SOME REGULARITY THEOREMS FOR CR MAPPINGS

G. P. BALAKUMAR AND KAUSHAL VERMA

Abstract. The purpose of this article is to study Lipschitz CR mappings from an

arXiv:1102.2335v2 [math.CV] 14 Feb 2011

h-extendible (or semi-regular) hypersurface in Cn . Under various assumptions on the

target hypersurface, it is shown that such mappings must be smooth. A rigidity result
for proper holomorphic mappings from strongly pseudoconvex domains is also proved.

1. Introduction
The purpose of this note is to prove regularity results for Lipschitz CR mappings from
h-extendible hypersurfaces in Cn .
Theorem 1.1. Let f : M → M ′ be a non-constant Lipschitz CR mapping between C ∞
smooth pseudoconvex finite type (in the sense of D’Angelo) hypersurfaces in Cn . Let
p ∈ M and p′ = f (p) ∈ M ′ . Assume that M is h-extendible at p, and that there is an
open neighbourhood U ′ ⊂ Cn of p′ and a C ∞ smooth defining function r ′ (z ′ ) for U ′ ∩ M ′
in U ′ which is plurisubharmonic on the pseudoconvex side of M ′ near p′ . If the Levi rank
of M ′ at p′ is at least n − 2 then f is C ∞ smooth in a neighbourhood of p.
This theorem, which is purely local in nature, is motivated by the regularity and rigidity
results for CR mappings obtained in [20], [21], [22] and [32]. Let B(p, ǫ), B(p′ , ǫ′ ) be
small open balls around p, p′ respectively. The pseudoconvex side of M near p is that
component of B(p, ǫ) \ M which is a pseudoconvex domain for small ǫ > 0. Denote
this by B − (p, ǫ) while the other component (the pseudoconcave part) will be denoted by
B + (p, ǫ). The same practice will be followed while referring to the respective components
of the complement of M ′ in B(p′ , ǫ′ ). Since M and M ′ are both pseudoconvex and of
finite type near p and p′ respectively, it follows that f admits a holomorphic extension to
the pseudoconvex side of M near p and this extension (which will still be denoted by f )
maps the pseudoconvex side of M near p into the pseudoconvex side of M ′ near p′ . In
fact, since f : M → M ′ is assumed to be Lipschitz, it follows that the extension is also
Lipschitz on B − (p, ǫ) ∪ M near p. Fix ǫ, ǫ′ > 0 so that
f : B − (p, ǫ) → B − (p′ , ǫ′ )
is holomorphic, f extends Lipschitz continuously up to M ∩ B(p, ǫ) and satisfies f (M ∩
B(p, ǫ)) ⊂ M ′ ∩ B(p′ , ǫ′ ). For brevity, let D and D ′ denote the pseudoconvex domains
B − (p, ǫ) and B − (p′ , ǫ′ ) respectively so that M ∩B(p, ǫ) and M ′ ∩B(p′ , ǫ′ ) are smooth open
finite type pieces on their boundaries. We may also assume that ǫ′ > 0 is small enough
so that the defining function r ′ (z ′ ) ∈ C ∞ (B(p′ , ǫ′ )) and D ′ = {r ′ (z ′ ) < 0} where r ′ (z ′ ) is
plurisubharmonic in D ′ . Note that while this condition holds for strongly pseudoconvex
1991 Mathematics Subject Classification. Primary: 32H40 ; Secondary : 32Q45, 32V10.
Key words and phrases. CR map, finite type, scaling method, Segre varieties.
The second author was supported in part by a grant from the UGC under DSA-SAP (Phase IV) and
the DST SwarnaJayanti Fellowship 2009-2010.
1
2 G. P. BALAKUMAR AND KAUSHAL VERMA

and convex finite type domains in Cn , not all (see [24]) pseudoconvex domains satisfy this
condition. Now observe that the map f : D → D ′ is not known to be proper though it
does extend continuously up to M ∩ B(p, ǫ). A sufficient condition for f to be proper
from D onto D ′ was considered by Bell-Catlin in [9] – namely, if f −1 (p′ ) = f −1 (f (p)) is
compact in M ∩ B(p, ǫ) then it is possible to choose D, D ′ in such a way that f extends
as a proper holomorphic mapping between these domains. Once this is established, the
C ∞ smoothness of f is a consequence of the techniques in [9]. The proof of theorem 1.1
therefore consists of showing that the fibre over p′ is compact in M ∩ B(p, ǫ). To do this,
we adapt the method of uniform scaling from [20],[21] and [22]. The domains D, D ′ and
the map f are scaled by composing f with a suitable family of automorphisms of Cn that
enlarge these domains near p, p′ respectively. This produces a scaled family f ν : Dν → Dν′
of maps between scaled domains. The scaled domains Dν converge in the Hausdorff sense
to a model domain of the form

D∞ = z ∈ Cn : 2ℜzn + P (′ z,′ z) < 0


(1.1)

where ′ z = (z1 , z2 , . . . , zn−1 ) and P (′ z,′ z) is a weighted homogeneous (the weights being
determined by the Catlin multitype of ∂D at p) real valued plurisubharmonic polynomial
without pluriharmonic terms of total weight 1, while the domains Dν′ converge to
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2 < 0 .

(1.2) D∞

Here 2m′ is the type of ∂D ′ at p′ and Q2m′ (z1 , z 1 ) is a subharmonic polynomial of degree
at most 2m′ without harmonic terms. D∞ ′
is manifestly of finite type while D∞ also has
the same property since p is assumed to be h-extendible. The local geometry of a smooth
pseudoconvex finite type hypersurface whose Levi rank is at least n − 2 guarantees the
existence of special polydiscs around points sufficiently close to the hypersurface on which
the defining function does not change by more than a prescribed amount – these are the
analogues of Catlin’s bidiscs and have been considered earlier in [18]. The holomorphic
mappings fν : Dν → Dν′ are shown to form a normal family by using these polydiscs as in
′
[43]; the limit map F : D∞ → D∞ is therefore holomorphic. The assumptions that f is
Lipschitz and that the defining function of ∂D ′ is plurisubharmonic on D ′ force F to be
non-degenerate. Furthermore, since f extends Lipschitz continuously up to M near p, it
is natural to expect that F imbibes some regularity near 0 ∈ ∂D∞ . This indeed happens
and it is possible to show that F is Hölder continuous up to ∂D∞ near the origin. The
main ingredient needed to do this is a stable rate of blow up of the Kobayashi metric on
Dν′ for all ν ≫ 1 and this follows by analyzing the behaviour of analytic discs in Dν′ whose
centers lie close to the origin. In particular, theorem 2.3 provides a stable lower bound
for the Kobayashi metric in Dν′ near the origin using ideas from [43]. Once F is known to
be Hölder continuous up to ∂D∞ near the origin, Webster’s theorem ([46]) implies that
F must be algebraic. Moreover, if f has a noncompact fibre over p′ , it can be shown that
the same must hold for that of F over F (0). This violates the invariance property of Segre
′
varieties associated to ∂D∞ and ∂D∞ and proves the compactness of the fibre of f over
′ ′
p ∈ ∂D .

It is known that smooth convex finite type hypersurfaces can also be scaled and morever
the stability of the Kobayashi metric on the scaled domains is also understood (see [21]).
Hence the same line of reasoning yields the following:
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 3

Corollary 1.2. With M as in theorem 1.1, let M ′ ⊂ Cn be a smooth convex finite type
hypersurface and f : M → M ′ a Lipschitz CR mapping. As before, let p ∈ M and
p′ = f (p) ∈ M ′ . Then f is C ∞ smooth in a neighbourhood of p.
More can be said when at least one of the hypersurfaces is strongly pseudoconvex. We
first consider the following local situation – let M ⊂ Cn be a C ∞ smooth pseudoconvex
finite type hypersurface and let p ∈ M be an h-extendible point. Let the Catlin multitype
of M at p be (1, m2 , . . . , mn ) where the mi ’s form an increasing sequence of even integers.
Then there exists a holomorphic coordinate system around p = 0 in which the defining
function for M takes the form:
r(z) = 2ℜzn + P (′ z,′ z) + R(z)
where P (′z,′ z) is a (1/mn , 1/mn−1 , . . . , 1/m2 )-homogeneous plurisubharmonic polynomial
of total weight one without pluriharmonic terms and the error R(z) has weight strictly
bigger than one. As usual, if J = (j1 , j2 , . . . , jn−1 ) is a multiindex of length n − 1, then
jn−1 jn−1
z denotes the monomial z1j1 z2j2 · · · zn−1
′ J
and ′ z J = z j11 z j22 · · · z n−1 .
Theorem 1.3. Let f : M → M ′ be a nonconstant Lipschitz CR mapping between real
hypersurfaces in Cn . Fix p ∈ M and p′ = f (p) ∈ M ′ . Suppose that M is C ∞ smooth
pseudoconvex and of finite type near p and that M ′ is C 2 strongly pseudoconvex near p′ . If
M is h-extendible at p, then the weighted homogeneous polynomial in the defining function
for M near p = 0 can be expressed as
P (′ z,′ z) = |P1 (′ z)|2 + |P2 (′ z)|2 + . . . + |Pn−1 (′ z)|2
where each Pi (′ z) for 1 ≤ i ≤ n − 1 is a weighted holomorphic polynomial of total weight
1/2. Moreover, the algebraic variety
′
V = z ∈ Cn−1 : P1 (′ z) = P2 (′ z) = . . . = Pn−1 (′ z) = 0
contains ′ 0 ∈ Cn−1 as an isolated point. In particular, there exist constants cj > 0 for
1 ≤ j ≤ n − 1 such that
P (′z,′ z̄) = c1 |z1 |mn + c2 |z2 |mn−1 + . . . + cn−1 |zn−1 |m2 + mixed terms
where the phrase ‘mixed terms’ denotes a sum of weight one monomials annihilated by at
least one of the natural quotient maps C[′ z,′ z̄] → C[′ z,′ z̄]/(zj z̄k ) for 1 ≤ j, k ≤ n − 1,
j 6= k.
′
As in theorem 1.1, the limit map F : D∞ → D∞ is holomorphic, nonconstant and hence
′ n
algebraic. Since D∞ ⋍ B , it is possible to show that F extends holomorphically across
0 ∈ ∂D∞ and F (0) = 0′ ∈ ∂D∞ ′
. The explicit description of the weighted homogeneous
′ ′
polynomial P ( z, z) follows from working with the extended mapping near the origin.
On the other hand, it is also natural to consider the case of CR mappings from strongly
pseudoconvex hypersurfaces and we have both a global and a local version – in the spirit
of theorem 1.1, for this case. First recall that a domain D ⊂ Cn is said to be regular at
p ∈ ∂D if there is a pair of open neighbourhoods V ⊂ U of p, constants M > 0, 0 < α ≤ 1
and β > 1 such that for any ζ ∈ V ∩ ∂D, there is a function φζ which is continuous on
U ∩ D, plurisubharmonic on U ∩ D and satisfies
−M|z − ζ|α ≤ φζ (z) ≤ −|z − ζ|β
4 G. P. BALAKUMAR AND KAUSHAL VERMA

for all z ∈ U ∩ D. It is known that the class of regular points includes open pieces of
strongly pseudoconvex boundaries, smooth weakly pseudoconvex finite type pieces in C2 ,
those that are smooth convex finite type in Cn (see [30]) and finally smooth pseudoconvex
finite type boundaries in Cn (see [17]). D is said to be regular if each of its boundary
points is regular.
Theorem 1.4. Let D ⊂ Cn be a bounded regular domain, D ′ ⊂ Cn a possibly unbounded
domain and f : D → D ′ a proper holomorphic mapping. Let p ∈ ∂D be a C 2 strongly
pseudoconvex point and p′ ∈ ∂D ′ be such that the boundary ∂D ′ is C ∞ smooth pseudocon-
vex and of finite type near p′ . Suppose that the Levi rank of ∂D ′ at p′ is n − 2 and assume
that p′ ∈ clf (p), the cluster set of p. Then p′ is also a strongly pseudoconvex point.
This fits in the paradigm, observed earlier by many other authors (for example [25]), that
a proper mapping does not increase the type of a boundary point. Note furthermore that
there are no other assumptions on the defining function for ∂D ′ near p′ as in theorem 1.1
except the Levi rank condition. The difficulty created by the lack of this assumption as
explained above is circumvented by the global properness of f – indeed, it is possible to
scale f to get a holomorphic limit F : Bn → D∞ ′
where D∞ ′
is as in (1.2). Now using the
fact that p ∈ ∂D is strongly pseudoconvex and hence regular, it is possible to peel off a
local correspondence from the global one f −1 : D ′ → D that extends continuously up to
∂D ′ near p′ and contains p in its cluster set by [13]. This local correspondence can be
scaled, using the Schwarz lemma for correspondences from [45] and the behaviour of the
scaled balls in the Kobayashi metric from [37]. This gives a well defined correspondence
′
from D∞ with values in Bn and this turns out to be the inverse for F . Thus F : Bn → D∞′
′ ′
is proper and this is sufficient to conclude that p ∈ ∂D must be strongly pseudoconvex.
Corollary 1.5. Let f : M → M ′ be a nonconstant Lipschitz CR mapping between real
hypersurfaces in Cn . Fix p ∈ M and p′ = f (p) ∈ M ′ . Suppose that M is C 2 strongly
pseudoconvex near p and that there is an open neighbourhood U ′ ⊂ Cn of p′ and a C ∞
smooth defining function r ′ (z ′ ) for U ′ ∩ M ′ in U ′ which is plurisubharmonic on the pseu-
doconvex side of M ′ near p′ . If the Levi rank of M ′ at p′ is at least n − 2 then p′ is a
strongly pseudoconvex point.
The first author would like to thank Hervé Gaussier for very patiently listening to the
material presented here.

2. Proof of theorem 1.1

Consider a smooth pseudoconvex finite type hypersurface M ⊂ Cn . Associated to each
p ∈ M are two well known invariants: one is the D’Angelo type
∆(p) = (∆n (p), ∆n−1 (p), . . . , ∆1 (p))
where the integer ∆q (p) is the q-type of M at p and is a measure of the maximal order
of contact of q dimensional varieties with M at p. To recall the definition (see [23]), let r
be a local defining function for M near p and let r̃(z) = r(z + p). Then for 1 ≤ q ≤ n,
∆q (p) = inf sup{ν(r̃ ◦ L ◦ τ )/ν(τ )}
L τ

where the infimum is taken over all linear embeddings L : Cn−q+1 → Cn and the supremum
is taken over all germs of holomorphic curves τ : (C, 0) → (Cn−q+1 , 0) mapping the
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 5

origin in C to the origin in Cn−q+1 and ν(f ) denotes the order of vanishing of f at the
origin. The smoothness of M at p implies that ∆n (p) = 1 and it can be seen that
2 ≤ ∆n−1 (p) ≤ ∆n−2 (p) ≤ . . . ≤ ∆1 (p) < ∞.
To quickly recall the Catlin multitype of M at p (see [16]), let Γn be the collection
of n-tuples of reals m = (m1 , m2 , . . . , mn ) such that 0 < m1 ≤ m2 ≤ . . . ≤ mn ≤ ∞.
Order Γn lexicographically. An element m of Γn is called distinguished provided there is
a holomorphic coordinate system w = φ(z) around p with mapped to the origin such that
if
(α1 + β1 )/m1 + (α2 + β2 )/m2 + . . . + (αn + βn )/mn < 1
β
then D α D r ◦ φ−1 (0) = 0; here α = (α1 , α2 , . . . , αn ) and β = (β1 , β2 , . . . , βn ) are n-tuples
β
and D α and D are the partial derivatives
∂ |α| /∂z1α1 ∂z2α2 . . . ∂znαn and ∂ |β| /∂ z̄1β1 ∂ z̄2β2 . . . ∂ z̄nβn
respectively. The Catlin multitype M(p) = (m1 , m2 , . . . , mn ) of M at p is defined to be
the largest amongst all distinguished elements. Since r(z) is a smooth defining function
for M near p, it can be seen that the first entry in M(p) is always one. If M(p) is finite,
i.e., mn < ∞, then there is a coordinate system around p = 0 such that the defining
function is of the form
(2.1) r(z) = 2ℜzn + P (′ z,′ z) + R(z)
where P (′z,′ z) is a (1/mn , 1/mn−1 , . . . , 1/m2 ) homogeneous polynomial of total weight
one which is plurisubharmonic and does not contain pluriharmonic terms and

γ
(2.2) |R(z)| . |z1 |mn + |z2 |mn−1 + . . . + |zn |m1
for some γ > 1. The homogeneity of P (′ z,′ z) with total weight one mentioned above
means that for t ≥ 0, P ◦ πt (′ z) = tP (′ z,′ z) where
πt (z1 , z2 , . . . , zn−1 ) = (t1/mn z1 , t1/mn−1 z2 , . . . , t1/m2 zn−1 ).
Other total weights µ > 0 occur when tµ (instead of t) can be factored from P ◦ πt (′ z).
The plurisubharmonicity of P (′ z,′ z) implies that each mk for 2 ≤ k ≤ n must be even.
Thus the variable zi is assigned a weight of 1/mn−i+1 for 1 ≤ i ≤ n − 1 and by definition
the weight of the monomial ′ z J ′ z K is
(j1 + k1 )/mn + (j2 + k2 )/mn−1 + . . . + (jn−1 + kn−1 )/m2
for (n − 1)-multiindices J = (j1 , j2 , . . . , jn−1 ) and K = (k1 , k2 , . . . , kn−1 ). A basic relation
between ∆(p) and M(p) proved in [16] is that mn+1−q ≤ ∆q for all 1 ≤ q ≤ n.
Call p ∈ M an h-extendible (or semi-regular) point if M(p) is finite and ∆(p) = M(p).
This happens if and only if (see [47]) there is a (1/mn , 1/mn−1, . . . , 1/m2 ) homogeneous
C 1 smooth real function a(′ z) on Cn−1 \ {0} such that a(′ z) > 0 whenever ′ z 6= 0 and
P (′ z,′ z) − a(′ z) is plurisubharmonic on Cn−1 and among other things, this is equivalent
to the model domain D∞ (as in (1.1)) being of finite type. We shall henceforth assume
that p = 0 and p′ = 0′ and that the respective defining functions r(z) and r ′ (z ′ ) satisfy
∂r/∂zn (0) 6= 0 and ∂r ′ /∂zn′ (0′ ) 6= 0. The holomorphic map f : D → D ′ is not necessarily
proper, but Hölder continuous with exponent δ ∈ (0, 1) on D near 0 ∈ ∂D by [42] and
f (0) = 0′ . Thus the assumption that f is Lipschitz is stronger than what is apriori known.
6 G. P. BALAKUMAR AND KAUSHAL VERMA

2.1. The Scaling Method applied to (D, D ′, f ). For z ∈ D close to the origin, note
that
dist(z, ∂D) . dist(f (z), ∂D ′ ) . dist(z, ∂D)
where the inequality on the right follows since f admits a Lipschitz extension to D near
the origin, while the left inequality follows by applying the Hopf lemma to r ′ ◦ f (z) which
is a negative plurisubharmonic function on D. To scale D, choose a sequence of points
pν = (′ 0, −δν ) in D along the inner normal at the origin, where δν > 0 and δν ց 0. Let
T ν be the dilation defined by
T ν : (z1 , z2 , . . . , zn−1 , zn ) 7→ (δν−1/mn z1 , δν−1/mn−1 z2 , . . . , δν−1/m2 zn−1 , δν−1 zn )
and note that T ν (pν ) = (′ 0, −1) while the domains Dν = T ν (D) are defined by
rν = δν−1 r ◦ (T ν )−1 (z) = 2ℜzn + P (′ z,′ z) + δν−1 R ◦ (T ν )−1 (z)
where

γ
δν−1 R ◦ (T ν )−1 (z) . δνγ−1 |z1 |mn + |z2 |mn−1 + . . . + |zn−1 |m2 + |zn |
by (2.1). On each compact set in Cn this error term converges to zero since γ > 1 and
hence the sequence of domains Dν converges in the Hausdorff metric to
D∞ = z ∈ Cn : 2ℜzn + P (′ z,′ z) < 0 .


