author supported in part by NSF Grants DMS 8601246 and DMS 8801587.

721
722 CHENG AND NI

(Throughout this paper the letter C will always denote a positive generic constant
which may vary from line to line.)
The first existence result concerning equation (1.1) seems due to Ni IN1]. He
established the following
THEOREM B. If K 0 and K < 0 in R 2 with K(x) > -Clxl near for some
-
constant > 2, then equation (1.1) possesses infinitely many solutions in 2. More
precisely, for every sufficiently small > O, there exists a solution u of (1.1) in 2 such
that
(1.3) u(x) g log lxl + O(1) near

Observe that as far as existence is concerned, Theorem B already complements

Theorem A. Nevertheless, Theorem B was improved by McOwen [M] by using a
different method, namely, the weighted Sobolev space approach, as follows.
THEOREM C. Under the same hypotheses of Theorem B and if K(O) < O, then for
every such that 0 < < (l- 2)/2, there exists a solution u of (1.1) on 2 and a
constant uoo such that
(1.4)
for every
near , then
u(x) + uoo + O(Ixl ) near
log Ixl
such that V > max{-1, 2- + 2}. If in addition that K < -Clxl
equation (1.1) admits no solution in 2 satisfying (1.3) with > (l
It turns out that Theorem C is still not sharp, even in the special case K -Ixl
near where > 2. As we shall soon see, one of the main results in this present
--
2)/2.

paper gives complete knowledge of all possible solutions of (1.1) on 2 in this case.
(In this paper we use the notation "f # near
"
to denote that "there exist two
positive constants C, C2 such that Cf > g > C2f near ".)
What happens if K < 0 and K -o 0 as x -o o faster than any power of Ixl? It was
already noticed by Ni I-N1] that if K < 0 in 2 and K -exp(-21xl 2) near
near ), equation (1.1) actually has a solution U Ixl 2 at .
then, besides the solutions given by Theorems B and C (which have logarithmic growth
In fact, to understand
the relation between these two kinds of solutions was the motivation of our present
study. We shall show that the solution U is the maximal solution of (1.1) in 2 (see
Remark 3.8 in Section 3 below).
Our main results are as follows. (The function K will be assumed to be locally
Holder continuous in 2 throughout the entire paper.)
THEOREM I. Suppose that
(i) K < 0 in 2 and there exists a sequence of bounded smooth domains {fi} such that
2 fi, i fi+ and K < 0 on fi,

1, 2,..., and
(ii) equation (1.1) possesses a solution v on g2.
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON 2 723

Then the function

U(x) sup(u(x)lu is a solution of (1.1) on [2}

is well defined everywhere in 2 and is a solution of (1.1) on 2.
Note that (i) in Theorem i holds if K < 0 in 2 and K < 0 near The function .
U given by (1.5) is the maximal solution of (1.1).
THEOREM II. Suppose that K < 0 in 2, and that K -Ixl near for some
> 2. Then the followin9 conclusions hold. -
(i) For each e (0, (l 2)/2), (1.1) possesses a unique solution u satisfying (1.3).
(ii) Let u be a solution of (1.1) on 2. Then either u =- U where U is given by (1.5)
or u =- u for some (0, (l 2)/2) where u is given by (i) above.
(iii) If ((l 2)/2) > > fl > 0, then u > ut in ,2. Furthermore, the asymptotic be-
havior of U near is given by

1-2
(1.6) U(x)
2
log Ixl- log log Ixl + O(1) near

THEOREM III. Suppose that K 0 and K < 0 in 2, and that K has compact
2. Then
support in
(i) for every harmonic function h and every constant > 0, there exists a unique
solution u of (1.1) with

(1.7) u(x) h(x) + log Ixl + O(1) near

(ii) for every solution u of (1.1) on 2, there exist a harmonic function h and a positive
constant such that (1.7) holds.
Theorem III says that in case K < 0, K 0 and K has compact support on E2,
the "solution set" of (1.1) may be indexed by had where h is harmonic and > 0,

