Advances in Mathematics 374 (2020) 107388

Contents lists available at ScienceDirect

Advances in Mathematics
www.elsevier.com/locate/aim

Boundary characterization of holomorphic isometric

embeddings between indefinite hyperbolic spaces
Xiaojun Huang a,1 , Jin Lu b,2 , Xiaomin Tang b,∗,3 , Ming Xiao c,4
a
Department of Mathematics, Rutgers University, New Brunswick, NJ 08903, USA
b
Department of Mathematics, Huzhou University, Huzhou, Zhejiang 313000, China
c
Department of Mathematics, University of California, San Diego, 9500 Gilman
Drive, La Jolla, CA 92093, USA

a r t i c l e i n f o a b s t r a c t

Article history: We provide in this paper a boundary characterization in

Received 1 November 2019 terms of the boundary CR invariants (or pseudo-Hermitian
Received in revised form 2 June 2020 invariants) for holomorphic isometric embeddings between
Accepted 16 August 2020
indefinite hyperbolic spaces of general codimensions.
Available online 27 August 2020
Communicated by the Managing © 2020 Elsevier Inc. All rights reserved.
Editors

MSC:
32H35
32H02
32V20

Keywords:
Indefinite hyperbolic spaces
Real hyperquadrics
Isometric mappings
Proper holomorphic mappings

* Corresponding author.
E-mail addresses: [email protected] (X. Huang), [email protected] (J. Lu),
[email protected] (X. Tang), [email protected] (M. Xiao).
1
Supported in part by NSF grant DMS-1665412 and DMS-2000050.
2
Supported in part by NNSF of China (Grant No. 11801172) and NSF of Zhejiang (Grant No.
LY20A010007).
3
Supported in part by NNSF of China (Grant No. 11971165).
4
Supported in part by NSF grant DMS-1800549.

https://doi.org/10.1016/j.aim.2020.107388
0001-8708/© 2020 Elsevier Inc. All rights reserved.
2 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

1. Introduction

For two integers n ≥ 2 and 0 ≤ l < n, we equip C n+1 with a Hermitian form Hl+1
with l + 1 negative and n − l positive eigenvalues. More precisely, write Ik,m for the
m × m diagonal matrix where its first k diagonal elements equal −1 and the rest equal
+1. Then the Hermitian form Hl+1 is given by Hl+1 (z, z) = zIl+1,n+1 z t for z ∈ C n+1 .
This naturally leads to the definition of the generalized ball Bln , which is a domain in
the projective space P n :

Bln = {[z0 , · · · , zn ] ∈ P n : |z0 |2 + · · · + |zl |2 > |zl+1 |2 + · · · + |zn |2 }.

Note when l = 0, B0n becomes the standard unit ball Bn in C n (embedded into P n ).
The generalized ball Bln is indeed an open orbit of the real form SU (l + 1, n + 1) of the
complex simple Lie group SL(n + 1, C) when acting on P n . Here SU (l + 1, n + 1) is the
special unitary group which consists of matrices preserving the Hermitian form Hl+1 on
C n+1 :
t
SU (l + 1, n + 1) = {A ∈ SL(n + 1, C) : AIl+1,n+1 A = Il+1,n+1 }.

The topological boundary ∂Bln of Bln , sometimes called the generalized sphere of
signature l, is the unique closed orbit under the action of SU (l + 1, n + 1) on P n . The
generalized sphere ∂Bln or its local realization, the real hyperquadric


l 
n−1
Hln = {(z, w) = (z1 , · · · , zn−1 , w) ∈ C n : Imw = − |zj |2 + |zj |2 }
j=1 j=l+1

serves as a basic model for Levi-nondegenerate hypersurfaces (see [5]) and plays a fun-
damental role in CR geometry. Note that when l = 0, H0n is the standard Heisenberg
hypersurface. Due to the special geometric structure of ∂Bln or Hln , many striking rigidity
phenomena have been discovered for mappings into the real hyperquadrics. Results along
these lines can be found for instance in [29,11,14,15,9,8,7,13] and references therein. In
particular, in [14] the first author defined a useful geometric invariant for a nonconstant
C 2 -smooth CR map F from ∂Bn to ∂BN (N ≥ n ≥ 2), called the geometric rank of F .
He proved that if n ≥ 2 and N ≤ 2n − 2, then the geometric rank must be identically
zero and furthermore F extends to a linear fractional holomorphic proper map from Bn
to BN .
The mapping property is of different flavor when l > 0. By studying local holomorphic
mappings from Hln to HlN , Baouendi-Huang [4] proved that any proper holomorphic
map from Bln to BlN extends to a totally geodesic embedding from P n to P N whenever
0 < l < n−1 and N ≥ n. After their work, many interesting results have been established.
Here we mention [1,2,10,26,13,12]. It remains an open problem to study the analytic
property of mappings into hyperquadrics in the general setting.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 3

On the other hand, the generalized ball has distinguished geometric feature as well.
Recall the automorphism group of Bln is given by SU (l + 1, n + 1) (see, for example,
section 1 in [5]). The generalized ball Bln possesses a canonical indefinite metric ωBln that
is invariant under the action of its automorphisms:

√ 
l 
n−1

ω Bln
¯
= − −1∂ ∂log |zj | −
2
|zj |2 .
j=0 j=l+1

When l = 0, the metric is identical with the (normalized) Poincaré metric on the unit ball.
A generalized ball equipped with the indefinite metric ωBln is often called an indefinite
hyperbolic space form.
In this paper, we give a complete characterization for local holomorphic isometric
embeddings between indefinite hyperbolic spaces in terms of a boundary CR invariant
of the maps. Let Ω be a connected open set of Bln and F a holomorphic map from Ω to
BlN . We say F is isometric if F ∗ (ωBN ) = ωBln on Ω.
l

Theorem 1.1. Let N ≥ n ≥ 3, 0 ≤ l ≤ n − 1, l ≤ l ≤ N − 1. Let U be an open subset

in P n containing some p ∈ ∂Bln and F be a holomorphic map from U into P N . Assume
U ∩ Bln is connected and F (U ∩ Bln ) ⊂ BlN , F (U ∩ ∂Bln ) ⊂ ∂BlN . Then the following are
equivalent.
(1) F is CR transversal and has geometric rank zero at generic points on U ∩∂Bln near p.
(2) F is an isometric embedding from (U ∩ Bln , ωBln ) to (BlN , ωBN ).
l

We recall that F is called CR transversal at p ∈ ∂Bln if TF (p) (∂BlN ) + dF (Tp P n ) =

TF (p) P N . We remark that if a map F as in the assumption of Theorem 1.1 exists and F
is CR transversal at some point near p, then we must have l ≥ l and N − l ≥ n − l. The
definition of geometric rank, which serves as a crucial invariant for holomorphic maps
between open pieces of the generalized spheres, will be given in Definition 3.3 of §3. It
can be routinely computed through the fourth order jets of the map. In the language of
pseudo-Hermitian geometry, the zero geometric rank at a point q̂ ∈ F (U ∩ ∂Bln ) is equiv-
(1,0)
alent to the condition that for any Xq̂ ∈ Tq̂ F (∂Bln ), the value at Xq̂ of the CR second
 (1,0)
fundamental form (Xq̂ , Xq̂ ) ∈ Tq̂ (∂BlN )/dF (T (1,0) (∂Bln )) of F (∂Bln ) ⊂ ∂BlN stays
 
in the null cone of the Levi form Lq̂ of ∂BlN at q̂, namely, L̂q̂ ( (Xq̂ , Xq̂ ), (Xq̂ , Xq̂ )) = 0.
See Proposition 3.5 for more discussions on this matter.
In general, in the zero geometric rank case, the second fundamental form does not
vanish identically. Instead, its image can have the largest possible real dimension 2(l −l),
that is nonzero unless l = l. This is indeed the main difficulty we will encounter in the
course of the proof of Theorem 1.1. The vanishing of CR second fundamental form is
linked with the linearity (or the total geodesy) of the map, while our main theorem
shows that the zero geometric rank condition, or equivalently the condition that the CR
second fundamental form stays in the null cone of the Levi form, is precisely the one
4 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

to characterize holomorphic isometric embeddings. For results related to the vanishing

of the CR second fundamental form, we refer the reader to [33] and many references
therein.

Remark 1.2.

1. The assumptions that F (U ∩ Bln ) ⊂ BlN and F (U ∩ ∂Bln ) ⊂ ∂BlN do not guarantee
F to be CR transversal in general even at a single boundary point, as the following
Example 1.3 shows.
2. Note the Levi form of the boundary of Bln , l > 0, has at least one negative eigenvalue
at each point. By applying the Lewy extension type theorem for mappings into
compact Kähler manifolds of Siu and Ivashkovich (see references in [4]), we see that
if F is holomorphic from U ∩ Bln , l > 0, into P N , then F extends to a holomorphic
mapping from a neighborhood of ∂Bln ∩ U to P N .
3. Let H be a holomorphic map in U such that F (U ∩ Bln ) ⊂ P n \ BlN . Note P n \ BlN ≈
−l −1 . We can thus regard H as a map from U ∩ Bl to BN −l −1 .
N n N
BN

Example 1.3. Let F (z) be the polynomial map from P 5 to P N (N ≥ 20) given by
F ([z0 , · · · , z5 ]) = [f, g, 0, · · · , 0]. Here
√ √ √ √ √ √
f = (z02 , 2z0 z1 , 2z0 z2 , z12 , z22 , z32 , z42 , z52 , 2z1 z2 , 2z3 z4 , 2z3 z5 , 2z4 z5 );
√ √ √ √ √ √ √ √ √
g = ( 2z0 z3 , 2z0 z4 , 2z0 z5 , 2z1 z3 , 2z1 z4 , 2z1 z5 , 2z2 z4 , 2z2 z4 , 2z2 z5 ).

Note

f 2 − g 2 = (|z0 |2 + |z1 |2 + |z2 |2 − |z3 |2 − |z4 |2 − |z5 |2 )2 .

Hence F (B25 ) ⊂ B11

N
and F (∂B25 ) ⊂ ∂B11
N
. The map F , however, is not CR transversal at
any boundary point of B25 .

When 1 ≤ l < 2l ≤ n − 1, the CR transversality automatically holds at F (q) for a

generic point q ∈ U ∩ ∂Bln (see [4] and [3]), and the geometric rank of F is always zero
at such a point q. Hence our main theorem gives, in this special case, a different proof of
the following theorem obtained in [2] (see also [26] for a different approach for a global
version of this theorem).

 N −1
Theorem 1.4 (Baouendi-Ebenfelt-Huang [2]). Let N ≥ n, 1 ≤ l ≤ n−1 2 , 1 ≤ l ≤ 2

and 1 ≤ l ≤ l < 2l. Let U be an open subset in P containing some p ∈ ∂Bl with U ∩ Bln
n n

being connected, and F a holomorphic map from U ∩ Bln into BlN . Assume that for any
sequence {qj }∞ ∞
j=1 ⊂ U ∩Bl that converges to ∂Bl , the limit set of {F (qj )}j=1 is contained
n n

in ∂Bl . Then F is an isometric embedding from (U ∩ Bl , ωBln ) into (Bl , ωBN ).

N n N
l
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 5

We remark that in [15], the first author proved a semi-rigidity theorem for proper
holomorphic map as the codimension increases. For holomorphic maps between general-
ized balls, the rigidity gradually disappears as the difference of the signature increases.
Our main theorem will provide a useful tool for such a study in the future as the rank
zero maps always appear in any signature difference case.

