Ob o (CI (B) ) Ob: Q CI CI An

JOURNAL OF 1lfE
AMERICAN MAlHEMATICAL SOCIETY

Volume 7, Number 3, July 1994
RIGIDITY OF HOLOMORPHIC MAPPINGS

AND A NEW SCHWARZ LEMMA AT mE BOUNDARY
DANIEL M. BURNS AND STEVEN G. KRA..."'ITZ
1. INTRODUCTION
The uniqueness portion of the classical Schwarz lemma has played an inter-
esting historical role. Usually first encountered in the classification of the con-
formal self-maps of the disc, it arises in more general considerations of automor-
phism groups, in the construction of the Caratheodory and Kobayashi/Royden
metrics, and in a variety of contexts in analysis on manifolds. Useful references
for this material are [AHL, KRl, KOB, GK1, GK2, YAU].
Generalizations of the disc Schwarz lemma to multiply-connected domains in
one complex variable are generated naturally and easily using the uniformization
theorem. The generalization to several complex variables requires insights of
a different nature. For example, the so-called Caratheodory-Cartan-Kaup-Wu
theorem [WU] says in part that if Q is a bounded domain in en and if CI> is
a holomorphic self-map of Q that fixes a point P E Q, then the holomorphic
Jacobian determinant det JacCl>(P) has modulus less than or equal to 1, and
equals 1 if and only if CI> is a biholomorphism of Q. An interesting and
important feature of this result is that it has a global hypothesis (that Q be
mapped to itself) and a local hypothesis (the condition on the behavior of the
mapping at P). The conclusion is then a strong global one. We will see this
paradigm repeated in the work that follows.
The purpose of this paper is to seek versions of the last stated result when
the point P lies in the boundary of Q. (A primitive version of such a result
appears in [KR2].) This problem, while of considerable intrinsic interest, is also
related to a variety of other work in the literature. We now indicate some of
these connections.
Our interest in this problem arose originally from a question of Warren
Wogen: Does there exist a holomorphic self-mapping CI> of the ball B in
e2 (i) which takes the point 1 = (1,0) E oB to itself (in a suitable sense)
and (ii) for which the set o(CI>(B)) has "high order of contact", as a set, with
the boundary oB of the target ball? Here part of the problem is to find a
suitable interpretation for the phrase "high order of contact". The mappings
Received by the editors May 3, 1993.
1991 Mathematics Subject Classification. Primary 32HlS, 32H20, 32H99.
Key words and phrases. Hoiomorphic mapping, boundary rigidity, Schwarz lemma
The first author was supported in part by grant #DMS9004149 from the National Science Foun-
dation. The second author was supported in part by grants #DMS8800523 and #DMS9101104
from the National Science Foundation.
@ 1994 American Mathematical Society
661
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

662 D. M. BURNS AND S. G. KRANTZ
<l>e(Zl' Z2) = ((1 - e)zl + e, (1 - e)z2) , e > 0, satisfy (i) and (ii) with the
boundary of the image ball having order of contact 1 with the boundary of the
target ball (the tangent planes agree but the second fundamental forms do not).
The mapping <1>0 is the identity and the order of contact is infinite. Wogen's
question may be interpreted as asking whether there are maps which are inter-
mediate to the <l>e' e > 0, and id = <1>0' The original interest in constructing
such maps was in finding counterexamples to certain assertions about composi-
tion operators (see [WOG]).
As our understanding of the problem developed, we also saw that it relates to a
result ofH. Alexander [ALE]: If UnB and U' nB are boundary neighborhoods
in the ball and if is a biholomorphic mapping of these neighborhoods which
extends C 2 to the boundary, then must be the restriction to Un B of a
biholomorphism of the entire ball. S. Pincuk [PIN] generalized this result to
bounded strictly pseudoconvex domains with real analytic boundaries and W.
Rudin [RU 1] reduced the hypothesis of C 2 to the boundary to an assumption
which is even weaker than continuity at a point. One interpretation of the
main result of the present paper is that the hypothesis of direct coincidence of
boundary neighborhoods in the results of Alexander, Pincuk, and Rudin may be
weakened to high order of contact of image boundary and target boundary. (See
also [GK3] for other more general versions of the Alexander phenomenon-in
particular, that paper characterizes not just the biholomorphic mappings of the
ball but mappings which are "approximately" biholomorphic mappings of the
ball.)
The organization of this paper is as follows: Section 2 presents a boundary
uniqueness result on the disc. The relationship of this result to the questions
just discussed is not immediately apparent, but is clarified later. Section 3 shows
how to extend the result of Section 2 to the ball in en. Section 4 shows how
to use variants of the Fornzss Imbedding Theorem and the Lempert theory of
extremal discs for the Kobayashi metric to derive a result for strictly pseudo-
convex domains. Section 5 discusses a generalization of the Schwarz uniqueness
theorems of the previous sections. Section 6 discusses geometric interpretations
of the main results and returns to the original question of order of contact. The
question of Wogen is then recalled and answered. Section 7 considers only very
briefly analogous problems on weakly pseudoconvex domains.
2. A RESULT ON THE DISC ABOUT BOUNDARY DERIVATIVES
Let D be the unit disc centered at 0 E e.