-
and is therefore completely understood. This in particular indicates that in this case
(1.1) does not possess a maximal solution on the entire E2. In case K < 0 in E2 with
K -Ixl near for some > 2, Theorem II asserts that the "solution set" of
(1.1) can be indexed by (0, (l 2)/2) with only one exception, namely the maximal
solution U. Thus in this case the solution set is also completely understood.
Moreover, as a consequence of this result and some further arguments, we are able
to obtain the following curious conclusion concerning the symmetry of solutions.
COROLLARY IV. Under the hypotheses of Theorem II, if in addition we assume
that K is radially symmetric, then all solutions of (1.1) on 2 are radially symmetric.
Remark. Note that no other conditions on K (such as the monotonicity assump-
tion) are imposed in Corollary IV. (Compare, e.g., results in [GNN] and [LN].)
724 CHENG AND NI

Furthermore, combining Corollary IV and Theorem II, we see that Theorem C is

not sharp even for radial K’s.
Our proofs are mainly built on comparison arguments based on the maximum
principle combined with careful estimates of maximal solutions. In a number of
cases below, it is clear that we could use the weighted Sobolev spaces method
instead. However, the maximum principle approach adopted here not only seems
simpler but also gives a unified treatment to our results. Moreover, the techniques
and methods used in this paper also apply to the conformal scalar curvature
equation
Au + Ku(n+2)/(n-2) --0
in R", n > 3. We refer the interested readers to [CN] for further details.
The organization of this paper is as follows. In Section 2, we shall establish
Theorem I as well as some properties of maximal solutions. Various asymptotic
behaviors of different maximal solutions are investigated in Section 3 and the proofs
of Theorems II, III and Corollary IV are included in Section 4. Finally, we make a
few concluding remarks in Section 5.

2. Existence and general properties of maximal solutions, in this section we

shall establish the existence of maximal solutions of (1.1) and some general proper-
ties, including the monotonicity property, of maximal solutions. We shall assume
throughout this section that equation (1.1) possesses a solution on R2, and we shall
denote such an arbitrary but fixed solution of (1.1) on 2 by v.
To prove Theorem I, we shall need the following lemmas. The first one follows
from direct calculation.
LEMMA 2.1. The function
Wa, b,z(X) b log(a 2 --Ix zl 2)
satisfies the equation
(2.2) Aw 4a2e-2be 2w 0 in Ba(z),

where Ba(z) is the ball of radius a centered at z.

The key ingredient in the proof of Theorem I is the following result which is also
of independent interest.
LEMMA 2.3. There exists at least one solution of the boundary value problem
,
(2.4) Au
u(x)
+ Ke
-
2u
0 in
as x O (i.e., as dist(x, cOf)
-
where f is a bounded smooth domain in 2 with K < 0 on Of. Furthermore, for any
solution u of (2.4) we have u > v in t).
0),
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON 2 725

Proof. For each positive integer N such that N > supnv, we first solve the
boundary value problem

(2.5)
+ Ke 2u 0 in f,
/.u N on cf,

and denote the solution by us. This may be done by using the standard barrier
method. For, the constant N is a supersolution of (2.5) while v serves as a subsolution.
Thus (2.5) possesses a solution us > v. (See, for instance, Theorem 2.3.1 in [$2].)
Moreover, by the maximum principle (see, e.g., Theorem 3.5 in [GT]) it follows that
(2.5) can have at most one solution since K < 0.
It is easy to see that us+l > us in f (again, by the maximum principle, see the
proof of (2.6) below); thus {us} is a monotonically increasing sequence as N
Since K < 0 on c3f, we have K < 0 on c3f, for e positive and sufficiently small
.
where
f {x fldist(x, c3f) > e}.

collection of balls B,,(Xk)

that

(2.6)
for all N and for each k 1,
- ,
Since Of is compact in f, there exist two positive constants a, 6 and a finite
k 1, 2,..., n such that cf Ba(xl) w w Ba(x,)
and K < -6 on Ba(Xk) for each k. Choosing b so large that 4aZe -2b < 6, we claim

Wa,b, xk > US in Ba(Xk)

n. This follows from the usual maximum principle

as follows. In Ba(Xk), we define q9 us w where w Wa, b, xk. Then

Atp --4a2e-2be 2w- Ke 2uN

where c(x) K(e 2" e2W)/(us w) < O. Since q9

-
--(4a2e -2b + K)e 2w- K(e 2" -e 2w) > -c(x)cp
on c3Ba(Xk), in particular
q9 < 0 near OBa(Xk). Therefore, the maximum principle implies that cp cannot assume
a nonnegative maximum inside Ba(Xk) and thus (2.6) holds.
From (2.6) it follows that the sequence {uslon} is bounded from above. On the
other hand, since Aus > 0 in f, we deduce from the maximum principle that the
sequence {usln,} is also bounded from above. Therefore, the monotone sequence
(usln,} is uniformly bounded in f, and standard arguments involving elliptic
regularity estimates (see, e.g., the proof of Theorem 2.10 in IN2-1) imply that the