Moreover, the mapping problem between indefinite hyperbolic spaces has been re-
cently discovered to be of critical importance in the study of mappings between bounded
symmetric domains. By using holomorphic double fibration, Ng [27] applied the results
for mappings between generalized balls in [4] to prove rigidity properties for proper
maps between the type I domains. Xiao-Yuan [31,30] established rigidity results for
IV IV
proper maps from the unit ball to the type IV domain Dm by regarding Dm as an iso-
metric submanifold of the indefinite hyperbolic space. We remark that every irreducible
bounded symmetric domain (equipped with Kähler-Einstein metric) can be isometri-
cally embedded into an indefinite hyperbolic space (with a normalizing constant) in a
canonical way. For instance, let Dp,q I
(p ≤ q) be the type I classical domain. Recall the
Borel embedding realizes Dp,q (p ≤ q) as an open subset in its compact dual Gp,q (the
I

Grassmannian of p-plane in C p+q ). And Gp,q can be holomorphically embedded into a

projective space P N for some appropriate N by the Plüker embedding P . Then P maps
I
Dp,q (p ≤ q) isometrically into BlN ⊂ P N for some appropriate l. We should also mention
closely related studies on various rigidity properties for holomorphic proper or isometric
maps, and CR mappings. To name a few, the readers are referred to the work by Eben-
felt [8], Ji [16], Kim-Zaitsev [18,19], Kim [17], Kossovskiy-Lamel [20], Lamel-Mir [22,21],
Mok [23,24], Mok-Ng [25], Yuan-Zhang [32] and references therein.
The paper is organized as follows. We analyze holomorphic isometries between indef-
inite hyperbolic spaces in Section 2. In Section 3, we recall the notion of geometric rank
for mappings between real hyperquadrics and set up notations and definitions needed
later. We also prove the equivalence of the geometric rank zero and Levi null cone con-
dition. Section 4 and Section 5 are devoted to proving the main theorem. The proof is
based on an induction argument. Compared with the method employed in [1] and [2], a
crucial lemma (Lemma 3.2, [14]) due to the first author cannot be applied anymore for τ
could be arbitrarily large, and this poses a major difficulty. Our main tools for the proof
include methods from CR geometry, normal form theory and most importantly the mov-
ing point trick introduced for studying maps between manifolds with huge group actions
in [14].

Acknowledgment

The authors thank the anonymous referees for their helpful comments, which have
greatly improved the readability of the paper. Part of the work was completed when the
first and the fourth authors were visiting Huzhou University in the part of the summers
of 2017, 2018 and 2019. The two authors thank the institute for the hospitality.
6 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

2. Isometries between indefinite hyperbolic spaces

We establish the following theorem in this section:

Theorem 2.1. Let F be a holomorphic map from an open connected subset Ω of Bln to
BlN . Assume Ω is contained in the affine cell U0 = {[z0 , · · · , zn ] ∈ P n : z0
= 0} and F (Ω)
is contained in the affine cell V0 = {[w0 , · · · , wN ] ∈ P N : w0
= 0}. Then the following
are equivalent:
(a). F is an isometric embedding from (Ω, ωBln ) to (BlN , ωBN ).
l
(b). After composing with automorphisms of Bln and BlN from the right and the left,
respectively, F equals to the following map in the local affine coordinates on U0 and V0 :

(z1 , · · · , zl , φ, zl+1 , · · · , zn , ψ).

Here φ, ψ are holomorphic maps with l − l and N − n − l + l components, respectively,

and satisfy φ ≡ ψ .

Here · denotes the usual Euclidean norm. We remark that if a map F as in (a)
exists, then we must have l ≥ l, N − l ≥ n − l. Before we prove the above theorem, we
fix some notations. We denote by δj,l the symbol which takes value -1 when 1 ≤ j ≤ l
and 1 otherwise. If l = 0, δj,0 is identically one for all j ≥ 1. We also denote by δj,l,l ,n
the symbol which takes value -1 when 1 ≤ j ≤ l or n ≤ j ≤ n + l − l − 1 and 1 otherwise.
When l = l, δj,l,l,n is the same as δj,l .
Let m ≥ 1. For two m-tuples x = (x1 , · · · , xm ), y = (y1 , · · · , ym ) of complex
m
numbers, we write x, y l = j=1 δj,l xj yj and |x|l = x, x̄ l . Also write x, y l,l ,n =
2
m
j=1 δj,l,l ,n xj yj and |x|l,l ,n = x, x̄ l,l ,n . Note if m ≤ n − 1, the two symbols ·, · and

2 
m l
·, · l,l ,n are identical. We use ·, · to denote the usual inner product: x, y = j=1 xj yj .
Denote by Il,m the m × m diagonal matrix whose jth diagonal element equals to
δj,l , 1 ≤ j ≤ m. Similarly we define Il ,m . Write Il,l ,n,m for the m × m diagonal ma-
trix whose jth diagonal element equals to δj,l,l ,n , 1 ≤ j ≤ m.

Proof of Theorem 2.1. It is easy to see (b) implies (a). We will show (a) implies (b).
Let F : Ω → BlN be as in the theorem. Write p0 = [1, 0, · · · , 0] ∈ U0 . By composing
F with automorphisms of Bln and BlN and shrinking Ω if necessary, we can assume
that p0 ∈ Ω, F (p0 ) = [1, 0, · · · , 0] ∈ V0 , and F (Ω) ⊂ V0 . Write F (z1 , · · · , zn ) =
(F1 (z1 , · · · , zn ), · · · , FN (z1 , · · · , zn )) in the local affine coordinates of U0 and V0 , which
are identified as complex Euclidean spaces. By the isometry assumption, we have
∂ ∂¯ log(1 − |F |2l ) = ∂ ∂¯ log(1 − |z|2l ). Since now F (0) = 0, by a standard reduction,
we get 1 − |F |2l = 1 − |z|2l or |F |2l = |z|2l . The conclusion then follows from the following
Proposition 2.2. 
t
Recall U (l , N ) = {A ∈ GL(N, C) : AIl ,N A = Il ,N }.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 7

Proposition 2.2. Let f = (f1 , · · · , fN ) and g = (g1 , · · · , gn ) be two holomorphic maps

on an open connected set V ⊂ C m . Assume that l ≤ n, l ≤ N , and |f |2l = |g|2l on V ,
and {g1 , · · · , gn } is a linearly independent set over C. Then l ≥ l and N − l ≥ n − l.
Moreover, there exists a matrix T ∈ U (l , N ) and two holomorphic maps φ, ψ with l − l
and N − n − l + l components, respectively, such that
(1): (f1 , · · · , fN ) = (g1 , · · · , gl , φ, gl+1 · · · , gn , ψ)T .
(2): φ ≡ ψ .

Proof. It follows from [28] that l ≥ l and N − l ≥ n − l. We will thus prove only the
latter part of the conclusion. To make notations simple, by reordering the components
of f , we assume |f |2l,l ,n+1 = |g|2l .
We write the vector space W = SpanC {g1 , · · · , gn , f1 , · · · , fN }. Since {g1 , · · · , gn } is a
linearly independent set over C, we can extend it a basis of W : {g1 , · · · , gn , ϕ1 , · · · , ϕk }.
Here k ≥ 0 (k = 0 means no ϕj s appear) and ϕ1 , · · · , ϕk are holomorphic functions on V .
Note there is an (n + k) × N matrix B such that (f1 , · · · , fN ) = (g1 , · · · , gn , ϕ1 , · · · ,
ϕk )B. Then it yields that

|f |2l,l ,n+1 = f Il,l ,n+1,N f¯t

(2.1)
= (g1 , · · · , gn , ϕ1 , · · · , ϕk )BIl,l ,n+1,N B̄ t (ḡ1 , · · · , ḡn , ϕ̄1 , · · · , ϕ̄k )t .

By assumption, (2.1) equals to |g|2l . We will need the following lemma.

Lemma 2.3. Let h1 , ..., hm be m linearly independent holomorphic functions in an open

connected set V . Assume that (h1 , ..., hm )C(h1 , ..., hm )t = 0, where C is a Hermitian
matrix, then C = 0. Consequently, if

(h1 , ..., hm )C1 (h1 , ..., hm )t = (h1 , ..., hm )C2 (h1 , ..., hm )t ,

where C1 , C2 are Hermitian matrices, then C1 = C2 .

Proof of Lemma 2.3. We only prove the first part of the lemma. The second part of the
statement is an easy consequence. We will prove by seeking a contradiction. Suppose
t
C
= 0. First write C = P DP , where D = diag(λ1 , ..., λs , 0, ..., 0), s > 0, is a diagonal
matrix, with all λi
= 0, λ1 ≥ ... ≥ λs , and P is a unitary matrix. We write (g1 , ..., gm ) :=
s
(h1 , ..., hm )P , which are also linearly independent. By hypotheses we have i=1 λi |gi |2 =
0. Clearly, λi cannot be all positive. Assume λ1 ≥ ... ≥ λν > 0 > λν+1 ≥ ... ≥ λs . It
yields that,


ν 
s
λi |gi |2 = (−λi )|gi |2 .
i=1 i=ν+1

It then follows from a lemma of D’Angelo ([6]) that the functions g1 , ..., gm are linearly
dependent. We thus get a contradiction. 
8 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Since (2.1) equals |g|2l , it follows from Lemma 2.3 that

BIl,l ,n+1,N B̄ t = diag(−1, · · · , −1, 1, · · · , 1, 0, · · · , 0) (2.2)

where on the right hand side of (2.2), there are l negative and n − l positive ones, and
k zeros on the diagonal. Now denote the rows of B by α1 , · · · , αn , β1 , · · · , βk , where
αi , βj , 1 ≤ i ≤ n, 1 ≤ j ≤⎛k are⎞N -dimensional row vectors. As a consequence of (2.2),
α1
⎜ ⎟
we have if we write D = ⎝ ... ⎠, then DIl,l ,n+1,N D̄t = Il,n . By page 386 of [4], we are
αn
able to extend {α1 , · · · , αn } to {α1 , · · · αn , αn+1 , · · · , αN } such that

AIl,l ,n+1,N Āt = Il,l ,n+1,N ,

where A is the N × N matrix whose jth row is αj , 1 ≤ j ≤ N . Thus A ∈ U (l, l , n + 1, N ).

Here U (l, l , n + 1, m) = {T ∈ GL(m, C) : T Il,l ,n+1,m T̄ t = Il,l ,n+1,m }. Consequently,
Il,l ,n+1,N Āt and C := (Il,l ,n+1,N Āt )−1 are also in U (l, l , n + 1, N ). Now set

fˆ = f Il,l ,n+1,N Āt = (g1 , · · · , gn , ϕ1 , · · · , ϕk )BIl,l ,n+1,N Āt .

Note by (2.2), αi , βj l,l ,n+1 = 0 for 1 ≤ i ≤ n, 1 ≤ j ≤ k. A direct computation verifies

that

t Il,n On×(N −n)
BIl,l ,n+1,N Ā = .
Ok×n M

Here Op×q denotes the p ×q zero matrix, M is a certain k×(N −n) matrix. Consequently,

fˆ = (−g1 , · · · , −gl , gl+1 , · · · , gn , h1 , · · · , hN −n ).

Write f˜ = fˆIl,N = f Il,l ,n+1,N Āt Il,N . Then

f˜ = (g1 , · · · , gl , gl+1 , · · · , gn , h1 , · · · , hN −n ).

Since Il,l ,n+1,N Āt Il,N ∈ U (l, l , n + 1, N ), we have |f˜|2l,l ,n+1 = |f |2l,l ,n+1 = |g|2l .
l −l N −n
j=1 |hj | j=l −l+1 |hj | . Writing φ = (h1 , · · · , hl −l ) and
2 2
This yields that =
˜
ψ = (hl −l+1 , · · · , hN −n ), we have f = f T = (g1 , · · · , gl , gl+1 , · · · , gn , φ, ψ)T and
φ 2 ≡ ψ 2 . Here T is the inverse of Il,l ,n+1,N Āt Il,N , which is still in U (l, l , n + 1, N ).
Then we reorder the components of f and f˜ back to obtain Proposition 2.2. 