Theorem 2.1. Let if> : D - D be a holomorphic function from the disc to itself
such that
if>(I;) = 1 + (t; - 1) + 0((1; - 1)4)
as I; - 1. Then if> ( 1;) == I; on the disc.
Remark 1. Results such as this one appear in the literature of conformal map-
pings (see, for instance, [VEL]) with the additional hypothesis that if> be univa-
lent (and often the function is assumed to be quite smooth-even analytic-in
a neighborhood of 1). The theorem presented here has no such hypothesis; so

RIGIDITY OF HOLOMORPHIC MAPPINGS 663
far as we know it is new. Even more surprising is that the exponent 4 is sharp:
simple geometric arguments show that the function
(C) = C - 1~(C - 1)3

satisfies the hypotheses of the theorem with 4 replaced by 3. Note also in
the proof that O(z - 1)4) can be replaced by o(z - 1)3). A similar remark
applies to Sections 3, 4, and 5 below.
X. Huang, in the papers [HUl, HU2, HU3], has explored additional condi-
tions under which the exponent in the error term may be decreased.
Proof of Theorem 2.1. Consider the holomorphic function
1 + (C)
g(C) = 1 _ (C) .
Then g maps the disc D to the right half plane. By the Herglotz representation
(this is just an application of the Banach-Alaoglu theorem; see [AHL)), there
must be a positive measure J.l on the interval [0, 211.') and an imaginary constant
~ such that
g(C) = -2
1
11.'
10
2x e i6 + C
i6
e - C
dJ.l(8)
'
+~ .
We use the hypothesis on to analyze the structure of g and hence that of
J.l. To wit,
4
g( C) = 1 + C+ O( C-
1)4 = 1 + C + O( C_ 1) 2 .
1-C-0(C-l) 1-C
From this and equation (*) we easily conclude that the measure J.l has the
form J.l = do + v , where do is (211.' times) the Dirac mass at the origin and v is
another positive measure on [0, 211.'). In fact, a nice way to verify the positivity
of v is to use the equation
-..
1+ C
-1r+O(C-l) =-2
11.' 0 e
+ Cd(do+V)(O)+~
C
2 1 1 2x e i6
6
l -
to derive a Fourier-Stieltjes expansion of do + v and then to apply the Herglotz

criterion [KAT, p. 38].
We may simplify equation (**) to
O(C -1) 2 = -2
1
11.'
10
2x e i6
e
6
l
+ Cdv(8) +~.
- C
Pass to the real part of the last equation, thus eliminating the constant ~.
Since v is a positive measure, we thus see that the real part of the integral on
the right of this last equation represents a positive harmonic function h on the
disc that satisfies
h(C) = O(C - 1)2 .
In particular, h takes a minimum at the point C = 1 and is O(lz - 112)
there as well. This contradicts Hopfs lemma [KRl] unless h == O. But h == 0

implies that v == 0 , hence that
g(') = 1+'
1- , .
Therefore 4>(') == , , and the theorem is proved. 0
Remark 2. Notice that the formulation of Theorem 2.1 is a boundary unique-

ness theorem in the vein of the uniqueness part of the classical Schwarz lemma.
However, the result sidesteps the original questions about "order of contact".
In fact the Riemann Mapping Theorem guarantees that, with any reasonable
definition of the phrase, there are univalent holomorphic maps of the disc to
the disc so that the image boundary has arbitrarily high (even infinite order)
contact with the target boundary. Thus Theorem 2.1 addresses some more
rigid structural phenomenon.
We defer our consideration of the original geometric question to 6.
3. A RESULT FOR THE BALL
Theorem 3.1. Let Been be the unit ball. Let cI> : B -+ B be a holomorphic
mapping of the ball to itself such that
cI>(z) = 1 + (z - 1) + O(lz _114)
as z -+ 1. (Here 1 denotes the distinguished boundary point 1 = (1 , 0, ... ,0)
of the ball.) Then 4>(z) == z on the ball.
Proof. There is no useful Herglotz representation on the ball (however, see
[AIZ] for related ideas): this is a deep fact which cannot be circumvented.
Thus we present a new argument that reduces the ball case to the disc case. For
simplicity we restrict attention to dimension two.
For each point a E B let ~ be the complex line joining a and 1. Let da
be the complex disc given by ~nB . Now for fixed a consider the holomorphic
function
'II: D--+B
'~(CO) .
Define also the mapping 4>a : B --+B , which is an automorphism of the ball
mapping do onto da and fixing 1. (That such maps exist follows from ele-
mentary geometric considerations, or see [RU2] for an explicit formula.) Finally
define
and
,,: do--+D,
" : (Zl ' 0) ~ Zl

Then the function
Ha :D--+D
-I
,~ "0 111 0 (4)a) 0 cI> 0 4>a 0 "'(')

is well defined. Moreover, it is straightforward to check that H satisfies the

hypotheses of Theorem 2.1 (since CI> agrees with the identity to high order,
the composition (Ja)-I CI> Ja is the identity to high order). It follows that
Ha (,) ==,.
Now set
Ga = (Ja)-I CI> Ja "'(') == (g~(,), g;(,)).
The statement that H a (,) ==, means that g~(,) == ,. But
Ig~(')12 + Ig;(')1 2< 1
for' ED. Letting 1'1- 1 yields then that Ig;({)I- O. Thus == O. It now
follows that the image of Ga already lies in do. As a result, it must be that CI>
g;
preserves da But this can only hold for every choice of a if CI> is the identity
mapping.
The proof is now complete. 0
4. GENERALIZATION OF THE RESULT TO STRONGLY CONVEX

AND TO STRONGLY PSEUDOCONVEX DOMAINS
It is clear that the methods of the last section will not apply to strongly
pseudoconvex domains. Indeed most strongly pseudoconvex domains have no
automorphisms except the identity (see [GKI]), so we are missing a major tool.
However, an inspection of the proof of Theorem 3.1 shows that the principal
geometric construct is of a family of mappings of the disc into the ball such
that the image analytic discs are holomorphic retracts of the entire domain (this
is where the global nature of the result, alluded to in the introduction, comes
in). Thanks to work of Lempert [LEMI], we know a large family of domains
for which such special analytic discs exist:
Proposition 4.1 (Lempert). Let 0 c en be a smoothly bounded, strictly convex
(in the real sense) domain with ~k -boundary (k ~ 6). Let Q E 0 and P E 80.
Then there exist unique holomorphic mappings
f/JP,Q : D--+o., "'P,Q : o.--+D
which are ~k-4 up to the boundary, and such that