as N .
sequence {us} converges monotonically to a solution u of the equation (1.1) in
Since e can be arbitrarily small, we conclude that the sequence {us}
actually converges monotonically to a solution u of (1.1) in f. The boundary
condition u(x) as dist(x, Of) 0 follows from the fact that u > us in f for
every N. Thus u is a solution of (2.4). Our last assertion that any solution of (2.4)
726 CnEN AND NI

must be bounded below by v follows from the usual maximum principle as in the
proof of (2.6) above. Q.E.D.
We are now ready to prove Theorem i.
Proof of Theorem i. For each i, let u be a solution of (2.4) with fl fl s. Once
again standard arguments involving the maximum principle (similar to the proof
of(2.6) above) show that u > u+l in fs, Thus for any fixed bounded domain f, the
sequence {u[i is sufficiently large} is monotonically decreasing on f. Since us > v
in fs for all i, we conclude from elliptic regularity estimates that the sequence
converges uniformly to a solution fi of(1.1) on fl. Since the convergence is monotone,

solution of(1.1) on
< by
U the
,
is a solution of (1.1) on the entire space 2, and > v. Since v is an arbitrary
we have > U where U is given by (1.5). On the other hand,
definition of U. Thus U and our proof is complete. Q.E.D.
Remark 2.7. It is clear that with obvious modifications in the above proof
Theorem I may be extended to the more general case where 2 is replaced by an
arbitrary open set fl in its statement.
An important and useful property of maximal solutions is the following mono-
tonicity property.
THEOREM 2.8. Suppose that the hypotheses of Theorem I hold and that 0 > K >
on 2. Let U (i resp.) be the maximal solution of Au + Ke 2u 0 (At/+ ie 2u 0
resp.) in R 2. Then U > l.
Proof. From the proof of Theorem I we see that ui -* U as -, go where us is a
solution of (2.4) with f fs guaranteed by Lemma 2.3. Thus it suffices to show
that us > t in f for every 1, 2, To this end, we set tO t u in fs. Then
0 AtO -I- Ie 20- Ke 2u’ < AtO q- c(x)tO
where c(x) =- g(e 20 e2U’)/(O ui) < O, and tO < 0 near cf. Therefore, tO < 0 in
fs by the maximum principle. Q.E.D.
Remark 2.9. Our next observation is that if the equation (1.1) is invariant under
a transformation T, then from the definition of the maximal solution it follows that
the maximal solution of (1.1) (if it exists) is also invariant under the same transforma-
tion T. Thus, in particular, if K(x) is symmetric with respect-to a point or a line,
then the corresponding maximal solution of (1.1) in 2 is also symmetric with
respect to the same point or the same line. A useful consequence is that if K is
radially symmetric, then so is the corresponding maximal solution U.
The last property of maximal solutions we include here is the following
PIOr’OSITION 2.10. Suppose that K < 0 in 2 and that U is the maximal solution
of (1.1) given by (1.5) which is well defined in R 2. Let u be an arbitrary solution of
(1.1) in 2. Then either u < U everywhere in 2 Or U U in 2.
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON 2 727

Proof. (The proof is standard; we include it here for the sake of completeness.)
It is clear that u < U in E2. Suppose that U(Xo) U(xo) for some Xo E2. Setting
q9 U U, c(x) K(e 2u e2V)/(u U), we have c(x) < 0 and Aq9 + c(x)q 0 in
E.
R Let B be an arbitrary ball containing Xo. We then have maxao 0 since
O(Xo) 0. Thus it follows from the strong maximum principle that q9 _= 0 in B, hence
in 2, since B can be arbitrarily large. This proves our assertion. Q.E.D.
3. Asymptotic behavior of maximal solutions. We shall study the asymptotic
behavior of maximal solutions of (1.1) for various K’s in this section. In particular,
we shall establish (1.6) and prove that, in case K decays exponentially near ,the
corresponding

pointed out in
maximal

the
solution

introduction,
has

there
polynomial growth.
First, we assume that K < 0 in 2 and that K -exp(-2 Ixl 2) near
are two kinds of solutions of (1.1) in
.
2:
As we
one
is a one-parameter family of solutions u(x) log Ixl + O(1) near for every

behavior is u Ixl 2 near

discussion.) Our result
.
positive ; the other consists of at least one solution u of which the asymptotic

here is
(See Proposition 4.21 and Section 5 for some further
the following

. THEOREM 3.1. Suppose that K < 0 in 2 and that K

Then the maximal solution U of (1.1) in 2 satisfies
-exp(-2 Ix] 2) near x

(3.2) U(x) Ixl 2 + O(1) near x .