3. Geometric rank, second fundamental form and Levi null cone

In this section, we give the definition of geometric rank of CR transversal maps in the
positive signature case and justify its invariant property. We also show the equivalence
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 9

between geometric rank zero condition and the null cone condition of the values of CR
second fundamental form associated with the image manifold of the mapping.
We first set up certain notations and terminologies which will also be needed in §4
and §5 for the proof of our main theorem. For 0 ≤ l ≤ n − 1, we define the generalized
Siegel upper-half space


l 
n−1
Sln = {(z, w) ∈ C n−1 × C : Im(w) > − |zj |2 + |zj |2 }.
j=1 j=l+1

The boundary of Sln is the standard hyperquadrics: Hln = {(z, w) ∈ C n−1 × C : Im(w) =
n−1 
j=1 δj,l |zj | }. We also define for l ≤ l ≤ N − 1
2


N −1
N −1
 ,n = {(z, w) ∈ C × C : Im(w) > δj,l,l ,n |zj |2 }.
N
Sl,l
j=1

We similarly define SlN , HlN , Hl,l  ,n . Now for (z, w) = (z1 , · · · , zn−1 , w) ∈ C , let
N n

Ψn (z, w) = [i + w, 2z, i − w] ∈ P . Then Ψn is the Cayley transformation which bi-

n

holomorphically maps the generalized Siegel upper-half space Sln and its boundary Hln
onto Bln \ {[z0 , · · · , zn ] : z0 + zn = 0} and ∂Bln \ {[z0 , · · · , zn ] : z0 + zn = 0}, respectively.
N N   N
Note Hl,l  ,n is identical to Hl if l = l. When l > l, Hl is holomorphically equivalent
N N N
to Hl,l ,n by a permutation of coordinates in C . We will more often work with Hl,l  ,n

instead of HlN , as it makes notations simpler.

We will write Aut(Hln ) and Aut0 (Hln ) for the (holomorphic) automorphism group of
Hln and the isotropy group of Hln at 0, respectively. Write Aut+ (Hln ) and Aut+ n
0 (Hl ) for
n n
the automorphisms in Aut(Hl ) and Aut0 (Hl ), respectively, that in addition preserves
sides (that is, maps Sln to Sln ). Clearly they are subgroups of Aut(Hln ) and Aut0 (Hln ),
N N + N + N
respectively. We define Aut(Hl,l  ,n ), Aut0 (Hl,l ,n ) and Aut (Hl,l ,n ) and Aut0 (Hl,l ,n )

similarly.
Recall we denote by (z, w) = (z1 , · · · , zn−1 , w) the coordinates of C n . Write u for the
real part of w and write

∂ ∂ ∂
Lj := 2iδj,l z¯j + , 1 ≤ j ≤ n − 1, T := . (3.1)
∂w ∂zj ∂u

Then {L1 , · · · , Ln−1 } forms a global basis for the CR tangent bundle T (1,0) Hln of Hln ,
where T is a tangent vector field of Hln transversal to T (1,0) Hln ⊕ T (0,1) Hln .
Let F = (f˜, g) = (f, φ, g) = (f1 , · · · , fn−1 , φ1 , · · · , φN −n , g) be a holomorphic map
from a neighborhood U of p0 ∈ Hln to C N , satisfying F (U ∩Sln ) ⊂ Sl,l N
 ,n and F (U ∩Hl ) ⊂
n

Hl,l ,n . We additionally assume M1 := U ∩ Hl is connected and F is CR transversal on

N n

M1 . We will define the geometric rank for such a map F .

First for each p ∈ M1 , we associate it with a map Fp defined by

Fp = τpF ◦ F ◦ σp0 = (f˜p , gp ) = (fp , φp , gp ). (3.2)

10 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Here for each p = (z0 , w0 ) ∈ M1 , we write σ(z

0
0 ,w0 )
∈ Aut+ (Hln ) for the map

0
σ(z0 ,w0 )
(z, w) = (z + z0 , w + w0 + 2iz, z¯0 l ),

0
and define τ(z0 ,w0 )
∈ Aut+ (Hl,l
n
 ,n ) by

F
τ(z0 ,w0 )
(ξ, η) = (ξ − f˜(z0 , w0 ), η − g(z0 , w0 ) − 2iξ, f˜(z0 , w0 ) l,l ,n ).

Then Fp is a holomorphic map in a neighborhood of 0 ∈ C n , which sends an open piece

of Hln into Hl,l  ,n with Fp (0) = 0. Moreover, F (U ∩ Sl ) ⊂ Sl,l ,n .
N n N

Note the fundamental commutator identities hold:

∂ ∂ ∂
[L̄j , Lj ] = 2iδj,l ( + ) = 2iδj,l , 1 ≤ j ≤ n − 1;
∂w ∂ w̄ ∂u (3.3)
[L̄j , Lk ], [T, Lk ], [Lj , Lk ], [Lk , Lk ] = 0, if 1 ≤ j
= k ≤ n − 1.

By the assumption that F (U ∩ M1 ) ⊂ Hl,l

N
 ,n , we have

Im g = f˜, f¯˜ l,l ,n on M1 . (3.4)

In the following, for a holomorphic map h = (h1 , · · · , hK ) from C n to C K , we write

∂ 2 h1 ∂ 2 hK
hzj = ( ∂h  
∂zj , · · · , ∂zj ), hwzj = hzj w = ( ∂w∂zj , · · · , ∂w∂zj ), 1 ≤ j ≤ n − 1. The notations
1 ∂hK

hw , hzj zk , hww are understood similarly. We apply L̄j Lj to (3.4) and obtain

λ(p) := (gp )w (0) = gw (p) − 2if˜w (p), f˜(p) l,l ,n = δj,l Lj (f˜), Lj (f˜) l,l ,n (p). (3.5)

Note this implies λ(p) is a real number. Recall the CR-transversality assumption is
equivalent to λ(p)
= 0 (see for example, [4]). Furthermore, since Fp preserves the sides,
we have λ(p) > 0 (see e.g. page 396 in [4]).
We apply L̄k , Lj , j
= k to (3.4) and get Lj (f˜), Lk (f˜) l,l ,n |p = 0. Let for 1 ≤ j ≤ n −1,

 ∂ f˜p   ∂fp,1 ∂fp,n−1 ∂φp,1 ∂φp,N −n 

Ej (p) := |0 = ,··· , , ··· , |0 = Lj (f˜)(p);
∂zj ∂zj ∂zj ∂zj ∂zj
 ∂ f˜p   ∂fp,1 ∂fp,n−1 ∂φp,1 ∂φp,N −n 
Ew (p) := |0 = ,··· , , ··· , |0 = T (f˜)(p).
∂w ∂w ∂w ∂w ∂w

Then

Ej (p), Ej (p) l,l ,n = δj,l λ(p), Ej (p), Ek (p) l,l ,n = 0, 1 ≤ j

= k ≤ n − 1. (3.6)

E (p)
Write E for the (n − 1) × (N − 1) matrix whose jth row is j
λ(p)
,1 ≤ j ≤ n − 1. Then
t
E satisfies EIl,l ,n,N −1 Ē = Il,n−1 . Here Il,n−1 and Il,l ,n,N −1 are as defined in §3.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 11

As in [4], we can choose (N − 1)-dimensional row vectors C1 (p), · · · , CN −n (p) such

that if we write
⎛ E (p) ⎞
1
λ(p)
⎜ ··· ⎟
⎜ ⎟
n−1 (p) ⎟
⎜ E
⎜
A(p) = ⎜ λ(p) ⎟
⎟
⎜ C1 (p) ⎟
⎝ ⎠
···
CN −n (p)

then

t
A(p)Il,l ,n,N −1 A(p) = Il,l ,n,N −1 , i.e., A(p) ∈ U (l, l , n, N − 1). (3.7)

Here recall U (l, l , n, m) = {T ∈ GL(m, C) : T Il,l ,n,m T̄ t = Il,l ,n,m }. Note that one
can choose Cj (p)’s in such a way that A(p) is smooth in p for p ≈ p0 by the standard
Gram-Schmidt process.
t
Next note B(p) := A−1 (p) = Il,l ,n,N −1 A(p) Il,l ,n,N −1 is also in U (l, l , n, N − 1).
Write

B(p) = (B1 (p), · · · , Bn−1 (p), B̂n (p), · · · , B̂N −1 (p)),

where Bj (p) s and B̂i (p) s are (N − 1)-dimensional column vectors. Note B1 (p), · · · ,
Bn−1 (p) only depend on E1 (p), · · · , En−1 (p). Indeed, we have

 t t
E1 (p) En−1 (p)
(B1 (p), · · · , Bn−1 (p)) = Il,l ,n,N −1  ,··· ,  Il,n−1 . (3.8)
λ(p) λ(p)

Define Fp∗ = (f˜p∗ , gp∗ ) = ((fp∗ )1 , · · · , (fp∗ )n−1 , (φ∗p )1 , · · · , (φ∗p )N −n , gp∗ ) by

 
1 B(p) 0
Fp∗ = Fp 1 . (3.9)
λ(p) 0 λ(p)

Then Fp∗ is a holomorphic map in a neighborhood of 0 ∈ C n , which sends an open piece

∗
of Hln into Hl,l
N
 ,n with Fp (0) = 0 and the following holds (see [4], [2] for more details).

⎧
⎪ ∗
⎨fp = z + O(|w| + |(z, w)| )
⎪ 2

φ∗p = O(|w| + |(z, w)|2 )

⎪
⎪
⎩g ∗ = w + O(|(z, w)|2 ).
p

Let
12 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

∂ f˜p∗ 1
a(p) = (a1 (p), · · · , an−1 (p), an (p), · · · , aN −1 (p)) := (0) =  Ew (p)B(p).
∂w λ(p)
(3.10)
Note
1 1
ak (p) =  Ew (p)Bk (p) for 1 ≤ k ≤ n − 1, and |a(p)|2l,l ,n = |Ew (p)|2l,l ,n .
λ(p) λ(p)
(3.11)
Set for 1 ≤ k, j ≤ n − 1,

∂ 2 (fp∗ )k 1 1
dkj (p) := |0 =  (f˜p )wzj (0)Bk (p) =  Lj (f˜w )(p)Bk (p),
∂zj ∂w λ(p) λ(p)
∂ 2 gp∗ 1 1
ck (p) := |0 = (gp )wzk (0) = 
Lk (gw − 2if˜w , f˜(p) l,l ,n,N ) |p ,
∂zk ∂w λ(p) λ(p)
 2 ∗
1 ∂ gp 1  
r(p) := Re 2
|0 = Re (gp )ww (0)
2 ∂w 2λ(p)
1   

= Re gww − 2if˜ww , f˜(p) l,l ,n,N |p .
2λ(p)

Write (ξ, η) = (ξ1 , · · · , ξN −1 , η) for the coordinates of C N and define

 
ξ − a(p)η η
Gp (ξ, η) = , , (3.12)
Qp (ξ, η) Qp (ξ, η)
 
where Qp (ξ, η) = 1 + 2iξ, a(p) l,l ,n + r(p) − ia(p), a(p) l,l ,n η. Then Gp ∈
∗∗ ∗
Aut+ N
0 (Hl,l ,n ). Let Fp be the composition of Fp with Gp :

Fp∗∗ = (f˜p∗∗ , gp∗∗ ) = (fp∗∗ , φ∗∗ ∗∗ ∗

p , gp ) := Gp ◦ Fp . (3.13)

We recall some notations (from [14,15] and [4]) for functions of weighted degree that will
be used in the remaining context of the paper. We assign the weight of z to be 1, and
assign the weight of u and w to be 2. We say a smooth  function h(z, z̄, u) on U ∩ Hln is
 
of quantity Owt (s) for 0 ≤ s ∈ N, if  h(tz,tz̄,t u) 
2

ts  is bounded for (z, u) on any compact

subset
 of U ∩Hln and t close to 0. Similarly, we say h is of quantity owt (s) for 0 ≤ s ∈ N,
 h(tz,tz̄,t2 u) 
if  ts
 converges to 0 uniformly for (z, u) on any compact subset of U ∩ Hn as t
 l
goes to 0.
In general, for a smooth function h(z, z̄, u) on U ∩ Hln , we denote h(k) (z, z̄, u) the
sum of terms-weighted degree k in the Taylor expansion of h at 0. And h(k) (z, z̄, u) also
sometimes denotes a weighted homogeneous polynomial of degree k, if h is not specified.
When h(k) (z, z̄, u) extends to a holomorphic polynomial of weighted degree k, we write
it as h(k) (z, w) or h(k) (z) if it depends only on z.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 13

By Lemma 2.2 in [4], we have the following normalization and CR Gauss-Codazzi

equation:

Lemma 3.1. For each p ∈ M , Fp∗∗ satisfies the normalization condition:

⎧
⎪ ∗∗ i ∗∗(1)
⎪
⎨fp = z + 2 ap (z)w + Owt (4)
∗∗(2)
φ∗∗ = φp (z) + Owt (3)
⎪ p
⎪
⎩g ∗∗ = w + O (5),
p wt

with

z̄, a∗∗(1)
p (z) l |z|2l = |φ∗∗(2)
p (z)|τ , τ = l − l. (3.14)

Remark 3.2. As mentioned in [4], there exists τp∗∗ ∈ Aut+ ∗∗ ∗∗

0 (Hl,l ,n ) such that Fp = τp ◦
N

∗∗(1)
Fp . From (3.14), we see, if we write ap (z) = zA(p), then A(p)Il,n−1 is a (n −1) ×(n −1)
Hermitian matrix.