Let r∞ (z) = 2ℜzn + P (′z,′ z). To scale D ′ recall that by [18], for each ζ near 0′ ∈ ∂D ′
there is a unique polynomial automorphism Φζ (z) : Cn → Cn with Φζ (ζ) = 0 such that
n−1 n−1 X 
X X X

r(Φ−1
ζ (z)) = r(ζ) + 2ℜzn + ajk (ζ)z1j z k1 + 2
|zα | + ℜ bαjk (ζ)z1j z k1 zα
j+k≤2m α=1 α=2 j+k≤m
j,k>0 j,k>0

+ O |zn ||z| + |z∗ |2 |z| + |z∗ ||z1 |m+1 + |z1 |2m+1 .

where for z = (z1 , . . . , zn ) ∈ Cn , we denote z∗ = (z2 , . . . , zn−1 ) ∈ Cn−2 . These automor-

phisms converge to the identity uniformly on compact subsets of Cn as ζ → 0. Further-
more, if for ζ = (ζ1 , ζ2 , . . . , ζn ) ∈ D ′ as above we consider the point ζ̃ = (ζ1 , ζ2 , . . . , ζn + ǫ)
where ǫ > 0 is chosen to ensure that ζ̃ ∈ ∂D ′ , then the actual form of Φζ (z) shows that
Φζ̃ (ζ) = (0, . . . , 0, −ǫ) – since the explicit description of these automorphisms will come
up later, we shall be content at this stage with merely collecting the relevant properties
needed to describe the scaling of D ′ . To define the distinguished polydiscs around ζ (more
precisely, biholomorphic images of polydiscs), let
Al (ζ) = max{|ajk (ζ)| : j + k = l}, 2 ≤ l ≤ 2m′
and
Bl′ (ζ) = max{|bαjk (ζ)| : j + k = l′ , 2 ≤ α ≤ n − 1}, 2 ≤ l′ ≤ m′ .
For each δ > 0 define

1/l
1/l′
τ (ζ, δ) = min δ/Al (ζ) , δ 1/2 /Bl′ (ζ) : 2 ≤ l ≤ 2m′ , 2 ≤ l′ ≤ m′


Since the type of ∂D ′ at the origin is 2m′ it follows that A2m′ (0) > 0 and hence A2m′ (ζ)
is positive for all ζ sufficiently close to the origin. Thus
′
δ 1/2 . τ (ζ, δ) . δ 1/2m
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 7

for ζ close to the origin – the upper bound being a consequence of the non-vanishing of
A2m′ (ζ) near the origin while the lower bound follows since the greatest possible exponent
of δ in the definition of τ (ζ, δ) is 1/2. Set
τ1 (ζ, δ) = τ (ζ, δ) = τ, τ2 (ζ, δ) = . . . = τn−1 (ζ, δ) = δ 1/2 , τn (ζ, δ) = δ
and define
R(ζ, δ) = {z ∈ Cn : |zk | < τk (ζ, δ), 1 ≤ k ≤ n}
which is a polydisc around the origin in Cn with polyradii τk (ζ, δ) along the zk direction
for 1 ≤ k ≤ n and let
Q(ζ, δ) = Φ−1


ζ R(ζ, δ)
which is a distorted polydisc around ζ. It was shown in [43] that these domains satisfy
the engulfing property, i.e., for all ζ in a small fixed neighbourhood of the origin, there
is a uniform constant C > 0 such that if η ∈ Q(ζ, δ), then Q(η, δ) ⊂ Q(ζ, Cδ) and
Q(ζ, δ) ⊂ Q(η, Cδ).
Consider the sequence p′ν = f (pν ) ∈ D ′ that converges to the origin and denote by w ′ν
the point on ∂D ′ chosen such that if p′ν = (p′ν ′ν ′ν ′ν ′ν ′ν ′ν
1 , p2 , . . . , pn ) then w = (p1 , p2 , . . . , pn +γν )
for some γν > 0. Note that
γν ≈ dist(p′ν , ∂D ′ )
for all large ν. Hence
δν = dist(pν , ∂D) ≈ dist(p′ν , ∂D ′ ) ≈ γν
for all large ν. Let g ν = Φw′ν (·) be the polynomial automorphism of Cn corresponding to
w ′ν ∈ D ′ as described above. Let us consider the holomorphic mappings
f ν = g ν ◦ f : D → g ν (D ′)
and define a dilation of coordinates in the target space by
B ν : (z1′ , z2′ , . . . , zn′ ) 7→ (τ1ν )−1 z1′ , (τ2ν )−1 z2′ , . . . , (τnν )−1 zn′

where τ1ν = τ (w ′ν , γν ), τjν = γ1/2

ν
for 2 ≤ j ≤ n − 1 and τnν = γν . Let D ν = T ν (D) and
D ′ν = (B ν ◦ g ν )(D ′ ) be the scaled domains and the scaled maps between them are
F ν = B ν ◦ f ν ◦ (T ν )−1 : Dν → Dν′ .
To understand the Hausdorff limit of the domains Dν′ , note first that B ν ◦g ν (p′ν ) = (′ 0, −1),
which implies that F ν (′ 0, −1) = (′ 0, −1) for all ν, and that rν′ , the defining function for
Dν′ is given by
n−1
X n−1
X
γν−1 r ′ ◦ (B ν ◦ g ν )−1 (z) = 2ℜzn + Qν (z1 , z 1 ) + |zα |2 + ℜ Sνα (z1 , z 1 )zα + O(τ1ν )


α=2 α=2

where X
Qν (z1 , z 1 ) = ajk (w ′ν )γν−1 (τ1ν )j+k z1j z k1
j+k≤2m′
j,k>0
and X
Sνα (z1 , z 1 ) = bαjk (w ′ν )γν−1/2 (τ1ν )j+k z1j z k1 .
j+k≤m′
j,k>0
8 G. P. BALAKUMAR AND KAUSHAL VERMA

By the definition of Al , Bl′ and τ1ν it follows that the largest coefficient in both Qν and
Sνα is at most one in modulus. It was shown in [18] that there exists a uniform ǫ > 0 such
that
bαjk (w ′ν )γν−1/2 (τ1ν )j+k . (τ1ν )ǫ
for all possible indices j, k, α and all large ν. Therefore some subsequence of this family
of defining functions converges together with all derivatives on compact sets to
′
r∞ (z) = 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2
where Q2m′ (z1 , z 1 ) is a polynomial of degree at most 2m′ without harmonic terms. Hence
the domains Dν′ converge to
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2 < 0

D∞
which, being the smooth limit of pseudoconvex domains, is itself pseudoconvex. In par-
ticular, it follows that Q2m′ (z1 , z 1 ) is subharmonic.
It is known that analytic discs in a bounded regular domain in Cn satisfy the so-called
attraction property (see [11]), i.e., if the centre of a given disc is close to a boundary
point, then a given subdisc around the origin cannot wander too far away from the same
boundary point. A quantitative version of this was proved by Berteloot-Coeuré in [14]
and forms the basis for controlling families of scaled mappings in C2 . Using ideas from
[12] and [18], the following analogue was proved in [43] and will be useful in this situation
as well – we include the statement for the sake of completeness.
Proposition 2.1. Let Ω ⊂ Cn be a bounded domain. Let ζ0 ∈ ∂Ω and suppose there is
an open neighbourhood V such that V ∩ ∂Ω is C ∞ smooth pseudoconvex of finite type and
the Levi rank is at least n − 2 on V ∩ ∂Ω. Fix a domain ω ⊂ Cm .
Then for any fixed point z0 ∈ ω and a compact K ⊂ ω containing z0 , there exist
constants ǫ(K), C(K) > 0 such that for any ξ ∈ V ∩ ∂Ω and 0 < ǫ < ǫ(K), every
holomorphic mapping F : ω → Ω with
|F (z0 ) − ζ0 | < ǫ(K) and F (z0 ) ∈ Q(ξ, ǫ)
also satisfies F (K) ⊂ Q(ξ, C(K), ǫ).
Now let {Kj } be an increasing sequence of relatively compact domains that exhaust D∞
such that each contains the base point (′ 0, −1). Fix K = Kµ and let f˜ν = f ◦ (T ν )−1 .
Then
f˜ν (′ 0, −1) = f (′ 0, −δν ) = f (pν ) = p′ν
and hence f˜ν (′ 0, −1) → 0 ∈ ∂D ′ . In particular, f˜ν (′ 0, −1) = p′ν ∈ Q(w ′ν , 2γν ) for all large
ν, by the construction of these distorted polydiscs. By the previous proposition
f˜ν (K) ⊂ Q(w ′ν , C(K)γν )
and therefore
F ν (K) = B ν ◦ g ν ◦ f˜ν (K) ⊂ B ν ◦ g ν (Q(w ′ν , C(K)γν ).
However,
B ν ◦ g ν (Q(w ′ν , C(K)γν ) = B ν (R(w ′ν , C(K)γν ))
which by definition is contained in a polydisk around the origin with polyradii
rk = τk (w ′ν , C(K)γν )/τk (w ′ν , γν )
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 9

for 1 ≤ k ≤ n. Note that rn = C(K) and that rk = (C(K))1/2 for 2 ≤ k ≤ n − 1. Since

C(K) > 1, and this may be assumed without loss of generality, the definition of τ (ζ, δ)
shows that for δ ′ < δ ′′ ,
′
(δ ′ /δ ′′ )1/2 τ (ζ, δ ′′ ) ≤ τ (ζ, δ ′ ) ≤ (δ ′ /δ ′′ )1/2m τ (ζ, δ ′′).
Therefore
τ1 (w ′ν , C(K)γν )/τ1 (w ′ν , γν ) ≤ (C(K))1/2
and hence r1 ≤ (C(K))1/2 . Thus {F ν } is uniformly bounded on each compact set in D∞
′
and is therefore normal. Let F : D∞ → D∞ be a holomorphic limit of some subsequence in
{F } and since F ( 0, −1) = ( 0, −1) for all ν by construction, it follows that F (′ 0, −1) =
ν ν ′ ′

(′ 0, −1). The maximum principle shows that F (D∞ ) ⊂ D∞ ′

.
Let R > 0 be arbitrary and fix z ∈ D∞ ∩ B(0, R). Note that since f preserves the
distance to the boundary, it follows that
(rν′ ◦ F ν )(z) = γν−1 (r ′ ◦ f ◦ Tν−1 )(z) . (δν /γν ) rν (z)
and hence
′
r∞ ◦ F (z) . r∞ (z)
since the constants appearing in the first estimate are independent of ν. This shows that
′
the cluster set of a finite boundary point of D∞ does not intersect D∞ and therefore F is
non-degenerate.

2.2. Boundedness of F ν (0). A crucial point in the proof of theorem 1.1 is to show that
near the origin, the family {F ν } of the scaled maps is uniformly Hölder continuous upto
the boundary i.e., the Hölder constant and exponent are stable under the scaling. A first
step in establishing this is to show that the images of the origin under these scaled maps
is stable.
Proposition 2.2. The sequence {F ν (0)} is bounded.
Proof. Recall the scaled maps
F ν = B ν ◦ g ν ◦ f ◦ (T ν )−1
where B ν , T ν were linear maps and f (0) = 0. So F ν (0) = (B ν ◦ g ν ◦ f )(0) = (B ν ◦ g ν )(0).
Now let us recall the explicit form of the map g ν = Φw′ν from [43], which is a polynomial
automorphism of Cn that reduces the defining function to a certain normal form as stated
in section 2.1. Let us recall this reduction procedure for any given domain Ω in Cn that is
smooth pseudoconvex of finite type 2m and of Levi rank (at least) n − 2 on a (small) open
boundary piece Σ in ∂D∞ . Let r be a smooth defining function for Σ with ∂r/∂zn (z) 6= 0
for all z in a small neighbourhood U in Cn of Σ such that the vector fields
Ln = ∂/∂zn , Lj = ∂/∂zj + bj (z, z̄)∂/∂zn

−1
where bj = ∂r/∂zn ∂r/∂zj , form a basis of CT (1,0) (U) and satisfy Lj r ≡ 0 for 1 ≤ j ≤
¯
n − 1 and ∂ ∂r(z)(Li , L̄j )2≤i,j≤n−1 has all its eigenvalues positive for each z ∈ U.

The reduction now consists of five steps – for ζ ∈ U, the map Φζ = φ5 ◦ φ4 ◦ φ3 ◦ φ2 ◦ φ1

where each φj is described below.
10 G. P. BALAKUMAR AND KAUSHAL VERMA

φ1 is the affine map given by

n−1
X n−1
X
φ1 (z1 , . . . , zn ) = (z1 − ζ1 , . . . , zn−1 − ζn−1 , zn + bζj zj − (ζn + bζj ζj )).
j=1 j=1

where the coefficients bζj = bj (ζ, ζ̄) are clearly smooth functions of ζ on U. Therefore, φ1
translates ζ to the origin and
r(φ−1(z)) = r(ζ) + ℜzn + terms of higher order.
where the constant term disappears when ζ ∈ Σ.
The remaining reductions remove the occurrence of harmonic (not just pluriharmonic)
monomials in the variables z1 , . . . , zn−1 of weights upto one in the weighted homogeneous
expansion of the defining function with respect to the weights given by the multitype for
Σ, i.e., the variable z1 is assigned a weight of 1/2m, zn a weight of 1 while the others are
assigned 1/2 each. Now, since the Levi form restricted to the subspace
L∗ = spanCn hL2 , . . . , Ln−1 i
(1,0)
of Tζ (∂Ω) is positive definite, we may diagonalize it via a unitary transform φ2 and
a dilation φ3 will then ensure that the quadratic part involving only z2 , z3 , . . . , zn−2 in
the Taylor expansion of r is |z2 |2 + |z3 |2 + . . . + |zn−2 |2 . The entries of the matrix that
represents the composite of the last two linear transformations are smooth functions of ζ
and in the new coordinates still denoted by z1 , . . . zn , the defining function is in the form
n−1 X
X m n−1
X
(aαj z1j bαj z̄1j )zα cα zα2


(2.3) r(z) = r(ζ) + ℜzn + ℜ + +ℜ
α=2 j=1 α=2

X n−1
X n−1
X X
aj,k z1j z̄1k + |zα |2 + ℜ bαj,k z1j z̄1k zα


+
2≤j+k≤2m α=2 α=2 2≤j+k≤m

+ O(|zn ||z| + |z∗ |2 |z| + |z∗ ||z1 |m+1 + |z1 |2m+1 )

A change in the normal variable zn to absorb the pluriharmonic terms here i.e., z1k , z̄1k , zα2 as
well as z1k zα , z̄1k z̄α , can be done according to the following standard change of coordinates
zj = tj (1 ≤ j ≤ n − 1),
zn = tn − P1 (t1 , . . . , tn−1 )
where
2m
X n−1 X
X m n−1
X
P1 (t1 , . . . , tn−1 ) = ak0 tk1 − aαk tα tk1 − cα t2α
k=2 α=2 k=1 α=2

with coefficients that are smooth functions of ζ.

Finally, just as we absorbed into the simplest pure term ℜzn that occurs in the Taylor
expansion, other pure terms not divisible by this variable in the last step, we may absorb
into the simplest of non-harmonic monomials that occur there namely, |zα |2 where 2 ≤
α ≤ n − 1 other non-harmonic terms of degree > 2 not divisible by them. Let us do this
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 11

to those of weight at most one, remaining in (2.3) rewritten in the t-coordinates, which
are of the form t̄j1 tα by applying the transform

t1 = w1 , tn = wn ,
tα = wα − P2 (w1 ) (2 ≤ α ≤ n − 1)

where
m
X
P2 (w1 ) = bαk w1k
k=1

with coefficients smooth in ζ, as before (since all these coefficients are simply the deriva-
tives of some order of the smooth defining function r evaluated at ζ). We then have the
sought for simplification of the Taylor expansion.

Now, suppose that we already have the reduced form holding at the origin, then (since
no further normalization would be required) Φ0 may be taken to be the identity. Note
that the normalizing map Φζ is not uniquely determined – even among the class of all
maps of the same form – owing to the (only) ambiguity in the choice of the diagonalizing
map φ2 ; however, this choice can certainly be done in a manner such that the coefficients
of that unitary matrix are smooth in ζ and satisfying the ‘initial condition’ that φ2 for
the origin is the identity. Thus in all, the map Φζ is smooth in the parameter ζ with Φ0
being the identity map. This implies that the family Φζ (·) is uniformly Lipschitz (where
ζ ∈ U) and converges uniformly on compact subsets of Cn to the identity as ζ approaches
the origin as mentioned in section 2.1. Next, the consequence of the simpler fact that the
map Φ(ζ, 0) = Φζ (0) is Lipschitz in a neighbourhood of the origin, to our setting is that

|g ν (0)| ≤ C1 |w ′ν |

with C1 independent of ν. Now

 
F ν (0) = B ν ◦ g ν (0) = (τ1ν )−1 (g ν (0))1 , . . . , (τn−1
ν
)−1 (g ν (0))n−1, (τnν )−1 (g ν (0))n .

1/2 1/2
From the fact that τ1ν & γν and τjν = γν for 2 ≤ j ≤ n − 1 while τnν = γν , we get that

B ν ◦ g ν (0) ≤ C2 |w ′ν |/γν

with C2 again independent of ν. Now, since f is Lipschitz upto M and δν . γν we have

|w ′ν | ≤ dist(w ′ν , p′ν ) + dist(p′ν , 0′)

= dist(p′ν , ∂D ′ ) + |f (pν ) − f (0)|
≤ C3 (γν + |pν |)
= C3 (γν + δν )
≤ C4 γ ν

with constants independent of ν as before, implying that {F ν (0)} is bounded.

12 G. P. BALAKUMAR AND KAUSHAL VERMA

2.3. Stability of the Kobayashi metric. For Ω, Σ, U with 0 ∈ Σ, as in the previous

section, recall from [43] (see also [12]), the M-metric defined for ζ ∈ U ∩ Ω by
Xn


MΩ (ζ, X) = | DΦζ (ζ)X k |/τk (ζ, ǫ(ζ)) = D(Bζ ◦ Φζ )(ζ)(X) l1
k=1
ǫ(ζ)
with Bζ = Bζ where ǫ(ζ) > 0 is such that ζ + (0, . . . , 0, ǫ(ζ)) lies on Σ and
Bζδ (z1 , . . . , zn ) = (τ1 )−1 z1 , . . . , (τn )−1 zn

where τ1 = τ (ζ, δ), τj = δ 1/2 for 2 ≤ j ≤ n − 1 and τn = δ. Now let us get to our scaled
domains Dν′ = B ν ◦ g ν , where we recall B ν = Bp′ν and g ν = Φw′ν and denote by ν Φζ
and ν Bζ the normalizing map and the dilation respectively, associated to (Dν′ , ζ); we shall
drop the left superscript ν for the domain D ′ . Since all the coefficients of the polynomial
automorphisms ν Φζ (·) depend smoothly on the derivatives of the defining functions rν ,
which converge in the C ∞ -topology on U, there exists L > 1 such that
1/L < | det D(ν Φζ (z) | < L


(2.4)
for all ν and ζ ∈ U. The main result of this section is the following stability theorem for
the Kobayashi metric.
Theorem 2.3. There exists a neighbourhood U of the origin such that
KDν′ (z, X) ≥ CMDν′ (z, X)
for all z ∈ U ∩ Dν′ , ν ≫ 1 where C is a positive constant independent of ν.
The proof of this requires two steps and follows the lines of the argument presented in
[43]. The first step consists in uniformly localizing the Kobayashi metric of the scaled
domains near the origin which translates to verifying the following uniform version of the
attraction of analytic discs near a local plurisubharmonic peak point.
Lemma 2.4. There exists a neighbourhood V ⊆ U of the origin and δ > 0 independent of j
such that for any j large enough and any analytic disc f : ∆ → Dj′ with f (0) = ζ ∈ V ∩D ′j ,
we have
f (∆δ ) ⊆ U ∩ Dj′
Proof. If the lemma were false, then for any neighbourhood V ⊂ U and any δ > 0, there
exists a sequence of analytic discs fj : ∆ → Dj′ with fj (0) = ζ j ∈ V ∩ Dj′ , converging to
the origin but fj (∆δ ) * U. Consider
f˜j = S −1 ◦ fj : ∆ → D ′
j j

where Sj = B j ◦ Φw′j are the scaling maps. Then ζ̃ j = f˜j (0) = Sj−1 (ζ j ) converges to the
origin. Indeed, the family Sj−1 = Φ−1wj
◦ (B j )−1 is equicontinuous at the origin since their
derivatives are
D(Φ−1 j −1
w j ) ◦ (B )
with D(Φ−1 w j ) being bounded in a neighbourhood of the origin and (B )
j −1
converging to
′
the zero map. Now, 0 ∈ ∂D being a plurisubharmonic peak point by [17], the simplest
version of the attraction property of analytic discs (see for instance lemma 2.1.1 in [33]),
gives for all j ≫ 1 that
f˜j (∆γ ) ⊆ U ∩ D ′
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 13

for some γ ∈ (0, 1). Rescale the f˜j ’s, so that we may take γ = 1. Now recall the sharper
version of the attraction property given by lemma 3.6 of [43] namely,
Lemma 2.5. Let W ⋐ U be a neighbourhood of the origin. There exist constants
α, A ∈ [0, 1] and K ≥ 1 such that for any analytic disc f : ∆N → U that satisfies
MD′ (f (t), f ′(t)) ≤ A, f (0) ∈ W and K N −1 ǫ(f (0)) < α also satisfies,
 
N
f (∆N ) ⊂ Q f (0), K ǫ(f (0))

We intend to apply this lemma to the analytic discs g̃j (t) = f˜j (r0 t) for which we only
need to verify
MD′ g̃j (t), g̃j′ (t) ≤ A

since g̃j (0) = f˜j (0) converges to the origin. Then

MD′ g̃j (t), g̃j′ (t) = MD′ f˜j (r0 t), r0 f˜j′ (r0 t)

= r0 MD′ f˜j (r0 t), f˜′ (r0 t)

j

≤ r0 C5 KD′ f˜j (r0 t), f˜j′ (r0 t)

by theorem 3.10 of [43], where C5 is a positive constant independent of j. At the last

step, we use the fact that f˜j ’s map ∆ into so small a neighbourhood of the origin where
KD′ ≈ MD′ . As the Kobayashi metric decreases under holomorphic mappings, we have
r0 C5 KD′ f˜j (r0 t), f˜j′ (r0 t) ≤ r0 C5 K∆ t, ∂/∂t

So, if we choose r0 such that

 
r0 C5 sup K∆ (t, ∂/∂t) ≤ A,
|t|≤r0

we will have completed the required verification of the hypothesis for the discs gj : ∆ →
D ′ ∩ U and the aforementioned lemma gives
f˜j (∆r0 ) = g̃j (∆) ⊂ Q g̃j (0), ǫ(g̃j (0)) = Q f˜j (0), ǫ(f˜j (0)) .



(2.5)
At this point observe also that r0 does not depend on δ. Next, fix any compact subdisc
K ⊂ ∆r0 and note by (2.5) that,
f˜j (K) ⊂ Q f˜j (0), ǫ(f˜j (0))

and subsequently,
(Sj ◦ f˜j )(K) ⊂ Sj Q(f˜j (0)), ǫ(f˜j (0))



−1 n
= (B j ◦ Φw′j ) ◦ Bζ̃ j ◦ Φζ˜′ j (∆ )
 
= Bp′j ◦ Φw′j ◦ Φ−1 ◦ Bζ̃−1 (∆n )


(2.6) ˜′ j j
ζ

where ζ̃ ′j = ζ̃ j + (0, . . . , 0, ǫ(ζ̃ j )). Now the fact that ǫ(ζ̃ j ) ≈ ǫ(pj ), as both pj and ζ̃ j
converge to the origin and corollary 2.8 of [18] – the fact that the distinguished radius τ
at any two points in U are comparable – together imply the boundedness of the sequence
τ (ζ̃ ′j , ǫ(ζ̃ j ))/τ (w ′j , ǫ(pj )), by C6 say. This when combined with (2.4) and (2.6), yields the
14 G. P. BALAKUMAR AND KAUSHAL VERMA

stability of our discs fj on compact subdiscs in ∆ i.e., we have for a positive constant
CK > C6 L2 (which may depend on K but not on j), that
fj (K) = (Sj ◦ f˜j )(K) ⊆ ∆√C × . . . × ∆√C × ∆C
K K K

Hence by Montel’s theorem, fj ’s converge to a map f∞ : ∆r0 → D ′ ∞ , where f∞ (0) = 0 ∈

′
∂D∞ . If φ is a local plurisubharmonic peak function at the origin, then the maximum
principle applied to the subharmonic function ψ = φ ◦ f∞ implies that f∞ ≡ constant,
since φ peaks precisely at one point. This contradicts our assumption that fj (∆δ ) * U
(as soon as δ is taken smaller than r0 ) and completes the proof of the lemma. 
Remark 2.6. The same line of argument gives proposition 2.1 by replacing the discs by
balls in Cm , covering any given compact K ⊂ ω by balls and using the engulfing property
of the special polydics Q.
The second and more technical step, is a quantitative form of the Schwarz lemma at the
boundary which generalizes lemma 3.6 of [43] mentioned earlier. To do this we require a
stable version of the engulfing property for the distinguished polydiscs Qν assosciated to
the scaled domain Dν′ . Fix a pair of neighbourhoods 0 ∈ W ⋐ V ⋐ U.
Lemma 2.7. There exist constants α, A ∈ [0, 1] and K > 1 such that for every analytic
disc
f : ∆N → V ∩ Dν′
that satisfies MD′ν f (t), f ′ (t) < A, f (0) ∈ W and K N −1 ǫ(f (0)) < α also satisfies,




f (∆N ) ⊂ Qν f (0), K N ǫ f (0) .