Proof. We define g(r) maxlxl=rK(x). Then g also satisfies the hypotheses of
Theorem I (since Theorem B guarantees condition (ii) in Theorem I). Letting t] be
the maximal solution of
Au + Ie 2u 0

in [2, we have, by Theorem 2.8, that t > U, and is radial in view of Remark 2.9.
Set w(r) IT(r) rE; then
(3.3) Aw (-g)e2r2e2W 4

lim sup,-,oo w(r) .

in 2. We assert that w is bounded above in 2. Suppose for contradiction that
Then there exists R > 0 such that (-ge2g2)e2WtR) 4 > 0
and w’(R) > O, and (3.3) implies that (rw,) > 0 at r R. Integrating from R to r
(r > R), we obtain that two(r) > Rwr(R) > O, i.e., w must be monotonically increasing
to in jR, ). Thus it follows from (3.3) that there exist two constants/ > 1 and
C > 0 such that

’Aw >/Ce 2w in r >/,

(3.4)
,w,(/) > 0.
Now, standard arguments lead to a contradiction. For, setting s log r, we have
728 CHENG AND NI

Wss > Cr2e 2w(s) and (3.4) becomes

Ce 2w in s > S,
(3.5)
.ws(S) > O,
where S log/. Multiplying both sides of (3.5) by ws and integrating, we obtain
w2 > Ce 2’, i.e., we > C in IS, ). Integrating again, we have

as s ,
e -(s) e -()
ff we

a contradiction, and our assertion is proved.

ds > C(s
while the left-hand side tends to exp(-w(S)) < as s
S)
-
This gives .
It remains to show that U- Ixl 2 is bounded below. This may be done by a
comparison argument using Theorem 2.8. Observe first that u(x) Ixl 2 C is a
solution of

(3.6) Au (4e2Ce-21xl’-)e2U 0

in R 2. Suppose that U* is the maximal solution of (3.6) in I 2. Then for sufficiently

large C we have K >-4e2Ce -21’12 in R2, and thus Theorem 2.8 implies that
U > U* > Ixl 2 C. This establishes (3.2). Q.E.D.
Remark 3.7. In fact, the solution u Ixl 2 is the maximal solution of the equation
Au 4e-21xl2e2U 0
in 2. In Section 4 we shall include a brief discussion (Proposition 4.21) which
classifies all the radial solutions of the above equation in 2.
Remark 3.8. in case K < 0 in 2 and K -exp(- 2 Ixl 2) near it was noticed ,
in IN1] that, besides the solutions given by Theorems B and C (which have

.
logarithmic growth near ), equation (1.1) actually has a solution U in I 2 with
U Ixl 2 0(1) near It then follows from Theorems 3.1 and 4.1 below that this
solution U must be the maximal solution of (1.1).
Our next goal is to establish (1.6) which not only will be useful in the future but
also seems to have some independent interest. (For instance, it improves a previous
nonexistence result in [M].)
THEOREM 3.9. Suppose that K < 0 in 2 and that K -Ixl near for some

(3.10) U(x)
2
log Ixl log log Ixl near
-
> 2. Then the maximal solution U of (1.1) has the following asymptotic behavior
1-2
+ O(1)
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON R2 729

Proof. We first consider the following special case. Suppose that/ is radial,
/ < 0in [0, or) and/(r) -Cr -z in JR, ), where R, C are two positive constants,
and let t be the maximal solution of Au +/e 2u 0 in 2. Then is radial by
Remark 2.9. Setting xl r and

w(x) O(x) -..2 log xl

2

in xl > R, we see immediately that w is radial and

Aw C Ixl-2e 2 0

in Ixl > R. Making the change of variables s log r, we have

Wss Ce2W
in s > log R. Multiplying this equation by w and integrating the resulting equation
from s to s l(> s), we obtain

(3.11) w2 (sl) w2 (s) + C(e 2wt,) e2Wt)).

