Академический Документы
Профессиональный Документы
Культура Документы
359
Introduction
Affine maximal hypersurfaces are extremals of the interior variation of affinely
invariant volume. The corresponding EulerLagrange equation is a fourth-order
PDE. Originally, these hypersurfaces were called affine minimal hypersurfaces.
Calabi calculated the second variation and proposed calling them affine maximal
(see [1]). For affine maximal surfaces, there are different versions of so called affine
Bernstein conjectures. One is Cherns conjecture (see [6]). Another is a problem
raised by Calabi (see [2]), called Calabis conjecture (see [12]). The two conjectures differ in the assumptions on the completeness of the affine maximal surface
considered. While Chern assumed that the surface is a convex graph over R 2 , which
means that the surface is Euclidean complete, Calabi assumed that the surface is
complete with respect to the Blaschke metric. In [13], the authors present a proof of
Cherns conjecture. Under an additional assumption, the first author gave a partial
answer to Calabis conjecture (see [8]). Calabis conjecture was recently solved,
see [9, 14].
Let x: M An+1 be an affine maximal hypersurface given by a locally strongly
convex function
xn+1 = f (x1 , x2 , . . . , xn )
360
defined in a domain
An . In this paper, we say that f is locally strongly
convex which means that the Hessian of the function f is positive definite. Then
x (M) is an affine maximal hypersurface if and only if f satisfies the following
fourth-order PDE:
2 1/(n+2)
f
det
= 0,
xi xj
where denotes the Laplacian with respect to the Blaschke metric G, which is
defined by
1
ij
G det(Gkl )
.
=
xi
xj
det(Gkl )
Following Calabi [3] and Pogorelov [11], we consider the Riemannian metric G#
on M, defined by
fij dxi dxj ,
G# =
where fij = 2 f /xi xj . This is a very natural metric for a convex graph. Our
main result can be stated as follows:
THEOREM. Let xn+1 = f (x1 , . . . , xn ) be a locally strongly convex function defined in a domain
An . If
M = {(x1 , . . . , xn , f (x1 , . . . , xn ) | (x1 , . . . , xn )
}
is an affine maximal hypersurface, and if M is complete with respect to the metric
G# , then, in the case where n = 2 or n = 3, M must be an elliptic paraboloid.
For the proof of the theorem, we first show that if the maximal hypersurface M
is complete with respect to the metric G# and if the norm of its Ricci curvature
Ric# G# is bounded, then M must be an elliptic paraboloid. Next, we use Hofers
Lemma to prove that Ric# G# must be bounded.
1. Preliminaries
Let An+1 be the unimodular affine space of dimension n + 1, M be a connected and oriented C manifold of dimension n, and x: M An+1 a locally
strongly convex hypersurface. We choose a local unimodular affine frame field x,
e1 , e2 , . . . , en , en+1 on M such that
e1 , . . . , en Tx M,
det(e1 , . . . , en , en+1 ) = 1,
Gij = ij ,
en+1 = Y,
361
where we denote the Blaschke metric and the affine normal vector field by Gij and
Y , respectively. Denote by U , Aij k and Bij the affine conormal vector field, the
FubiniPick tensor and the affine Weingarten tensor with respect to the frame field
x, e1 , . . . , en , and by Rij denote the Ricci curvature. We have the following local
formulas (see [10]):
Aij k U,k Bij U,
(1.1)
U,ij =
U = nL1 U,
Aiik = 0,
Rij =
Amli Amlj +
(1.2)
(1.3)
n2
n
Bij + L1 ij ,
2
2
(1.4)
where L1 denotes the affine mean curvature, and , denotes the covariant differentiation with respect to the Blaschke metric (note that with respect to the orthonormal
j
frame field x, e1 , . . . , en , we have Aij k = Aik ). A locally strongly convex
hypersurface is called an affine maximal hypersurface if L1 = 0 everywhere.
Let x: M An+1 be given by a locally strongly convex function
xn+1 = f (x1 , . . . , xn ).
We choose the following unimodular affine frame field:
f
,
e1 = 1, 0, . . . , 0,
x1
f
,
e2 = 0, 1, 0, . . . , 0,
x2
.........
f
,
en = 0, 0, . . . , 1,
xn
en+1 = (0, 0, . . . , 0, 1).