Recall the definition of Al (z) and τ (z, δ) from section 2.1. Let Aνl (z) and τ ν (z, δ) de-
note the corresponding quantities for the domain Dν′ and A∞ ∞
l (z) and τ (z, δ) that for
D∞ . Then it is clear that Al (z) → Al (z) and consequently τ (z, δ) → τ ∞ (z, δ), both
′ ν ∞ ν

convergences being uniform for z in V . Let δ > 0 be given. Let

l
M = sup τ ν (z, δ) : z ∈ V, 2 ≤ l ≤ 2m′ , ν ≥ 1


and 0 < ǫ < δ/M. Then there exists by the uniform convergence of the A’s, an Nδ
independent of z ∈ V such that for all 2 ≤ l ≤ 2m and all ν > Nδ ,
(2.7) A∞ ν ∞
l (z) − ǫ < Al (z) < Al (z) + ǫ

The right inequality gives for all 2 ≤ l ≤ 2m that

l
l
l
τ ∞ (z, δ) Aνl (z) < τ ∞ (z, δ) A∞ ∞
l (z) + τ (z, δ) ǫ < δ + δ

by the definition of τ ∞ and ǫ . The definition of τ ν now makes this read as

(2.8) τ ∞ (z, δ) < 21/2 τ ν (z, δ)
for all ν > Nδ and z ∈ V . Meanwhile the left inequality at (2.7) similarly gives,

l
l ν
l
τ ν (z, δ) A∞ ν ν
l (z) < τ (z, δ) A (z) + τ (z, δ) ǫ < 2δ

which implies τ ν (z, δ) < 21/2 τ ∞ (z, δ). Combining this with corollary 2.8 of [18] applied
to τ ∞ , we get for z ∈ Q∞ (z ′ , δ) and all ν large that
(2.9) 1/C τ ν (z ′ , δ) < τ ν (z, δ) < Cτ ν (z ′ , δ)
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 15

for some C > 0, independent of z ′ and ν. This will lead to the following uniform engulfing
property of these polydiscs.
Lemma 2.8. There exists a positive constant C such that for all ν large, z ′ ∈ U and
z ′′ ∈ Qν (z ′ , δ) we have
Qν (z ′′ , δ) ⊂ Qν (z ′ , Cδ)
Qν (z ′ , δ) ⊂ Qν (z ′′ , Cδ)
Proof. Let us recall what Qν is and rewrite for instance the first statement as
ν δ

−1 n

−1 n

Bz ′′ ◦ ν Φz ′′ ∆ ⊂ ν BzCδ ν
′ ◦ Φz ′ ∆
which is equivalent to saying that the map
ν
(2.10) BzCδ ν ν
′ ◦ Φz ′ ◦ ( Φz ′′ )
−1
◦ (ν Bzδ′′ )−1
maps the unit polydisc into itself. Now recall from (2.4), that the sequence of Jacobians
of ν Φz ′ ◦ ν Φ−1 2
z ′′ is uniformly bounded above by L . Unravelling the definition of the scalings
ν Cδ ν δ
Bz ′ and Bz ′′ , then shows that the map defined in (2.10) carries the unit polydisc into
the polydisc of polyradius given by
L2 τ ν (z ′′ , δ)/τ ν (z ′ , Cδ), L2 δ 1/2 /(Cδ)1/2 , . . . , L2 δ 1/2 /(Cδ)1/2 , L2 δ/Cδ


(2.11)
Now the uniform comparability of the τ ’s obtained in (2.9) gives C7 > 0 independent of
ν and z ′ , such that
τ ν (z ′′ , δ) < C7 τ ν (z ′ , δ)
whereas looking at the definition of the τ directly gives
τ ν (z ′ , Cδ) > C 1/2m τ ν (z ′ , δ)
Combining these two we see that the first component at (2.11) is bounded above by
L2 C7 /C 1/2m . Therefore, choosing C to be such that C 1/2 > L2 and C 1/2m > L2 C7 ensures
that the map at (2.10) leaves the unit polydisc invariant, completing the verification of
the lemma. 
Before passing, let us record one more stable estimate concerning the τ ’s to be of use
later, namely
′
(2.12) τ ν (z, δ) ≤ C8 δ 1/2m
for C8 > 0 (independent of ν), which follows as in section 2.1 from the fact that the
Aν2m′ (z)’s are uniformly bounded below by a positive constant on a neighbourhood of the
origin (which as usual we may assume to be U).
Next, observing that all the constants in the calculations in the proof of lemma 3.6 of
[43] are stable owing to (2.9), lemma 2.8 and the stability of the estimates on the various
coefficients of ν Φ’s given in lemma 3.4 of [43] – because all these coefficients are nothing
but the derivatives of some order (less than 2m) of the defining functions of the scaled
domains, which converge in the C ∞ -topology on U– lemma 2.7 follows as well.
Proof of theorem 2.3. If the theorem were false, there would exist a subsequence νj of N
and ζ j ∈ D ′νj converging to the origin such that
 
KDν′ j ζ j , X MDν′ j (ζ j , X) < 1/j 2

16 G. P. BALAKUMAR AND KAUSHAL VERMA

which after re-indexing the D ′νj ’s as D ′j and using lemma 2.4 reads as :
There exists a sequence ζ j ∈ U ∩ D ′j with ζ j converging to the origin and a sequence of
analytic discs
fj : ∆ → U ∩ Dj′
such that   
j
fj (0) = ζ and fj′ (0) j
= Rj X MDj (ζ , X)
′

with Rj ≥ j 2 .
The idea as in [12] and [43] now consists of scaling these analytic discs appropriately to
enlarge their domains to discs of growing radius while ensuring their normality also, so
′
that taking their limit produces an entire curve lying inside D∞ to contradict its Brody
+
hyperbolicity. To work this out, define Mj (t) : ∆1/2 → R by
Mj (t) = MDj′ fj (t), fj′ (t)

Then
 
fj (0), fj′ (0) j j


Mj (0) = MDj′ = MDj′ ζ , Rj X MDj′ (ζ , X)
= Rj ≥ j 2
Recall the following lemma from [12].
Lemma 2.9. Let (X, d) be a complete metric space and let M : X → R+ be a locally
bounded function. Then for all σ > 0 and for all u ∈ X satisfying M(u) > 0, there exists
v ∈ X such that
(i) d(u, v) ≤ 2/σM(u)
(ii) M(v) ≥ M(u)
(iii) M(x) ≤ 2M(v) if d(x, v) ≤ 1/σM(v)
Apply this lemma to Mj (t) on ∆1/2 with u = 0 and σ = 1/j, to get aj ∈ ∆1/2 such that
|aj | ≤ 2j/Mj (0) and
(2.13) Mj (aj ) ≥ Mj (0) ≥ j 2
and furthermore,


(2.14) Mj (t) ≤ 2Mj (aj ) on ∆ aj , j/Mj (aj )
(for all j big enough so that the discs above lie inside ∆1/2 ). Now, scaling the unit disc
with respect to the points aj which approach the origin (since the contention is that the
derivatives of the fj ’s near the origin, in the M-metric, are blowing up rapidly), we get
the discs ∆j = sj (∆) where sj = bj ◦ φaj with φaj the translation that transfers aj to the
origin and bj being the map that dilates by a factor of 2A−1 Mj (aj ) with A the constant
of lemma 2.7. The scaled analytic discs gj : ∆j → U ∩ Dj′ are then given by
gj (t) = (fj ◦ s−1
j )(t) = fj (aj + cj t),

where cj = A/2Mj (aj ). Note that

gj′ (t) = cj fj′ (aj + cj t)
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 17

which implies
 
gj (t), gj′ (t) cj t), fj′ (aj


MD′j = cj Mj fj (aj + + cj t)
= cj Mj (aj + cj t)


Now note that aj + cj t lies in ∆ aj , j/Mj (aj ) for all |t| < j as 0 < A < 1 and therefore
by (2.13), (2.14) and the definition of cj , we have for t ∈ ∆j that
MD′j gj (t), gj′ (t) = cj Mj (aj + cj t)




< cj 2Mj (aj )
(2.15) =A
We wish to apply lemma 2.7 to these gj ’s. First we must that verify their centres lie close
enough to the origin; to that end write
|gj (0)| ≤ |fj (aj ) − fj (0)| + |fj (0)|
and note that the first term on the right can be made arbitrarily small by the equicon-
tinuity of the fj ’s which map into the bounded neighbourhood U. After passing to a
subsequence if necessary to ensure that ǫ(gj (0)) ≤ α/K j−1, we have with (2.15), verified
all the criteria of lemma 2.7 and therefore
 
j N
(2.16) gj (∆N ) ⊂ Q gj (0), K ǫ(gj (0))
for all j ≥ N. Let ηj = gj (0) and ηj′ = ηj + (0, . . . , 0, ǫj ) with ǫj > 0 such that rj (ηj′ ) = 0
and note that
ηj ∈ Qj ηj′ , Cǫj

for a uniform constant C ≥ 1 since ǫj ≈ rj (ηj ), uniformly in j. Consequently using lemma

2.7, (2.16) becomes
 
(2.17) gj (∆N ) ⊂ Qj ηj′ , CK N ǫj
for all j ≥ N. Now if we let
hj = j Bηj ◦ j Φηj′ ◦ gj : ∆j → S̃j (U ∩ D ′ )
where S̃j is the map
j
Bηj ◦ j Φηj′ ◦ Bp′j ◦ Φw′j
then what (2.17) translates to, for the map hj is that
hj (∆N ) ⊂ ∆(CK N )1/2 × . . . × ∆(CK N )1/2 × ∆CK N
for all N ≤ j, exactly as in the proof of lemma 2.4. Also, note that the domains S̃j (U ∩D ′ )
′
after passing to a subsequence if necessary, converge to a domain D̃∞ of the same form as
′
D∞ , by the same argument as in section 2.1 together with (2.12). Thus Montel’s theorem,
a diagonal sequence argument and an application of the maximum principle to the limit,
gives rise to an entire curve
′
h : C → D̃∞ .
Indeed, note that
hj (0) = j Bηj ◦ j Φηj′ (ηj ) = (0, . . . , 0, −1)


18 G. P. BALAKUMAR AND KAUSHAL VERMA

′
for all j and so h(0) = (0, . . . , 0, −1), which lies in D̃∞ all of whose boundary points –
including the point at infinity – are peak points. Now, to check that h is nonconstant, we
examine the sequence of their derivatives at the origin. We write Xj for gj′ (0) = cj fj′ (aj )
(which tend to 0 but are bounded below in the M-metric).
h′j (0) l1 = j Bηj′ ◦ D(j Φηj′ ) (ηj )(Xj ) l1


(2.18)
≥ j Bηj ◦ D(j Φηj ) (ηj )(Xj ) l1




= MD′j gj (0), Xj
= cj Mj (aj ) = A/2
where the lower bound here is a consequence of (2.9) applied to τ j for the points ηj and
ηj′ and the fact that the family
D(ν Φz ′ ) ◦ D(µ Φ−1
z ′′ )
is uniformly bounded below in norm. Passing to the limit in (2.18), now gives
|h′ (0)|l1 ≥ C(A/2) > 0
and we reach the contradiction mentioned earlier namely to the Brody hyperbolicity of
′
∂ D̃∞ (see for instance, lemma 3.8 of [43]). 
Finally, let us note the consequence of theorem (2.3) in the form that we shall make use
of
Corollary 2.10. There is neighbourhood U of the origin and a positive constant C such
that
1/2m′
KD′ν (z, X) ≥ C |X|/ dist(z, ∂D ′ν )
for all z ∈ U ∩ Dν′ and all ν ≫ 1.
This comes from the facts that ǫν (z) ≈ dist(z, ∂D ′ν ) and for some C9 > 0 we have for all
ν that
′
MD′ν (z, X) ≥ C9 |X|l1 /(ǫν (z))1/2m
which in turn follows from (2.12).
2.4. Uniform Hölder continuity of the scaled maps near the origin. In this sec-
tion, we recall from [20], the arguments which show that the family of scaled maps is
uniformly Hölder continuous upto the boundary – we already know that each scaled map
is Hölder continuous upto the origin.
Theorem 2.11. There exist positive constants r, C such that
′
|F ν (z ′ ) − F ν (z ′′ )| ≤ C|z ′ − z ′′ |1/2m
for any z ′ , z ′′ ∈ D̄ν ∩ B(0, r) and all large ν.
By the previous section we may assume that {F ν (0)} converges to a (finite) boundary
′
point q ∈ ∂D∞ . Recall that the origin (respectively the ℜzn axis) is a common boundary
point (respectively, the common normal at the origin) for all the scaled domains. At this
special boundary point, we first show that the family of scaled mappings is equicontinuous
– we already know that this holds on compact subsets of D∞ – or equivalently that, given
any neighbourhood of q, there exists a small ball about the origin, such that every scaled
map carries the piece of its domain intercepted by this ball into that given neighbourhood.
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 19

Then, the uniform lower bound on the Kobayashi metric near q will sharpen the uniform
boundary distance decreasing property to uniform Hölder continuity, of the scaled maps.
In particular then, the family {F ν } ∪ {F } is equicontinuous near the origin, upto the
respective boundary.

Lemma 2.12. For any ǫ > 0 there exists δ > 0, such that |F ν (z) − q| < ǫ for any
z ∈ Dν ∩ B(0, δ) and all large ν.

Proof. Suppose to obtain a contradiction, that the assertion were false. Then there exists
ǫ0 > 0 and a sequence aν ∈ Dν such that aν → 0 and |F ν (aν ) − q| ≥ ǫ0 . Since every F ν
is continuous upto the boundary ∂Dν , we can also choose a sequence of points bν ∈ Dν
lying on the common inner normal to Dν at the origin i.e., bν = (0, −βν ) with βν > 0,
such that bν → 0 and |F ν (bν ) − q| → 0. Let sν = |aν − bν |. It is not difficult to see that
there exists a constant C > 0 such that for all ν, there is a smooth path γν : [0, 3sν ] → Dν
with the following properties:
(i) γν (0) = aν , γν (3sν ) = bν
(ii) dist(γν (t), ∂Dν ) ≥ Ct, for t ∈ [0, sν ],
dist(γν (t), ∂Dν ) ≥ Csν , for t ∈ [sν , 2sν ],
dist(γν (t), ∂Dν ) ≥ C(3sν − t), for t ∈ [2sν , 3sν ],
(iii) |dγν (t)/dt| ≤ C, for t ∈ [0, sν ].
By corollary (2.10) there is a positive constant C such that for any w ∈ Dν′ ∩ B(q, 2α) for
some α > 0, and any vector X ∈ Cn the lower bound
′
(2.19) KDν′ (w, X) ≥ C dist(w, ∂Dν′ )−1/2m |X|

holds for all large ν. Let η = min{α/4, ǫ0/4}. Since F ν (bν ) lies in B(q, α) for ν large
enough, we can choose tν ∈ [0, 3sν ] such that F ν ◦ γν (tν ) ∈ ∂B(q, 2η) and F ν ◦ γν ((tν , 3sν ])
is contained in B(q, 2η). From section 2.1 we know that

(2.20) dist(F ν (z), ∂Dν′ ) ≤ C(R) dist(z, ∂Dν )

for any R > 0 and z ∈ Dν ∩ B(0, R) with F ν (z) ∈ Dν′ ∩ (B(q, 2α)). Fix r > 0 and let
z ∈ Dν ∩ B(0, r) be such that F ν (z) ∈ Dν′ ∩ B(q, 2τ ). Since the Kobayashi metric is
decreasing under holomorphic mappings, we get from (2.19) and (2.20) that

KDν (z, X) ≥ KDν′ (F ν (z), dF ν (z)X)

′
≥ C dist(F ν (z), ∂Dν′ )−1/2m |dF ν (z)X|
′
≥ C dist(z, ∂Dν )−1/2m |dF ν (z)X|.

On the other hand it can be seen that for all ν,

KDν (z, X) ≤ |X|/dist(z, ∂Dν )

and this implies the uniform estimate

′
(2.21) |dF ν (z)| ≤ Cdist(z, ∂Dν )−1+1/2m
20 G. P. BALAKUMAR AND KAUSHAL VERMA

for all large ν and z ∈ Dν ∩ B(0, r) such that F ν (z) ∈ Dν ∩ B(p, 2α). Therefore,
Z 3sν
ν ν ν
|F (γ (tν )) − F (bν )| ≤ |dF ν (γ ν (t))||dγν /dt|dt
tν
Z 3sν
′
≤C dist(γν (t), ∂Dν )−1+1/2m dt
tν
1/2m′
≤ Csν →0
as ν → ∞, which is a contradiction.

Proof of theorem 2.11. It was just shown that F ν (z) lies in B(q, 2α) for any z ∈ D̄ν ∩
B(0, δ) if δ > 0 is chosen small enough. Hence (2.21) holds for any w = F ν (z) and a
similar integration argument as above, then gives the estimate asserted in the theorem
with a uniform constant. 
2.5. Compactness of f −1 (0). Recall that to establish theorem 1.1, we were to prove
that f −1 (0) is compact in M. Suppose to obtain a contradiction that this were false.
Then the intersection
f −1 (0) ∩ M ∩ ∂B(0, ǫ) 6= φ.
for all ǫ > 0 small. Since the scaled mappings differ from f by a biholomorphic change of
coordinates on the domain and the target, the same holds for them as well, i.e.,
(F ν )−1 (F ν (0)) ∩ ∂Dν ∩ ∂B(0, ǫ) 6= φ.
Let us show that this property passes to the limit as well, i.e.,
F −1 (q) ∩ ∂D∞ ∩ ∂B(0, ǫ) 6= φ
for all ǫ > 0 small – so small that Theorem 2.11 holds for r = ǫ. Let aν be in
(F ν )−1 (F ν (0)) ∩ ∂Dν ∩ ∂B(0, ǫ) and aν → a ∈ ∂D∞ ∩ ∂B(0, ǫ). Let b be a point in
D∞ near a. Since Dν → D∞ , there exists an integer N such that for all ν > N, b lies in
D ν and
|F (a) − q| ≤ |F (a) − F (b)| + |F (b) − F ν (b)| + |F ν (b) − F ν (aν )| + |F ν (aν ) − q|
By taking ν large enough, the second term can be made arbitrarily small by the conver-
gence of F ν at b and the last term as well by the convergence of F ν (aν ) = F ν (0) to q.
The equicontinuity of the family {F ν } ∪ {F } given by theorem 2.11, then assures that
the same can be done with the third and the first terms, by taking b sufficiently close to
a and ν larger if necessary. Thus, F −1 (q) is not compact in any neighbourhood of the
origin in ∂D∞ .
On the other hand, by starting from the standpoint of F being a holomorphic mapping
between algebraic domains that extends continuously upto a boundary piece Σ of ∂D∞ ,
we may apply the theorem of Webster to establish the algebraicity of F . Thus F extends
as an analytic set and by the invariance property of Segre varieties (see [19]) as a locally
finite to one holomorphic map near 0 ∈ ∂D∞ . Contradiction.
To make this work, we only have to show that F does not map an open piece of ∂D∞
′
into the weakly pseudoconvex points on ∂D∞ . Let us denote by w(Γ) the set of all weakly
pseudoconvex points of a given smooth hypersurface Γ. Since
∂D∞ = z ∈ Cn : 2ℜzn + P (′z,′ z̄) = 0

 SOME REGULARITY THEOREMS FOR CR MAPPINGS 21

it follows that


w(∂D∞ ) = Z × C ∩ ∂D∞
where Z is the real algebraic variety in Cn−1 defined by the vanishing of the determinant
of the complex Hessian of P in Cn−1 . The finite type assumption on ∂D∞ implies that
w(∂D∞ ) is a real algebraic set of dimension at most 2n − 2. Recall that
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z̄1 ) + |z2 |2 + . . . + |zn−1 |2 = 0 ,

F (Σ) ⊂ ∂D∞
and denote Q2m′ by Q for brevity.
Proposition 2.13. The map F extends to an algebraic map in Cn .
Proof. Suppose now that there exists a strictly pseudoconvex point a ∈ ∂D∞ such that
′
F (a) is also a strictly pseudoconvex point in ∂D∞ . Then by [41], F is a smooth CR-
diffeomorphism near a and extends locally biholomorphically across a by the reflection
principle in [39]. Then Webster’s theorem [46] assures us that F is algebraic. So, we
may as well let Σ be the set of strictly pseudoconvex points, assume that F maps it into
′
w(∂D∞ ) and argue only to obtain a contradiction. As before note that
′
) = z ∈ Cn : ∆Q(z1 , z̄1 ) = 0 ∩ ∂D∞′

w(∂D∞ .
′
Then w(∂D∞ ) admits a semi-analytic stratification by real analytic manifolds of dimension
′
2n − 2 and 2n − 3. Using the specific form of ∂D∞ , this can be explicitly described as
follows. Let us denote by V = V (∆Q) the real algebraic variety in C defined by the
polynomial ∆Q. The set of singular points Sng(V ), near which V may fail to be a smooth
′
curve is finite. Let a ∈ w(∂D∞ ) be such that π1 (a) ∈ Reg(V ) where π1 is the natural
surjection onto the z1 -axis. Then V is a smooth curve near π1 (a) and after a biholomorphic
change of coordinates, we may assume that in its vicinity V coincides with {ℑz1 = 0}.
Note that the fibre of π1 over a in w(∂D∞ ′
) namely π1−1 (π1 (a)) ∩ w(∂D∞
′
), is given by
n o n o
n 2 2 n
z ∈ C : 2ℜzn + |z2 | + . . . + |zn−1 | = −Q(π1 (a), π1 (a)) ∩ z ∈ C : z1 = π1 (a)

which is evidently equivalent to ∂Bn−1 ⊂ Cn−1 . This description of the fibres of π1

′
restricted to w(∂D∞ ) persists in a neighbourhood of a and then the real analytic strata
S2n−2 ⊂ w(∂D∞ ) of dimension 2n − 2 is locally biholomorphic to ∂Bn−1 × {ℑz1 = 0}. The
′
′
complement of S2n−2 in w(∂D∞ ), which we will denote by S2n−3 is locally biholomorphic
n−1
to ∂B × Sng(V ) and this evidently has dimension 2n − 3. Observe also that S2n−2 is
generic since its complex tangent space has dimension n − 2 as a complex vector space.
Now, if
F (Σ) ∩ S2n−2 6= φ
then by the continuity of F , there is an open piece of Σ that is mapped by F into S2n−2 .
Denote this piece by Σ̃. We may assume that 0 ∈ Σ̃ and that F (0) = 0 ∈ S2n−2 . Using
an idea from Lemma 3.2 in [28], let L be a 2 dimensional complex plane that intersects
′
w(∂D∞ ) in a totally real submanifold of real dimension 2 – this is possible by the genericity
of S2n−2 – and this also holds for all translates La′ of L passing through a′ in a sufficiently
small neighbourhood U ′ of the origin. We may therefore find a non-negative, strictly
plurisubharmonic function φa′ on U ′ that vanishes on
Sa′ = L′a′ ∩ w(∂D∞
′
) ∩ U ′.
22 G. P. BALAKUMAR AND KAUSHAL VERMA

Indeed by a change of coordinates, we may assume that

L = spanC h∂/∂z1 , ∂/∂zn i = {z2 = . . . = zn−1 = 0}
and then
φa′ = |z2 − a′2 |2 + . . . + |zn−1 − a′n−1 |2 + |ℑz1 |2 + (r∞
′
(z, z̄))2
furnishes an example. By the continuity of F and the openness of Σ̃ in ∂D∞ , we can pick
b ∈ D∞ so near the origin that F (b) ∈ U ′ and ∂Ab ⊂ Σ̃ where
Ab = {z ∈ Cn : zn = bn } ∩ F −1 (LF (b) )
Since the pull-back of an analytic set under a holomorphic map is again analytic of no lesser
dimension, Ab is a positive dimensional analytic set. Also, b ∈ Ab and F (∂Ab ) ⊂ SF (b) .
Therefore, ψb = φF (b) ◦ F is a non-negative, plurisubharmonic funtion on Ab that vanishes
on ∂Ab . By the maximum principle ψb ≡ 0 on Ab which implies that F maps all of Ab
′
into SF (b) . Since F maps D∞ into D∞ , this is a contradiction.
To finish, note that the the remaining possibility is F (Σ) ⊂ S2n−3 . The above argument
can be repeated in this case as well – we will only need to replace |ℑz1 |2 as a subharmonic
function vanishing on a curve-segment of Reg(V ) by |p(z1 )|2 where p is a holomorphic
polynomial that vanishes on the finite set Sng(V ). 