We next claim that A(p) is independent of the choice of Cj (p). To see this, we first
recall

f˜p∗ − a(p)gp∗
f˜p∗∗ =   .
1 + 2if˜p∗ , a(p) l,l ,n + r(p) − ia(p), a(p) l,l ,n gp∗

Then,

∂ 2 (fp∗∗ )k
Pjk = |0 = dkj (p) − ak (p)cj (p) − δjk (ia(p), a(p) l,l ,n + r(p)). (3.15)
∂zj ∂w

Here δjk is the Kronecker symbol. Note each term in (3.15) is independent of Cj (p).

Definition 3.3. The rank of the (n − 1) × (n − 1) matrix A(p) = −2i(Pjk )1≤j,k≤(n−1) ,

denoted by RkF (p), is called the geometric rank of F at p. In particular, F is said to
have geometric rank zero at p if RkF (p) = 0, which occurs if and only if A(p) = 0.

Since A(p) is smooth on p, we see that RkF (p) is a lower semi-continuous function in
p ∈ U ∩ Hln . Furthermore, with the same proof as that for Lemma 2.2 (A), (B) in [15],
we have the following invariant property of geometric rank:

Proposition 3.4. Let F1 be holomorphic in a small neighborhood U ⊂ C n of p ∈ Hln as

above. That is F1 (U ∩ Hln ) ⊂ Hl,l N
 ,n and F1 (U ∩ Sl ) ⊂ Sl,l ,n . Moreover, F1 is CR-
n N

transversal along U ∩ Hl . Assume that F2 = τ ◦ F1 ◦ σ with σ ∈ Aut+ (Hln ) and τ ∈

n

Aut+ (Hl,l
N n N
 ,n ). Then F2 is CR-transversal and side-preserving map from Hl to Hl,l ,n ,

and RkF2 (p) = RkF1 (σ(p)).

14 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

We next define the geometric ranks for maps between generalized spheres. Let F be a
holomorphic map from a small neighborhood U of q ∈ ∂Bln to C N . Assume F (U ∩ Bln ) ⊂
BlN and F (U ∩ ∂Bln ) ⊂ ∂BlN , and in addition F is CR-transversal along U ∩ ∂Bln . We
can find some Cayley transformations Φq that biholomorphically maps Sln and Hln to
Bln \ V and ∂Bln \ V for some variety V with q ∈ / V . Write p = Φ−1
q (q) ∈ Hl .
n

Similarly, we can find some Cayley transformation ΨF (q) that biholomorphically maps
 ,n and Hl,l ,n to Bl \ W and ∂Bl \ W for some variety W with F (q) ∈
N N N N
Sl,l / W . Set
−1
F̂ = ΨF (q) ◦ F ◦ Φq and regard it as a germ of map at p ∈ Hl . We then define the
n

geometric rank of F at q, denoted by RkF (q), to be the geometric rank RkF̂ (p) of F̂ at
p. By the above proposition, RkF (q) is independent of the choices of Φq and ΨF (q) , and
thus it is well-defined. Note RkF (q) is a lower semi-continuous function in q ∈ U ∩ ∂Bln .
We next give a description of the geometric rank zero condition in terms of the CR
second fundamental form and the Levi null cone. The reader is refereed to [10] for many
notations and background on this matter.
Let M̂ ⊂ C N be a Levi non-degenerate hypersurface with signature l . Let M ⊂ M̂
be a Levi non-degenerate submanifold of hypersurface type of signature l and of CR
dimension n − 1. Let θ̂ be a contact form of M̂ and T̂ be its corresponding Reeb vector
field. Let {L̂1 , · · · , L̂N −1 } be a frame of T (1,0) M̂ near q̂ ∈ M ⊂ M̂ . We can assume that
{L̂1 , · · · , L̂n−1 } are tangent to M when restricted to M . Let {θ̂1 , · · · , θ̂N −1 , θ̂} be the dual
frame of {L̂1 , · · · , L̂N −1 , T̂ }. Then the Levi matrix (ĝαβ ) is given by dθ̂ = iĝαβ θ̂α ∧ θ̂β .
We normalize the frames such that (ĝαβ ) = Il,l ,n,N −1 . Let ω̂kj be the Webster connection
with respect to this frame (see [2]). Identify the CR normal bundle N = T (1,0) M̂ /T (1,0) M
along M as the orthogonal complement (with respect to the Levi form of M̂) of T (1,0) M
−1 j j
in T (1,0) M̂ restricted to M . Then N has a frame {L̂a }N σ
a=n . Write ω̂k = ω̂kσ θ̂ . Then
 (1,0) (1,0) (1,0)  (1,0)
the CR second fundamental form : Tq̂ M × Tq̂ M → Tq̂ M̂ Tq̂ M of M in
 N −1 a
its ambient space M̂ is given by (L̂α , L̂β ) = a=n ω̂αβ L̂a . Notice the concept of CR
second fundamental form is invariant under holomorphic change of coordinates.
 ,n and M = F (U ∩Hl ). For any p ∈ M ⊂ M̂ , as in Remark 3.2,
N n
Now, we take M̂ = Hl,l
after a holomorphic change of coordinates, we can assume that p = 0 and M is the image
of Hln under F which satisfies the normalization in Lemma 3.1. Then by the computation
(2) (2)  n−1 n−1
in [9], writing φ(2) = (φ1 , · · · , φN −n ), we have ( α=1 bα ∂z∂α |0 , α=1 bα ∂z∂α |0 ) =
N −n (2)
j=1 φj (b) ∂ξj+n−1 |0 . Here (ξ, η) = (ξ1 , · · · , ξN , η) is the coordinates of the target
∂

Euclidean space C N . Recall that the rank zero condition is equivalent to |φ(2) |τ ≡ 0.
With these set-ups, we have the following:

Proposition 3.5. Let F be a holomorphic map in an open set U and maps M := U ∩ Hln
N
into Hl,l  ,n . Assume F is CR transversal along M . Then the following two statements

are equivalent:
(1): F has geometric rank zero at every q ∈ M .
 
(2): For any q̂ ∈ F (M ) ⊂ HlN , and any Xq̂ ∈ T (1,0) M , then L̂q̂ ( (Xq̂ , Xq̂ ), (Xq̂ , Xq̂ ))
= 0. Here L̂q̂ is the Levi form of the ambient space at q̂. Namely, the value of the second
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 15

fundamental form of M in its ambient space is in the null cone of the Levi form of the
ambient space.

Claim. A holomorphic map F = (f, g) = (z, φ, ψ, w) : Hln → Hl,l

N
 ,n with φ ≡ ψ has
 
geometric rank zero (here φ, ψ have l − l and N − n − l + l components, respectively).

Proof. We first prove it is the case at p = 0. Note φ, φ(0) = ψ, ψ(0) . Consequently,

f, f(0) l,l ,n = 0. Write F (0) = ( q , 0) and take τ0 ∈ Aut+ (Hl,l
q , qn ) = ( N
 ,n ) to be

τ0 (ξ, η) = (ξ − q, η −2iξ, q l,l ,n ). Set F1 = τ0 ◦F . Then F1 (0) = 0 and F1 = (z, φ1 , ψ1 , w)



with φ1 ≡ ψ1 . We replace F by F1 and still write the new map as F . This will not
change the geometric rank at p = 0. (See Proposition 3.4.) Then notice there exist
holomorphic functions ϕ1 , · · · , ϕk which has no constant terms or linear terms in z
such that SpanC {z1 , · · · , zn−1 , φ, ψ} = SpanC {z1 , · · · , zn−1 , ϕ1 , · · · , ϕk }. By the proof
of Proposition 2.2, we can find some matrix T ∈ U (l, l , n − 1, N − 1) such that F2 =
(fT, w) = (z, φ̂, ψ̂, w), and the components of φ̂, ψ̂ are linear combinations of ϕj ’s. In
particular, they have no linear terms in z. Then it is easy to verify by definition that
the geometric rank of F2 is zero at p = 0. By Proposition 3.4, F also has geometric
rank zero at p = 0. To study the geometric rank at a point p ≈ 0, we note there exists
σ ∈ Aut+ (Hln ) such that σ(0) = p. Moreover, there exists τ ∈ Aut+ (Hl,l N
 ,n ) such that

G := τ ◦F ◦σ = (z, φ, ψ, w). By the preceding argument, RkG (0) = 0. By Proposition 3.4,

we have RkF (p) = RkF (σ(0)) = RkG (0) = 0. 

4. Proof of Theorem 1.1

In this section, we give a proof of Theorem 1.1. We will first work with maps between
hyperquadrics instead of generalized spheres. This makes it easier to apply techniques
from CR geometry. The following result is crucial to establish Theorem 1.1.

Theorem 4.1. Let U ⊂ C n a small (connected) neighborhood of 0. Let F be a holomorphic

map from U to C N such that F (0) = 0 and F (U ∩ Hln ) ⊂ Hl,l N
 ,n , F (U ∩ Sl ) ⊂ Sl,l ,n .
n N

Then the following statements are equivalent.

(1) F is CR transversal at 0 and F has geometric rank zero near 0 along Hln .
(2) There exists some τ ∈ Aut+ N
0 (Hl,l ,n ) such that the following holds near 0:

τ ◦ F = (z, φ, ψ, w),

where φ and ψ are holomorphic maps near 0 with l − l and N − n − l + l components,

respectively, satisfying φ ≡ ψ .

At the end of §3, we have shown that (2) implies (1). We will therefore prove only the
converse implication in this section.
16 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

4.1. Some preliminaries

Assume F satisfies the assumption in (1). By Lemma 3.1 and the zero geometric rank
condition, we can compose F with some element in Aut+ N ˜
0 (Hl,l ,n ) to make F = (f , g) =
(f, ϕ, g) satisfy the following normalization:
⎧
⎪
⎪
⎨f = z + Owt (4)
φ = φ(2) (z) + Owt (3) (4.1)
⎪
⎪
⎩g = w + O (5).
wt

For p ∈ M near 0, let Fp be as in (3.2) and λ(p) as in (3.5). Recall λ(p) ≈ 1 is a

real number for p ≈ 0. Let Fp∗ = ( √1λ f˜p B(p), λ1 gp ) be as in (3.9). We next let B(p) =
(B1 (p), · · · , Bn−1 (p), B̂n (p), · · · , B̂N −1 (p)) ∈ U (l, l , n, N − 1) be as in §3. In particular,
(3.8) holds:

 t t
E1 (p) En−1 (p)
(B1 (p), · · · , Bn−1 (p)) = Il,l ,n,N −1  ,··· ,  Il,n−1 . (4.2)
λ(p) λ(p)

Let a(p) be as in (3.10) and (3.11). Then we have

1
aj (p) = T (f˜), δj,l Lj (f˜) l,l ,n , if 1 ≤ j ≤ n − 1; (4.3)
λ(p)
1
aj (p) =  T (f˜), B̂j (p) , if n ≤ j ≤ N − 1. (4.4)
λ(p)
 
∂ 2 gp∗
Let r(p) = 12 Re ∂w2 |0 be as in §3 and write

 
(fp∗ , gp∗ ) := 1 + 2if˜p∗ , a(p) l,l ,n + r(p) − ia(p), a(p) l,l ,n gp∗ . (4.5)

Let Fp∗∗ = (f˜p∗∗ , gp∗∗ ) be as in (3.13). That is,

f˜p∗ − a(p)gp∗ gp∗

f˜p∗∗ = , gp∗∗ = . (4.6)
(fp∗ , gp∗ ) (fp∗ , gp∗ )

Then the normalization in Lemma 3.1 holds.