Then the Blaschke metric is given by (see [10])
2 1/(n+2) 2
f
f
dxi dxj .
G = det
xi xj
xi xj
The affine conormal vector field U can be identified with
2 1/(n+2)
f
f
f
,...,
,1 .
U = det
xi xj
x1
xn
362
1
ij
G det(Gkl )
.
=
xi
xj
det(Gkl )
Denote
2 1/(n+2)
f
,
= det
xi xj
and
=
2G
.
(1.6)
(1.7)
,i ,j2
2 ,j ,j i
,
,i =
2
2 ,j2 i + 2 ,j ,j ii
,j ,i ,j i
4G
4
+
2
.
=
2
3
In the case where (p) = 0, it is easy to get, at p,
2
2 ,ij
.
(1.8)
Now we assume that (p) = 0 and choose an orthonormal frame field such that,
at p M, ,1 = , ,i = 0, i > 1. Then
2
4
2 ,j2 i + 2 ,j ,j ii
,1
,1
,11
4
+2 3.
(1.9)
=
363
i>1
1
2
,11
+
,ii
n 1 i>1
=
Similarly,
A2ml1
n
2 + 2
n 1 ,11
2
+2
2
,1i
i>1
2
,1i
.
(1.10)
i>1
n
A2 .
n 1 111
(1.11)
(1.12)
2
A2ml1 ,1
(n 2)
(n 2)
2
,11 ,1
2
,11 ,1
(n 2)A111
4
(n 2)2 (n 1) ,1
.
8n
2
n1
4
2
,11
,1
(n 2)2 (n 1) ,1
+ 2
.
(n + 2)
2
8n
3
Note that
2
2
2
4
2
,11
,1
,1
,11
,i
i>1 ,1i
=4
+4
+ 3 4
.
2
Then (1.14) and (1.15) together give us
2
,i
n2 n 2
,i
n
,i
+
2(n 1)
2(n 1)
4
n2 2 ,1
(n 2)2 (n 1)
.
+ 2
8n
2(n 1) 3
3
,1
(1.13)
(1.14)
(1.15)
(1.16)
364
(1.17)
,i .
4
(1.18)
Denote by # and Ric# the Laplacian and the Ricci curvature with respect to
the metric G# , resp. By definition of Laplacian and a direct calculation, we get
2G#
= 2G ,
# r = r
(1.19)
n 2 < , r >G#
,
2
(1.20)
fi (0) = 0,
fij (0) = ij .
Denote by r(p0 , p) the geodesic distance function from p0 with respect to the
metric G# . For any positive number a, let Ba (p0 ) = {p M | r(p0 , p) a}.
Consider the function
F = (a 2 r 2 )2
365
0.
2
(a 2 r 2 )2 a 2 r 2
(2.1)
(2.2)
+
(a 2 r 2 )2 a 2 r 2
=
24r 2 r2G
4r2G
4rr
+
+ 2
.
2
2
2
2
2
(a r )
a r
a r2
(2.3)
Since
2(n 2)r < , r >G#
4r # r
4rr
+ 2
=
2
2
2
2
a r
a r
a r2
2
4r
# r
+
a2 r 2
1
24
2G#
3
and
2G#
= 2G ,
r2G#
= r2G ,
r2G# = 1,
we have
24r 2 r2G#
4r2G#
4r# r
+
+
+
(a 2 r 2 )2
(a 2 r 2 )
(a 2 r 2 )
+
=
1
24
24(n 2)2 r 2
+ 2
r2G#
(a r 2 )2
4
4r# r
24(1 + (n 2)2 )r 2
+
+
+
(a 2 r 2 )2
(a 2 r 2 )
(a 2 r 2 )
1
24
.
(2.4)
366
Recall that (M, G# ) is a complete Riemannian manifold with the Ricci curvature bounded from below by a constant K, K > 0. We have
24(1 + (n 2)2 )r 2
4n
4(n 1) K r
+ 2
+
+
(a 2 r 2 )2
(a r 2 )
(a 2 r 2 )
1
24
. (2.6)
3
4
2,i
2
3
4
1
6
,i i
2,i
3
+
2
36r 2
+
(a 2 r 2 )2
1
12
1
12
,
(2.7)
288
192 K r
2016r 2
+
+
.