(I) f/JP,Q(O) = Q,
f/Jp,Q(I) = P;
(2) "'P,Qof/Jp,Q(')=Cforevery 'ED;
(3) if f/J : D - 0 is holomorphic, and if "'p ,Q f/J(') =, for, ED, then
f/J(') == f/JP,Q(')
Note that (2) implies that f/Jp ,Q is extremal for the infinitesimal Kobayashi
e
metric on 0 in the direction := f/J~,Q(O) E TQ(O) , while "'P,Q is the ex-
tremal for the "dual" infinitesimal Caratheodory metric at e.
Proof. The statements of the proposition follow from [LEMI, Theoreme 3] and
[LEM2]. 0
Given Proposition 4.1, the proof of Theorem 3.1 gives the following result.

666 D. M. BURNS AND S. O. KRANTZ
Theorem 4.2. Let n be a smoothly bounded, strictly convex domain in Cn with

~6 boundary, and let PEa n be fixed. If : n --+ n is a holomorphic mapping
such that
4
(z) = z + O(lz - PI ),
then (z) z. =
Proof. Indeed, the maps rp P ,Q above replace fjJ a 0 'I' in the proof of Theorem
3.1, and the 'I'P , Q play the role of " 0 711 0 (fjJa) -I there. The uniqueness
argument at the end of the proof of Theorem 3.1 is here achieved by (3) of
Proposition 4.1. 0
Unfortunately, Lempert's results fail for a general strictly pseudoconvex do-
main (see [SIB]), but they remain true in a neighborhood of P, in directions
approximately C-tangent to an at P. More precisely, we have the following.
Proposition 4.3. Let n be a smoothly bounded, strongly pseudoconvexdomain
in Cn with ~6 boundary, and let PEOn be fixed. Then there exists Q En,
an open neighborhood V c n of Q, and holomorphic mappings
rpP,Q': D-D., 'l'P,Q' : D.--+D
as in Proposition 4.1 above, for every Q' E V . Furthermore, given any neighbor-
hood U c cn of P, one can assume that both Q and the images rpp,Q,(D) lie
in On u.
Proof. Given U, that the stationary disc rp P ,Q exists for some Q E U n n is
shown in [LEMI, p. 468]. The existence of the open set V :3 Q, and of the
corresponding rp P , Q" Q' E V , follows from [LEM 1, Proposition 10, p. 446].
To get the dual functions 'I'P ,Q' , we need the following preparatory geometrical
lemma.
Lemma 4.4. Let nand P be as in Proposition 4.3 above. Then there exist a
neighborhood n' of D., a bounded, strictly convex domain n" c Cn , and a
holomorphic map F : n' --+ Cn such that:
(I) for a suitable open neighborhood W of F(P) , F is biholomorphic,
F: F-1(W) --+ W;
(2) F(n) c n" ;
(3) F(an) n W = an" n W.
Admitting the lemma for the moment, we return to the proof of Proposition
4.3. We now choose the Q and V more carefully, namely so that V c U n
F- 1(W) n n, where W, F are as in the lemma. Then the mappings F 0 rp P ,Q'
are stationary discs for the strictly convex domain n" c Cn Thus there exist
dual mappings
o
'l'F(P),F(Q') :
fi" --+ D
and the mappings 'I'P, Q' = 'I'~(P), F(Q') 0 F will have all the desired properties in
the statement of Proposition 4.3, as follows directly from Proposition 4.1. 0
Using Proposition 4.3, the proof of Theorem 4.2 extends directly to give:

Theorem 4.5. Let .0 be a connected, smoothly bounded, strongly pseudoconvex

domain. Let P E 8.0, and let : .0 --+ .0 be a holomorphic map such that
(z) = z + O(lz - PI 4 ). Then (z) == z.
Proof. Find Q, V as in Proposition 4.3 above. The proof as in Theorems 3.1
and 4.2 shows that is the identity on each disc rp P ,Qt (D); in particular,
(Q') = Q' for all Q' E V. By analytic continuation, (z) == z. 0
We finish this section with the proof of Lemma 4.4. We start with FOI1UeSS'S
embedding theorem [FOR], which gives a neighborhood .0' of 0, a holo-
morphic embedding G : 0' '--+ eN (for some large N), a smoothly bounded
strictly convex set O~ c eN such that 8.Q~ intersects G(.Q') transversally,
and G(O) = .Q~ n G(.Q'). Let T be the tangent space to G(.Q') at G(P); T
is isomorphic to en, say E : T ....::::.. en linearly. Let n : eN --+ T be the
Hermitian orthogonal projection onto T (with respect to the usual euclidean
metric on eN), so that n is e-linear. We claim that the composition
F : .0' ..!!... eN ....!!.. T ~ en
will be the desired map, if we shrink .0' about 0 as necessary. To see this,
notice first that property (1) of Lemma 4.4 follows from the strict convexity of
O~ and the fact that G is an embedding. (We may be required to shrink Q'
here.) It is an easy exercise in the implicit function theorem to see that, in a
neighborhood of P, F is a local diffeomorphism of 80 onto a strictly convex
hypersurface in en . Since Eon is linear, the compact set E 0 n(O~) := O~ is
convex and F(O) c .0';. It is again a simple exercise in the implicit function
theorem to see that the boundary of O~ is smooth and strictly convex near
F(P). Thus there is a closed euclidean ball B which is tangent to 80~ at
F(P) and which contains o.~. Translate B a distance e > 0 in the direction
of the inward euclidean normal to 80~ at F(P) , to a new ball called Be. For
e sufficiently small, let He denote the (real) convex hull of BUF(O) in en ,so
that the boundary of He coincides with F(80') in a neighborhood of F(P).
The domain He is convex, but not smooth and not strictly convex. We can
enlarge He slightly to a smoothly bounded, strictly convex set .0" c en whose
boundary still agrees with F(80) in a neighborhood of F(P). That this last
is possible is clear from a picture, but is elementary and tedious to write down
explicitly, so we omit this verification. The .0" just constructed satisfies all the
claims of Proposition 4.4. 0
We close this section with the remark that Lemma 4.4 is an attempt to re-
embed Q in en as a strictly convex domain, which is certainly impossible.
The point here is that F is globally defined, and does re-embed 0 as a strictly
convex domain, at least in a neighborhood of P.
5. ANOTHER DIRECT EXTENSION
For P #- Q it is interesting to consider the setting of Theorem 4.2 with the