We pause to fix some notations. Write N for the set of non-negative integers. For a
|α|+|β|
multi-index α = (α1 , · · · , αn−1 ) ∈ N n−1 and β ∈ N, we write Dzα Dw
β
= α1 ∂ αn−1 β .
∂z1 ···∂zn−1 ∂w
In the following context, we will introduce a notion of weighted degree in p ∈ Hln . Let
h(z, w, p) be a smooth function in W × V . Here (z, w) ∈ W and W is an open set in
C n−1 × C, while p ∈ V and V is an open subset of Hln containing 0. We say h ∈ Owt,p (k)
with k ≥ 0 if for every α ∈ N n−1 , β ∈ N and (z0 , w0 ) ∈ W , it holds that H(p) :=
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 17

Dzα Dwβ
h(z0 , w0 , p) is in O wt (k). That is, writing p = (p̂, pn ) = (p1 , · · · , pn−1 , pn = u +
 ¯ 2 
iv), we have  H(tp̂,ttkp̂,t u)  is bounded for p on any compact subset of V and t close to
0. Sometimes even h is independent of (z, w), we will still use the notion Owt,p (k) to
distinguish the variables (z, w) and p. We next prove the following proposition.

Proposition 4.2.

λ(p) = 1 + Owt,p (3), (4.7)

aj (p) = Owt,p (2) if 1 ≤ j ≤ n − 1, (4.8)

aj (p) = Owt,p (1) if n ≤ j ≤ N − 1, (4.9)

r(p) = Owt,p (1). (4.10)


Proof of Proposition 4.2. Note by (4.1), T g(p) = gw (p) = 1 + Owt,p (3), f (p) = Owt,p (1),
T f (p) = fw (p) = Owt,p (2), and also ϕ(p) = Owt,p (2), T ϕ(p) = ϕw (p) = Owt,p (1). And


Lj ϕ(p) = Owt,p (1) for 1 ≤ j ≤ n − 1. We have by (3.5), λ(p) = 1 + Owt,p (3). It follows
from (4.3) and (4.7) that

aj (p) = Owt,p (2) if 1 ≤ j ≤ n − 1.

When n ≤ j ≤ N −1, since T f˜(p) = Owt,p (1), we conclude by (4.4) that aj (p) = Owt,p (1).

Note gww (p) = Owt,p (1), f˜(p) = Owt,p (1). Using (4.7), we have r(p) = Owt,p (1). 

4.2. A crucial proposition

The key step to prove Theorem 4.1 is to establish the following Proposition 4.3. The
proof of the proposition heavily relies on the moving point trick (see [14]). Recall τ = l −l.

Proposition 4.3. Let F be as above. Fix an integer s ≥ 5. Assume

⎧
⎪
⎪ ∗∗ (t−1)
≡ 0,
⎨(fp )
∗∗ (t)
(gp ) ≡ 0, (4.11)
⎪
⎪
⎩(ϕ∗∗ )(s1 ) , (ϕ∗∗ )(s2 ) ≡ 0,
p p τ

for all p ∈ Hln close to 0 and all 4 ≤ s1 + s2 = t < s. Then (4.11) holds at p = 0 for
s1 + s2 = t = s.

Proof of Proposition 4.3. We split the proof into several lemmas (Lemma 4.4–4.14). Re-
call we denote by N the set of non-negative integers. Let α ∈ N n−1 , β ∈ N. We say
(α, β) ∈ E if |α| + β = 1. Fix an integer ŝ ≥ 2. Write Iŝ for the collection of indices
(α, β) ∈ N n−1 × N that satisfies |α| + 2|β| = ŝ and (α, β) ∈
/ E.
18 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Lemma 4.4. Let s ≥ 5 be as in Proposition 4.3. Fix s1 ≥ 2, s2 ≥ 2 with s1 + s2 < s. Then

for any (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 , and any 1 ≤ j ≤ n − 1, the following hold:

Lj Lα1 T β1 ϕ(s1 +1) , Lα2 T β2 ϕ(s2 ) τ (p) = −Lα1 T β1 ϕ(s1 ) , L̄j Lα2 T β2 ϕ(s2 +1) τ (p); (4.12)
L̄j Lα1 T β1 ϕ(s1 +1) , Lα2 T β2 ϕ(s2 ) τ (p) = −Lα1 T β1 ϕ(s1 ) , Lj Lα2 T β2 ϕ(s2 +1) τ (p). (4.13)

Proof of Lemma 4.4. We start with the hypothesis (4.11), which implies that

β1 ˜∗∗ β2 ˜∗∗
Dzα1 Dw fp , Dzα2 Dw fp l,l ,n (0) = 0, (4.14)

where (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . Note that by (4.5), (4.6), we have

f˜p∗∗ = f˜p∗ [1 − 2if˜p∗ , a(p) l,l ,n − r(p)gp∗ ] − a(p)gp∗ + Owt,p (2). (4.15)

Recall gp∗ =
gp
λ(p) . Also λ(p) = 1 + Owt,p (3) and

Dzα Dw
β
gp (0) = Lα T β g(p) − 2iLα T β f˜(p), f˜(p) l,l ,n . (4.16)

It then follows from the assumption that whenever |α| + 2β ≤ s − 2, and (α, β) ∈
/ E, we
have

β ∗
Dzα Dw gp (0) = Owt,p (2).

Recall by Proposition 4.2, a(p) = Owt,p (1) and r(p) = Owt,p (1). We then obtain from
(4.14) and (4.15) that

β1 ˜∗ β2 ˜∗
Dzα1 Dw fp , Dzα2 Dw fp l,l ,n (0)
β1 ˜∗ β2 ˜∗ ˜∗
+Dzα1 Dw fp , −2iDzα2 Dw (fp fp , ā l,l ,n ) l,l ,n (0)
β1 ˜∗ ˜∗ β2 ˜∗
+−2iDzα1 Dw (fp fp , ā l,l ,n ), Dzα2 Dw fp l,l ,n (0) (4.17)
β1 ˜∗ β2
+Dzα1 Dw fp , −Dzα2 Dw (r(p)f˜p∗ gp∗ ) l,l ,n (0)

+−Dzα1 Dw
β1
(r(p)f˜p∗ gp∗ ), Dzα2 Dw
β2 ˜∗
fp l,l ,n (0) = Owt,p (2).

t
On the other hand, recall f˜p∗ =  1 f˜p B(p),
λ(p)
and B(p)Il,l ,n,N −1 B(p) = Il,l ,n,N −1 . This
implies for any α̂1 , α̂2 ∈ N n−1 , and β̂1 , β̂2 ∈ N,

β̂1 ˜∗ β̂2 ˜∗
Dzα̂1 Dw fp , Dzα̂2 Dw fp l,l ,n (0)
1 1 (4.18)
β̂2 ˜
= Dα̂1 Dβ̂1 f˜p , Dzα̂2 Dw fp l,l ,n (0) = Lα̂1 T β̂1 f˜(p), Lα̂2 T β̂2 f˜(p) l,l ,n .
λ(p) z w λ(p)
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 19

Now letting |αˆ1 | +2β̂1 ≥ 2 or |αˆ2 | +2β̂2 ≥ 2 and |αˆ1 | +2β̂1 +|αˆ2 | +2β̂2 ≤ s −1, we have
L T β̂1 f (p) = Owt,p (1) or Lα̂2 T β̂2 f (p) = Owt,p (1), and Lα̂1 T β̂1 ϕ(p), Lα̂2 T β̂2 ϕ(p) τ =
α̂1

Owt,p (1). Thus the quantity in (4.18) belongs to Owt,p (1). But r(p) = Owt,p (1), and
aj (p) = Owt,p (1) for all j. This implies the last four terms on the left hand side of (4.17)
belong to Owt,p (2). Hence we obtain from (4.17) that the weighted degree one part (in p)
in expansion of Dα1 Dβ1 f˜∗ , Dzα2 Dw
z w p
β2 ˜∗
f l,l ,n (0) equals 0. By (4.18) and (4.7), we conclude
p
that the weighted degree one part in expansion of Lα1 T β1 f˜(p), Lα2 T β2 f˜(p) l,l ,n equals
zero.
Using again the fact

Dzα̂ Dw
β̂
fp (0) = Lα̂ T β̂ f (p) = Owt,p (1),

whenever 2 ≤ |α̂| + 2β̂ ≤ s − 2, we have Lα1 T β1 f (p), Lα2 T β2 f (p) l belongs to

Owt,p (2). Hence we conclude the weighted degree one part in the expansion of
Lα1 T β1 ϕ(p), Lα2 T β2 ϕ(p) τ equals 0. This means

Lα1 T β1 ϕ(s1 +1) (p), Lα2 T β2 ϕ(s2 ) (p) τ + Lα1 T β1 ϕ(s1 ) (p), Lα2 T β2 ϕ(s2 +1) (p) τ = 0.
(4.19)

We finally apply Lj and Lj to (4.19) and obtain the two equations in Lemma 4.4. 

Write ej ∈ N n−1 , 1 ≤ j ≤ n − 1, for the (n − 1)-tuple whose jth component equals

n−1
1 and all other components equal 0. Write α1 = j=1 kj1 ej , i.e. α1 = (k11 , · · · , kn−1
1
).
n−1 2
Similarly, write α2 = j=1 kj ej . We have the following lemmas.

Lemma 4.5. Let s1 ≥ 2, s2 ≥ 2 and s1 + s2 = s ≥ 5. Assume (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 ,

β1 ≥ 1, and (α1 , β1 − 1) ∈
/ E. If there is some 1 ≤ j0 ≤ n − 1, such that kj10 = 0 and
2
kj0 = 0, then

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ (p) = −Lα1 T β1 −1 ϕ(s1 −2) , Lα2 T β2 +1 ϕ(s2 +2) τ (p).

Proof of Lemma 4.5. We have

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = T Lα1 T β1 −1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ

δj0 ,l (4.20)
= (L̄j0 Lj0 − Lj0 L̄j0 )Lα1 T β1 −1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ .
2i

Here we have used the identity [L̄j , Lj ] = 2iδj,l T . By the assumption kj10 = 0,

Lj0 L̄j0 Lα1 T β1 −1 ϕ(s1 ) = Lj0 Lα1 T β1 −1 L̄j0 ϕ(s1 ) = 0.

Furthermore, note (α1 + ej0 , β1 − 1) ∈

/ E. We apply Lemma 4.4 twice to obtain (4.20)
equals
20 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

δj0 ,l
− Lj0 Lα1 T β1 −1 ϕ(s1 −1) , Lj0 Lα2 T β2 ϕ(s2 +1) τ
2i
(4.21)
δj ,l
= 0 Lα1 T β1 −1 ϕ(s1 −2) , L̄j0 Lj0 Lα2 T β2 ϕ(s2 +2) τ .
2i

Again noting

L̄j0 Lj0 = Lj0 L̄j0 + 2iδj0 ,l T and Lj0 L̄j0 Lα2 T β2 ϕ(s2 +1) = Lj0 Lα2 T β2 L̄j0 ϕ(s2 +1) = 0,

we have (4.21) equals to the following

δj0 ,l
(−2iδj0 ,l )Lα1 T β1 −1 ϕ(s1 −2) , T Lα2 T β2 ϕ(s2 +2) τ
2i
=−Lα1 T β1 −1 ϕ(s1 −2) , Lα2 T β2 +1 ϕ(s2 +2) τ .

This proves Lemma 4.5. 

Lemma 4.6. Let s, s1 , s2 be as in Lemma 4.5 and (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . Then

∗ ∗
Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ (p) = CLα̂1 T β̂1 ϕ(s1 ) , Lα̂2 T β̂2 ϕ(s2 ) τ (p).