3. Proof of theorem 1.3

We begin with the scaling template for (D, D ′ , f ) in section 2.1, maintaining as far as
possible the notations therein. For clarity and completeness, let us briefly describe the
scaling of D ′ which is simpler this time as M ′ is strongly pseudoconvex. As before, let
pν = (′ 0, −δν ) ∈ D and note that p′ν = f (pν ) converges to the origin which is a strongly
pseudoconvex point on ∂D ′ . Let w ν ∈ ∂D ′ be such that
|w ν − pν | = dist(p′ν , ∂D ′ ) = γν
Furthermore, since ∂D ′ is strongly pseudoconvex near the origin, we may choose a srtongly
plurisubharmonic function in a neighbourhood of the origin that serves as a defining
function for D ′ . Arguing as in Section 2.1, it follows that f preserves the distance to the
boundary, i.e.,
δν ≈ γν
′ν
for ν ≫ 1. For each w , lemma 2.2 in [40] provides a degree two polynomial automorphism
of Cn that firstly, transfers w ′ν and the normal to ∂D ′ there, to the origin and the ℜzn -axis
respectively and secondly, ensures that the second order terms in the Taylor expansion of
the defining function r ′ ◦ (g ν )−1 of the domain g ν (D ′ ) constitute a hermitian form that
coincides with the standard one i.e., |z1 |2 + . . . + |zn−1 |2 , upon restricion to the complex
tangent space. Define the dilations
B ν (′ z, zn ) = (γν−1/2 ′ z, γν−1 zn )
and note that the scaled domains Dν′ = (B ν ◦ g ν )−1 (D ′ ) are defined by
γν−1 rν′ (z) = 2ℜ(g ν (z))n + |(g ν (z))1 |2 + . . . + |(g ν (z))n−1 |2 + O(γν1/2 ).
These converge in the Hausdorff metric to
′
D∞ = H = {z ∈ Cn : 2ℜzn + |z1 |2 + . . . + |zn−1 |2 < 0},
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 23

which is the unbounded manifestation of the ball, in view of the fact that the g ν ’s converge
uniformly on compact subsets of Cn to the identity map.
Standard arguments as in [40] show that the scaled maps
F ν = B ν ◦ g ν ◦ f ◦ (T ν )−1
converge to a map F : D∞ → H. If some point z0 of D∞ is sent by F to w0 ∈ ∂H ∪ ∞,
then composing F with a local peak function at w0 , we get a function holomorphic on a
neighbourhood of z0 and peaking precisely at z0 . By the maximum principle, F (z) ≡ w0 .
However, F ν (′ 0, −1) = B ν ◦ g ν ◦ f ◦ T ν (′ 0, −1) = B ν ◦ g ν (f (pν )) = B ν (′ 0, −γν ) = (′ 0, −1).
Hence, F (′ 0, −1) = (′ 0, −1) 6= w0 . Thus, F maps D∞ into D∞ ′
and is again as in section
2, a non-degenerate, locally proper map extending continuously upto the boundary in a
neighbourhood of the origin. By theorem 2.1 in [19], F extends holomorphically across
the origin. By composing with a suitable automorphism of the ball, we may also assume
F (0) = 0. Then, ℜ(Fn (z)) is a pluriharmonic function that is negative on D∞ and attains
a maximum at the origin. So, by the Hopf lemma, we must have α = ∂(ℜFn )/∂xn (0) > 0
which combined with the fact that DF preserves the complex tangent space and thereby
′
the complex normal at the origin (to the hypersurfaces ∂D∞ and ∂D∞ , which themselves
correspond under F near the origin), implies that
Fn (z) = αzn + g(z)
for some holomorphic function g with g(z) = o(|z|). Now, let us compare the two defining
functions for D∞ , near the origin:
2ℜ(Fn (z)) + |F1 (z)|2 + . . . + |Fn−1 (z)|2 = h(z, z̄)(2ℜzn + P (′ z,′ z̄))
for a non-vanishing real analytic function h(z, z̄). Contemplate a weighted homogeneous
expansion of the above equation with respect to the weight (1/mn , . . . , 1/m2 ) given by
the multitype of ∂D∞ . Note firstly that on the left, pluriharmonic terms arise precisely
from ℜ(Fn (z)). Next, note that the lowest possible weight for any term on the right is one
and the non-pluriharmonic component of this weight is h(0)P (′ z,′ z̄). What this means
for the left, is that each Fj must expand as
Fj (z) = Pj (′ z) + (terms of weight > 1/2)
where each Pj is either weighted homogeneous of weight 1/2 or identically zero and
(3.1) h(0)P (′z,′ z̄) = |P1 (′ z)|2 + . . . + |Pn−1(′ z)|2
Clearly, all the Pj ’s cannot be zero as P is non-zero. In fact, the finite type character
of ∂D∞ forces all of them to be non-zero, as follows. After a rearrangement if necessary,
assume that Pj ∈ C[′ z] = C[z1 , . . . , zn−1 ] is non-zero precisely when 1 ≤ j ≤ m ≤ n − 1.
Then the common zero set V ⊂ Cn−1 , of these Pj ’s gives rise to the complex analytic
variety i(V ) in ∂D∞ where i : Cn−1 → Cn is the natural inclusion. The finite type
constraint compels this variety and needless to say V , to be discrete. Furthermore, the
weighted homogeneity reduces it to {0} : If (z1 , . . . , zn−1 ) is a non-trivial zero of the Pj ’s
and t ∈ C, then
Pj (et/mn z1 , . . . , et/m2 zn−1 ) = et/2 Pj (z1 , . . . , zn−1 ) = 0
for all 1 ≤ j ≤ m and so the entire curve defined by
γ(t) = (et/mn z1 , . . . , et/m2 zn−1 )
24 G. P. BALAKUMAR AND KAUSHAL VERMA

lies inside V . Now, consider the ideal I generated by these polynomials P1 , . . . , Pm , which
is a C-algebra whose transcendence degree cannot exceed m ≤ n − 1. On the other hand,
by the Nullstellensatz, for a large integer N the algebraically independent monomials,
z1N , . . . , zn−1
N
must all lie in I, forcing its transcendence degree over C and hence m to equal
n − 1. The upshot therefore, is that P is the squared norm of a weighted homogeneous
polynomial endomorphism P̂ of Cn−1 with P̂ −1(0) = 0. Now, put z2 = . . . = zn−1 = 0 in
(3.1). This cannot reduce the left hand side there to zero, otherwise the z1 -axis will lie
inside the zero set of P and hence in ∂D∞ . What this means for the right hand side of
(3.1), is that pure z1 -term(s) must occur. Since all polynomials therein are homogeneous
of same weight, we conclude that every pure z1 -term there must only be of the form
c|z1 |mn for some c > 0. Needless to say, the same holds for all the other variables zj for
2 ≤ zj ≤ n − 1, as well and we have the last statement of theorem 1.3, namely
P (′ z,′ z̄) = c1 |z1 |mn + c2 |z2 |mn−1 + . . . + cn−1 |zn−1 |m2 + mixed terms
with all cj ’s being positive and the mixed terms comprising of weight one monomials in
′
z,′ z̄ each of which is annihilated by at least one of the natural quotient maps C[′ z,′ z̄] →
C[′ z,′ z̄]/(zj z̄k ), 1 ≤ j, k ≤ n − 1 where j 6= k.

4. Proof of theorem 1.4

Proof. As before by the lower semi-continuity of rank, ∂D ′ is of rank at least n − 2 in a
neighbourhood Γ′ ⊂ ∂D ′ of p′ which we may assume also to be of finite type and pseudo-
convex, consequently regular. Let us apply Theorem A of [13] to the proper holomorphic
correspondence f −1 : D ′ → D. The hypothesis on clf −1 (Γ′ ) there holds, since ∂D is glob-
ally regular. Therefore by that theorem, f −1 extends continuously upto Γ′ as a proper
correspondence. However, we are not certain of the splitting of f −1 near p into branches
– for that p we will have to lie away from the branch locus of f −1 . Nevertheless, there
exist neighbourhoods U ′ of p′ and U of p and a (local) correspondence
−1
floc : U ′ ∩ D ′ → U ∩ D,
−1
extending continuously upto the boundary such that the graph of floc is contained in that
of f −1 and clf −1 (p′ ) = {p} where the last condition comes from the fact that f is finite
loc
to one upto the boundary, by that theorem again. Assume that both p = 0 and p′ = 0
and choose a sequence p′ν = (′ 0, −δν ) ∈ D ′ ∩ U ′ on the inner normal approaching the
origin. Since f : D → D ′ is proper and 0 ∈ clf (0) there exists a sequence pν ∈ D with
−1
pν → 0 such that f (pν ) = p′ν . Moreover, by the continuity of floc upto the boundary
and the condition clf −1 (0) = {0}, we may assume after shrinking U, U ′ if necessary, that
loc
−1 ′ν
pν ∈ D ∩ U with floc (p ) = {pν }. Now scale D with respect to {pν } and D ′ with respect
to {p′ν } – to scale D ′ , we only consider the dilations
T ν (z1 , . . . , zn ) = (τ1ν )−1 z1 , (τ2ν )−1 , . . . , (τnν )−1 zn

1/2
where τ1ν = τ (0, δν ), τjν = δν for 2 ≤ j ≤ n − 1 and τnν = δν , while for D we use the
composition B ν ◦ g ν where g ν and B ν are as in section 3. As before, the limiting domains
for Dν = (B ν ◦ g ν )(U ∩ D) and Dν′ = T ν (U ′ ∩ D ′ ) are the ball Bn and
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z̄1 ) + |z2 |2 + . . . + |zn−1 |2 < 0 ,

D∞
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 25

respectively, where this time Q2m′ is homogeneous of degree 2m′ and coincides with the
polynomial of this degree in the (homogeneous) Taylor expansion of the defining function
r ′ for ∂D ′ near the origin. Normality of the scaled mappings
F ν = T ν ◦ f ◦ (B ν ◦ g ν )−1
follows as before by proposition 2.1. But section 2 can no longer guarantee nondegeneracy
of a limit map, owing to the lack of a clear cut boundary distance conservation property
of f . However, the existence of the inverse as a proper correspondence paves the way for a
different approach. To begin with, recall from [35], the notion of normality for correspon-
dences and theorem 3 therein, the version of Montel’s theorem for proper holomorphic
′
correspondences with varying domains and ranges. Let Kµ be an exhaustion of D∞ by
′
compact subsets containing ( 0, −1). To establish the normality of scaled correspondences
−1 ν
(floc ) = (B ν ◦ g ν ) ◦ floc
−1
◦ (T ν )−1 ,
−1 ν
it suffices by theorem 3 of [35] to show that (floc ) (Kµ ) ⋐ D∞ for each µ ∈ N. To this
˜−1 ν
end, fix a Kµ , let (floc ) denote the correspondence floc −1
◦(T ν )−1 and note that (f˜loc
−1 ν
) (Kµ )
ν
is connected (since any of its components must contain p ). Now (viewing the origin in
∂D as the sequence pν that approaches it from the interior of D), we recall the Schwarz
lemma for correspondences given by theorem 1.2 of [45], which assures us that the images
of Kµ under these correspondences (f˜loc −1 ν
) , will be contained in Kobayashi balls about pν
of a fixed size, i.e., there exists R > 0 independent of ν such that
(f˜−1 )ν (Kµ ) ⊂ B K (pν , R),
loc D

where BΩK (p, R) denotes the Kobayashi ball centered at p ∈ Ω of radius R, for any given
domain Ω. Then, since biholomorphisms preserve Kobayashi balls,
) (Kµ ) ⊂ (B ν ◦ g ν )((f˜loc
−1 ν −1 ν
) (Kµ )) ⊂ (B ν ◦ g ν )(BDK K
(′ 0, −1), R


(floc (pν , R)) = BD ν

K
(′ 0, −1), R ⊂ BBKn (′ 0, −1), R+1 for all ν large by lemma 4.4 of



while BD ν
[37], giving−1the
stability of the images of Kµ under scaling. Noting that ( 0, −1), ( 0, −1) ∈ Graph(floc )ν
′ ′
−1 ν
for all ν, we conclude that {(floc ) } must admit a subsequence that converges to a cor-
respondence that is inverse to F , where F : Bn → D∞ ′
is a limit of {F ν } (which means
that the composition of these correspondences contains the graph of the identity). By
[13], both F and the inverse correspondence F −1 extend continuously upto the respective
boundaries and in particular F is finite to one on the boundary. Then, F is smooth upto
the boundary by [9], preserves boundaries and then is algebraic again as in section 2.6.
and thereafter by [5] extends holomorphically past the boundary. Now recall that the
determinant of the Levi forms are related as
2
λBn (z) = λD∞
′ (F (z))|JF (z)|

for all z ∈ ∂Bn , which readily gives the strict pseudoconvexity of ∂D∞ ′
. In particular
therefore, Q2m′ (z1 , z̄1 ) must be |z1 | which gives the strict pseudoconvexity of p′ in ∂D ′ . 
2

Proof of corollary 1.5. By the proof of Theorem 1.1 in section 2, we have that f −1 (f (p))
is compact in M and as described in [9], it is possible to choose neighbourhoods U of
p and U ′ of p′ in Cn such that if D and D ′ are the pseudoconvex sides of U ∩ M and
U ′ ∩ M ′ respectively then f extends to be a proper map from D into D ′ , putting us in
the situation of theorem 1.4. 
26 G. P. BALAKUMAR AND KAUSHAL VERMA

References
[1] M. S. Baouendi, L. P. Rothschild: Germs of CR maps between real analytic hypersurfaces, Invent.
Math. 93 (1988), 481–500.
[2] M. S. Baouendi, H. Jacobowitz, F. Treves: On the analyticity of CR mappings, Ann. of Math. (2)
122 (1985), 365–400.
[3] E. Bedford: Proper holomorphic mappings, Bull. Amer. Math. Soc. (N.S) 10 (1984), 157–175.
[4] E. Bedford, E. Barletta: Existence of proper mappings from domains in C2 , Indiana Univ. Math.
J. 39 (1990), no. 2, 315–338.
[5] E.Bedford, S.Bell: Extension of proper holomorphic mappings past the boundary, Manuscripta
Math. 50 (1985), 1-10.
[6] E. Bedford, J. E. Fornaess: Local extension of CR functions from weakly pseudoconvex boundaries,
Michigan Math. J. 25 (1978), 259–262.
[7] S. Bell: CR maps between hypersurfaces in Cn , in Several Complex Variables and Complex Ge-
ometry, Part 1 (Santa Cruz, Calif., 1989), Proc. Sympos. Pure Math. 52, Part 1, Amer. Math.
Soc., Providence, (1991), 13–22.
[8] S. Bell: Local regularity of CR homeomorphisms, Duke Math. J. 57 (1988), 295–300.
[9] S. Bell, D. Catlin: Regularity of CR mappings, Math. Z. 199 (1988), 357–368.
[10] S. Bell, R, Narasimhan: Proper holomorphic mappings of complex spaces, in Several Complex
Variables (W. Barth, R. Narasimhan Eds.), Encyclopaedia Math. Sci. 69, Springer-Verlag, Berlin
(1990).
[11] F. Berteloot: Attraction des disques analytiques et continuité höldérienne dpplications holomorphes
propres, in Topics in Complex Analysis (Warsaw, 1992), Banach Center Publ. 31, Polish Acad.
Sci., Warsaw, (1995), 91–98.
[12] F. Berteloot: Principle de Bloch et estimations de la metrique de Kobayashi des domains in C2 ,
J. Geom. Anal. 1 (2003), 29–37.
[13] F. Berteloot, A. Sukhov: On the continuous extension of holomorphic correspondences, Ann.
Scuola Norm. Sup. Pisa Cl. Sci. (4) 24 (1997), no. 4, 747–676.
[14] F. Berteloot, G. Coeuré : Domaines de C2 , pseudoconvexes et de type fini ayant un groupe non
compact d’automorphismes, Ann. Inst. Fourier (Grenoble) 41 (1991), no. 1, 7786.
[15] D. Catlin: Estimates of invariant metrics on pseudoconvex domains of dimension two, Math. Z.
200 (1989), 429–466.
[16] D. Catlin: Boundary invariants of pseudoconvex domains, Ann. of Math. (2) 120 (1984), no. 3,
529586.
[17] S. Cho: A lower bound on the Kobayashi metric near a point of finite type in Cn , J. Geom. Anal.
2 (1992), no. 4, 317–325.
[18] S. Cho: Boundary behavior of the Bergman kernel function on some pseudoconvex domains in Cn ,
Trans. Amer. Math. Soc. 345 (1994), no. 2, 803–817.
[19] B. Coupet, S. Pinchuk: Holomorphic equivalence problem for weighted homogeneous rigid domains
in Cn+1 Complex analysis in modern mathematics (Russian), 5770, FAZIS, Moscow, 2001.
[20] B. Coupet, A. Sukhov: On CR mappings between pseudoconvex hypersurfaces of finite type in C2 ,
Duke Math. J. 88 (1997), no. 2, 281–304.
[21] B. Coupet, H. Gaussier, A. Sukhov: Regularity of CR maps between convex hypersurfaces of finite
type, Proc. Amer. Math. Soc. 127 (1999), no. 11, 3191–3200.
[22] B. Coupet, S. Pinchuk, A. Sukhov: On boundary rigidity and regularity of holomorphic mappings,
Internat. J. Math. 7 91996), 617–643.
[23] J. D’Angelo: Real hypersurfaces, orders of contact, and applications, Ann. of Math. 115 (1982),
615–637.
[24] K. Diederich, J. E. Fornaess: Pseudoconvex domains: An example with nontrivial Nebenhülle,
Math. Ann. 225 (1977), 275-292.
[25] K. Diederich, J. E. Fornaess: Proper holomorphic images of strictly pseudoconvex domains, Math.
Ann. 259 (1982), no. 2, 279–286.
[26] K. Diederich, G. Herbort: Pseudoconvex domains of semiregular type, Contributions to complex
analysis and analytic geometry, 127161, Aspects Math., E26, Vieweg, Braunschweig, 1994.
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 27

[27] K. Diederich, S. Pinchuk: Proper holomorphic maps in dimension 2 extend, Indiana Univ. Math.
J. 44 (1995), 1089–1126.
[28] K. Diederich, S. Pinchuk: Regularity of continuous CR maps in arbitrary dimension, Michigan
Math. J. 51 (2003), 111–140.
[29] K. Diederich, S. Webster: A reflection principle for degenerate real hypersurfaces, Duke Math. J.
47 (1980), 835–845.
[30] J. E. Fornaess, N. Sibony: Construction of P.S.H. functions on weakly pseudoconvex domains,
Duke Math. J. 58 (1989), no. 3, 633–655.
[31] F. Forstneric: Proper holomorphic mappings: A survey, in Several Complex Variables, Proceedings
of the special year at the Mittag-Leffler Institute (Ed. J. E. Fornaess) (1993), 297–363, Princeton
University Press, Princeton, NJ.
[32] H. Gaussier: Smoothness of Cauchy Riemann maps for a class of real hypersurfaces, Publ. Mat.
45 (2001), no. 1, 79–94.
[33] H. Gaussier: Tautness and complete hyperbolicity of domains in Cn , Proc. Amer. Math. Soc. 127
(1999), no. 1, 105116.
[34] G. Herbort: On the invariant differential metrics near pseudoconvex boundary points where the
Levi form has corank one, Nagoya Math. J. 130 (1993), 25–54.
[35] Klingenberg, W.; Pinchuk, S.: Normal families of proper holomorphic correspondences, Math. Z.
207 (1991), 1, 9196.
[36] S. Krantz: Convexity in complex analysis, Several Complex Variables and Complex Geometry,
Part 1 (Santa Cruz, Calif., 1989), Proc. Sympos. Pure Math. 52, Part 1, Amer. Math. Soc.,
Providence, (1991), 119–137.
[37] P. Mahajan, K. Verma: Some aspects of the Kobayashi and Carathéodory metrics on pseudoconvex
domains, To appear in J. Geom. Anal.
[38] J. Merker: On envelopes of holomorphy of domains covered by Levi-flat hats and the reflection
principle, Ann. Inst. Fourier (Grenoble) 52 (2002), no. 5, 1443–1523.
[39] S. Pinchuk: On the analytic continuation of holomorphic mappings, Math. USSR Sb. 27, 375–392.
[40] S. Pinchuk: Holomorphic inequivalence of certain classes of domains in Cn , Mat. Sb. (N.S.)
111(153) (1980),1, 6794, 159.
[41] S. Pinchuk, Sh. Tsyganov: Smoothness of CR mappings of strictly pseudoconvex hypersurfaces,
Izv. Akad. Nauk. SSSR Ser. Mat. 53 (1989), 1120–1129.
[42] A. Sukhov: On boundary behaviour of holomorphic mappings, (Russian), Mat. Sb. 185 (1994),
131–142; English transl. in Russian Acad. Sci. Sb. Math. 83 (1995), 471–483.
[43] Do D. Thai, Ninh V. Thu: Characterization of domains in Cn by their noncompact automorphism
groups, Nagoya Math. J. 196 (2009), 135–160.
[44] J. M. Trépreau: Sur le prolongement holomorphe des fonctions CR definies sur une hypersurface
reélle de classe C 2 dans Cn , Invent. Math. 43 (1977), 53–68.
[45] K. Verma: A Schwarz lemma for correspondences and applications, Publ. Mat. 47 (2003), no. 2,
373–387,
[46] S. Webster: On the mapping problem for algebraic real hypersurfaces, Invent. Math. 43 (1977),
53–68.
[47] Yu, Ji Ye: Weighted boundary limits of the generalized Kobayashi-Royden metrics on weakly pseu-
doconvex domains, Trans. Amer. Math. Soc. 347 (1995), no. 2, 587614.
[48] Yu, Ji Ye: Peak functions on weakly pseudoconvex domains, Indiana Univ. Math. J. 43 (1994), no.
4, 12711295.

Department of Mathematics, Indian Institute of Science, Bangalore 560 012, India

E-mail address: [email protected]

Department of Mathematics, Indian Institute of Science, Bangalore 560 012, India

E-mail address: [email protected]
 SOME REGULARITY THEOREMS FOR CR MAPPINGS

G. P. BALAKUMAR AND KAUSHAL VERMA

Abstract. The purpose of this article is to study Lipschitz CR mappings from an

arXiv:1102.2335v2 [math.CV] 14 Feb 2011

h-extendible (or semi-regular) hypersurface in Cn . Under various assumptions on the

target hypersurface, it is shown that such mappings must be smooth. A rigidity result
for proper holomorphic mappings from strongly pseudoconvex domains is also proved.