2
(a r 2 )2 a 2 r 2
a2 r 2
(2.8)
(a 2 r 2 )2 2304a 2 + 192 Ka 3
= b1 a 2 + b2 a 3 ,
(2.9)
where b1 = 2304 and b2 = 192 K. It is obvious from (1.17) that (2.9) also holds
for n = 2. Hence, at any interior point of Ba (p0 ), we have
b1
a2 1
r2
a2
1
2 + b2
a 1
r2 2
a2
Let a , then
0.
It follows that
2
f
= 1,
det
xi xj
and
G# = G.
(2.10)
367
a
H (p).
2
Now we assume that ||Ric# ||2G# is not bounded above. Then there is a sequence of
points p. M such that ||Ric# ||2G# (p. ) . Let B1 (p. ) be the geodesic ball with
center p. and radius 1. Consider a family of functions /(.): B2 (p. ) R, . N,
defined by
/(.) = ||Ric# ||G# + + L,
(2.11)
x B-. (q. ),
(2.12)
(2.13)
The restriction of the hypersurface x to the balls B-. (q. ) defines a family M(.)
of maximal hypersurfaces. For every ., we normalize M(.) as follows:
Step 1. By adding a linear function we may assume that, at ql , (x1 , . . . , xn ) =
(0, . . . , 0) and
f (0) = 0,
fi (0) = 0.
(2.14)
j
where ai (.) are constants. Choosing ai (.) appropriately and using an obvious
, we may assume that, for every ., we have fij (0) = ij . Note that,
notation f, /
368
is invariant.
under the parameter transformation (2.14), /
Step 2. We take an affine transformation by
xi (.),
xi (.) = a(.)
1 i n,
xn+1 (.),
xn+1 (.) = (.)
where (.) and a(.) are constants. It is easy to verify that each M(.)
again is
a locally strongly convex maximal hypersurface. We now choose (.) = a(.)2 ,
, one can see that
(q. ). Using again an obvious notation f
, /
(.) = /
f
ij (.) = fij (.),
(.) =
/
1
(.).
/
(.)
The first equation is trivial. We calculate the second one. From (1.6), (1.21) we
can easily get
,
= 1
||G # =
||Ric
#
1 #
Ric G# ,
.
= 1L
L
(2.15)
d(.)(q. ).
x B
(2.16)
f
(.)
(0) = 0,
i
2 f
(.)
(0) = ij ,
i j
(2.17)
(.) = 1,
/
369
(2.18)
(.)(x) 4,
/
d(.) ,
d(.)(0).
x B
(2.19)
as . .
(2.20)
fij (0) = ij ,
(ii)
Ric# G# + +
f il f j m f km fij k flmn 4.
2
i 1/4n2 }. Then there is constant C1 > 0 such
Denote D := {(1 , . . . , n )|
that, for (1 , . . . , n ) D, the following estimates hold:
(1)
fii 4n,
(2) 1/C1 det(fij ) C1 ,
(3) Define do by do2 = 1/7n2 (4n)n1 C1 then
do { i2 < 1/7n2 } D, where
do is the geodesic ball with center 0 and radius do with respect to the metric
G# .
Proof. (1) Consider an arbitrary curve
ai2 = 1, s 0 .
4 = 1 = a1 s, . . . , n = an s
By assumption, we have
f il f j m f kn fij k flmn 4,
fii (0) = n.
Since f il f j m f kn fij k flmn is independent of the choice of coordinates 1 , . . . , n ,
for any point (s) we may assume that fij = i ij . Then
il
f f
jm
f fij k flmn =
kn
2
fij k
1
2
fij k
3 .
i j k
fii
It follows that
2
1
f
iik 3
fij2 k 4
( fii )
( fii )3
370
and, hence,
1
fii (x(s)))
=
fiik (x(s))ak
ds
( fii (x(s)))3/2
2
1/2
1/2
f (x(s))
2
n iik
a
k
( fii (x(s)))3
2 n.