Here C is a nonzero constant. Moreover, if s = 2s∗ is even, then s∗1 = s∗2 = s∗ . If

s = 2s∗ + 1 is odd, then s∗1 = s∗ , s∗2 = s∗ + 1. And (α̂1 , β̂1 ) ∈ Is∗1 , (α̂2 , β̂2 ) ∈ Is∗2 .

Proof of Lemma 4.6. We first notice that the following equations follow from (3.3) and
an induction argument:

L̄j Lkj = Lkj L̄j + 2ikδj,l T Lk−1

j ; L̄kj Lj = Lj L̄kj + 2ikδj,l T L̄k−1
j . (4.22)

The proof of Lemma 4.6 for s = 5 is slightly different and we will leave it to §5.
We will therefore assume s ≥ 6 in the following context of proof. Furthermore, since
A, B̄ τ = B, Ā τ for two vectors A, B, we will assume s1 ≥ s2 − 1.
We prove by induction on m = s1 − s2 ≥ 0. If m = −1 or 0, i.e. s1 = s2 − 1 or s1 = s2 ,
the conclusion is trivial. Now suppose the conclusion holds for −1 ≤ m ≤ k with k ≥ 0
and consider the case where m = k + 1. In this case, s1 − s2 ≥ 1 and thus s1 ≥ 4. We
have two different cases (A) and (B).
(A) If α1
= 0, we let i0 be the smallest integer such that ki10
= 0 and write α̃1 =
α1 − ei0 . Then we have by Lemma 4.4,

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ

=Li0 Lα̃1 T β1 ϕ(s1 −1) , Lα2 T β2 ϕ(s2 ) τ (4.23)

= − Lα̃1 T β1 ϕ(s1 −1) , L̄i0 Lα2 T β2 ϕ(s2 +1) τ .

 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 21

n−1
Write α2 = j=1 kj2 ej . Note if ki20 = 0, then L̄i0 Lα2 T β2 ϕ(s2 +1) = 0, and the conclusion
is trivially true. Now assume ki20
= 0 and write α̃2 = α2 − ki20 ei0 . Then by (4.22), we
have (4.23) equals

k2
− Lα̃1 T β1 ϕ(s1 −1) , L̄i0 Li0i0 Lα̃2 T β2 ϕ(s2 +1) τ
k2 k2 −1
= − Lα̃1 T β1 ϕ(s1 −1) , (Li0i0 L̄i0 + 2iki20 δi0 ,l T Li0i0 )Lα̃2 T β2 ϕ(s2 +1) τ
k2 −1
=2iki20 δi0 ,l Lα̃1 T β1 ϕ(s1 −1) , Li0i0 Lα̃2 T β2 +1 ϕ(s2 +1) τ .

Now (s1 − 1) − (s2 + 1) = k − 1 and thus the proof is finished by the inductive hypothesis.
(B) If α1 = 0, then β1 ≥ 2. We have by (3.3) and Lemma 4.4,

T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ

δ1,l
= (L̄1 L1 − L1 L̄1 )T β1 −1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ
2i
δ1,l
= L̄1 L1 T β1 −1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ
2i
−δ1,l
= L1 T β1 −1 ϕ(s1 −1) , L1 Lα2 T β2 ϕ(s2 +1) τ .
2i
Again now (s1 −1) −(s2 +1) = k −1 and the proof is done by the inductive hypothesis.
This establishes Lemma 4.6. 

Lemma 4.7. (a). Let s1 , s2 , s be as in Lemma 4.5. Let (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . If
α1
= α2 , then

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ (p) = 0.

(b). Let s1 , s2 , s be as in Lemma 4.5 and assume s is odd. Let (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈
Is2 . Then

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ (p) = 0.

Proof of Lemma 4.7. Again we will treat the case for s = 5 separately and leave its
n−1
proof to §5. We therefore assume here s ≥ 6. Writing αi = j=1 kji ej , 1 ≤ i ≤ 2, by
assumption there is some 1 ≤ j ≤ n − 1 such that kj1
= kj2 . To make the notation simple,
we assume, without loss of generality, that kj1 > kj2 , and will further assume j = 1. We
will first need the following claim.

Claim 4.8. If Lα1 T β1 = L21 , or Lα1 T β1 = L1 Lj , j

= 1, or Lα1 T β1 = L1 T , then

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = 0.

22 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Proof. We prove the three cases separately.

(I). If Lα1 T β1 = L21 , we have s1 = 2. Since s ≥ 6, we have s2 ≥ 4, i.e. |α2 | + 2β2 ≥ 4.
Note k12 = 0 or 1. First assume k12 = 0.
(1). If α2
= 0, then there is some 2 ≤ j ≤ n − 1 such that kj2
= 0. Write α̃2 = α2 − ej .
We have by Lemma 4.4,

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = L21 ϕ(2) , Lj Lα̃2 T β2 ϕ(s2 ) τ

= − L̄j L21 ϕ(3) , Lα̃2 T β2 ϕ(s2 −1) τ = 0.

(2). If α2 = 0, then β2 ≥ 2. We have by (3.3) and Lemma 4.4,

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = L21 ϕ(2) , T T β2 −1 ϕ(s2 ) τ

−δ2,l 2 (2) −δ2,l 2 (2)
= L1 ϕ , (L̄2 L2 − L2 L̄2 )T β2 −1 ϕ(s2 ) τ = L1 ϕ , L̄2 L2 T β2 −1 ϕ(s2 ) τ
2i 2i
δ2,l −δ2,l
= L2 L21 ϕ(3) , L2 T β2 −1 ϕ(s2 −1) τ = L1 L2 ϕ(2) , L̄1 L2 T β2 −1 ϕ(s2 ) τ = 0.
2i 2i

Now consider the case k12 = 1, write α̃2 = α2 − e1 . Then by Lemma 4.4 and (4.22),

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = L21 ϕ(2) , L1 Lα̃2 T β2 ϕ(s2 ) τ

= − L̄1 L21 ϕ(3) , Lα̃2 T β2 ϕ(s2 −1) τ = −4iδ1,l L1 T ϕ(3) , Lα̃2 T β2 ϕ(s2 −1) τ .

It is reduced to case (III).

(II). If Lα1 T β1 = L1 Lj , j
= 1. We again have s1 = 2, s2 ≥ 4. Note also we must have
k1 = 0. If there exist i
= j such that ki2
= 0, then write α̃2 = α2 − ei and by Lemma 4.4,
2

L1 Lj ϕ(2) , Li Lα̃2 T β2 ϕ(s2 ) τ = −L̄i L1 Lj ϕ(3) , Lα̃2 T β2 ϕ(s2 −1) τ = 0.

Now assume ki2 = 0 for all i

= j and kj2
= 0. Then by Lemma 4.4 and (3.3),

L1 Lj ϕ(2) , Lα2 T β2 ϕ(s2 ) τ = L1 Lj ϕ(2) , Lj Lα2 −ej T β2 ϕ(s2 ) τ

= − L1 L̄j Lj ϕ(3) , Lα2 −ej T β2 ϕ(s2 −1) τ = −2iδj,l L1 T ϕ(3) , Lα2 −ej T β2 ϕ(s2 −1) τ .

Again it is reduced to case (III).

Finally consider α2 = 0. Then β2 ≥ 2. We have again by (3.3) and Lemma 4.4,

−δ1,l
L1 Lj ϕ(2) , T β2 ϕ(s2 ) τ = L1 Lj ϕ(2) , L̄1 L1 T β2 −1 ϕ(s2 ) τ
2i
δ1,l 2 δ1,l 2 (2)
= L1 Lj ϕ(3) , L1 T β2 −1 ϕ(s2 −1) τ = − L ϕ , L̄j L1 T β2 −1 ϕ(s2 ) τ = 0.
2i 2i 1

(III). If Lα1 T β1 = L1 T , then s1 = 3, s2 ≥ 3 and k12 = 0. We have several subcases:

 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 23

(1). If |α2 | ≥ 2, then there exist j

= 1, such that kj2
= 0. By Lemma 4.4,

L1 T ϕ(3) , Lα2 T β2 ϕ(s2 ) τ = L1 T ϕ(3) , Lj Lα2 −ej T β2 ϕ(s2 ) τ

= − L̄j L1 T ϕ(4) , Lα2 −ej T β2 ϕ(s2 −1) τ = 0.

(2). If |α2 | = 1 and β2 ≥ 2, then the same computation as in the preceding case (1)
yields

L1 T ϕ(3) , Lα2 T β2 ϕ(s2 ) τ = 0.

(3). Next consider the case |α2 | = 1, and β2 = 1. Writing Lα2 = Lj , j

= 1, we have

L1 T ϕ(3) , Lj T ϕ(3) τ
−δj,l
= L1 T ϕ(3) , (L̄j Lj − Lj L̄j )Lj ϕ(3) τ (4.24)
2i
−δj,l δj,l
= L1 T ϕ(3) , L̄j L2j ϕ(3) τ + L1 T ϕ(3) , Lj L̄j Lj ϕ(3) τ .
2i 2i
Note by Lemma 4.4 and (3.3) respectively, we have

L1 T ϕ(3) , L̄j L2j ϕ(3) τ = −Lj L1 T ϕ(4) , L2j ϕ(2) τ = Lj T ϕ(3) , L̄1 L2j ϕ(3) τ = 0;

L1 T ϕ(3) , Lj (L̄j Lj )ϕ(3) τ = −2iδj,l L1 T ϕ(3) , Lj T ϕ(3) τ .

We substitute the above two equations into (4.24) to get

L1 T ϕ(3) , Lj T ϕ(3) τ = −L1 T ϕ(3) , Lj T ϕ(3) τ .

Hence L1 T ϕ(3) , Lj T ϕ(3) τ = 0.

(4). We finally consider the case where |α2 | = 0. Then we must have β2 ≥ 2. In this case,
the proof is similar to the case (I).(2).
This proves Claim 4.8. 

Now we come back to the proof of part (a), Lemma 4.7. Recall k11 > k12 . Suppose
k12= 0. If Lα1 T β1 ∈ {L21 , L1 L2 , · · · , L1 Ln−1 , L1 T }, then we are done by Claim 4.8.
Otherwise, i.e. Lα1 T β1 ∈
/ {L21 , L1 L2 , · · · , L1 Ln−1 , L1 T }, we have (α1 − e1 , β1 ) ∈
/ E and
by Lemma 4.4,

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = L1 Lα1 −e1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ

= − Lα1 −e1 T β1 ϕ(s1 −1) , L̄1 Lα2 T β2 ϕ(s2 +1) τ = 0.

Now we assume k12 ≥ 1. Note k11 − (k12 − 1) ≥ 2. We can keep applying Lemma 4.4
and (4.22) to move (k12 − 1) L1 ’s to Lα2 T β2 ϕ(s2 ) , and annihilate (k12 − 1) L1 s in Lα2 .
24 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Then we get a new inner product Lα̃1 T β̃1 ϕ(s̃1 ) , Lα̃2 T β̃2 ϕ(s̃2 ) τ . Here Lα̃2 contains only
one L1 and Lα̃1 has at least two L1 ’s. If Lα̃1 T β̃1 = L21 , then the conclusion follows
from Claim 4.8. Otherwise, we can apply Lemma 4.4 again and move one more L1 from
Lα̃1 T β̃1 ϕ(s̃1 ) to Lα̃2 T β̃2 ϕ(s̃2 ) , and get new inner product Lα̂1 T β̂1 ϕ(ŝ1 ) , Lα̂2 T β̂2 ϕ(ŝ2 ) τ ,
where Lα̂2 contains no L1 and Lα̂1 has at least one L1 . This is reduced to the known
case we considered before (k12 = 0).
For part (b) of Lemma 4.7, note if s = s1 + s2 is odd, we must have α1
= α2 . Thus
the conclusion follows from part (a). 

Lemma 4.9. Let s1 , s2 , s be as in Lemma 4.5. Let (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . Then one
of the following must hold.

1. Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = 0.

∗
2. Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = CT ŝ ϕ(s ) , T ŝ ϕ(s∗ ) τ ,
where s = 2s∗ = 4ŝ, with ŝ ≥ 2, s∗ ≥ 4.
∗
3. Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ = CLj T ŝ ϕ(s +1)
, Lj T ŝ ϕ(s∗ +1) τ ,

where s = 2s∗ + 2 = 4ŝ + 2, with s∗ = 2ŝ ≥ 2, and 1 ≤ j ≤ n − 1. Here, as before, C

denotes a nonzero constant which may be different in different contexts.