modified hypothesis that
(z) = Q + p,(z - P) + O(lz _ P1 4 ) ;

here p, is some local holomorphic transformation which takes the tangent plane
to on at P to the tangent plane to on at Q. Of course we cannot hope to
conclude under these circumstances that ~ is the identity. To see what the
correct conclusion might be, consider two special cases:
(1) If n is the ball, then ~ may be composed with a unitary rotation taking
Q to P. This reduces the situation to the result of 3. We conclude that ~ is
a rotation.
~2) If n is strongly pseudoconvex and there is a biholomorphism T of n
whose extension to the boundary takes Q to P then T 0 ~ satisfies the hy-
potheses of Theorem 4.5. We conclude that ~ is a biholomorphism of n.
A generic strongly pseudoconvex n has no nontrivial biholomorphisms (see
[OK1]); so the second case above can be considered descriptive but not prescrip-
tive. Moreover, for any positive integer m it is a simple matter to construct a
strongly pseudoconvex domain n with boundary points P, Q such that on
near Q is (after a rigid motion) a 2m-order perturbation of on near P-thus,
it can be arranged that a rigid motion of n maps n into n, maps P to Q,
and agrees with the identity to order 2m, yet the mapping is not biholomor-
phic. In summary, a generalization of our results in the direction we have been
discussing must take a more restrictive form. We have the following:
Theorem S.l. Let n c en be a smoothly bounded, strongly pseudoconvex do-
main with C(f) boundary. Suppose P, Q E on and ~ is a local biholomorphic
equivalence of on sending P to Q. If F : n - n is a holomorphic map such
that
F(z) = ~(z) + O(lz _ P1 4 )
near P, then ~ continues analytically to a biholomorphism of nand F(z) ==
~(z) on n.
Proof. We first note that the localized argument in the proof of Theorem 4.5
can be applied to ~ -1 0 F in a neighborhood of P to show that F == near
P in o. We would like to use analytic continuation now, but must distinguish
between two cases.
(1) on is not spherical, i.e., is nowhere locally CR-equivalentto alBn In this
case, the theorem of Kruzhilin-Vitushkin [VIT] says that the local holomorphic
equivalence ~ can be analytically continued along arbitrary continuous curves
in on, as a local biholomorphic equivalence sending on to itself. By analytic
continuation, then, we conclude that F extends C(f) to on, and has nonvan-
ishing Jacobian determinant. The mapping F is therefore a biholomorphism,
since it is a self-map of n.
(2) on is spherical, i.e., locally CR-equivalent to alBn at some (and hence
every, by analytic continuation) point. Then by a result of [BU], the universal
cover n
is biholomorphic to lBn The universal covering of a sufficiently n'
small open neighborhood n' of n is also realizable as an open set lBn C
n' c en, and we let P E alBn be an inverse image of P for this extended
covering map. The mapping F lifts to F : lBn - Bn , and what we have proved
above now shows that in a neighborhood of P, F gives a local biholomorphic
equivalence of alBn to itself. But it is known from [CHM] that such a local
equivalence is the restriction of a global automorphism ci> of lBn By analytic

RIGIDITY OF HOLOMORPmC MAPPINGS 669
continuation, F == b on :IBn , and b induces the claimed analytic continuation

ofq,toO.D
Remark 3. In conclusion, we note that these two cases behave differently with re-
spect to analytic continuation of boundary equivalences, namely, the theorem of
Kruzhilin-Vitushkin is false if 80 is spherical. We do not know, incidentally,
whether in general the analytic continuation given by the Kruzhilin-Vitushkin
theorem is single valued.
6. GEOMETRIC INTERPRETATIONS
AND RETURN TO THE QUESTION OF ORDER OF CONTACT
An elegant theorem of Ian Graham [GRA] calculates the asymptotic bound-