1. Introduction
The purpose of this note is to prove regularity results for Lipschitz CR mappings from
h-extendible hypersurfaces in Cn .
Theorem 1.1. Let f : M → M ′ be a non-constant Lipschitz CR mapping between C ∞
smooth pseudoconvex finite type (in the sense of D’Angelo) hypersurfaces in Cn . Let
p ∈ M and p′ = f (p) ∈ M ′ . Assume that M is h-extendible at p, and that there is an
open neighbourhood U ′ ⊂ Cn of p′ and a C ∞ smooth defining function r ′ (z ′ ) for U ′ ∩ M ′
in U ′ which is plurisubharmonic on the pseudoconvex side of M ′ near p′ . If the Levi rank
of M ′ at p′ is at least n − 2 then f is C ∞ smooth in a neighbourhood of p.
This theorem, which is purely local in nature, is motivated by the regularity and rigidity
results for CR mappings obtained in [20], [21], [22] and [32]. Let B(p, ǫ), B(p′ , ǫ′ ) be
small open balls around p, p′ respectively. The pseudoconvex side of M near p is that
component of B(p, ǫ) \ M which is a pseudoconvex domain for small ǫ > 0. Denote
this by B − (p, ǫ) while the other component (the pseudoconcave part) will be denoted by
B + (p, ǫ). The same practice will be followed while referring to the respective components
of the complement of M ′ in B(p′ , ǫ′ ). Since M and M ′ are both pseudoconvex and of
finite type near p and p′ respectively, it follows that f admits a holomorphic extension to
the pseudoconvex side of M near p and this extension (which will still be denoted by f )
maps the pseudoconvex side of M near p into the pseudoconvex side of M ′ near p′ . In
fact, since f : M → M ′ is assumed to be Lipschitz, it follows that the extension is also
Lipschitz on B − (p, ǫ) ∪ M near p. Fix ǫ, ǫ′ > 0 so that
f : B − (p, ǫ) → B − (p′ , ǫ′ )
is holomorphic, f extends Lipschitz continuously up to M ∩ B(p, ǫ) and satisfies f (M ∩
B(p, ǫ)) ⊂ M ′ ∩ B(p′ , ǫ′ ). For brevity, let D and D ′ denote the pseudoconvex domains
B − (p, ǫ) and B − (p′ , ǫ′ ) respectively so that M ∩B(p, ǫ) and M ′ ∩B(p′ , ǫ′ ) are smooth open
finite type pieces on their boundaries. We may also assume that ǫ′ > 0 is small enough
so that the defining function r ′ (z ′ ) ∈ C ∞ (B(p′ , ǫ′ )) and D ′ = {r ′ (z ′ ) < 0} where r ′ (z ′ ) is
plurisubharmonic in D ′ . Note that while this condition holds for strongly pseudoconvex
1991 Mathematics Subject Classification. Primary: 32H40 ; Secondary : 32Q45, 32V10.
Key words and phrases. CR map, finite type, scaling method, Segre varieties.
The second author was supported in part by a grant from the UGC under DSA-SAP (Phase IV) and
the DST SwarnaJayanti Fellowship 2009-2010.
1
2 G. P. BALAKUMAR AND KAUSHAL VERMA

and convex finite type domains in Cn , not all (see [24]) pseudoconvex domains satisfy this
condition. Now observe that the map f : D → D ′ is not known to be proper though it
does extend continuously up to M ∩ B(p, ǫ). A sufficient condition for f to be proper
from D onto D ′ was considered by Bell-Catlin in [9] – namely, if f −1 (p′ ) = f −1 (f (p)) is
compact in M ∩ B(p, ǫ) then it is possible to choose D, D ′ in such a way that f extends
as a proper holomorphic mapping between these domains. Once this is established, the
C ∞ smoothness of f is a consequence of the techniques in [9]. The proof of theorem 1.1
therefore consists of showing that the fibre over p′ is compact in M ∩ B(p, ǫ). To do this,
we adapt the method of uniform scaling from [20],[21] and [22]. The domains D, D ′ and
the map f are scaled by composing f with a suitable family of automorphisms of Cn that
enlarge these domains near p, p′ respectively. This produces a scaled family f ν : Dν → Dν′
of maps between scaled domains. The scaled domains Dν converge in the Hausdorff sense
to a model domain of the form

D∞ = z ∈ Cn : 2ℜzn + P (′ z,′ z) < 0


(1.1)

where ′ z = (z1 , z2 , . . . , zn−1 ) and P (′ z,′ z) is a weighted homogeneous (the weights being
determined by the Catlin multitype of ∂D at p) real valued plurisubharmonic polynomial
without pluriharmonic terms of total weight 1, while the domains Dν′ converge to
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2 < 0 .

(1.2) D∞

Here 2m′ is the type of ∂D ′ at p′ and Q2m′ (z1 , z 1 ) is a subharmonic polynomial of degree
at most 2m′ without harmonic terms. D∞ ′
is manifestly of finite type while D∞ also has
the same property since p is assumed to be h-extendible. The local geometry of a smooth
pseudoconvex finite type hypersurface whose Levi rank is at least n − 2 guarantees the
existence of special polydiscs around points sufficiently close to the hypersurface on which
the defining function does not change by more than a prescribed amount – these are the
analogues of Catlin’s bidiscs and have been considered earlier in [18]. The holomorphic
mappings fν : Dν → Dν′ are shown to form a normal family by using these polydiscs as in
′
[43]; the limit map F : D∞ → D∞ is therefore holomorphic. The assumptions that f is
Lipschitz and that the defining function of ∂D ′ is plurisubharmonic on D ′ force F to be
non-degenerate. Furthermore, since f extends Lipschitz continuously up to M near p, it
is natural to expect that F imbibes some regularity near 0 ∈ ∂D∞ . This indeed happens
and it is possible to show that F is Hölder continuous up to ∂D∞ near the origin. The
main ingredient needed to do this is a stable rate of blow up of the Kobayashi metric on
Dν′ for all ν ≫ 1 and this follows by analyzing the behaviour of analytic discs in Dν′ whose
centers lie close to the origin. In particular, theorem 2.3 provides a stable lower bound
for the Kobayashi metric in Dν′ near the origin using ideas from [43]. Once F is known to
be Hölder continuous up to ∂D∞ near the origin, Webster’s theorem ([46]) implies that
F must be algebraic. Moreover, if f has a noncompact fibre over p′ , it can be shown that
the same must hold for that of F over F (0). This violates the invariance property of Segre
′
varieties associated to ∂D∞ and ∂D∞ and proves the compactness of the fibre of f over
′ ′
p ∈ ∂D .

It is known that smooth convex finite type hypersurfaces can also be scaled and morever
the stability of the Kobayashi metric on the scaled domains is also understood (see [21]).
Hence the same line of reasoning yields the following:
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 3

Corollary 1.2. With M as in theorem 1.1, let M ′ ⊂ Cn be a smooth convex finite type
hypersurface and f : M → M ′ a Lipschitz CR mapping. As before, let p ∈ M and
p′ = f (p) ∈ M ′ . Then f is C ∞ smooth in a neighbourhood of p.
More can be said when at least one of the hypersurfaces is strongly pseudoconvex. We
first consider the following local situation – let M ⊂ Cn be a C ∞ smooth pseudoconvex
finite type hypersurface and let p ∈ M be an h-extendible point. Let the Catlin multitype
of M at p be (1, m2 , . . . , mn ) where the mi ’s form an increasing sequence of even integers.
Then there exists a holomorphic coordinate system around p = 0 in which the defining
function for M takes the form:
r(z) = 2ℜzn + P (′ z,′ z) + R(z)
where P (′z,′ z) is a (1/mn , 1/mn−1 , . . . , 1/m2 )-homogeneous plurisubharmonic polynomial
of total weight one without pluriharmonic terms and the error R(z) has weight strictly
bigger than one. As usual, if J = (j1 , j2 , . . . , jn−1 ) is a multiindex of length n − 1, then
jn−1 jn−1
z denotes the monomial z1j1 z2j2 · · · zn−1
′ J
and ′ z J = z j11 z j22 · · · z n−1 .
Theorem 1.3. Let f : M → M ′ be a nonconstant Lipschitz CR mapping between real
hypersurfaces in Cn . Fix p ∈ M and p′ = f (p) ∈ M ′ . Suppose that M is C ∞ smooth
pseudoconvex and of finite type near p and that M ′ is C 2 strongly pseudoconvex near p′ . If
M is h-extendible at p, then the weighted homogeneous polynomial in the defining function
for M near p = 0 can be expressed as
P (′ z,′ z) = |P1 (′ z)|2 + |P2 (′ z)|2 + . . . + |Pn−1 (′ z)|2
where each Pi (′ z) for 1 ≤ i ≤ n − 1 is a weighted holomorphic polynomial of total weight
1/2. Moreover, the algebraic variety
′
V = z ∈ Cn−1 : P1 (′ z) = P2 (′ z) = . . . = Pn−1 (′ z) = 0
contains ′ 0 ∈ Cn−1 as an isolated point. In particular, there exist constants cj > 0 for
1 ≤ j ≤ n − 1 such that
P (′z,′ z̄) = c1 |z1 |mn + c2 |z2 |mn−1 + . . . + cn−1 |zn−1 |m2 + mixed terms
where the phrase ‘mixed terms’ denotes a sum of weight one monomials annihilated by at
least one of the natural quotient maps C[′ z,′ z̄] → C[′ z,′ z̄]/(zj z̄k ) for 1 ≤ j, k ≤ n − 1,
j 6= k.
′
As in theorem 1.1, the limit map F : D∞ → D∞ is holomorphic, nonconstant and hence
′ n
algebraic. Since D∞ ⋍ B , it is possible to show that F extends holomorphically across
0 ∈ ∂D∞ and F (0) = 0′ ∈ ∂D∞ ′
. The explicit description of the weighted homogeneous
′ ′
polynomial P ( z, z) follows from working with the extended mapping near the origin.
On the other hand, it is also natural to consider the case of CR mappings from strongly
pseudoconvex hypersurfaces and we have both a global and a local version – in the spirit
of theorem 1.1, for this case. First recall that a domain D ⊂ Cn is said to be regular at
p ∈ ∂D if there is a pair of open neighbourhoods V ⊂ U of p, constants M > 0, 0 < α ≤ 1
and β > 1 such that for any ζ ∈ V ∩ ∂D, there is a function φζ which is continuous on
U ∩ D, plurisubharmonic on U ∩ D and satisfies
−M|z − ζ|α ≤ φζ (z) ≤ −|z − ζ|β
4 G. P. BALAKUMAR AND KAUSHAL VERMA

for all z ∈ U ∩ D. It is known that the class of regular points includes open pieces of
strongly pseudoconvex boundaries, smooth weakly pseudoconvex finite type pieces in C2 ,
those that are smooth convex finite type in Cn (see [30]) and finally smooth pseudoconvex
finite type boundaries in Cn (see [17]). D is said to be regular if each of its boundary
points is regular.
Theorem 1.4. Let D ⊂ Cn be a bounded regular domain, D ′ ⊂ Cn a possibly unbounded
domain and f : D → D ′ a proper holomorphic mapping. Let p ∈ ∂D be a C 2 strongly
pseudoconvex point and p′ ∈ ∂D ′ be such that the boundary ∂D ′ is C ∞ smooth pseudocon-
vex and of finite type near p′ . Suppose that the Levi rank of ∂D ′ at p′ is n − 2 and assume
that p′ ∈ clf (p), the cluster set of p. Then p′ is also a strongly pseudoconvex point.
This fits in the paradigm, observed earlier by many other authors (for example [25]), that
a proper mapping does not increase the type of a boundary point. Note furthermore that
there are no other assumptions on the defining function for ∂D ′ near p′ as in theorem 1.1
except the Levi rank condition. The difficulty created by the lack of this assumption as
explained above is circumvented by the global properness of f – indeed, it is possible to
scale f to get a holomorphic limit F : Bn → D∞ ′
where D∞ ′
is as in (1.2). Now using the
fact that p ∈ ∂D is strongly pseudoconvex and hence regular, it is possible to peel off a
local correspondence from the global one f −1 : D ′ → D that extends continuously up to
∂D ′ near p′ and contains p in its cluster set by [13]. This local correspondence can be
scaled, using the Schwarz lemma for correspondences from [45] and the behaviour of the
scaled balls in the Kobayashi metric from [37]. This gives a well defined correspondence
′
from D∞ with values in Bn and this turns out to be the inverse for F . Thus F : Bn → D∞′
′ ′
is proper and this is sufficient to conclude that p ∈ ∂D must be strongly pseudoconvex.
Corollary 1.5. Let f : M → M ′ be a nonconstant Lipschitz CR mapping between real
hypersurfaces in Cn . Fix p ∈ M and p′ = f (p) ∈ M ′ . Suppose that M is C 2 strongly
pseudoconvex near p and that there is an open neighbourhood U ′ ⊂ Cn of p′ and a C ∞
smooth defining function r ′ (z ′ ) for U ′ ∩ M ′ in U ′ which is plurisubharmonic on the pseu-
doconvex side of M ′ near p′ . If the Levi rank of M ′ at p′ is at least n − 2 then p′ is a
strongly pseudoconvex point.
The first author would like to thank Hervé Gaussier for very patiently listening to the
material presented here.

2. Proof of theorem 1.1

Consider a smooth pseudoconvex finite type hypersurface M ⊂ Cn . Associated to each
p ∈ M are two well known invariants: one is the D’Angelo type
∆(p) = (∆n (p), ∆n−1 (p), . . . , ∆1 (p))
where the integer ∆q (p) is the q-type of M at p and is a measure of the maximal order
of contact of q dimensional varieties with M at p. To recall the definition (see [23]), let r
be a local defining function for M near p and let r̃(z) = r(z + p). Then for 1 ≤ q ≤ n,
∆q (p) = inf sup{ν(r̃ ◦ L ◦ τ )/ν(τ )}
L τ

where the infimum is taken over all linear embeddings L : Cn−q+1 → Cn and the supremum
is taken over all germs of holomorphic curves τ : (C, 0) → (Cn−q+1 , 0) mapping the
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 5

origin in C to the origin in Cn−q+1 and ν(f ) denotes the order of vanishing of f at the
origin. The smoothness of M at p implies that ∆n (p) = 1 and it can be seen that
2 ≤ ∆n−1 (p) ≤ ∆n−2 (p) ≤ . . . ≤ ∆1 (p) < ∞.
To quickly recall the Catlin multitype of M at p (see [16]), let Γn be the collection
of n-tuples of reals m = (m1 , m2 , . . . , mn ) such that 0 < m1 ≤ m2 ≤ . . . ≤ mn ≤ ∞.
Order Γn lexicographically. An element m of Γn is called distinguished provided there is
a holomorphic coordinate system w = φ(z) around p with mapped to the origin such that
if
(α1 + β1 )/m1 + (α2 + β2 )/m2 + . . . + (αn + βn )/mn < 1
β
then D α D r ◦ φ−1 (0) = 0; here α = (α1 , α2 , . . . , αn ) and β = (β1 , β2 , . . . , βn ) are n-tuples
β
and D α and D are the partial derivatives
∂ |α| /∂z1α1 ∂z2α2 . . . ∂znαn and ∂ |β| /∂ z̄1β1 ∂ z̄2β2 . . . ∂ z̄nβn
respectively. The Catlin multitype M(p) = (m1 , m2 , . . . , mn ) of M at p is defined to be
the largest amongst all distinguished elements. Since r(z) is a smooth defining function
for M near p, it can be seen that the first entry in M(p) is always one. If M(p) is finite,
i.e., mn < ∞, then there is a coordinate system around p = 0 such that the defining
function is of the form
(2.1) r(z) = 2ℜzn + P (′ z,′ z) + R(z)
where P (′z,′ z) is a (1/mn , 1/mn−1 , . . . , 1/m2 ) homogeneous polynomial of total weight
one which is plurisubharmonic and does not contain pluriharmonic terms and

γ
(2.2) |R(z)| . |z1 |mn + |z2 |mn−1 + . . . + |zn |m1
for some γ > 1. The homogeneity of P (′ z,′ z) with total weight one mentioned above
means that for t ≥ 0, P ◦ πt (′ z) = tP (′ z,′ z) where
πt (z1 , z2 , . . . , zn−1 ) = (t1/mn z1 , t1/mn−1 z2 , . . . , t1/m2 zn−1 ).
Other total weights µ > 0 occur when tµ (instead of t) can be factored from P ◦ πt (′ z).
The plurisubharmonicity of P (′ z,′ z) implies that each mk for 2 ≤ k ≤ n must be even.
Thus the variable zi is assigned a weight of 1/mn−i+1 for 1 ≤ i ≤ n − 1 and by definition
the weight of the monomial ′ z J ′ z K is
(j1 + k1 )/mn + (j2 + k2 )/mn−1 + . . . + (jn−1 + kn−1 )/m2
for (n − 1)-multiindices J = (j1 , j2 , . . . , jn−1 ) and K = (k1 , k2 , . . . , kn−1 ). A basic relation
between ∆(p) and M(p) proved in [16] is that mn+1−q ≤ ∆q for all 1 ≤ q ≤ n.
Call p ∈ M an h-extendible (or semi-regular) point if M(p) is finite and ∆(p) = M(p).
This happens if and only if (see [47]) there is a (1/mn , 1/mn−1, . . . , 1/m2 ) homogeneous
C 1 smooth real function a(′ z) on Cn−1 \ {0} such that a(′ z) > 0 whenever ′ z 6= 0 and
P (′ z,′ z) − a(′ z) is plurisubharmonic on Cn−1 and among other things, this is equivalent
to the model domain D∞ (as in (1.1)) being of finite type. We shall henceforth assume
that p = 0 and p′ = 0′ and that the respective defining functions r(z) and r ′ (z ′ ) satisfy
∂r/∂zn (0) 6= 0 and ∂r ′ /∂zn′ (0′ ) 6= 0. The holomorphic map f : D → D ′ is not necessarily
proper, but Hölder continuous with exponent δ ∈ (0, 1) on D near 0 ∈ ∂D by [42] and
f (0) = 0′ . Thus the assumption that f is Lipschitz is stronger than what is apriori known.
6 G. P. BALAKUMAR AND KAUSHAL VERMA

2.1. The Scaling Method applied to (D, D ′, f ). For z ∈ D close to the origin, note
that
dist(z, ∂D) . dist(f (z), ∂D ′ ) . dist(z, ∂D)
where the inequality on the right follows since f admits a Lipschitz extension to D near
the origin, while the left inequality follows by applying the Hopf lemma to r ′ ◦ f (z) which
is a negative plurisubharmonic function on D. To scale D, choose a sequence of points
pν = (′ 0, −δν ) in D along the inner normal at the origin, where δν > 0 and δν ց 0. Let
T ν be the dilation defined by
T ν : (z1 , z2 , . . . , zn−1 , zn ) 7→ (δν−1/mn z1 , δν−1/mn−1 z2 , . . . , δν−1/m2 zn−1 , δν−1 zn )
and note that T ν (pν ) = (′ 0, −1) while the domains Dν = T ν (D) are defined by
rν = δν−1 r ◦ (T ν )−1 (z) = 2ℜzn + P (′ z,′ z) + δν−1 R ◦ (T ν )−1 (z)
where

γ
δν−1 R ◦ (T ν )−1 (z) . δνγ−1 |z1 |mn + |z2 |mn−1 + . . . + |zn−1 |m2 + |zn |
by (2.1). On each compact set in Cn this error term converges to zero since γ > 1 and
hence the sequence of domains Dν converges in the Hausdorff metric to
D∞ = z ∈ Cn : 2ℜzn + P (′ z,′ z) < 0 .