Solving this differential inequality with fii (0) = n, we get
1
d(
3/2
( fii (x(s)))
1
1
.
s n
( fii (x(s)))1/2
n
From the assumption we have s 1/2n , then (1) follows.
(2) Again consider an arbitrary curve
ai2 = 1, s 0 .
4 = 1 = a1 s, . . . , n = an s
By assumption, we have
ij
f i j
4.
2
It follows that
2
i 2 4.
( fii )
By (1) we get
1 d(x(s))
4 n.
ds
Solving this differential inequality with (0) = 1, we obtain
4 ns ln(x(s)) 4 ns.
Recall that s 1/2n, then (2) follows.
(3) Denote by min , max the minimal and maximal eigenvalues of (fij ). Then,
from (1) and (2), we have max 4n and
1
n1
det(fij ) min n1
min .
max (4n)
C1
(2.21)
xi2 2
1
xi2 ,
n1
C1 (4n)
(2.22)
371
as . .
It is easy to see that under the conditions (i ) and (ii) in Lemma 3, the estimates
(1), (2) and (3) in Lemma 3 still hold. By the same argument as above, we conclude
that there is a ball around p and a subsequence .k , such that f(.k ) converges to
f on the ball and, correspondingly, all derivatives. As a limit, we get a maximal
hypersurface M , which contains a geodesic ball of radius do around p. Then we
return to the original parameters. Note that the geodesic distance is independent of
the choice of the parameters. It is obvious that M and M agree on the common
part. We repeat this procedure to extend M to be defined on
2do , etc. In this way,
we may extend M to be a maximal hypersurface defined in a domain
Rn ,
# . Using (2.18) and (2.19), we get
which is complete with respect to the metric G
# ||G # 4,
||Ric
/(0)
= 1.
372
defined in a domain
An . If x(M) is an affine maximal hypersurface and if
x(M) is complete with respect to the metric G# , is it an elliptic paraboloid?
Acknowledgements
The first author would like to thank Professors U. Simon and L. Vrancken for many
valuable discussions.
Both authors are partially supported by 973 project, NSFC 10271083 grant and
a Chinese-German exchange project of NSFC and DFG.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
Calabi, E.: Hypersurfaces with maximal affinely invariant area, Amer. J. Math. 104 (1982),
91126.
Calabi, E.: Convex affine maximal surfaces, Results in Math. 13 (1988), 209223.
Calabi, E.: Improper affine hyperspheres of convex type and a generalization of a theorem by
K. Jogens, Michigan Math. J. 5 (1958), 105126.
Calabi, E.: Affine differential geometry and holomorphic curves, in: Lecture Notes in
Math. 1422, Springer, New York, 1990, pp. 1521.
Caffareli, L. A. and Guitierrez, C. E.: Properties of the solutions of the linearrized Monge
Ampere equations, Amer. J. Math. 119 (1997), 423465.
Chern, S. S.: Affine minimal hypersurfaces, in: Proc. Japanese-U.S. Seminar, 1977, Tokyo,
1978, pp. 1730.
Hofer, H.: Pseudoholomorphic curves in symplectizations with applications to the Weinstein
conjecture in dimension three, Invent. Math. 114 (1993), 515563.
Li, A.-M.: Some theorems in affine differential geometry, Acta Math. Sinica (NS) 5, 345354.
Li, A.-M. and Jia, F.: The Calabi conjecture on affine maximal surfaces, Results in Math. 40
(2001), 256272.
Li, A.-M., Simon, U. and Zhao G.: Global Affine Differential Geometry of Hypersurfaces, De
Gruyter, Berlin, 1993.
Pogorelov, A. V.: The Minkowski Multidimensional Problem, Wiley, New York, 1978.
Simon, U.: Affine differential geometry, in: Handbook of Differential Geometry, NorthHolland, Amsterdam, 2000, pp. 905961.
Trudinger, N. and Wang, X. J.: The Bernstein problem for affine maximal hypersurfaces, Invent.
Math. 140 (2000), 399422.
Trudinger, N. and Wang, X. J.: Affine complete locally convex hypersurfaces, Invent. Math.
150 (2002), 4560.
Yau, S. T. and Schoen, R.: Differential Geometry, Science Press, Beijing, 1988 [in Chinese].