ary behavior of the infinitesimal Caratheodory and Kobayashi metrics near a
strongly pseudoconvex boundary point. The work of Fefferman in [PEP] cal-
culates the asymptotic boundary behavior of the infinitesimal Bergman metric
near a strongly pseudoconvex boundary point. In both papers the proofs make
it clear that if two strongly pseudoconvex domains have local defining functions
which agree to fourth order at a strongly pseudoconvex boundary point P then
the Caratheodory (resp. Kobayashi, resp. Bergman) metric has local boundary
behavior at P which is the same for both domains. Thus we see that the hy-
potheses of Theorem 4.5 are precisely what are needed to guarantee that the
Caratheodory (resp. Kobayashi, resp. Bergman) metric near the distinguished
boundary point P has asymptotic boundary behavior which is identical to that
for the pullback metric (under the mapping q,) near P. To what extent can
this reasoning be reversed? That is, if cl> is a self-map of 0, taking a bound-
ary point P to itself and such that q, preserves asymptotically some invariant
metric near P, then must q, be an automorphism of O?
Interestingly, the problem is more subtle than this discussion suggests. For,
the fourth order contact that we have been discussing is enough to guarantee
identical local behavior of the metrics (for instance, it is enough to pick up the
first term of the asymptotic expansion for the Bergman kernel at P), but it does
not suffice to capture the global information that is necessary for the truth of
the theorems we have been considering. Thus, at this time we do not have a
purely geometric method for proving the theorems of Sections 3 and 4.
Now let us return to our original question concerning "geometric order of con-
tact" of the image of a mapping with its target boundary. To give a more natural
geometrical notion of contact here for our problem we consider an anisotropic
notion appropriate to the strong pseudoconvexity of our domains. Let 0 as
before be a smoothly bounded strongly pseudoconvex domain, and fix P E 80 .
Let p be a defining function for 8n near P. Normalize the coordinate sys-
tem at P so that z = 0 at P, the z l-direction is the complex normal di-
rection, and z' := (z2' ... ,zn) are the complex tangential directions. Define
c5(z) = (Izl + Iz'1 4)1/4. We say that a function WN = WN(z) is of weight
N at P if IWN(z)1 :$ Cc5(z/ near P, for some C> o. Similarly, in anal-
ogy with Landau's notation, we will denote by wN(z) a function such that
lim z .... o \wN(z)\ /c5 N(z) = o. These notions are independent of the coordinate
normalizations.

Now let cz, be a holomorphic map which is ~k on 0 near P E ao, cz,(0) c

0' , cz,(P) = Q E ao' , and let p' be a defining function for a 0.' at Q.
Definition 6.1. The set cz,(aO) is said to have geometric contact with 80' of
weight N at cz,(P) = Q if
(*) p' 0 cz, = h . p + W N
where h is a ~k-l positive function on Q near P and WN is as above.
Remark 4. In this definition, k must be at least N for the definition to make
intrinsic sense. See below for some remarks on more efficient measures of
differentiability here.
Now let us relate this geometric notion of contact with the analytic osculation
discussed earlier.
Proposition 6.2. Let Bn be the unit ball in en, n ~ 2, and cz, a holomorphic
mapping, cz,: Bn --. Bn. Suppose cz, is ~6 to the boundary near P E 8Bn and
Q = cz,(P) E aB n . If cz,(aBn) has geometric contact of weight 6 with 8B n at
Q, then there is a global biholomorphism '1': Bn --. Bn such that
cz,(z) = 'I'(z) + o(lz _ P1 3 )
near P.
Remark 5. The Riemann Mapping Theorem shows that the proposition is false
when n = 1. Concerning the sharpness of the proposition, and of Theorem
3.1, see Example 6.3 below.
Proof of Proposition 6.2. Without loss of generality, by composing cz, with au-
tomorphisms of Bn , we can assume that P = Q = (1 , 0, ... , 0). Conjugating
with a Cayley transform, we can also replace Bn with ~n = {( z , w) E Cn - 1 X e I
1m w > IzI2}, and P with o. The correspondence can be given explicitly by
.zl - 1 Zj+l
W = -I Z + 1 ' Zj = Z + l' j = 1, ... , n - 1.
1 1
The coordinates (z, w) at 0 are, suitably renumbered, normalized as in Def-
inition 6.1, and the proof of the proposition is an examination of our hy-
pothesis (*) with N = 6 in terms homogeneous with respect to the dilations
T6 : (z, w) --. (Jz, J 2w) , J > o.
Again, composing with automorphisms of ~n' we can assume without loss
of generality that cz, has an expansion near 0:
cz,(z, w) = (z + A(z, w), w + B(z, w)
where
N
A( z , w) = I: a" (z , w) + w (N + 1),
N
B(z, w) = I: b,,(z , w) + w(N + 1) .
,,=3

Here, a l l ' bll are homogeneous of degree 1/ for the T,J-dilations. Also W(k) ,
as distinct from Ie, denotes an error term in an obvious sense. Similarly, h
can be written
N
h=l+2: hll+ w (N+l).
11=2
We will use N:::; 6 in these expansions.
Let allo be the first nonvanishing term in A; then (*) shows that bll0 +1 is
the first nonvanishing term of B. Thus
~(z , w) = (z + allo + h.w.t., w + bllo +1 + h.w.t.)

where "h.w.t." is "higher weight terms" and 1/0 ~ 2. The part of equation
(*) homogeneous of weight 1/0 + 1 (where 1/0 < N), and with P = Q = 0,
p = p 1 = Im w - 1z 12 ,IS
. Just
.
h ll _ I (lmw -Izl )
o
2
= -2 Re(all 0
,z) + bll0 +1
h
were, ~
lor V, v E IV
'"..n-I (
,V, V
')
= ",n-I -'
L...j=1 VjV j
Now the operation
evaluated at 1m w = Iz1 2 , is the operator" L" on formal power series that

was defined in formula (2.6) of [CHM]. Equation (*)11 +1 says precisely that
o
L(all ,bll + l ) = o. We shall consider 1/0 = 2, 3 directly; the cases 1/0 = 4, 5,
o 0
which we shall also need, will then follow from Lemma 2.1 of [CHM].
The case 1/0 = 2. In this case we have
hi = (a, z) + (a, z), some a E en ,

a 2 = Q(z) + pw, Q quadratic in z, pEen,
b3 = C(z) + w(z, ga), C cubic, Y E en .
Then equation (*) 3 becomes
( (a,2iz)w _ (a,2i Z}W) _ ((a, z )1z 12 + (-a, z-) 1z 12) + ((a,2iz}w _ (a,2iZ}W)
= -((Q(z), z} + (Q(z) , z} ) - (w(P, z) + w(P, z})
C(Z) _ C(Z)) (W(Z, y) _ W(z, W})

+ ( 2i 2i + 2i 2i
Comparing like monomials on either side (let us label these with the notation

(p, q) ), we find that
(3,0) 0= C~:) or C(z) = 0;

w(z, a} w(z, y}
(2,0) 2i = 2i or a = y;
(3, 1) -(z, a}(z , z) = -(Q(z) , z}
or Q(z) = (z, a}z;
(1, 1) (a, z}w (z, a}'ll! _ ( ) _- ( P) .