Since
.
ws > 0, w is monotonically increasing and thus has a limit at say, w 2 ,
as s From the fact that (3.5) has no solution in [S, ), we conclude that w < 0
near or, and thus 2 < 0. If 2 < 0 then w < 2 near (since w > 0), and it follows
that w(s) < C + 2s near s c. This implies that

+ 2)log r +C

.
-
near which contradicts Theorem C and the definition of 0. Hence 2 0. Since
w < 0 near ,
it is bounded from above near The nonexistence part in Theorem

monotone near .
C then implies that w is not bounded below. Thus w
Now, letting s
as s since w is
in (3.11), we have ws2 Ce 2w(s) near i.e., ,
we C

for s sufficiently large, say, s > So. Integrating the last equation from So to s, we
obtain

-
e -wt) e -wt) + C(s So).
Hence for large s we have
C
2Cs e -’) s,
730 CHENG AND NI

-
for s large. This proves that/ satisfies (3.10).
In the general case K -Ixl near o, it is easy to see that there exist two smooth
radial functions K1, K2 with the following properties.
(i) 0 > K1 > K > K 2 in [2.
(ii) K l(r) -C r -t near and K2(r -C2 r-t near oo for some positive con-
stants C, C2.
Since by what we have just proved the maximal solution U of the equation
Au + Ke 2u 0, 1, 2, in 2 must satisfy (3.10), the maximal solution U of (1.1)
must also satisfy (3.10) by the monotonicity property Theorem 2.8. Q.E.D.
4. Complete classifications of all solutions. The main purpose of this section is
to prove Theorems II, III and Corollary IV (stated in the Introduction). We shall
also include some partial results when K in (1.1) decays exponentially near .
Proof of Theorem 11. First, note that the first half of part (iii) follows from
standard comparison arguments using the maximum principle which is similar to
the proof of (2.6) above. Also, note that the asymptotic behavior (1.6) of U was
already established in Theorem 3.9. Moreover, the existence part of (i) is contained
in Theorem C, while the uniqueness part of (i) follows from the following general
theorem.
THEOREM 4.1. Let u, v be two solutions of (1.1) in ,2 with lu vl bounded in 2.
Suppose that K 0 and K < 0 in 2. Then u v in 2.
Proof. We define w u v. Then
A(w 2) 2wAw + 2 IDwl 2 > 2wAw
-2K(u v)2[_(e 2u e-V)/(u v)] > 0,
i.e., w 2 is bounded and subharmonic in E2. Thus w 2 is constant by Liouville’s
theorem. This constant can only be zero since K 0; hence u v. Q.E.D.
It remains to prove part (ii) of Theorem II. Suppose that u U. Then u < U in
,,
E2 by Proposition 2.10. Setting ff U u and letting ff be the spherical mean of
,
we have A A, > 0 in 2, i.e., rris nondecreasing in r > 0. Since K < 0 near
o, A, > 0 near o. Thus is not constant, and we obtain easily that ,(r) > e log r
near r o for some e > 0. That is,

(4.2) g(r)< U(r)-elogr< -e logr

2’
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON R2 731

for r large by Theorem 3.9 where

Since u is subharmonic and
, U are the spherical means of u, U respectively.

u*(r) max u(x)

Ixl
< max
Ixl
U(x) O(log r)

near r o (again, by Theorem 3.9), using a result of Hayman’s (Theorem 2, p. 77

and Remark (ii), p. 75, in [H-I), we conclude from (4.2) that there exist a sequence
rn o and a constant C* such that
(4.3) u*(rn) <
/-2
2 )
e log r, + C*
Next we claim that u* is subharmonic in 2. Let x be an arbitrary point in R2.
Then there exists a point Xo with xol r xl and U(Xo) u*(r). Since u is sub-
harmonic and u* is radial,

u*(Ixl)-u(xo)< ls,u < IS’--]l

for all R < r, where SR and S are the circles of radius R centered at Xo and x,
respectively. Thus u* is subharmonic in 2.
Consider the function

-
1-2
0(Ixl)- u*(Ixl)- log xl- C*
2

in x 0. it is subharmonic in 2\{0 and 0(r,) < 0 for all n. Therefore, q9 < 0 in

{x 21rx < Ixl < r) by the usual maximum principle since q also satisfies the
mean-value inequality. Letting n o, we have q9 < 0 in Ixl > rx, i.e.,