Let r∞ (z) = 2ℜzn + P (′z,′ z). To scale D ′ recall that by [18], for each ζ near 0′ ∈ ∂D ′
there is a unique polynomial automorphism Φζ (z) : Cn → Cn with Φζ (ζ) = 0 such that
n−1 n−1 X 
X X X

r(Φ−1
ζ (z)) = r(ζ) + 2ℜzn + ajk (ζ)z1j z k1 + 2
|zα | + ℜ bαjk (ζ)z1j z k1 zα
j+k≤2m α=1 α=2 j+k≤m
j,k>0 j,k>0

+ O |zn ||z| + |z∗ |2 |z| + |z∗ ||z1 |m+1 + |z1 |2m+1 .

where for z = (z1 , . . . , zn ) ∈ Cn , we denote z∗ = (z2 , . . . , zn−1 ) ∈ Cn−2 . These automor-

phisms converge to the identity uniformly on compact subsets of Cn as ζ → 0. Further-
more, if for ζ = (ζ1 , ζ2 , . . . , ζn ) ∈ D ′ as above we consider the point ζ̃ = (ζ1 , ζ2 , . . . , ζn + ǫ)
where ǫ > 0 is chosen to ensure that ζ̃ ∈ ∂D ′ , then the actual form of Φζ (z) shows that
Φζ̃ (ζ) = (0, . . . , 0, −ǫ) – since the explicit description of these automorphisms will come
up later, we shall be content at this stage with merely collecting the relevant properties
needed to describe the scaling of D ′ . To define the distinguished polydiscs around ζ (more
precisely, biholomorphic images of polydiscs), let
Al (ζ) = max{|ajk (ζ)| : j + k = l}, 2 ≤ l ≤ 2m′
and
Bl′ (ζ) = max{|bαjk (ζ)| : j + k = l′ , 2 ≤ α ≤ n − 1}, 2 ≤ l′ ≤ m′ .
For each δ > 0 define

1/l
1/l′
τ (ζ, δ) = min δ/Al (ζ) , δ 1/2 /Bl′ (ζ) : 2 ≤ l ≤ 2m′ , 2 ≤ l′ ≤ m′


Since the type of ∂D ′ at the origin is 2m′ it follows that A2m′ (0) > 0 and hence A2m′ (ζ)
is positive for all ζ sufficiently close to the origin. Thus
′
δ 1/2 . τ (ζ, δ) . δ 1/2m
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 7

for ζ close to the origin – the upper bound being a consequence of the non-vanishing of
A2m′ (ζ) near the origin while the lower bound follows since the greatest possible exponent
of δ in the definition of τ (ζ, δ) is 1/2. Set
τ1 (ζ, δ) = τ (ζ, δ) = τ, τ2 (ζ, δ) = . . . = τn−1 (ζ, δ) = δ 1/2 , τn (ζ, δ) = δ
and define
R(ζ, δ) = {z ∈ Cn : |zk | < τk (ζ, δ), 1 ≤ k ≤ n}
which is a polydisc around the origin in Cn with polyradii τk (ζ, δ) along the zk direction
for 1 ≤ k ≤ n and let
Q(ζ, δ) = Φ−1


ζ R(ζ, δ)
which is a distorted polydisc around ζ. It was shown in [43] that these domains satisfy
the engulfing property, i.e., for all ζ in a small fixed neighbourhood of the origin, there
is a uniform constant C > 0 such that if η ∈ Q(ζ, δ), then Q(η, δ) ⊂ Q(ζ, Cδ) and
Q(ζ, δ) ⊂ Q(η, Cδ).
Consider the sequence p′ν = f (pν ) ∈ D ′ that converges to the origin and denote by w ′ν
the point on ∂D ′ chosen such that if p′ν = (p′ν ′ν ′ν ′ν ′ν ′ν ′ν
1 , p2 , . . . , pn ) then w = (p1 , p2 , . . . , pn +γν )
for some γν > 0. Note that
γν ≈ dist(p′ν , ∂D ′ )
for all large ν. Hence
δν = dist(pν , ∂D) ≈ dist(p′ν , ∂D ′ ) ≈ γν
for all large ν. Let g ν = Φw′ν (·) be the polynomial automorphism of Cn corresponding to
w ′ν ∈ D ′ as described above. Let us consider the holomorphic mappings
f ν = g ν ◦ f : D → g ν (D ′)
and define a dilation of coordinates in the target space by
B ν : (z1′ , z2′ , . . . , zn′ ) 7→ (τ1ν )−1 z1′ , (τ2ν )−1 z2′ , . . . , (τnν )−1 zn′

where τ1ν = τ (w ′ν , γν ), τjν = γ1/2

ν
for 2 ≤ j ≤ n − 1 and τnν = γν . Let D ν = T ν (D) and
D ′ν = (B ν ◦ g ν )(D ′ ) be the scaled domains and the scaled maps between them are
F ν = B ν ◦ f ν ◦ (T ν )−1 : Dν → Dν′ .
To understand the Hausdorff limit of the domains Dν′ , note first that B ν ◦g ν (p′ν ) = (′ 0, −1),
which implies that F ν (′ 0, −1) = (′ 0, −1) for all ν, and that rν′ , the defining function for
Dν′ is given by
n−1
X n−1
X
γν−1 r ′ ◦ (B ν ◦ g ν )−1 (z) = 2ℜzn + Qν (z1 , z 1 ) + |zα |2 + ℜ Sνα (z1 , z 1 )zα + O(τ1ν )


α=2 α=2

where X
Qν (z1 , z 1 ) = ajk (w ′ν )γν−1 (τ1ν )j+k z1j z k1
j+k≤2m′
j,k>0
and X
Sνα (z1 , z 1 ) = bαjk (w ′ν )γν−1/2 (τ1ν )j+k z1j z k1 .
j+k≤m′
j,k>0
8 G. P. BALAKUMAR AND KAUSHAL VERMA

By the definition of Al , Bl′ and τ1ν it follows that the largest coefficient in both Qν and
Sνα is at most one in modulus. It was shown in [18] that there exists a uniform ǫ > 0 such
that
bαjk (w ′ν )γν−1/2 (τ1ν )j+k . (τ1ν )ǫ
for all possible indices j, k, α and all large ν. Therefore some subsequence of this family
of defining functions converges together with all derivatives on compact sets to
′
r∞ (z) = 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2
where Q2m′ (z1 , z 1 ) is a polynomial of degree at most 2m′ without harmonic terms. Hence
the domains Dν′ converge to
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z 1 ) + |z2 |2 + . . . + |zn−1 |2 < 0

D∞
which, being the smooth limit of pseudoconvex domains, is itself pseudoconvex. In par-
ticular, it follows that Q2m′ (z1 , z 1 ) is subharmonic.
It is known that analytic discs in a bounded regular domain in Cn satisfy the so-called
attraction property (see [11]), i.e., if the centre of a given disc is close to a boundary
point, then a given subdisc around the origin cannot wander too far away from the same
boundary point. A quantitative version of this was proved by Berteloot-Coeuré in [14]
and forms the basis for controlling families of scaled mappings in C2 . Using ideas from
[12] and [18], the following analogue was proved in [43] and will be useful in this situation
as well – we include the statement for the sake of completeness.
Proposition 2.1. Let Ω ⊂ Cn be a bounded domain. Let ζ0 ∈ ∂Ω and suppose there is
an open neighbourhood V such that V ∩ ∂Ω is C ∞ smooth pseudoconvex of finite type and
the Levi rank is at least n − 2 on V ∩ ∂Ω. Fix a domain ω ⊂ Cm .
Then for any fixed point z0 ∈ ω and a compact K ⊂ ω containing z0 , there exist
constants ǫ(K), C(K) > 0 such that for any ξ ∈ V ∩ ∂Ω and 0 < ǫ < ǫ(K), every
holomorphic mapping F : ω → Ω with
|F (z0 ) − ζ0 | < ǫ(K) and F (z0 ) ∈ Q(ξ, ǫ)
also satisfies F (K) ⊂ Q(ξ, C(K), ǫ).
Now let {Kj } be an increasing sequence of relatively compact domains that exhaust D∞
such that each contains the base point (′ 0, −1). Fix K = Kµ and let f˜ν = f ◦ (T ν )−1 .
Then
f˜ν (′ 0, −1) = f (′ 0, −δν ) = f (pν ) = p′ν
and hence f˜ν (′ 0, −1) → 0 ∈ ∂D ′ . In particular, f˜ν (′ 0, −1) = p′ν ∈ Q(w ′ν , 2γν ) for all large
ν, by the construction of these distorted polydiscs. By the previous proposition
f˜ν (K) ⊂ Q(w ′ν , C(K)γν )
and therefore
F ν (K) = B ν ◦ g ν ◦ f˜ν (K) ⊂ B ν ◦ g ν (Q(w ′ν , C(K)γν ).
However,
B ν ◦ g ν (Q(w ′ν , C(K)γν ) = B ν (R(w ′ν , C(K)γν ))
which by definition is contained in a polydisk around the origin with polyradii
rk = τk (w ′ν , C(K)γν )/τk (w ′ν , γν )
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 9

for 1 ≤ k ≤ n. Note that rn = C(K) and that rk = (C(K))1/2 for 2 ≤ k ≤ n − 1. Since

C(K) > 1, and this may be assumed without loss of generality, the definition of τ (ζ, δ)
shows that for δ ′ < δ ′′ ,
′
(δ ′ /δ ′′ )1/2 τ (ζ, δ ′′ ) ≤ τ (ζ, δ ′ ) ≤ (δ ′ /δ ′′ )1/2m τ (ζ, δ ′′).
Therefore
τ1 (w ′ν , C(K)γν )/τ1 (w ′ν , γν ) ≤ (C(K))1/2
and hence r1 ≤ (C(K))1/2 . Thus {F ν } is uniformly bounded on each compact set in D∞
′
and is therefore normal. Let F : D∞ → D∞ be a holomorphic limit of some subsequence in
{F } and since F ( 0, −1) = ( 0, −1) for all ν by construction, it follows that F (′ 0, −1) =
ν ν ′ ′

(′ 0, −1). The maximum principle shows that F (D∞ ) ⊂ D∞ ′

.
Let R > 0 be arbitrary and fix z ∈ D∞ ∩ B(0, R). Note that since f preserves the
distance to the boundary, it follows that
(rν′ ◦ F ν )(z) = γν−1 (r ′ ◦ f ◦ Tν−1 )(z) . (δν /γν ) rν (z)
and hence
′
r∞ ◦ F (z) . r∞ (z)
since the constants appearing in the first estimate are independent of ν. This shows that
′
the cluster set of a finite boundary point of D∞ does not intersect D∞ and therefore F is
non-degenerate.

2.2. Boundedness of F ν (0). A crucial point in the proof of theorem 1.1 is to show that
near the origin, the family {F ν } of the scaled maps is uniformly Hölder continuous upto
the boundary i.e., the Hölder constant and exponent are stable under the scaling. A first
step in establishing this is to show that the images of the origin under these scaled maps
is stable.
Proposition 2.2. The sequence {F ν (0)} is bounded.
Proof. Recall the scaled maps
F ν = B ν ◦ g ν ◦ f ◦ (T ν )−1
where B ν , T ν were linear maps and f (0) = 0. So F ν (0) = (B ν ◦ g ν ◦ f )(0) = (B ν ◦ g ν )(0).
Now let us recall the explicit form of the map g ν = Φw′ν from [43], which is a polynomial
automorphism of Cn that reduces the defining function to a certain normal form as stated
in section 2.1. Let us recall this reduction procedure for any given domain Ω in Cn that
is smooth pseudoconvex of finite type 2m and of Levi rank (at least) n − 2 on a (small)
open boundary piece Σ in ∂Ω and containing the origin, say. Let r be a smooth defining
function for Σ with ∂r/∂zn (z) 6= 0 for all z in a small neighbourhood U in Cn of Σ such
that the vector fields
Ln = ∂/∂zn , Lj = ∂/∂zj + bj (z, z̄)∂/∂zn

−1
where bj = ∂r/∂zn ∂r/∂zj , form a basis of CT (1,0) (U) and satisfy Lj r ≡ 0 for 1 ≤ j ≤
¯
n − 1 and ∂ ∂r(z)(Li , L̄j )2≤i,j≤n−1 has all its eigenvalues positive for each z ∈ U.

The reduction now consists of five steps – for ζ ∈ U, the map Φζ = φ5 ◦ φ4 ◦ φ3 ◦ φ2 ◦ φ1

where each φj is described below.
10 G. P. BALAKUMAR AND KAUSHAL VERMA

φ1 is the affine map given by

n−1
X n−1
X
φ1 (z1 , . . . , zn ) = (z1 − ζ1 , . . . , zn−1 − ζn−1 , zn + bζj zj − (ζn + bζj ζj )).
j=1 j=1

where the coefficients bζj = bj (ζ, ζ̄) are clearly smooth functions of ζ on U. Therefore, φ1
translates ζ to the origin and
r(φ−1(z)) = r(ζ) + ℜzn + terms of higher order.
where the constant term disappears when ζ ∈ Σ.
The remaining reductions remove the occurrence of harmonic (not just pluriharmonic)
monomials in the variables z1 , . . . , zn−1 of weights upto one in the weighted homogeneous
expansion of the defining function with respect to the weights given by the multitype for
Σ, i.e., the variable z1 is assigned a weight of 1/2m, zn a weight of 1 while the others are
assigned 1/2 each. Now, since the Levi form restricted to the subspace
L∗ = spanCn hL2 , . . . , Ln−1 i
(1,0)
of Tζ (∂Ω) is positive definite, we may diagonalize it via a unitary transform φ2 and
a dilation φ3 will then ensure that the quadratic part involving only z2 , z3 , . . . , zn−2 in
the Taylor expansion of r is |z2 |2 + |z3 |2 + . . . + |zn−2 |2 . The entries of the matrix that
represents the composite of the last two linear transformations are smooth functions of ζ
and in the new coordinates still denoted by z1 , . . . zn , the defining function is in the form
n−1 X
X m n−1
X
(aαj z1j bαj z̄1j )zα cα zα2


(2.3) r(z) = r(ζ) + ℜzn + ℜ + +ℜ
α=2 j=1 α=2

X n−1
X n−1 X
X
aj,k z1j z̄1k 2
ℜ bαj,k z1j z̄1k zα


+ + |zα | +
2≤j+k≤2m α=2 α=2 j+k≤m
j,k>0

+ O(|zn ||z| + |z∗ |2 |z| + |z∗ ||z1 |m+1 + |z1 |2m+1 )

A change in the normal variable zn to absorb the pluriharmonic terms here i.e., z1k , z̄1k , zα2 as
well as z1k zα , z̄1k z̄α , can be done according to the following standard change of coordinates
φ4 given by
zj = tj (1 ≤ j ≤ n − 1),
zn = tn − P1 (t1 , . . . , tn−1 )
where
2m
X n−1 X
X m n−1
X
P1 (t1 , . . . , tn−1 ) = ak0 tk1 − aαk tα tk1 − cα t2α
k=2 α=2 k=1 α=2
with coefficients that are smooth functions of ζ.
Finally, just as we absorbed into the simplest pure term ℜzn that occurs in the Taylor
expansion, other pure terms not divisible by this variable in the last step, we may absorb
into (some among) the simplest of non-pluriharmonic (and non-harmonic) monomials
occurring there namely, |zα |2 where 2 ≤ α ≤ n−1 other harmonic (but non-pluriharmonic)
terms of degree at least two which are not divisible by them. Let us do this to those of
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 11

weight at most one, remaining in (2.3) rewritten in the t-coordinates, which are of the
form t̄j1 tα by applying the transform φ5 given by

t1 = w1 , tn = wn ,
tα = wα − P2 (w1 ) (2 ≤ α ≤ n − 1)

where
m
X
P2 (w1 ) = bαk w1k
k=1

with coefficients smooth in ζ, as before (since all these coefficients are simply the deriva-
tives of some order of the smooth defining function r evaluated at ζ). We then have the
sought for simplification of the Taylor expansion.

Now, suppose that we already have the reduced form holding at one boundary point say
the origin, then (since no further normalization would be required) Φ0 may be taken to be
the identity. Note that the normalizing map Φζ is not uniquely determined – even among
the class of all maps of the same form – owing to the (only) ambiguity in the choice of the
diagonalizing map φ2 ; however, this choice can certainly be done in a manner such that
the coefficients of that unitary matrix are smooth in ζ and satisfying the ‘initial condition’
that φ2 for the origin is the identity. Thus in all, the map Φζ is smooth in the parameter ζ
with Φ0 being the identity map. This implies that the family Φζ (·) is uniformly Lipschitz
(where ζ ∈ U) and converges uniformly on compact subsets of Cn to the identity as ζ
approaches the origin as mentioned in section 2.1. Next, the consequence of the simpler
fact that the map Φ(ζ, 0) = Φζ (0) is Lipschitz in a neighbourhood of the origin, to our
setting is that
|g ν (0)| ≤ C1 |w ′ν |
with C1 independent of ν. Now
 
ν ν ν ν −1 ν ν −1 ν ν −1 ν
F (0) = B ◦ g (0) = (τ1 ) (g (0))1 , . . . , (τn−1 ) (g (0))n−1, (τn ) (g (0))n .

1/2 1/2
From the fact that τ1ν & γν and τjν = γν for 2 ≤ j ≤ n − 1 while τnν = γν , we get that

B ν ◦ g ν (0) ≤ C2 |w ′ν |/γν

with C2 again independent of ν. Now, since f is Lipschitz upto M and δν . γν we have

|w ′ν | ≤ dist(w ′ν , p′ν ) + dist(p′ν , 0′)

= dist(p′ν , ∂D ′ ) + |f (pν ) − f (0)|
≤ C3 (γν + |pν |)
= C3 (γν + δν )
≤ C4 γ ν

with constants independent of ν as before, implying that {F ν (0)} is bounded.

12 G. P. BALAKUMAR AND KAUSHAL VERMA

2.3. Stability of the Kobayashi metric. For Ω, Σ, U with 0 ∈ Σ, as in the previous

section, recall from [43] (see also [12]), the M-metric defined for ζ ∈ U ∩ Ω by
Xn


MΩ (ζ, X) = | DΦζ (ζ)X k |/τk (ζ, ǫ(ζ)) = D(Bζ ◦ Φζ )(ζ)(X) l1
k=1
ǫ(ζ)
with Bζ = Bζ where ǫ(ζ) > 0 is such that ζ + (0, . . . , 0, ǫ(ζ)) lies on Σ and
Bζδ (z1 , . . . , zn ) = (τ1 )−1 z1 , . . . , (τn )−1 zn

where τ1 = τ (ζ, δ), τj = δ 1/2 for 2 ≤ j ≤ n − 1 and τn = δ. Now let us get to our scaled
domains Dν′ = B ν ◦ g ν (D ′ ), where we recall B ν = Bp′ν and g ν = Φw′ν and denote by ν Φζ
and ν Bζ the normalizing map and the dilation respectively, associated to (Dν′ , ζ); we shall
drop the left superscript ν for the domain D ′ . Since all the coefficients of the polynomial
automorphisms ν Φζ (·) depend smoothly on the derivatives of the defining functions rν ,
which converge in the C ∞ -topology on U, there exists L > 1 such that
1/L < | det D(ν Φζ (z) | < L


(2.4)
for all ν and ζ ∈ U. The main result of this section is the following stability theorem for
the Kobayashi metric.
Theorem 2.3. There exists a neighbourhood U of the origin such that
KDν′ (z, X) ≥ CMDν′ (z, X)
for all z ∈ U ∩ Dν′ , ν ≫ 1 where C is a positive constant independent of ν.
The proof of this requires two steps and follows the lines of the argument presented in
[43]. The first step consists in uniformly localizing the Kobayashi metric of the scaled
domains near the origin which translates to verifying the following uniform version of the
attraction of analytic discs near a local plurisubharmonic peak point.
Lemma 2.4. There exists a neighbourhood V ⊆ U of the origin and δ > 0 independent of j
such that for any j large enough and any analytic disc f : ∆ → Dj′ with f (0) = ζ ∈ V ∩D ′j ,
we have
f (∆δ ) ⊆ U ∩ Dj′
Proof. If the lemma were false, then for any neighbourhood V ⊂ U and any δ > 0, there
exists a sequence of analytic discs fj : ∆ → Dj′ with fj (0) = ζ j ∈ V ∩ Dj′ , converging to
the origin but fj (∆δ ) * U. Consider
f˜j = S −1 ◦ fj : ∆ → D ′
j j

where Sj = B j ◦ Φw′j are the scaling maps. Then ζ̃ j = f˜j (0) = Sj−1 (ζ j ) converges to the
origin. Indeed, the family Sj−1 = Φ−1wj
◦ (B j )−1 is equicontinuous at the origin since their
derivatives are
D(Φ−1 j −1
w j ) ◦ (B )
with D(Φ−1 w j ) being bounded in a neighbourhood of the origin and (B )
j −1
converging to
′
the zero map. Now, 0 ∈ ∂D being a plurisubharmonic peak point by [17], the simplest
version of the attraction property of analytic discs (see for instance lemma 2.1.1 in [33]),
gives for all j ≫ 1 that
f˜j (∆γ ) ⊆ U ∩ D ′
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 13

for some γ ∈ (0, 1). Rescale the f˜j ’s, so that we may take γ = 1. Now recall the sharper
version of the attraction property given by lemma 3.6 of [43] namely,
Lemma 2.5. Let W ⋐ U be a neighbourhood of the origin. There exist constants
α, A ∈ [0, 1] and K ≥ 1 such that for any analytic disc f : ∆N → U that satisfies
MD′ (f (t), f ′(t)) ≤ A, f (0) ∈ W and K N −1 ǫ(f (0)) < α also satisfies,
 
N
f (∆N ) ⊂ Q f (0), K ǫ(f (0))

We intend to apply this lemma to the analytic discs g̃j (t) = f˜j (r0 t) for which we only
need to verify
MD′ g̃j (t), g̃j′ (t) ≤ A

since g̃j (0) = f˜j (0) converges to the origin. Then

MD′ g̃j (t), g̃j′ (t) = MD′ f˜j (r0 t), r0 f˜j′ (r0 t)

= r0 MD′ f˜j (r0 t), f˜′ (r0 t)

j

≤ r0 C5 KD′ f˜j (r0 t), f˜j′ (r0 t)

by theorem 3.10 of [43], where C5 is a positive constant independent of j. At the last

step, we use the fact that f˜j ’s map ∆ into so small a neighbourhood of the origin where
KD′ ≈ MD′ . As the Kobayashi metric decreases under holomorphic mappings, we have
r0 C5 KD′ f˜j (r0 t), f˜j′ (r0 t) ≤ r0 C5 K∆ t, ∂/∂t

So, if we choose r0 such that

 
r0 C5 sup K∆ (t, ∂/∂t) ≤ A,
|t|≤r0

we will have completed the required verification of the hypothesis for the discs gj : ∆ →
D ′ ∩ U and the aforementioned lemma gives
f˜j (∆r0 ) = g̃j (∆) ⊂ Q g̃j (0), ǫ(g̃j (0)) = Q f˜j (0), ǫ(f˜j (0)) .



(2.5)
At this point observe also that r0 does not depend on δ. Next, fix any compact subdisc
K ⊂ ∆r0 and note by (2.5) that,
f˜j (K) ⊂ Q f˜j (0), ǫ(f˜j (0))

and subsequently,
(Sj ◦ f˜j )(K) ⊂ Sj Q(f˜j (0)), ǫ(f˜j (0))



−1 n
= (B j ◦ Φw′j ) ◦ Bζ̃ j ◦ Φζ˜′ j (∆ )
 
= Bp′j ◦ Φw′j ◦ Φ−1 ◦ Bζ̃−1 (∆n )


(2.6) ˜′ j j
ζ

where ζ̃ ′j = ζ̃ j + (0, . . . , 0, ǫ(ζ̃ j )). Now the fact that ǫ(ζ̃ j ) ≈ ǫ(pj ), as both pj and ζ̃ j
converge to the origin and corollary 2.8 of [18] – the fact that the distinguished radius τ
at any two points in U are comparable – together imply the boundedness of the sequence
τ (ζ̃ ′j , ǫ(ζ̃ j ))/τ (w ′j , ǫ(pj )), by C6 say. This when combined with (2.4) and (2.6), yields the
14 G. P. BALAKUMAR AND KAUSHAL VERMA

stability of our discs fj on compact subdiscs in ∆ i.e., we have for a positive constant
CK > C6 L2 (which may depend on K but not on j), that
fj (K) = (Sj ◦ f˜j )(K) ⊆ ∆√C × . . . × ∆√C × ∆C
K K K

Hence by Montel’s theorem, fj ’s converge to a map f∞ : ∆r0 → D ′ ∞ , where f∞ (0) = 0 ∈

′
∂D∞ . If φ is a local plurisubharmonic peak function at the origin, then the maximum
principle applied to the subharmonic function ψ = φ ◦ f∞ implies that f∞ ≡ constant,
since φ peaks precisely at one point. This contradicts our assumption that fj (∆δ ) * U
(as soon as δ is taken smaller than r0 ) and completes the proof of the lemma. 
Remark 2.6. The same line of argument gives proposition 2.1 by replacing the discs by
balls in Cm , covering any given compact K ⊂ ω by balls and using the engulfing property
of the special polydics Q.
The second and more technical step, is a quantitative form of the Schwarz lemma at the
boundary which generalizes lemma 3.6 of [43] mentioned earlier. To do this we require a
stable version of the engulfing property for the distinguished polydiscs Qν assosciated to
the scaled domain Dν′ . Fix a pair of neighbourhoods 0 ∈ W ⋐ V ⋐ U.
Lemma 2.7. There exist constants α, A ∈ [0, 1] and K > 1 such that for every analytic
disc
f : ∆N → V ∩ Dν′
that satisfies MD′ν f (t), f ′ (t) < A, f (0) ∈ W and K N −1 ǫ(f (0)) < α also satisfies,




f (∆N ) ⊂ Qν f (0), K N ǫ f (0) .