2i - 2i - -w z, a - W ,z,
hence a = -2iP , some pEen.

Before proceeding, we recall that the automorphisms of ~n with 0 E a~n a
fixed point and normalized as in (**) have the form
( z+wP W)
(z, w) ~ 1 _ 2i(z, P} + ew' 1 - 2i(z, P} + ew '
where pEen is arbitrary and Ime = -IPI 2. The calculations performed

above tell us that, after composing with one of these automorphisms of ~n ' we
may assume that (**) holds with Vo ~ 3 .
The ease Vo = 3. Now we have
h2 = q(z) + aW + a'll! + q(z) , q quadratic, a E en ;
a3 = C(z) + wL(z) , C cubic, L linear;
b4 = Q(z) + wS(z) + yw, Q quartic, S quadratic, j
2
E C.
Thus (*)4 becomes
Im(qw + aw 2 ) - 2Req Izl2 - Im(qw) - 2Re(aw Iz12) -lwl 2 Ima

= Im(Q + wS + yw 2 ) - 2Re[(C(z), z} + w(L(z), z}] .
Comparing like terms now yields
(4,0) Q - O,
(3,0) S - q;
(2,0) a = y;
(3,1) C(z) = q(z) . z ;
(1, 1) Ima = O,
qw
(2,1) -aw(z,z}-Ti = -w(L(z),z};
hence q == 0 and L(z) = az.
Note that we have now that q = S = 0 and C = z . q = O. The upshot
of our calculation is that the mapping' 4> agrees in form (up to order vo ) with
an automorphism of ~n that fixes O. Composing 4> with the inverse of said
automorphism (as we are obviously free to do), we have that 4> satisfies (**)
with Vo = 4.
The eases Vo = 4, 5. Applying Lemma 2.1 of [CHM, p. 233], we find that
(a 4 ,bs)=(0,0) and (a s ,b6 )=0.

Thus we have proved that the geometric order of contact 6 implies that
cI>(Z, W) = (Z, W) + (terms of homogeneity at least six) .
This is almost the desired result. In terms of the usual notion of order, this
result says that there is a constant a E C such that, without loss of generality,
cI>(z, w) = (z + aw 3 , w) + O(lz _ P14) .
We can now use a global argument related to Theorem 3.1 above. Consider
the mapping of the upper half plane K := {1m w > O} to itself given by
w - w + B(O, w). By the above we know that B(O, w) is O(lwI 4 ) at w = 0
and so by the conjugated form of the boundary Schwarz lemma, B(O, w) == O.
Now consider the map of D into Un given by
w-+cI>(O, w) = (A(O, w), w + B(O, w)) = (A(O, w), w) .
By the uniqueness argument at the end of the proof of Theorem 3.1, it follows
that A(O, w) == O. Hence, the coefficient a above is 0, which proves the
proposition. 0
Remark 6. Of course we could assume above that cI>( 8l1!n) had contact of weight
N ~ 7, and the proposition would follow from purely local calculations. It
would certainly not be sharp then.
In the above proposition, we can weaken considerably the notion of differen-
tiability on cI> and h as follows: Say that cI> is admissibly differentiable at P
of weight N, if in coordinates normalized at P = 0 as above, there exists an ap-
proximation of .9J(z, w) = (A(z, w), B(z, w)) where A - av(z, w) = E::C/
wN(z, w) and B - E:=obv(z, w) = wN(z, w). Here the approximations are
to hold on admissible approach regions at 0 in the sense of Koranyi. Note that
the av ' bv are assumed to be To-homogeneous as in the proof above. One can
similarly define nontangential differentiability of order N, using nontangential
approach regions. The corresponding concepts for the real function h we leave
to the reader. Proposition 6.2 follows if cI> is admissibly differentiable of weight
6 and nontangentially differentiable of order 3 at P. Presumably one does not
need all derivatives in z of cI> up to order 6 for the validity of Proposition
6.2. We have chosen to work with the notion of contact expressed in terms
of weight because of its naturality with respect to CR-geometry, as in [CHM].
Example 6.3 below shows that, with this measure of contact, Theorem 3.1 and
Proposition 6.2 are sharp.
Example 6.3. We will construct self-maps of the Siegel domain U2 to itself
which describe with some precision how sharp the above results are. We start
from the Riemann Mapping Theorem in C, and take V to be a smoothly
bounded, simply connected open set c K (the upper half plane in C) such that
Vn8K in C contains a neighborhood of 0 in the real axis. Let fIJ : K - V be
a Riemann mapping (i.e., biholomorphism). Since fIJ is one-to-one, fIJ' (w) ::F 0,
for all w E K. Since K is simply connected, we can choose a branch of
JflJI(W) globally on K. By the Pick-Schwarz lemma, applied to the map fIJ,

we know that
Iq/(w)1 2 1
(#)
Ilmql(w)1 2 $ IImwI2 .
By (#), the mapping of ex Jf' c e2 to e2 given by
F: (z, w)----+h/qI'(w)z, qI(w)
sends ~2 to itself. Without loss of generality, we may choose qI in such a way

that
qI(W) = W +AW 3 + O(lwI 4 ) ,
where A is real, A > o. Then, in a neighborhood of 0, we have