(4.4) u(x) <

/-2
2 )
e log xl + c
for Ixl
Now we can define

(4.5) v(x) [-K(y)e 2"ty] log Ix Yl dy

for x 2. Since 0 < -K(y)e 2" < ClY1-2tx+ at o by (4.4), (4.5) is well defined.
By dividing 2 into three regions
{Y e NIlyl > 2 Ixl},
(y 2121xl > lyl > Ixl/2}
732 CHENG AND NI

and
{y e ellyl < Ixl/2},
we may estimate (4.5) and obtain

(4.6) v(x) [-K(y)e 2"(’)] dy log ]xl + O(1)

near o. (The argument here is somewhat standard and is thus omitted.) From (4.5)
we also have
Av Ke 2u in R 2.

Therefore, A(u v) 0 and u v is harmonic in R2. On the other hand, combining

(4.4) and (4.6), obtain
we

(4.7) (u v)(x) < C log xl

near oo for some positive constant C. This implies that u v + Constant, and near
oe we have, by (4.6),

(4.8) u(x) 0 log Ixl + O(1)

for some > 0. We conclude from Theorem C that e < (l 2)/2 and u satisfies (1.3).
This finishes the proof of Theorem II. Q.E.D.
Remark 4.9. Hayman’s theorem (used above to obtain (4.3)) is essential to our
argument. In fact, a close examination of the proof of Hayman’s theorem shows
that we actually have

u*(r) (r) + o(1)

as r -. oo ((2.10), p. 79 in [HI). This already implies (4.4) which is really what we
need in the above proof. A number of extensions and generalizations of Hayman’s
theorem could also be used here instead, for instance, Theorem 12 in [B] (p. 491)
and Theorem 4 in ILl (p. 172), as well as results in many deep papers on this subject
including [EHH] and others.
Proof of Theorem III. We first prove part (i). Let h be a given harmonic function
in 2 and let e be a given positive number. Set u h + w; then equation (1.1)
becomes

(4.10) Aw q- Re 2w --0
in 2, where/ -= Ke 2h 0 and is still of compact support. Theorem 2.2 in [KN]
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON R2 733

(or Theorem C in the Introduction) now implies that (4.10) has a solution w in 2
with w(x) log Ixl / O(1) near o. This proves the existence part of (i). (In fact,
it follows from Theorem C that u has a more precise asymptotic behavior, namely,

(4.11) u(x) h(x) + log Ixl -4- Uoo -4- O(Ixl )

near c, for every y > -1.) The uniqueness part of (i) follows from Theorem 4.1
immediately.
We now come to the proof of part (ii). Let u be a solution of (1.1) in 2. Since K
has compact support, we have Au 0 near c. Thus (the spherical mean of u) is
also harmonic near Ixl c, and there exist two constants, and C such that

(4.12) (r) a log r +C

for r > R. Also, by Green’s theorem and (1.1) we have, for r large,
r’(r) c3--- 2r
(-K)e2U > 0

since K 0 and K has compact support, where v denotes the unit outer normal to
St. Next, writing
v(x) [-K(y)e 2)] log Ix Yl dy,
we see that v is a Newtonian potential and

Av Au
in [2. Direct computation shows that

(4.13) v(x) log Ixl + O(1)

near oz. Thus the function h(x) u(x) v(x) is harmonic in 2, and (4.13) implies
(1.7). Q.E.D.
We remark that in (1.7), by Liouville’s theorem for harmonic functions, and
h(x) h(O)are uniquely determined by u.
Proof of Corollary IV. Let u be a solution of (1.1) where K satisfies the hy-
.
potheses of Theorem II. Then, by Theorem II (ii), either u U in which case u is
radially symmetric by Remark 2.9, or u u for some Let Ro be the rotation in
2 by angle 0. Then u(Rox) is also a solution of (1.1) since K is radially symmetric.
Since u(Rox) also satisfies (1.3), we have

lug(x)- u(Rox)l-- O(1) near c,

734 CHENG AND NI

which in turn implies that

(4.14) u,(x) u,(Rox)
for all x [2 by Theorem 4.1. Since (4.14) holds for every angle 0, we conclude that
u, is radially symmetric, and Corollary IV is proved.
To end this section, we shall include a classification of all radial solutions of (1.1)
in case K is radial and decays exponentially near oz. To be more precise, we consider
the case K -e -2r2 near oz.
Let u be a radial solution of (1.1). Suppose that u U. Then u < U everywhere
in 2. Setting q9 U u, we see that q is a positive radial function in 2 (by Remark
2.9) and that Ao > 0 in 2 with the strict inequality holds somewhere in 2. Thus
q(r) is monotonically nondecreasing in r and qg(r) > e log r near oz for some e > 0.
Substituting this estimate into the equation for qg, we obtain, by Theorem 3.1, that