Recall the definition of Al (z) and τ (z, δ) from section 2.1. Let Aνl (z) and τ ν (z, δ) de-
note the corresponding quantities for the domain Dν′ and A∞ ∞
l (z) and τ (z, δ) that for
D∞ . Then it is clear that Al (z) → Al (z) and consequently τ (z, δ) → τ ∞ (z, δ), both
′ ν ∞ ν

convergences being uniform for z in V . Let δ > 0 be given. Let

l
M = sup τ ν (z, δ) : z ∈ V, 2 ≤ l ≤ 2m′ , ν ≥ 1


and 0 < ǫ < δ/M. Then there exists by the uniform convergence of the A’s, an Nδ
independent of z ∈ V such that for all 2 ≤ l ≤ 2m and all ν > Nδ ,
(2.7) A∞ ν ∞
l (z) − ǫ < Al (z) < Al (z) + ǫ

The right inequality gives for all 2 ≤ l ≤ 2m that

l
l
l
τ ∞ (z, δ) Aνl (z) < τ ∞ (z, δ) A∞ ∞
l (z) + τ (z, δ) ǫ < δ + δ

by the definition of τ ∞ and ǫ . The definition of τ ν now makes this read as

(2.8) τ ∞ (z, δ) < 21/2 τ ν (z, δ)
for all ν > Nδ and z ∈ V . Meanwhile the left inequality at (2.7) similarly gives,

l
l ν
l
τ ν (z, δ) A∞ ν ν
l (z) < τ (z, δ) A (z) + τ (z, δ) ǫ < 2δ

which implies τ ν (z, δ) < 21/2 τ ∞ (z, δ). Combining this with corollary 2.8 of [18] applied
to τ ∞ , we get for z ∈ Q∞ (z ′ , δ) and all ν large that
(2.9) 1/C τ ν (z ′ , δ) < τ ν (z, δ) < Cτ ν (z ′ , δ)
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 15

for some C > 0, independent of z ′ and ν. This will lead to the following uniform engulfing
property of these polydiscs.
Lemma 2.8. There exists a positive constant C such that for all ν large, z ′ ∈ U and
z ′′ ∈ Qν (z ′ , δ) we have
Qν (z ′′ , δ) ⊂ Qν (z ′ , Cδ)
Qν (z ′ , δ) ⊂ Qν (z ′′ , Cδ)
Proof. Let us recall what Qν is and rewrite for instance the first statement as
ν δ

−1 n

−1 n

Bz ′′ ◦ ν Φz ′′ ∆ ⊂ ν BzCδ ν
′ ◦ Φz ′ ∆
which is equivalent to saying that the map
ν
(2.10) BzCδ ν ν
′ ◦ Φz ′ ◦ ( Φz ′′ )
−1
◦ (ν Bzδ′′ )−1
maps the unit polydisc into itself. Now recall from (2.4), that the sequence of Jacobians
of ν Φz ′ ◦ ν Φ−1 2
z ′′ is uniformly bounded above by L . Unravelling the definition of the scalings
ν Cδ ν δ
Bz ′ and Bz ′′ , then shows that the map defined in (2.10) carries the unit polydisc into
the polydisc of polyradius given by
L2 τ ν (z ′′ , δ)/τ ν (z ′ , Cδ), L2 δ 1/2 /(Cδ)1/2 , . . . , L2 δ 1/2 /(Cδ)1/2 , L2 δ/Cδ


(2.11)
Now the uniform comparability of the τ ’s obtained in (2.9) gives C7 > 0 independent of
ν and z ′ , such that
τ ν (z ′′ , δ) < C7 τ ν (z ′ , δ)
whereas looking at the definition of the τ directly gives
τ ν (z ′ , Cδ) > C 1/2m τ ν (z ′ , δ)
Combining these two we see that the first component at (2.11) is bounded above by
L2 C7 /C 1/2m . Therefore, choosing C to be such that C 1/2 > L2 and C 1/2m > L2 C7 ensures
that the map at (2.10) leaves the unit polydisc invariant, completing the verification of
the lemma. 
Before passing, let us record one more stable estimate concerning the τ ’s to be of use
later, namely
′
(2.12) τ ν (z, δ) ≤ C8 δ 1/2m
for C8 > 0 (independent of ν), which follows as in section 2.1 from the fact that the
Aν2m′ (z)’s are uniformly bounded below by a positive constant on a neighbourhood of the
origin (which as usual we may assume to be U).
Next, observing that all the constants in the calculations in the proof of lemma 3.6 of
[43] are stable owing to (2.9), lemma 2.8 and the stability of the estimates on the various
coefficients of ν Φ’s given in lemma 3.4 of [43] – because all these coefficients are nothing
but the derivatives of some order (less than 2m′ ) of the defining functions of the scaled
domains, which converge in the C ∞ -topology on U– lemma 2.7 follows as well.
Proof of theorem 2.3. If the theorem were false, there would exist a subsequence νj of N
and ζ j ∈ D ′νj converging to the origin such that
 
KDν′ j ζ j , X MDν′ j (ζ j , X) < 1/j 2

16 G. P. BALAKUMAR AND KAUSHAL VERMA

which after re-indexing the D ′νj ’s as D ′j and using lemma 2.4 reads as :
There exists a sequence ζ j ∈ U ∩ D ′j with ζ j converging to the origin and a sequence of
analytic discs
fj : ∆ → U ∩ Dj′
such that   
j
fj (0) = ζ and fj′ (0) j
= Rj X MDj (ζ , X)
′

with Rj ≥ j 2 .
The idea as in [12] and [43] now consists of scaling these analytic discs appropriately to
enlarge their domains to discs of growing radius while ensuring their normality also, so
′
that taking their limit produces an entire curve lying inside D∞ to contradict its Brody
+
hyperbolicity. To work this out, define Mj (t) : ∆1/2 → R by
Mj (t) = MDj′ fj (t), fj′ (t)

Then
 
fj (0), fj′ (0) j j


Mj (0) = MDj′ = MDj′ ζ , Rj X MDj′ (ζ , X)
= Rj ≥ j 2
Recall the following lemma from [12].
Lemma 2.9. Let (X, d) be a complete metric space and let M : X → R+ be a locally
bounded function. Then for all σ > 0 and for all u ∈ X satisfying M(u) > 0, there exists
v ∈ X such that
(i) d(u, v) ≤ 2/σM(u)
(ii) M(v) ≥ M(u)
(iii) M(x) ≤ 2M(v) if d(x, v) ≤ 1/σM(v)
Apply this lemma to Mj (t) on ∆1/2 with u = 0 and σ = 1/j, to get aj ∈ ∆1/2 such that
|aj | ≤ 2j/Mj (0) and
(2.13) Mj (aj ) ≥ Mj (0) ≥ j 2
and furthermore,


(2.14) Mj (t) ≤ 2Mj (aj ) on ∆ aj , j/Mj (aj )
(for all j big enough so that the discs above lie inside ∆1/2 ). Now, scaling the unit disc
with respect to the points aj which approach the origin (since the contention is that the
derivatives of the fj ’s near the origin, in the M-metric, are blowing up rapidly), we get
the discs ∆j = sj (∆) where sj = bj ◦ φaj with φaj the translation that transfers aj to the
origin and bj being the map that dilates by a factor of 2A−1 Mj (aj ) with A the constant
of lemma 2.7. The scaled analytic discs gj : ∆j → U ∩ Dj′ are then given by
gj (t) = (fj ◦ s−1
j )(t) = fj (aj + cj t),

where cj = A/2Mj (aj ). Note that

gj′ (t) = cj fj′ (aj + cj t)
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 17

which implies
 
gj (t), gj′ (t) cj t), fj′ (aj


MD′j = cj Mj fj (aj + + cj t)
= cj Mj (aj + cj t)


Now note that aj + cj t lies in ∆ aj , j/Mj (aj ) for all |t| < j as 0 < A < 1 and therefore
by (2.13), (2.14) and the definition of cj , we have for t ∈ ∆j that
MD′j gj (t), gj′ (t) = cj Mj (aj + cj t)




< cj 2Mj (aj )
(2.15) =A
We wish to apply lemma 2.7 to these gj ’s. First we must that verify their centres lie close
enough to the origin; to that end write
|gj (0)| ≤ |fj (aj ) − fj (0)| + |fj (0)|
and note that the first term on the right can be made arbitrarily small by the equicon-
tinuity of the fj ’s which map into the bounded neighbourhood U. After passing to a
subsequence if necessary to ensure that ǫ(gj (0)) ≤ α/K j−1, we have with (2.15), verified
all the criteria of lemma 2.7 and therefore
 
j N
(2.16) gj (∆N ) ⊂ Q gj (0), K ǫ(gj (0))
for all j ≥ N. Let ηj = gj (0) and ηj′ = ηj + (0, . . . , 0, ǫj ) with ǫj > 0 such that rj (ηj′ ) = 0
and note that
ηj ∈ Qj ηj′ , Cǫj

for a uniform constant C ≥ 1 since ǫj ≈ rj (ηj ), uniformly in j. Consequently using lemma

2.8, (2.16) becomes
 
(2.17) gj (∆N ) ⊂ Qj ηj′ , CK N ǫj
for all j ≥ N. Now if we let
hj = j Bηj ◦ j Φηj′ ◦ gj : ∆j → S̃j (U ∩ D ′ )
where S̃j is the map
j
Bηj ◦ j Φηj′ ◦ Bp′j ◦ Φw′j
then what (2.17) translates to, for the map hj is that
hj (∆N ) ⊂ ∆(CK N )1/2 × . . . × ∆(CK N )1/2 × ∆CK N
for all N ≤ j, exactly as in the proof of lemma 2.4. Also, note that the domains S̃j (U ∩D ′ )
′
after passing to a subsequence if necessary, converge to a domain D̃∞ of the same form as
′
D∞ , by the same argument as in section 2.1 together with (2.12). Thus Montel’s theorem,
a diagonal sequence argument and an application of the maximum principle to the limit,
gives rise to an entire curve
′
h : C → D̃∞ .
Indeed, note that
hj (0) = j Bηj ◦ j Φηj′ (ηj ) = (0, . . . , 0, −1)


18 G. P. BALAKUMAR AND KAUSHAL VERMA

′
for all j and so h(0) = (0, . . . , 0, −1), which lies in D̃∞ all of whose boundary points –
including the point at infinity – are peak points. Now, to check that h is nonconstant, we
examine the sequence of their derivatives at the origin. We write Xj for gj′ (0) = cj fj′ (aj )
(which tend to 0 but are bounded below in the M-metric).
h′j (0) l1 = j Bηj′ ◦ D(j Φηj′ ) (ηj )(Xj ) l1


(2.18)
≥ j Bηj ◦ D(j Φηj ) (ηj )(Xj ) l1




= MD′j gj (0), Xj
= cj Mj (aj ) = A/2
where the lower bound here is a consequence of (2.9) applied to τ j for the points ηj and
ηj′ and the fact that the family
D(ν Φz ′ ) ◦ D(µ Φ−1
z ′′ )
is uniformly bounded below in norm. Passing to the limit in (2.18), now gives
|h′ (0)|l1 ≥ C(A/2) > 0
and we reach the contradiction mentioned earlier namely to the Brody hyperbolicity of
′
∂ D̃∞ (see for instance, lemma 3.8 of [43]). 
Finally, let us note the consequence of theorem (2.3) in the form that we shall make use
of
Corollary 2.10. There is neighbourhood U of the origin and a positive constant C such
that
1/2m′
KD′ν (z, X) ≥ C |X|/ dist(z, ∂D ′ν )
for all z ∈ U ∩ Dν′ and all ν ≫ 1.
This comes from the facts that ǫν (z) ≈ dist(z, ∂D ′ν ) and for some C9 > 0 we have for all
ν that
′
MD′ν (z, X) ≥ C9 |X|l1 /(ǫν (z))1/2m
which in turn follows from (2.12).
2.4. Uniform Hölder continuity of the scaled maps near the origin. In this sec-
tion, we recall from [20], the arguments which show that the family of scaled maps is
uniformly Hölder continuous upto the boundary – we already know that each scaled map
is Hölder continuous upto the origin.
Theorem 2.11. There exist positive constants r, C such that
′
|F ν (z ′ ) − F ν (z ′′ )| ≤ C|z ′ − z ′′ |1/2m
for any z ′ , z ′′ ∈ D̄ν ∩ B(0, r) and all large ν.
By the previous section we may assume that {F ν (0)} converges to a (finite) boundary
′
point q ∈ ∂D∞ . Recall that the origin (respectively the ℜzn axis) is a common boundary
point (respectively, the common normal at the origin) for all the scaled domains. At this
special boundary point, we first show that the family of scaled mappings is equicontinuous
– we already know that this holds on compact subsets of D∞ – or equivalently that, given
any neighbourhood of q, there exists a small ball about the origin, such that every scaled
map carries the piece of its domain intercepted by this ball into that given neighbourhood.
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 19

Then, the uniform lower bound on the Kobayashi metric near q will sharpen the uniform
boundary distance decreasing property to uniform Hölder continuity, of the scaled maps.
In particular then, the family {F ν } ∪ {F } is equicontinuous near the origin, upto the
respective boundary.

Lemma 2.12. For any ǫ > 0 there exists δ > 0, such that |F ν (z) − q| < ǫ for any
z ∈ Dν ∩ B(0, δ) and all large ν.

Proof. Suppose to obtain a contradiction, that the assertion were false. Then there exists
ǫ0 > 0 and a sequence aν ∈ Dν such that aν → 0 and |F ν (aν ) − q| ≥ ǫ0 . Since every F ν
is continuous upto the boundary ∂Dν , we can also choose a sequence of points bν ∈ Dν
lying on the common inner normal to Dν at the origin i.e., bν = (0, −βν ) with βν > 0,
such that bν → 0 and |F ν (bν ) − q| → 0. Let sν = |aν − bν |. It is not difficult to see that
there exists a constant C > 0 such that for all ν, there is a smooth path γν : [0, 3sν ] → Dν
with the following properties:
(i) γν (0) = aν , γν (3sν ) = bν
(ii) dist(γν (t), ∂Dν ) ≥ Ct, for t ∈ [0, sν ],
dist(γν (t), ∂Dν ) ≥ Csν , for t ∈ [sν , 2sν ],
dist(γν (t), ∂Dν ) ≥ C(3sν − t), for t ∈ [2sν , 3sν ],
(iii) |dγν (t)/dt| ≤ C, for t ∈ [0, sν ].
By corollary (2.10) there is a positive constant C such that for any w ∈ Dν′ ∩ B(q, 2α) for
some α > 0, and any vector X ∈ Cn the lower bound
′
(2.19) KDν′ (w, X) ≥ C dist(w, ∂Dν′ )−1/2m |X|

holds for all large ν. Let η = min{α/4, ǫ0/4}. Since F ν (bν ) lies in B(q, α) for ν large
enough, we can choose tν ∈ [0, 3sν ] such that F ν ◦ γν (tν ) ∈ ∂B(q, 2η) and F ν ◦ γν ((tν , 3sν ])
is contained in B(q, 2η). From section 2.1 we know that

(2.20) dist(F ν (z), ∂Dν′ ) ≤ C(R) dist(z, ∂Dν )

for any R > 0 and z ∈ Dν ∩ B(0, R) with F ν (z) ∈ Dν′ ∩ (B(q, 2α)). Fix r > 0 and let
z ∈ Dν ∩ B(0, r) be such that F ν (z) ∈ Dν′ ∩ B(q, 2τ ). Since the Kobayashi metric is
decreasing under holomorphic mappings, we get from (2.19) and (2.20) that

KDν (z, X) ≥ KDν′ (F ν (z), dF ν (z)X)

′
≥ C dist(F ν (z), ∂Dν′ )−1/2m |dF ν (z)X|
′
≥ C dist(z, ∂Dν )−1/2m |dF ν (z)X|.

On the other hand it can be seen that for all ν,

KDν (z, X) ≤ |X|/dist(z, ∂Dν )

and this implies the uniform estimate

′
(2.21) |dF ν (z)| ≤ Cdist(z, ∂Dν )−1+1/2m
20 G. P. BALAKUMAR AND KAUSHAL VERMA

for all large ν and z ∈ Dν ∩ B(0, r) such that F ν (z) ∈ Dν ∩ B(p, 2α). Therefore,
Z 3sν
ν ν ν
|F (γ (tν )) − F (bν )| ≤ |dF ν (γ ν (t))||dγν /dt|dt
tν
Z 3sν
′
≤C dist(γν (t), ∂Dν )−1+1/2m dt
tν
1/2m′
≤ Csν →0
as ν → ∞, which is a contradiction.

Proof of theorem 2.11. It was just shown that F ν (z) lies in B(q, 2α) for any z ∈ D̄ν ∩
B(0, δ) if δ > 0 is chosen small enough. Hence (2.21) holds for any w = F ν (z) and a
similar integration argument as above, then gives the estimate asserted in the theorem
with a uniform constant. 
2.5. Compactness of f −1 (0). Recall that to establish theorem 1.1, we were to prove
that f −1 (0) is compact in M. Suppose to obtain a contradiction that this were false.
Then the intersection
f −1 (0) ∩ M ∩ ∂B(0, ǫ) 6= φ.
for all ǫ > 0 small. Since the scaled mappings differ from f by a biholomorphic change of
coordinates on the domain and the target, the same holds for them as well, i.e.,
(F ν )−1 (F ν (0)) ∩ ∂Dν ∩ ∂B(0, ǫ) 6= φ.
Let us show that this property passes to the limit as well, i.e.,
F −1 (q) ∩ ∂D∞ ∩ ∂B(0, ǫ) 6= φ
for all ǫ > 0 small – so small that Theorem 2.11 holds for r = ǫ. Let aν be in
(F ν )−1 (F ν (0)) ∩ ∂Dν ∩ ∂B(0, ǫ) and aν → a ∈ ∂D∞ ∩ ∂B(0, ǫ). Let b be a point in
D∞ near a. Since Dν → D∞ , there exists an integer N such that for all ν > N, b lies in
D ν and
|F (a) − q| ≤ |F (a) − F (b)| + |F (b) − F ν (b)| + |F ν (b) − F ν (aν )| + |F ν (aν ) − q|
By taking ν large enough, the second term can be made arbitrarily small by the conver-
gence of F ν at b and the last term as well by the convergence of F ν (aν ) = F ν (0) to q.
The equicontinuity of the family {F ν } ∪ {F } given by theorem 2.11, then assures that
the same can be done with the third and the first terms, by taking b sufficiently close to
a and ν larger if necessary. Thus, F −1 (q) is not compact in any neighbourhood of the
origin in ∂D∞ .
On the other hand, by starting from the standpoint of F being a holomorphic mapping
between algebraic domains that extends continuously upto a boundary piece Σ of ∂D∞ ,
we may apply the theorem of Webster to establish the algebraicity of F . Thus F extends
as an analytic set and by the invariance property of Segre varieties (see [19]) as a locally
finite to one holomorphic map near 0 ∈ ∂D∞ . Contradiction.
To make this work, we only have to show that F does not map an open piece of ∂D∞
′
into the weakly pseudoconvex points on ∂D∞ . Let us denote by w(Γ) the set of all weakly
pseudoconvex points of a given smooth hypersurface Γ. Since
∂D∞ = z ∈ Cn : 2ℜzn + P (′z,′ z̄) = 0

 SOME REGULARITY THEOREMS FOR CR MAPPINGS 21

it follows that


w(∂D∞ ) = Z × C ∩ ∂D∞
where Z is the real algebraic variety in Cn−1 defined by the vanishing of the determinant
of the complex Hessian of P in Cn−1 . The finite type assumption on ∂D∞ implies that
w(∂D∞ ) is a real algebraic set of dimension at most 2n − 2. Recall that
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z̄1 ) + |z2 |2 + . . . + |zn−1 |2 = 0 ,

F (Σ) ⊂ ∂D∞
and denote Q2m′ by Q for brevity.
Proposition 2.13. The map F extends to an algebraic map in Cn .
Proof. Suppose now that there exists a strictly pseudoconvex point a ∈ ∂D∞ such that
′
F (a) is also a strictly pseudoconvex point in ∂D∞ . Then by [41], F is a smooth CR-
diffeomorphism near a and extends locally biholomorphically across a by the reflection
principle in [39]. Then Webster’s theorem [46] assures us that F is algebraic. So, we
may as well let Σ be the set of strictly pseudoconvex points, assume that F maps it into
′
w(∂D∞ ) and argue only to obtain a contradiction. As before note that
′
) = z ∈ Cn : ∆Q(z1 , z̄1 ) = 0 ∩ ∂D∞′

w(∂D∞ .
′
Then w(∂D∞ ) admits a semi-analytic stratification by real analytic manifolds of dimension
′
2n − 2 and 2n − 3. Using the specific form of ∂D∞ , this can be explicitly described as
follows. Let us denote by V = V (∆Q) the real algebraic variety in C defined by the
polynomial ∆Q. The set of singular points Sng(V ), near which V may fail to be a smooth
′
curve is finite. Let a ∈ w(∂D∞ ) be such that π1 (a) ∈ Reg(V ) where π1 is the natural
surjection onto the z1 -axis. Then V is a smooth curve near π1 (a) and after a biholomorphic
change of coordinates, we may assume that in its vicinity V coincides with {ℑz1 = 0}.
Note that the fibre of π1 over a in w(∂D∞ ′
) namely π1−1 (π1 (a)) ∩ w(∂D∞
′
), is given by
n o n o
n 2 2 n
z ∈ C : 2ℜzn + |z2 | + . . . + |zn−1 | = −Q(π1 (a), π1 (a)) ∩ z ∈ C : z1 = π1 (a)

which is evidently equivalent to ∂Bn−1 ⊂ Cn−1 . This description of the fibres of π1

′
restricted to w(∂D∞ ) persists in a neighbourhood of a and then the real analytic strata
S2n−2 ⊂ w(∂D∞ ) of dimension 2n − 2 is locally biholomorphic to ∂Bn−1 × {ℑz1 = 0}. The
′
′
complement of S2n−2 in w(∂D∞ ), which we will denote by S2n−3 is locally biholomorphic
n−1
to ∂B × Sng(V ) and this evidently has dimension 2n − 3. Observe also that S2n−2 is
generic since its complex tangent space has dimension n − 2 as a complex vector space.
Now, if
F (Σ) ∩ S2n−2 6= φ
then by the continuity of F , there is an open piece of Σ that is mapped by F into S2n−2 .
Denote this piece by Σ̃. We may assume that 0 ∈ Σ̃ and that F (0) = 0 ∈ S2n−2 . Using
an idea from Lemma 3.2 in [28], let L be a 2 dimensional complex plane that intersects
′
w(∂D∞ ) in a totally real submanifold of real dimension 2 – this is possible by the genericity
of S2n−2 – and this also holds for all translates La′ of L passing through a′ in a sufficiently
small neighbourhood U ′ of the origin. We may therefore find a non-negative, strictly
plurisubharmonic function φa′ on U ′ that vanishes on
Sa′ = L′a′ ∩ w(∂D∞
′
) ∩ U ′.
22 G. P. BALAKUMAR AND KAUSHAL VERMA

Indeed by a change of coordinates, we may assume that

L = spanC h∂/∂z1 , ∂/∂zn i = {z2 = . . . = zn−1 = 0}
and then
φa′ = |z2 − a′2 |2 + . . . + |zn−1 − a′n−1 |2 + |ℑz1 |2 + (r∞
′
(z, z̄))2
furnishes an example. By the continuity of F and the openness of Σ̃ in ∂D∞ , we can pick
b ∈ D∞ so near the origin that F (b) ∈ U ′ and ∂Ab ⊂ Σ̃ where
Ab = {z ∈ Cn : zn = bn } ∩ F −1 (LF (b) )
Since the pull-back of an analytic set under a holomorphic map is again analytic of no lesser
dimension, Ab is a positive dimensional analytic set. Also, b ∈ Ab and F (∂Ab ) ⊂ SF (b) .
Therefore, ψb = φF (b) ◦ F is a non-negative, plurisubharmonic funtion on Ab that vanishes
on ∂Ab . By the maximum principle ψb ≡ 0 on Ab which implies that F maps all of Ab
′
into SF (b) . Since F maps D∞ into D∞ , this is a contradiction.
To finish, note that the the remaining possibility is F (Σ) ⊂ S2n−3 . The above argument
can be repeated in this case as well – we will only need to replace |ℑz1 |2 as a subharmonic
function vanishing on a curve-segment of Reg(V ) by |p(z1 )|2 where p is a holomorphic
polynomial that vanishes on the finite set Sng(V ). 