F(z, w) = (z + ~AW2z, W + AW 3 ) + O(l(z, w)1 4 )
J
Here we have chosen the branch of qI' (w) which is 1 at w = o. For this
map, F(fJ'l4) has contact of weight 5 but not weight 6; the case Vo = 5 above
breaks down here. This example (extended to en) also shows that Theorem
3.1 is sharp for all n. Note that the map F actually extends analytically past
o.
It is enlightening to attempt to carry out the proof of Proposition 6.2 for
n = 1 to see that the coefficients of 4> are not determined by the contact
equation, no matter what order of contact is assumed. Note also that the higher
dimensional example above has contact of F(fJ~2) with fJ~2 along the entire
curve (0, u), U = Rew E R., near U = O. The contact weight is sharp at all
points of this curve.
7. DISCUSSION OF THE WEAKLY PSEUDOCONVEX CASE AND CLOSING REMARKS

The analysis we have been discussing does not carry over to the case of
weakly pseudoconvex domains. A simple example serves to illustrate the wealth
of questions and problems available: Consider the domains
Om = {(z\, z2): Izl + IZ212m < I} ,
for m a positive integer. What conditions near 1 = (1, 0) on a holomor-

phic mapping 4> : Om --+ Om will force
to be the identity mapping? The
key observation is that Om covers the ball B via the mapping Vlm(z\, z2) =
(z\ ' (z2)m) . For any 0= (0\ ' ( 2 ) E B with O 2 =F 0, the analytic disc do: (see
3) lifts to m extremal discs da.,j' j = 1, ... , m, in Om. Then the analysis
of 3 may be carried out with these extremal discs-provided that the mapping
4> has Taylor jet at P which is sufficiently restricted so that after it is pushed
down with the function VIm off the branch locus, it satisfies the conditions of
Theorem 3.1. Note in particular that the correct condition on the Taylor jet
will be nonisotropic and will depend on m. X. Huang [HUl, HU2, HU3] has
explored these matters in some detail.

An argument similar to the one just outlined may be carried out near any
weakly pseudoconvex boundary point that locally "covers" a strongly pseudo-
convex boundary point. This is in fact a rather restrictive class, and has recently
been described by Barletta and Bedford [BABE] (however, see also [GAY]).
ACKNOWLEDGMENT
We thank John Bland and Robert E. Greene for helpful discussions about
the general nature of this problem and AI Baemstein for a useful conversation
regarding the proof of Theorem 2.1. We also thank J. Moser and N. Sibony for
their timely interest and suggestions.
REFERENCES
[AHL] L. Ahlfors, Conformal invariants, McGraw-Hill, New York, 1973.

[ALE] H. Alexander, Holomorphic mappings from the ball and polydisc, Math. Ann. 209 (1974),
249-256.
[AIZ] L. A. Aizenberg, Multidimensional analogues of the Carleman formula with integration
over boundary sets of maximal dimension, Akad. Nauk. SSSR Sibirsk. Otdel. Inst. Fiz.,
Krasnoyarsk, 1985, pp. 12-22,272. (Russian)
[BABE] E. Barletta and E. Bedford, Existence of proper mappings from domains in e2 , Indiana
Univ. Math. J. 39 (1990), 315-338.
[BU] D. Bums, A multi-valued Hartogs theorem and developing maps, preprint.
[BSW] D. Bums. S. Shnider, and R. Wells, On deformations of strictly pseudoconvex domains,
Invent. Math. 46 (1978), 237-253.
[CHM] S. S. Chern and J. Moser, Real hypersurj'aces in complex manifolds, Acta Math. 133 (1975),
219-271.
[FEF] C. Feiferman, The Bergman kernel and biholomorphic mappings ofpseudoconvex domains,
Invent. Math. 26 (1974), 1-65.
[FOR] J. Fonuess, Strictly pseudoconvex domains in convex domains, Amer. J. Math. 98 (1976),
529-569.
[GAV] E. Gavosto, Thesis, Washington University, 1990.
[GRA] I. Graham, Boundary behavior ofthe Caratheodory and Kobayashi metrics on strongly pseu-
doconvex domains in en with smooth boundary, Trans. Amer. Math. Soc. 207 (1975),
219-240.
[GK1] R. E. Greene and S. G. Krantz, Stability properties of the Bergman kernel and curvature
properties of bounded domains, Recent Progress in Several Complex Variables, Princeton
Univ. Press, Princeton, NJ, 1982.
[GKl] __ , Deformation of complex structures. estimates for the equation, and stability of thea
Bergman kernel, Adv. Math. 43 (1982), 1-86.
[GK3] __ , Methods for studying the automorphism groups of weakly pseudoconvex domains,
Proc. Internat. Conf. on Complex Geometry (Cetraro, Italy), Mediterranean Press, Cal-
abria, 1991.
[HU1] X. Huang, Some applications of Bel/'s theorem to weakly pseudoconvex domains, Pacific J.
Math. (to appear).
[HU2] __ , A boundary rigidity problem of holomorphic mappings on weakly pseudoconvex do-
mains, preprint.
[HU3] _ , Preservation principle of extremal mappings and its applications, Illinois J. Math. (to
appear).
[KAT) Y. Katznelson, An introduction to harmonic analysis, Dover, New York, 1976.