Aq9 >C Ce -2(r2-u) > C- Ce -2+(1) > C Cr -2 >C>0

for r large, say, r > R. That is, q satisfies the inequality
(4.15) (rqg’)’> Cr in JR, ),
(tp’(R) > O.
Integrating (4.15) twice we obtain that qg(r) > Cr 2
near r oz, i.e., for r large, the
following holds

(4.16) u(r) < U(r)- Cr 2

for some constant C > 0. From (4.16) it is easy to see that the number

(4.17) a(u)
Jo s[- K(s)]e 2uts) ds

is finite.
On the other hand, since u solves the equation
1
u" +-u’ + Ke2U 0 in (0, oz),
(4.18)
u’(O) o,
it is not hard to obtain the following integral representation of u:

(4.19) u(r)=u(O)+fs[-K(s)]e2"’S)log() ds.

 CONFORMAL GAUSSIAN CURVATURE EQUATION ON R2 735

It then follows from (4.17) and standard computations that

(4.20)

0(u) log r + O(1)

near r
Conversely, by Theorem B, we see that for every positive real number e there
exists a solution u of (1.1) in N2 satisfying (1.3). Again, if Ro denotes the rotation in
[2 by angle 0, then u(Rox) is also a solution of (1.1) since K is radially symmetric.
Once more, Theorem 4.1 guarantees u(x) u(Rox) for all x in N2 since
]u,(x)- u,(Rox)] 0(1)
near o. This proves that u, must be radially symmetric.
Summing up the above discussion, we obtain a complete understanding of all
radial solutions of (1.1) in 2. It may be stated as follows.

and that K -e -2r2 near .

PROPOSITION 4.21. Suppose that K is radially symmetric and nonpositive in 2
Then the following conclusions hold.
(i) For each positive real number 0, (1.1) possesses a unique radial solution u,
satisfyin9 (1.3).
(ii) Let u be a radial solution of(1.1) in 2. Then either u =- U where U is the maximal
solution given by (1.5), or u u, for some positive real number where u, is 9iven
by (i) above.
(iii) If > fl > 0 then u, > ut in 2. Furthermore, the maximal solution U is also
radially symmetric with its asymptotic behavior near go given by (3.2).
Remark 4.22. From the point of view of ordinary differential equations, we may
consider the following initial value problem

1
u" +- u’ + Ke 2u =0 inr>0,
(4.23)
u’(0) 0, u(0) a

and denote the solution by u(r; a). Under the hypotheses of Proposition 4.21, it is
not hard to obtain the following conclusions.
(i) If a > b, then u(r; a) > u(r; b) wherever both solutions are defined.
(ii) There exists a finite real number a, such that
(1) u(r; a,) U(r) for all r > O,
(2) u(r; a) exists in [0, oz) for every a < a,, and
(3) u(r; a) becomes o at some finite point for every a > a,.
736 CIaENG ANO I

(iii) The asymptotic behavior of u(r; a) is given by

u(r; a,) r2 + O(1) near

and

u(r; a) (a) log r + O(1) near

where

(4.24) (a)
f s[- K(s)]e 2uts;a ds

and (a) is finite for every a < a,.

(iv) The function (a) given by (4.24) is a one-to-one, onto, continuous and strictly
increasing function from (-, a,) to (0, ).
This remark provides us with a clear picture of all possible solutions of (4.23). A
similar remark also holds under the hypotheses of Corollary IV. We shall omit the
details.