3. Proof of theorem 1.3

We begin with the scaling template for (D, D ′ , f ) in section 2.1, maintaining as far as
possible the notations therein. For clarity and completeness, let us briefly describe the
scaling of D ′ which is simpler this time as M ′ is strongly pseudoconvex. As before, let
pν = (′ 0, −δν ) ∈ D and note that p′ν = f (pν ) converges to the origin which is a strongly
pseudoconvex point on ∂D ′ . Let w ν ∈ ∂D ′ be such that
|w ν − p′ν | = dist(p′ν , ∂D ′ ) = γν
Furthermore, since ∂D ′ is strongly pseudoconvex near the origin, we may choose a strongly
plurisubharmonic function in a neighbourhood of the origin that serves as a defining
function for D ′ . Arguing as in Section 2.1, it follows that f preserves the distance to the
boundary, i.e.,
δν ≈ γν
′ν
for ν ≫ 1. For each w , lemma 2.2 in [40] provides a degree two polynomial automorphism
of Cn that firstly, transfers w ′ν and the normal to ∂D ′ there, to the origin and the ℜzn -axis
respectively and secondly, ensures that the second order terms in the Taylor expansion of
the defining function r ′ ◦ (g ν )−1 of the domain g ν (D ′ ) constitute a hermitian form that
coincides with the standard one i.e., |z1 |2 + . . . + |zn−1 |2 , upon restriction to the complex
tangent space. Define the dilations
B ν (′ z, zn ) = (γν−1/2 ′ z, γν−1 zn )
and note that the scaled domains Dν′ = (B ν ◦ g ν )(D ′ ) are defined by
γν−1 rν′ (z) = 2ℜ(g ν (z))n + |(g ν (z))1 |2 + . . . + |(g ν (z))n−1 |2 + O(γν1/2 ).
These converge in the Hausdorff metric to
′
D∞ = H = {z ∈ Cn : 2ℜzn + |z1 |2 + . . . + |zn−1 |2 < 0},
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 23

which is the unbounded manifestation of the ball, in view of the fact that the g ν ’s converge
uniformly on compact subsets of Cn to the identity map.
Standard arguments as in [40] show that the scaled maps
F ν = B ν ◦ g ν ◦ f ◦ (T ν )−1
converge to a map F : D∞ → H. If some point z0 of D∞ is sent by F to w0 ∈ ∂H ∪ {∞},
then composing F with a local peak function at w0 , we get a function holomorphic on a
neighbourhood of z0 and peaking precisely at z0 . By the maximum principle, F (z) ≡ w0 .
However, F ν (′ 0, −1) = B ν ◦ g ν ◦ f ◦ T ν (′ 0, −1) = B ν ◦ g ν (f (pν )) = B ν (′ 0, −γν ) = (′ 0, −1).
Hence, F (′ 0, −1) = (′ 0, −1) 6= w0 . Thus, F maps D∞ into D∞ ′
and is again as in section
2, a non-degenerate, locally proper map extending continuously upto the boundary in a
neighbourhood of the origin. By theorem 2.1 in [19], F extends holomorphically across
the origin. By composing with a suitable automorphism of the ball, we may also assume
F (0) = 0. Then, ℜ(Fn (z)) is a pluriharmonic function that is negative on D∞ and attains
a maximum at the origin. So, by the Hopf lemma, we must have α = ∂(ℜFn )/∂xn (0) > 0
which combined with the fact that DF preserves the complex tangent space and thereby
′
the complex normal at the origin (to the hypersurfaces ∂D∞ and ∂D∞ , which themselves
correspond under F near the origin), implies that
Fn (z) = αzn + g(z)
for some holomorphic function g with g(z) = o(|z|). Now, let us compare the two defining
functions for D∞ , near the origin:
2ℜ(Fn (z)) + |F1 (z)|2 + . . . + |Fn−1 (z)|2 = h(z, z̄)(2ℜzn + P (′ z,′ z̄))
for a non-vanishing real analytic function h(z, z̄). Contemplate a weighted homogeneous
expansion of the above equation with respect to the weight (1/mn , . . . , 1/m2 ) given by
the multitype of ∂D∞ . Note firstly that on the left, pluriharmonic terms arise precisely
from ℜ(Fn (z)). Next, note that the lowest possible weight for any term on the right is one
and the non-pluriharmonic component of this weight is h(0)P (′ z,′ z̄). What this means
for the left, is that each Fj must expand as
Fj (z) = Pj (′ z) + (terms of weight > 1/2)
where each Pj is either weighted homogeneous of weight 1/2 or identically zero and
(3.1) h(0)P (′z,′ z̄) = |P1 (′ z)|2 + . . . + |Pn−1(′ z)|2
Clearly, all the Pj ’s cannot be zero as P is non-zero. In fact, the finite type character
of ∂D∞ forces all of them to be non-zero, as follows. After a rearrangement if necessary,
assume that Pj ∈ C[′ z] = C[z1 , . . . , zn−1 ] is non-zero precisely when 1 ≤ j ≤ m ≤ n − 1.
Then the common zero set V ⊂ Cn−1 , of these Pj ’s gives rise to the complex analytic
variety i(V ) in ∂D∞ where i : Cn−1 → Cn is the natural inclusion. The finite type
constraint compels this variety and needless to say V , to be discrete. Furthermore, the
weighted homogeneity reduces it to {0} : If (z1 , . . . , zn−1 ) is a non-trivial zero of the Pj ’s
and t ∈ C, then
Pj (et/mn z1 , . . . , et/m2 zn−1 ) = et/2 Pj (z1 , . . . , zn−1 ) = 0
for all 1 ≤ j ≤ m and so the entire curve defined by
γ(t) = (et/mn z1 , . . . , et/m2 zn−1 )
24 G. P. BALAKUMAR AND KAUSHAL VERMA

lies inside V . Now, consider the ideal I generated by these polynomials P1 , . . . , Pm , which
is a C-algebra whose transcendence degree cannot exceed m ≤ n − 1. On the other hand,
by the Nullstellensatz, for a large integer N the algebraically independent monomials,
z1N , . . . , zn−1
N
must all lie in I, forcing its transcendence degree over C and hence m to equal
n − 1. The upshot therefore, is that P is the squared norm of a weighted homogeneous
polynomial endomorphism P̂ of Cn−1 with P̂ −1(0) = 0. Now, put z2 = . . . = zn−1 = 0 in
(3.1). This cannot reduce the left hand side there to zero, otherwise the z1 -axis will lie
inside the zero set of P and hence in ∂D∞ . What this means for the right hand side of
(3.1), is that terms involving z1 (and z̄1 ) must occur. Since all polynomials therein are
homogeneous of same weight, we conclude that terms involving z1 , z̄1 there, must only
be of the form c|z1 |mn for some c > 0. Needless to say, the same holds for all the other
variables zj for 2 ≤ zj ≤ n − 1, as well and we have the last statement of theorem 1.3,
namely
P (′ z,′ z̄) = c1 |z1 |mn + c2 |z2 |mn−1 + . . . + cn−1 |zn−1 |m2 + mixed terms
with all cj ’s being positive and the mixed terms comprising of weight one monomials in
′
z,′ z̄ each of which is annihilated by at least one of the natural quotient maps C[′ z,′ z̄] →
C[′ z,′ z̄]/(zj z̄k ), 1 ≤ j, k ≤ n − 1 where j 6= k.

4. Proof of theorem 1.4

Proof. As before by the lower semi-continuity of rank, ∂D ′ is of rank at least n − 2 in a
neighbourhood Γ′ ⊂ ∂D ′ of p′ which we may assume also to be of finite type and pseudo-
convex, consequently regular. Let us apply Theorem A of [13] to the proper holomorphic
correspondence f −1 : D ′ → D. The hypothesis on clf −1 (Γ′ ) there holds, since ∂D is glob-
ally regular. Therefore by that theorem, f −1 extends continuously upto Γ′ as a proper
correspondence. However, we are not certain of the splitting of f −1 near p into branches
– for that p we will have to lie away from the branch locus of f −1 . Nevertheless, there
exist neighbourhoods U ′ of p′ and U of p and a (local) correspondence
−1
floc : U ′ ∩ D ′ → U ∩ D,
−1
extending continuously upto the boundary such that the graph of floc is contained in that
of f −1 and clf −1 (p′ ) = {p} where the last condition comes from the fact that f is finite
loc
to one upto the boundary, by that theorem again. Assume that both p = 0 and p′ = 0
and choose a sequence p′ν = (′ 0, −δν ) ∈ D ′ ∩ U ′ on the inner normal approaching the
origin. Since f : D → D ′ is proper and 0 ∈ clf (0) there exists a sequence pν ∈ D with
−1
pν → 0 such that f (pν ) = p′ν . Moreover, by the continuity of floc upto the boundary
and the condition clf −1 (0) = {0}, we may assume after shrinking U, U ′ if necessary, that
loc
−1 ′ν
pν ∈ D ∩ U with floc (p ) = {pν }. Now scale D with respect to {pν } and D ′ with respect
to {p′ν } – to scale D ′ , we only consider the dilations
T ν (z1 , . . . , zn ) = (τ1ν )−1 z1 , (τ2ν )−1 , . . . , (τnν )−1 zn

1/2
where τ1ν = τ (0, δν ), τjν = δν for 2 ≤ j ≤ n − 1 and τnν = δν , while for D we use the
composition B ν ◦ g ν where g ν and B ν are as in section 3. As before, the limiting domains
for Dν = (B ν ◦ g ν )(U ∩ D) and Dν′ = T ν (U ′ ∩ D ′ ) are the ball Bn and
′
= z ∈ Cn : 2ℜzn + Q2m′ (z1 , z̄1 ) + |z2 |2 + . . . + |zn−1 |2 < 0 ,

D∞
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 25

respectively, where this time Q2m′ is homogeneous of degree 2m′ and coincides with the
polynomial of this degree in the (homogeneous) Taylor expansion of the defining function
r ′ for ∂D ′ near the origin. Normality of the scaled mappings
F ν = T ν ◦ f ◦ (B ν ◦ g ν )−1
follows as before by proposition 2.1. But section 2 can no longer guarantee nondegeneracy
of a limit map, owing to the lack of a clear cut boundary distance conservation property
of f . However, the existence of the inverse as a proper correspondence paves the way for a
different approach. To begin with, recall from [35], the notion of normality for correspon-
dences and theorem 3 therein, the version of Montel’s theorem for proper holomorphic
′
correspondences with varying domains and ranges. Let Kµ be an exhaustion of D∞ by
′
compact subsets containing ( 0, −1). To establish the normality of scaled correspondences
−1 ν
(floc ) = (B ν ◦ g ν ) ◦ floc
−1
◦ (T ν )−1 ,
−1 ν
it suffices by theorem 3 of [35] to show that (floc ) (Kµ ) ⋐ D∞ for each µ ∈ N. To this
˜−1 ν
end, fix a Kµ , let (floc ) denote the correspondence floc −1
◦(T ν )−1 and note that (f˜loc
−1 ν
) (Kµ )
ν
is connected (since any of its components must contain p ). Now (viewing the origin in
∂D as the sequence pν that approaches it from the interior of D), we recall the Schwarz
lemma for correspondences given by theorem 1.2 of [45], which assures us that the images
of Kµ under these correspondences (f˜loc −1 ν
) , will be contained in Kobayashi balls about pν
of a fixed size, i.e., there exists R > 0 independent of ν such that
(f˜−1 )ν (Kµ ) ⊂ B K (pν , R),
loc D

where BΩK (p, R) denotes the Kobayashi ball centered at p ∈ Ω of radius R, for any given
domain Ω. Then, since biholomorphisms preserve Kobayashi balls,
) (Kµ ) ⊂ (B ν ◦ g ν )((f˜loc
−1 ν −1 ν
) (Kµ )) ⊂ (B ν ◦ g ν )(BDK K
(′ 0, −1), R


(floc (pν , R)) = BD ν

K
(′ 0, −1), R ⊂ BBKn (′ 0, −1), R+1 for all ν large by lemma 4.4 of



while BD ν
[37], giving−1the
stability of the images of Kµ under scaling. Noting that ( 0, −1), ( 0, −1) ∈ Graph(floc )ν
′ ′
−1 ν
for all ν, we conclude that {(floc ) } must admit a subsequence that converges to a cor-
respondence that is inverse to F , where F : Bn → D∞ ′
is a limit of {F ν } (which means
that the composition of these correspondences contains the graph of the identity). By
[13], both F and the inverse correspondence F −1 extend continuously upto the respective
boundaries and in particular F is finite to one on the boundary. Then, F is smooth upto
the boundary by [9], preserves boundaries and then is algebraic again as in section 2.6.
and thereafter by [5] extends holomorphically past the boundary. Now recall that the
determinant of the Levi forms are related as
2
λBn (z) = λD∞
′ (F (z))|JF (z)|

for all z ∈ ∂Bn , which readily gives the strict pseudoconvexity of ∂D∞ ′
. In particular
therefore, Q2m′ (z1 , z̄1 ) must be |z1 | which gives the strict pseudoconvexity of p′ in ∂D ′ . 
2

Proof of corollary 1.5. By the proof of Theorem 1.1 in section 2, we have that f −1 (f (p))
is compact in M and as described in [9], it is possible to choose neighbourhoods U of
p and U ′ of p′ in Cn such that if D and D ′ are the pseudoconvex sides of U ∩ M and
U ′ ∩ M ′ respectively then f extends to be a proper map from D into D ′ , putting us in
the situation of theorem 1.4. 
26 G. P. BALAKUMAR AND KAUSHAL VERMA

References
[1] M. S. Baouendi, L. P. Rothschild: Germs of CR maps between real analytic hypersurfaces, Invent.
Math. 93 (1988), 481–500.
[2] M. S. Baouendi, H. Jacobowitz, F. Treves: On the analyticity of CR mappings, Ann. of Math. (2)
122 (1985), 365–400.
[3] E. Bedford: Proper holomorphic mappings, Bull. Amer. Math. Soc. (N.S) 10 (1984), 157–175.
[4] E. Bedford, E. Barletta: Existence of proper mappings from domains in C2 , Indiana Univ. Math.
J. 39 (1990), no. 2, 315–338.
[5] E.Bedford, S.Bell: Extension of proper holomorphic mappings past the boundary, Manuscripta
Math. 50 (1985), 1-10.
[6] E. Bedford, J. E. Fornaess: Local extension of CR functions from weakly pseudoconvex boundaries,
Michigan Math. J. 25 (1978), 259–262.
[7] S. Bell: CR maps between hypersurfaces in Cn , in Several Complex Variables and Complex Ge-
ometry, Part 1 (Santa Cruz, Calif., 1989), Proc. Sympos. Pure Math. 52, Part 1, Amer. Math.
Soc., Providence, (1991), 13–22.
[8] S. Bell: Local regularity of CR homeomorphisms, Duke Math. J. 57 (1988), 295–300.
[9] S. Bell, D. Catlin: Regularity of CR mappings, Math. Z. 199 (1988), 357–368.
[10] S. Bell, R, Narasimhan: Proper holomorphic mappings of complex spaces, in Several Complex
Variables (W. Barth, R. Narasimhan Eds.), Encyclopaedia Math. Sci. 69, Springer-Verlag, Berlin
(1990).
[11] F. Berteloot: Attraction des disques analytiques et continuité höldérienne dpplications holomorphes
propres, in Topics in Complex Analysis (Warsaw, 1992), Banach Center Publ. 31, Polish Acad.
Sci., Warsaw, (1995), 91–98.
[12] F. Berteloot: Principle de Bloch et estimations de la metrique de Kobayashi des domains in C2 ,
J. Geom. Anal. 1 (2003), 29–37.
[13] F. Berteloot, A. Sukhov: On the continuous extension of holomorphic correspondences, Ann.
Scuola Norm. Sup. Pisa Cl. Sci. (4) 24 (1997), no. 4, 747–676.
[14] F. Berteloot, G. Coeuré : Domaines de C2 , pseudoconvexes et de type fini ayant un groupe non
compact d’automorphismes, Ann. Inst. Fourier (Grenoble) 41 (1991), no. 1, 7786.
[15] D. Catlin: Estimates of invariant metrics on pseudoconvex domains of dimension two, Math. Z.
200 (1989), 429–466.
[16] D. Catlin: Boundary invariants of pseudoconvex domains, Ann. of Math. (2) 120 (1984), no. 3,
529586.
[17] S. Cho: A lower bound on the Kobayashi metric near a point of finite type in Cn , J. Geom. Anal.
2 (1992), no. 4, 317–325.
[18] S. Cho: Boundary behavior of the Bergman kernel function on some pseudoconvex domains in Cn ,
Trans. Amer. Math. Soc. 345 (1994), no. 2, 803–817.
[19] B. Coupet, S. Pinchuk: Holomorphic equivalence problem for weighted homogeneous rigid domains
in Cn+1 Complex analysis in modern mathematics (Russian), 5770, FAZIS, Moscow, 2001.
[20] B. Coupet, A. Sukhov: On CR mappings between pseudoconvex hypersurfaces of finite type in C2 ,
Duke Math. J. 88 (1997), no. 2, 281–304.
[21] B. Coupet, H. Gaussier, A. Sukhov: Regularity of CR maps between convex hypersurfaces of finite
type, Proc. Amer. Math. Soc. 127 (1999), no. 11, 3191–3200.
[22] B. Coupet, S. Pinchuk, A. Sukhov: On boundary rigidity and regularity of holomorphic mappings,
Internat. J. Math. 7 91996), 617–643.
[23] J. D’Angelo: Real hypersurfaces, orders of contact, and applications, Ann. of Math. 115 (1982),
615–637.
[24] K. Diederich, J. E. Fornaess: Pseudoconvex domains: An example with nontrivial Nebenhülle,
Math. Ann. 225 (1977), 275-292.
[25] K. Diederich, J. E. Fornaess: Proper holomorphic images of strictly pseudoconvex domains, Math.
Ann. 259 (1982), no. 2, 279–286.
[26] K. Diederich, G. Herbort: Pseudoconvex domains of semiregular type, Contributions to complex
analysis and analytic geometry, 127161, Aspects Math., E26, Vieweg, Braunschweig, 1994.
 SOME REGULARITY THEOREMS FOR CR MAPPINGS 27

[27] K. Diederich, S. Pinchuk: Proper holomorphic maps in dimension 2 extend, Indiana Univ. Math.
J. 44 (1995), 1089–1126.
[28] K. Diederich, S. Pinchuk: Regularity of continuous CR maps in arbitrary dimension, Michigan
Math. J. 51 (2003), 111–140.
[29] K. Diederich, S. Webster: A reflection principle for degenerate real hypersurfaces, Duke Math. J.
47 (1980), 835–845.
[30] J. E. Fornaess, N. Sibony: Construction of P.S.H. functions on weakly pseudoconvex domains,
Duke Math. J. 58 (1989), no. 3, 633–655.
[31] F. Forstneric: Proper holomorphic mappings: A survey, in Several Complex Variables, Proceedings
of the special year at the Mittag-Leffler Institute (Ed. J. E. Fornaess) (1993), 297–363, Princeton
University Press, Princeton, NJ.
[32] H. Gaussier: Smoothness of Cauchy Riemann maps for a class of real hypersurfaces, Publ. Mat.
45 (2001), no. 1, 79–94.
[33] H. Gaussier: Tautness and complete hyperbolicity of domains in Cn , Proc. Amer. Math. Soc. 127
(1999), no. 1, 105116.
[34] G. Herbort: On the invariant differential metrics near pseudoconvex boundary points where the
Levi form has corank one, Nagoya Math. J. 130 (1993), 25–54.
[35] Klingenberg, W.; Pinchuk, S.: Normal families of proper holomorphic correspondences, Math. Z.
207 (1991), 1, 9196.
[36] S. Krantz: Convexity in complex analysis, Several Complex Variables and Complex Geometry,
Part 1 (Santa Cruz, Calif., 1989), Proc. Sympos. Pure Math. 52, Part 1, Amer. Math. Soc.,
Providence, (1991), 119–137.
[37] P. Mahajan, K. Verma: Some aspects of the Kobayashi and Carathéodory metrics on pseudoconvex
domains, To appear in J. Geom. Anal.
[38] J. Merker: On envelopes of holomorphy of domains covered by Levi-flat hats and the reflection
principle, Ann. Inst. Fourier (Grenoble) 52 (2002), no. 5, 1443–1523.
[39] S. Pinchuk: On the analytic continuation of holomorphic mappings, Math. USSR Sb. 27, 375–392.
[40] S. Pinchuk: Holomorphic inequivalence of certain classes of domains in Cn , Mat. Sb. (N.S.)
111(153) (1980),1, 6794, 159.
[41] S. Pinchuk, Sh. Tsyganov: Smoothness of CR mappings of strictly pseudoconvex hypersurfaces,
Izv. Akad. Nauk. SSSR Ser. Mat. 53 (1989), 1120–1129.
[42] A. Sukhov: On boundary behaviour of holomorphic mappings, (Russian), Mat. Sb. 185 (1994),
131–142; English transl. in Russian Acad. Sci. Sb. Math. 83 (1995), 471–483.
[43] Do D. Thai, Ninh V. Thu: Characterization of domains in Cn by their noncompact automorphism
groups, Nagoya Math. J. 196 (2009), 135–160.
[44] J. M. Trépreau: Sur le prolongement holomorphe des fonctions CR definies sur une hypersurface
reélle de classe C 2 dans Cn , Invent. Math. 43 (1977), 53–68.
[45] K. Verma: A Schwarz lemma for correspondences and applications, Publ. Mat. 47 (2003), no. 2,
373–387,
[46] S. Webster: On the mapping problem for algebraic real hypersurfaces, Invent. Math. 43 (1977),
53–68.
[47] Yu, Ji Ye: Weighted boundary limits of the generalized Kobayashi-Royden metrics on weakly pseu-
doconvex domains, Trans. Amer. Math. Soc. 347 (1995), no. 2, 587614.
[48] Yu, Ji Ye: Peak functions on weakly pseudoconvex domains, Indiana Univ. Math. J. 43 (1994), no.
4, 12711295.

Department of Mathematics, Indian Institute of Science, Bangalore 560 012, India

E-mail address: [email protected]

Department of Mathematics, Indian Institute of Science, Bangalore 560 012, India

E-mail address: [email protected]