[KOB] S. Kobayashi, Hyperbolic manifolds and holomorphic mappings, Marcel Dekker, New York,
1970.
[KRl] S. G. Krantz, Function theory of several complex variables, Wiley, New York, 1982.
[KR2] __ , A new compactness principle in complex analysis, Univ. Autonoma de Madrid, 1987.
[KOR] J. Kohn and H. Rossi, On the extension of holomorphic junctions from the boundary of a
complex manifold, Ann. of Math. (2) 81 (1965),451-472.
[KRV] N. Kruzhilin and A. Vitushkin, Extension oflocal mappings ofpseudoconvex surfaces, Ookl.
Akad. Nauk. SSSR 270 (1983), 271-274.
[LEMl] L. Lempert, La metrique Kobayashi et les representation des domains sur la boule, Bull.
Soc. Math. France 109 (1981), 427-474.
[LEM2] _ , Holomorphic retracts and intrinsic metrics in convex domains, Anal. Math. 8 (1982),
257-261.
[LEM3] _ , Intrinsic distances and holomorphic retracts, Complex Analysis and Applications,
Sofia, 1984.
[PIN] S. Pincuk, On the analytic continuation ofhoiomorphic mappings, Mat. USSR-Sb. 27 (1975),
416-435.
[RUl] W. Rudin, Holomorphic maps that extend to automorphisms of a ball, Proc. Amer. Math.
Soc. 81 (1981), 429-432.
[RU2] _ , Function theory of the unit ball in e" , Springer-Verlag, Berlin and New York, 1980.
[SIB] N. Sibony, Unpublished notes.
[VEL] J. Velling, Thesis, Stanford, 1985.
[VIT] A. G. Vitushkin, Real analytic hypersurfaces of complex manifolds, Uspekhi Mat. Nauk. 40
(1985), 3-31.
[WOO] W. Wogen, Composition operators acting on spaces of holomorphic junctions on domains in
e" , Operator Theory: Operator Algebras and Applications, Part 2 (Durham, New Hamp-
shire, 1988), Proc. Sympos. Pure Math., vol. 51, Part 2, Amer. Math. Soc., Providence, RI,
1990,pp.361-366.
[WU] H. H. Wu, Normal families of holomorphic mappings, Acta Math. 119 (1967), 193-2333.
[YAU] S. T. Yau, A generalized Schwarz lemma for Kahler manifolds, Amer. J. Math. 100 (1978),
197-204.
ABSTRACT. A rigidity theorem for holomorphic mappings, in the nature of the

uniqueness statement of the classical one-variable Schwarz lemma, is proved
at the boundary of a strongly pseudoconvex domain. The result reduces to an
interesting, and apparently new, result even in one complex dimension. The
theorem has a variety of geometric and analytic interpretations.
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF MICHIGAN, ANN ARBOR, MICHIGAN 48109

DEPARTMENT OF MATHEMATICS, Box 1146, WASHINGTON UNIVERSITY IN ST. LoUIS, ST. LoUIS,
MISSOURI 63130

Ob o (CI (B) ) Ob: Q CI CI An

Uploaded by

Copyright:

Available Formats

Ob o (CI (B) ) Ob: Q CI CI An

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ob o (CI (B) ) Ob: Q CI CI An

Uploaded by

Copyright:

Available Formats

JOURNAL OF 1lfE

AMERICAN MAlHEMATICAL SOCIETY

RIGIDITY OF HOLOMORPHIC MAPPINGS

DANIEL M. BURNS AND STEVEN G. KRA..."'ITZ

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

2. A RESULT ON THE DISC ABOUT BOUNDARY DERIVATIVES

Let D be the unit disc centered at 0 E e.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

(C) = C - 1~(C - 1)3

to derive a Fourier-Stieltjes expansion of do + v and then to apply the Herglotz

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

implies that v == 0 , hence that

Therefore 4>(') == , , and the theorem is proved. 0

Remark 2. Notice that the formulation of Theorem 2.1 is a boundary unique-

" : (Zl ' 0) ~ Zl

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

is well defined. Moreover, it is straightforward to check that H satisfies the

4. GENERALIZATION OF THE RESULT TO STRONGLY CONVEX

which are ~k-4 up to the boundary, and such that

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

Theorem 4.2. Let n be a smoothly bounded, strictly convex domain in Cn with

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

Theorem 4.5. Let .0 be a connected, smoothly bounded, strongly pseudoconvex

For P #- Q it is interesting to consider the setting of Theorem 4.2 with the

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

continuation, F == b on :IBn , and b induces the claimed analytic continuation

An elegant theorem of Ian Graham [GRA] calculates the asymptotic bound-

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

Now let cz, be a holomorphic map which is ~k on 0 near P E ao, cz,(0) c

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

~(z , w) = (z + allo + h.w.t., w + bllo +1 + h.w.t.)

evaluated at 1m w = Iz1 2 , is the operator" L" on formal power series that

hi = (a, z) + (a, z), some a E en ,

Then equation (*) 3 becomes

C(Z) _ C(Z)) (W(Z, y) _ W(z, W})

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

(p, q) ), we find that

(3,0) 0= C~:) or C(z) = 0;

(1, 1) (a, z}w (z, a}'ll! _ ( ) _- ( P) .

hence a = -2iP , some pEen.

where pEen is arbitrary and Ime = -IPI 2. The calculations performed

Im(qw + aw 2 ) - 2Req Izl2 - Im(qw) - 2Re(aw Iz12) -lwl 2 Ima

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

cI>(z, w) = (z + aw 3 , w) + O(lz _ P14) .

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

By (#), the mapping of ex Jf' c e2 to e2 given by

F: (z, w)----+h/qI'(w)z, qI(w)

sends ~2 to itself. Without loss of generality, we may choose qI in such a way

where A is real, A > o. Then, in a neighborhood of 0, we have

7. DISCUSSION OF THE WEAKLY PSEUDOCONVEX CASE AND CLOSING REMARKS

Om = {(z\, z2): Izl + IZ212m < I} ,

for m a positive integer. What conditions near 1 = (1, 0) on a holomor-

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

[AHL] L. Ahlfors, Conformal invariants, McGraw-Hill, New York, 1973.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

ABSTRACT. A rigidity theorem for holomorphic mappings, in the nature of the

DEPARTMENT OF MATHEMATICS, UNIVERSITY OF MICHIGAN, ANN ARBOR, MICHIGAN 48109

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use