Proof of Lemma 4.9. If s is odd, then the inner product equals 0 by part (b) of
Lemma 4.7. Now assume s is even and thus s ≥ 6. By Lemma 4.6, we can assume
s1 = s2 = s ≥ 3. And by Lemma 4.7, we can assume α1 = α2 = α, for otherwise the
inner product again equals 0. Consequently, β1 = β2 = β. We will prove by induction on
m = |α|. If m = 0, 1, the conclusion (2) or (3) holds trivially. Now suppose the conclusion
holds for 0 ≤ m ≤ k for some k ≥ 1, and consider the case m = k + 1 ≥ 2 (note in the
case m = 2, we must have β ≥ 1 due to the fact s1 = s2 ≥ 3). Pick 1 ≤ j0 ≤ n − 1 such
n−1
that, writing α = j=1 kj ej , we have j0
= 0. If α = kj0 ej0 , then kj0 = k + 1 ≥ 2. In this
case by Lemma 4.4 and (4.22),

k k
Lα T β ϕ(s) , Lα T β ϕ(s) τ = Lj0j0 T β ϕ(s) , Lj0j0 T β ϕ(s) τ
k k −1
= − L̄j0 Lj0j0 T β ϕ(s+1) , Lj0j0 T β ϕ(s−1) τ
k −1 k −1
= − 2ikj0 δj0 ,l Lj0j0 T β+1 ϕ(s+1) , Lj0j0 T β ϕ(s−1) τ
k −2 k −1
=2ikj0 δj0 ,l Lj0j0 T β+1 ϕ(s) , L̄j0 Lj0j0 T β ϕ(s) τ
k −2 k −2
=4kj0 (kj0 − 1)Lj0j0 T β+1 ϕ(s) , Lj0j0 T β+1 ϕ(s) τ .

This is reduced to the case m ≤ k.

If there is 1 ≤ i0
= j0 ≤ n−1 such that ki0
= 0, kj0
= 0. We write α̃ = α−ki0 ei0 −kj0 ej0
and compute, again by using Lemma 4.4 and (4.22),
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 25

k k k k
Lα T β ϕ(s) , Lα T β ϕ(s) τ = Li0i0 Lj0j0 Lα̃ T β ϕ(s) , Li0i0 Lj0j0 Lα̃ T β ϕ(s) τ
k k k −1 k
= − L̄i0 Li0i0 Lj0j0 Lα̃ T β ϕ(s+1) , Li0i0 Lj0j0 Lα̃ T β ϕ(s−1) τ
k −1 k k −1 k
= − 2iki0 δi0 ,l Li0i0 Lj0j0 Lα̃ T β+1 ϕ(s+1) , Li0i0 Lj0j0 Lα̃ T β ϕ(s−1) τ
k −1 k −1 k −1 k
=2iki0 δi0 ,l Li0i0 Lj0j0 Lα̃ T β+1 ϕ(s) , Li0i0 L̄j0 Lj0j0 Lα̃ T β ϕ(s) τ
k −1 k −1 k −1 k −1
=4ki0 kj0 δi0 ,l δj0 ,l Li0i0 Lj0j0 Lα̃ T β+1 ϕ(s) , Li0i0 Lj0j0 Lα̃ T β+1 ϕ(s) τ .

This is again reduced to the case m ≤ k. By induction, we see the conclusion holds. 

We now continue to prove Proposition 4.3.

Lemma 4.10. f (s−1) ≡ 0, g (s) ≡ 0.

Proof of Lemma 4.10. We split the proof into two parts, depending on whether s is odd
or even.
(I). First assume s is odd. Write s = 2s∗ − 1, s∗ ≥ 3.
(I.a). We first prove that f (s−1) (z, w) ≡ 0. Fix some j0 with 1 ≤ j0 ≤ n − 1. Write
∗
s −1
(2s∗ −2) ∗
−k−1
fj0 = a(2k) (z)ws .
k=0

Here a(2k) (z) depends on j0 . But we will not write it as a subscript to simplify the
notation. By the assumption of Proposition 4.3, for 1 ≤ j, k ≤ n − 1, it holds for p ≈ 0
on Hln that
∗
∂ ∂ s −2
(f ∗∗ )j (0) = 0.
∂zk ∂ws∗ −2 p

By (4.8), (4.15), we have if 1 ≤ j ≤ n − 1,

 
(fp∗∗ )j = (fp∗ )j 1 − 2if˜p∗ , a(p) l,l ,n − r(p)gp∗ + Owt,p (2). (4.25)

Moreover, if 1 ≤ j ≤ n − 1,

1 1 ˜
(fp∗ )j =  f˜p Bj (p) = fp , δj,l Lj f˜(p) l,l ,n . (4.26)
λ(p) λ(p)

For 1 ≤ j ≤ N − n, (φ∗p )j =  1 f˜p , B̂j (p) l,l ,n .

λ(p)
And λ(p) = 1 + Owt,p (3). Recall if
1 ≤ j ≤ n − 1, aj (p) = Owt,p (2) and for j ≥ n, aj (p) = T (f˜)(p), B̂j (p) + Owt,p (2) =
(3)
ϕw (p), bj (p) + Owt,p (2). Here bj (p) are the last (N − n) component of B̂j (p). Hence
we have
26 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

∗
∂ ∂ s −2 ∗
(f ∗∗ )j (0) = Lk T s −2 f˜(p), δj0 ,l Lj0 f˜(p) l,l ,n
∂zk ∂ws∗ −2 p 0
 −n−1
N
∗
˜ ˜
+ Lk f (p), δj0 ,l Lj0 f (p) l,l ,n − 2i T s −2 f˜p , B̂j (p)
j=n

(3) s∗ −2
·δ j,l,l ,n ϕw (p), bj (p) − r(p)T g(p) + Owt,p (2).

We choose k
= j0 . Note then Lk f˜(p), Lj0 f˜(p) l,l ,n = Owt,p (2). We thus have the
∗
weighted degree one part in p in the expansion of Lk T s −2 f˜(p), δj0 ,l Lj0 f˜(p) l,l ,n must
be zero. This implies, writing p = (p̂, pn ) = (p1 , · · · , pn−1 , pn ) ∈ Hln ,

∗ ∗
2iδk,l δj0 ,l (s∗ − 1)!p̄k a(0) + δj0 ,l a(2) −2 −3) (2)
zk (p̂) + δj0 ,l Lk T (p), ϕzj0 (p̂) τ ≡ 0. (4.27)
s
ϕ(2s

Collecting the anti-holomorphic part in p̂ in (4.27), we have

∗ ∗
2iδk,l δj0 ,l (s∗ − 1)!p̄k a(0) + δj0 ,l Lk T s −2 −3) (2)
ϕ(2s , ϕzj0 (p̂) τ ≡ 0.

We apply L̄k to this equation to get

∗ ∗
2iδk,l δj0 ,l (s∗ − 1)!a(0) + δj0 ,l Lk T s −2
ϕ(2s −3)
, Lk Lj0 ϕ(2) τ ≡ 0.
∗ ∗
By Lemma 4.7, Lk T s −2 ϕ(2s −3) , Lk Lj0 ϕ(2) τ ≡ 0. Thus we conclude a(0) = 0. Since
1 ≤ j0 ≤ n − 1 is arbitrary, this conclusion in the case s∗ = 3 implies the following fact.


Corollary 4.11. fww (p) = Owt,p (1).

Now we prove by induction that all a(2k) ≡ 0 for 1 ≤ k ≤ s∗ − 1. Suppose we have

already proved a(2i) ≡ 0 for 1 ≤ i < k, with some 1 ≤ k ≤ s∗ − 1. Next we aim to prove
a(2k) (z) ≡ 0. For that, note by assumption, for any α ∈ N n−1 with |α| = 2k − 1, we have
(s∗ −k−1) ∗∗
Dzα Dw (fp )j (0) = 0.
n−1
As before, we have, writing α = μ=1 kμ eμ ,

∗ ∗
−k−1)
(s
0 = Dzα Dw (fp∗∗ )j0 (0) = Lα T s −k−1 f˜(p), δj0 ,l Lj0 f˜(p) l,l ,n
  −n+1
N
∗
+ c̃(kμ )Lμ f˜(p), δj0 ,l Lj0 f˜(p) l,l ,n − 2i Lα−eμ T s −k−1 f˜p , B̂j (p)
(4.28)
kμ ≥1 j=n

∗
(3) −k−1
· δj,l,l ,n ϕw (p), bj (p) − r(p)Lα−eμ T s g(p) + Owt,p (2)


Here the sum kμ ≥1 is taken over those μ satisfying kμ ≥ 1, and c̃(kμ ) is some integer
depending on kμ (the value of c̃(kμ ) is determined by the Leibniz rule).
We will need to use the following facts.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 27


Claim 4.12. Let F be as in Proposition 4.3. Then gww (p) = Owt,p (2).

Proof. This is trivial if s > 5 in Proposition 4.3 and therefore we only need to prove for
the case s = 5. We will postpone the proof to (I.b). 

Corollary 4.13. r(p) = Owt,p (2).

Proof of Corollary 4.13. Recall

1   

r(p) = Re gww (p) − 2if˜ww (p), f˜(p) l,l ,n,N .
2λ(p)

Then the conclusion follows easily from Corollary 4.11 and Claim 4.12. 

(3)
Write ϕw (z, w) = d(1) (z). Also recall a(2j) (z) ≡ 0 for 1 ≤ j < k. We conclude by
collecting the weighted degree one terms in (4.28) that

−n+1
N
(s∗ − k − 1)!Dzα a(2k) (p̂) + ĉj d(1) (p̂), bj (0) ≡ 0. (4.29)
j=n

Here ĉj ’s are constants, which may be 0. For instance, if kj0 = 0, then all ĉj = 0. We
further collect holomorphic terms in (4.29) to see (s∗ − k − 1)!Dzα a(2k) (p̂) ≡ 0. Thus
we conclude Dzα a(2k) (p̂) ≡ 0 for every α ∈ N n−1 with |α| = 2k − 1. This yields that
a(2k) (z) ≡ 0. By induction, we see f (s−1) (z, w) ≡ 0.

(I.b). Next we will prove g (s) ≡ 0. Before that we first prove Claim 4.12. For that, we
write

g (5) (z, w) = c(1) (z)w2 + c(3) (z)w + c(5) (z).

gp∗
Recall gp∗∗ = (fp∗ ,gp∗ ) (see (4.5), (4.6)), and gp∗ = 1
λ(p) gp . We then have (using (4.7))

 
gp∗∗ = gp + gp − 2if˜p∗ , a(p) l,l ,n − r(p)gp + Owt,p (2). (4.30)

Proof of Claim 4.12. By the assumption of Proposition 4.3, (gp∗∗ )(4) ≡ 0, and thus

2
Dw (gp∗∗ )(0) = 0. (4.31)

Recall by (4.16), Dwβ

gp (0) = T β g(p) − 2iT β f˜(p), f˜(p) l,l ,n . Then it follows from Corol-
lary 4.11 that if β ≤ 2, Dw β
gp (0) = T β g(p) + Owt,p (2). Consequently, equations (4.30)
and (4.31) yield that

 
T 2 g(p) + T g(p) − 2iT f˜p∗ (0), a(p) l,l ,n − r(p)T g(p) + Owt,p (2) = 0.
28 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

By the same argument as before, we obtain

 −n+1
N 
(3)
T 2 g(p) + T g(p) − 2i T f˜p (0), B̂j (p) · δj,l,l ,n · ϕw (p), bj (p) − r(p)T g(p)
j=n

+ Owt,p (2) = 0.
(4.32)

Recall p = (p̂, pn ) = (p1 , · · · , pn−1 , pn ) and note

1   

r(p) = Re gww (p) − 2if˜ww (p), f˜(p) l,l ,n,N
2λ(p)
1 
= Re{gww (p)} + Owt,p (2)
2
1 
= c(1) (p̂) + c(1) (p̂) + Owt,p (2).
2
Here we have used (4.7) and Corollary 4.11.
Collecting the weighted degree one terms in (4.32), we have

−n−1
N
1  (1) 
2c(1) (p̂) + ĉj d(1) (p̂), bj (0) − c (p̂) + c(1) (p̂) ≡ 0.
j=n
2

Here ĉj ’s are some constants. We further collect holomorphic terms to get 32 c(1) (p̂) = 0.
We thus have c(1) (z) = 0 and this proves Claim 4.12. 