5. Concluding remarks. From Proposition 4.21 we see that if K < 0 is radial

with K -exp(- 2r 2) near oz, then (1.1) possesses exactly one radial solution (the

.
maximal solution) which has polynomial growth (, r 2) near ,
while all other radial
solutions have logarithmic growth near A natural question arises: are there other
(nonradial) solutions? It turns out that the answer is positive. That is, Corollary IV
does not hold in this case. For it is not hard to see, by Theorem 2.2 in I-KN], that
equation (1.1) with K < 0 in 2 and K(x) -exp(-2lxl 2) near also possesses,
among other things, solutions of the form u(x) h(x) + log(1 + Ixl 2) / O(1)
where h(x) is a harmonic function in R 2 with Ih(x)[ < Clxl near .
However, a
complete classification (like Theorem II or III) in this case has not yet been obtained.
We have some partial results in this direction, and we hope to report our progress
in a future paper.

REFERENCES
[A] L.V. AHL’FORS, An extension of Schwarz’s lemma, Trans. Amer. Math. Soc. 43 (1938), 359-364.
[B] P.D. BArtRV, The minimum modulus of all integral and subharmonic functions, Proc. London
Math. Soc. (3) 12 (1962), 445-495.
[CL] K.-S. C-mNG AND J.-T. LIN, On the elliptic equation Au K(x)u and Au K(x)e 2u, Trans.
Amer. Math. Soc. 3t)4 (1987), 639-668.
[CN] K.-S. CrEN6 AND W.-M. NI, in preparation.
[EHH] M. ESSEN, W. K. HAYMAN, AND A. HErBER, Slowly 9rowin9 subharmonic functions I, Comment.
Math. Helv. 52 (1977), 329-356.

equations in ,
[GNN] B. GIOAS, W.-M. N, AND L. NRNBERG, Symmetry of positive solutions of nonlinear elliptic
Adv. Math. Supp. Stud. 7A (1981), 369-402.
 CONFORMAL GAUSSIAN CURVATURE EQUATION ON R2 737

[GT] D. GILBARG AND N. S. TRUDINGER, Elliptic Partial Differential Equations of Second Order, 2nd
edition, Grundlehren Math. Wiss. 224, Springer-Verlag, Berlin, 1983.
[HI W. K. HAYMAN, Slowly growing integral and subharmonic functions, Comment. Math. Helv.
34 (1960), 75-84.
[K] J. KAZDAN, Prescribing the Curvature of a Riemannian Manifold, NSF-CBMS Regional
Conference Lecture Notes 57 (1985).
[Ke] J. B. KELLER, On the solutions of Au f(u), Comm. Pure Appl. Math. 10 (1957), 503-510.
[KN] C. E. KENIG AND W.-M. NI, On the elliptic equation Lu k / Ke 2u 0, Ann. Scuola Norm.
Sup. Pisa CI. Sci. (4) 12 (1985), 191-224.
[L] J. L. LEWIS, Some theorems on the cos r2 inequality, Trans. Amer. Math. Soc. 167 (1972),
171-189.
[LN] Y. LI AND W.-M. NI, On the existence and symmetry properties of finite total mass solutions of
Matukuma equation, Eddington equation and their generalizations, Arch. Rational Mech.
Anal. 108 (1989), 175-194.
[M] R. McOWEN, On the equation Au + Ke 2u f and prescribed negative curvature in 2, j. Math.
Anal. Appl. 103 (1984), 365-370.
IN1] W.-M. NI, On the elliptic equation Au / Ke 2= 0 and conformal metrics with prescribed
Gaussian curvatures, Invent. Math. 66 (1982), 343-352.
IN2] On the elliptic equation Au + Kutn+2)/tn-2) --0, its generalizations and applications in
geometry, Indiana Univ. Math. J. 31 (1982), 493-529.
[o] O. A. OLEINIK, On the equation Au + k(x)e 0, Russian Math. Surveys 33 (1978), 243-244.
[Os] R. OSSERMAN, On the inequality Au > f(u), Pacific J. Math. 7 (1957), 1641-1647.
is1] D. H. SATTINGER, Conformal metrics in 2 with prescribed curvature, Indiana Univ. Math. J.
22 (1972), 1-4.
[$2] Topics in Stability and Bifurcation Theory, Lecture Notes in Math. 309, Springer-
Yerlag, Berlin, 1973.
[w] H. WITTICH, Ganze Losungen der Differentialgleichung Au e", Math. Z. 49 (1944), 579-582.
CHENG: INSTITUTE OF APPLIED MATHEMATICS, NATIONAL CHUNG CHENG UNIVERSITY, CHIAYI, TAIWAN
62117
NI: SCHOOL OF MATHEMATICS, UNIVERSITY OF MINNESOTA, MINNEAPOLIS, MINNESOTA 55455