We are now at the position to prove g (s) ≡ 0. For that we write

∗
s −1
∗
−j−1
g (s) (z, w) = c(2j+1) (z)ws .
j=0

We will first prove c(1) (z) = 0 (this is already established for s = 5). By assumption,
s∗ −1 ∗∗
Dw (gp )(0) = 0. Recall Dw β
gp (0) = T β g(p) − 2iT β f˜(p), f˜(p) l,l ,n . It follows from
s∗ −1 s∗ −1 ∗
(I.a) that T f (p) = Owt,p (1) and thus Dw gp (0) = T s −1 g(p) + Owt,p (2). Then by
(4.30), Proposition 4.2, and Corollary 4.13.
∗
s −1 ∗∗
0 = Dw (gp )(0)
 −n+1
N 
∗ ∗
−1
T s −2 f˜∗ (p), B̂j (p) · δj,l,l ,n · ϕw (p), bj (p)
(3)
= Ts g(p) + T g(p) − 2i
j=n

+ Owt,p (2).

(3)
Recall ϕw (z, w) = d(1) (z). Collecting weighted degree one terms in the above equation,
we get
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 29

−n−1
N
(s∗ − 1)!c(1) (p̂) + ĉj d(1) (p̂), bj (0) = 0 for some constants ĉj s.
j=n

We then collect holomorphic terms to see c(1) (z) = 0.

Now suppose we have already proved c(2j+1) (z) ≡ 0 for 0 ≤ j < k with 1 ≤ k ≤ s∗ −1,
s∗ −1 ∗∗
and aim to prove c(2k+1) (z) ≡ 0. For that we use the fact that Dzα Dw (gp )(0) = 0
for any α ∈ N n−1 with |α| = 2k. By a similar argument as above, using (4.30) and (I.a),
and collecting holomorphic weighted degree one terms, we see Dzα c(2k+1) (z) ≡ 0. This
implies c(2k+1) (z) ≡ 0. By induction, we see all c(2k+1) (z) = 0 and thus g (s) ≡ 0.
(II). Finally we consider the case s = 2s∗ is even. Here s∗ ≥ 3.
(II.a). We start with the proof of f (s−1) ≡ 0. For that, we fix some 1 ≤ j0 ≤ n − 1
and write
∗
s −1
∗
(s−1) −j−1
fj0 (z, w) = a(2j+1) (z)ws .
j=0

∗
s −1 ∗∗
To prove a(1) (z) = 0, we notice, by assumption, that Dw (fp )j0 (0) = 0. By (4.25),
Proposition 4.2, and Corollary 4.13, we have
∗ ∗
s −1 ∗∗ s −1 ∗
Dw (fp )j0 (0) = Dw (fp )j0 (0) + Owt,p (2)
∗
−1
= T s f˜(p), δj0 ,l Lj0 f˜(p) l,l ,n + Owt,p (2) ≡ 0.

This implies δj0 ,l (s∗ − 1)!a(1) (p̂) ≡ 0. We thus have a(1) (z) ≡ 0.
Now suppose we have proved a(2j+1) (z) ≡ 0 for 1 ≤ j < k with 1 ≤ k ≤ s∗ − 1, and
s∗ −1 ∗∗
consider the case j = k. By assumption, Dzα Dw (fp )(0) ≡ 0 with |α| = 2k. We use
the same argument as before to derive Dzα a (2k+1)
(p̂) ≡ 0 for any |α| = 2k. Thus we have
a(2k+1) (z) ≡ 0. By induction, we have f (s−1) ≡ 0.
(II.b). It remains to prove g (s) ≡ 0. Write
∗

s
∗
(s) −k
g (z, w) = c(2k) (z)ws .
k=0

∗
s −1 ∗∗
We first prove that c(0) (z) ≡ 0. By assumption, Dzj Dw (gp )(0) ≡ 0, for any 1 ≤ j ≤
∗ ∗
s −1 s −1 s∗ −1 ˜
n − 1. By (4.16), Dz D j
gp (0) = Lj T
w g(p) − 2iLj T f (p), f˜(p) l,l ,n . It follows
∗ ∗ ∗
from (II.a) that Lj T s −1 f (p) = Owt,p (1) and thus Dzj Dw
s −1
gp (0) = Lj T s −1
g(p) +
Owt,p (2). Then by (4.30) and Proposition 4.2, we get
∗
s −1 ∗∗ ∗
−1
 s∗ −2 ˜∗

Dzj Dw (gp )(0) = Lj T s g(p) + T g(p) − 2iDzj Dw fp (0), a(p) l,l ,n
(4.33)
+ Owt,p (2) ≡ 0.

Note for any α ∈ N n−1 and β ∈ N,

30 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

β ˜∗
Dzα Dw fp (0), a(p) l,l ,n
1    1 t
= √ Dzα Dw fp (0) · B(p) · Il,l ,n,N · √ T f˜ · B(p)
β ˜
λ λ
1 β ˜
 t t
= Dzα Dw fp (0) · B(p) · Il,l ,n,N · B(p) T f˜ (4.34)
λ
=Dzα Dβ f˜p (0), T f˜ l,l ,n + Owt,p (2)
w

=Lα T β f˜(p), T f˜ l,l ,n + Owt,p (2)

=Lα T β ϕ(s−3) (p), T ϕ(3) τ + Owt,p (2); and
∗
−1
Lj T s g(p) = 2iδj,l s∗ !pj c(0) + (s∗ − 1)!c(2)
zj (p̂). (4.35)

Collecting weighted degree one terms in (4.33) and using (4.34), (4.35), we obtain

∗
2iδj,l s∗ !p̄j c(0) + (s∗ − 1)!c(2)
zj (p̂) = 2iLj T
s −2
ϕ(s−3) (p), T ϕ(3) (p) τ . (4.36)

We apply L̄j to (4.36) and get

∗
δj,l s∗ !c(0) = Lj T s −2
ϕ(s−3) (p), Lj T ϕ(3) (p) τ . (4.37)

We then have two cases:

Case I. If s∗ = 2σ is even, σ ≥ 2. Then by Lemma 4.5, we have

δj,l ∗ δj,l (−1)σ−2

c(0) = ∗
Lj T s −2 ϕ(s−3) , Lj T ϕ(3) τ = Lj T σ ϕ(2σ+1) , Lj T σ−1 ϕ(2σ−1) τ .
s ! s∗ !
(4.38)

Note σ ≥ 2. By Lemma 4.4 and (3.3), we obtain

δj,l (−1)σ−1 σ (2σ)

c(0) = T ϕ , L̄j Lj T σ−1 ϕ(2σ) l,l ,n
s∗ !
δj,l (−1)σ−1
= (−2iδj,l )T σ ϕ(2σ) , T σ ϕ(2σ) l,l ,n (4.39)
s∗ !
2i(−1)σ
= μ.
s∗ !

Here μ = T σ ϕ(2σ) , T σ ϕ(2σ) τ ∈ R.

Now we use the following fundamental identity:

Im{g} = f˜, f˜ l,l ,n , on Im w = z, z l .

We collect the weighted degree s terms in the above equation and recall we already know
f (t) (z, w) ≡ 0 for t ≤ s − 1. We then see
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 31


Im{g (s) } = ϕ(s1 ) , ϕ(s2 ) τ , on Im w = z, z l ; or (4.40)
s1 ,s2 ≥2, s1 +s2 =s
∗

s
∗ 
−k
Im{ c(2k) (z)ws }= ϕ(s1 ) , ϕ(s2 ) τ , on Im w = z, z l . (4.41)
k=0 s1 ,s2 ≥2, s1 +s2 =s

∗
Writing w = u + iv, we collect terms of us in equation (4.41). Since by (4.39), c(0) is an
imaginary number, we get
∗
s−2
c(0) 1 1
=  T j ϕ(2j) , ∗ T s∗ −j ϕ(s−2j) τ
i j=2
j! (s − j)!
∗
(4.42)
s −2
1
= ∗
T j ϕ(2j) , T s∗ −j ϕ(s−2j) τ .
j=2
j!(s − j)!

By Lemma 4.5, we have T j ϕ(2j) , T s∗ −j ϕ(s−2j) τ = (−1)σ−j μ, for all 2 ≤ j ≤ s∗ − 2.

Then (4.42) is reduced to
∗
s−2
c(0) (−1)σ−j μ
= . (4.43)
i j=2
j!(s∗ − j)!

We combine (4.39) and (4.43) to get

∗
s −2 s−2 ∗  ∗
(−1)j μs∗ ! j s
2μ = ∗
= (−1) μ . (4.44)
j=2
j!(s − j)! j=2
j

But
∗
s −2  ∗
s
(−1)j = 2s∗ − 2.
j=2
j

We thus obtain from (4.44) that 2μ = (2s∗ − 2)μ. This implies μ = 0. Consequently, by
(4.39), c(0) = 0.
Case II. If s∗ = 2σ + 1 is odd, here σ ≥ 1, then by (4.37) and Lemma 4.5, we have

δj,l ∗ δj,l (−1)σ−1

c(0) = ∗
Lj T s −2 ϕ(s−3) , Lj T ϕ(3) τ = Lj T σ ϕ(2σ+1) , Lj T σ ϕ(2σ+1) τ .
s ! s∗ !
(4.45)
(0)
Consequently, c is real, and we cannot use the method in Case I. We instead apply
T 2σ L̄1 L1 to (4.40) and evaluate at 0 to obtain

1 2σ 
T L̄1 L1 g (s) (0) = T 2σ L̄1 L1 ϕ(s1 ) , ϕ(s2 ) τ (0). (4.46)
2i
s1 ,s2 ≥2,s1 +s2 =s
32 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Note L̄1 L1 g (s) = 2iδ1,l T g (s) and L̄1 L1 ϕ(s1 ) , ϕ(s2 ) τ = L1 ϕ(s1 ) , L1 ϕ(s2 ) τ +2iδ1,l T ϕ(s1 ) ,
ϕ(s2 ) τ . We thus derive from (4.46) that


2σ   
2σ
δ1,l T 2σ+1 (s)
g (0) = L1 T j ϕ(s1 ) , L1 T 2σ−j ϕ(s2 ) τ (0)
j=0 s1 ,s2 ≥2,s1 +s2 =s
j


2σ   
2σ
+ 2iδ1,l T j+1 ϕ(s1 ) , T 2σ−j ϕ(s2 ) τ (0)
j=0 s1 ,s2 ≥2,s1 +s2 =s
j
(4.47)

2σ−1
2σ

= L1 T j ϕ(2j+1) , L1 T 2σ−j ϕ(s−2j−1) τ
j=1
j


2σ−2
2σ

+ 2iδ1,l T j+1 ϕ(2j+2) , T 2σ−j ϕ(s−2j−2) τ .
j=1
j

Note the left hand side δ1,l T 2σ+1 g (s) (0) equals δ1,l c(0) (s∗ !). If σ = 1, then the right hand
side equals 2μ∗ , where

μ∗ = L1 T ϕ(3) , L1 T ϕ(3) τ .

∗ δ ∗ ∗ ∗
By (4.45), we have c(0) = s1,l
∗ ! μ . Then (4.47) is reduced to μ = 2μ . Thus μ = c
(0)
= 0.
Now assume σ ≥ 2. Write

μ1 := L1 T σ ϕ(2σ+1) , L1 T σ ϕ(2σ+1) τ ;
−δ1,l σ (2σ)
μ2 := T σ ϕ(2σ) , T σ+1 ϕ(2σ+2) τ = T ϕ , L̄1 L1 T σ ϕ(2σ+2) τ
2i
δ1,l δ1,l
= L1 T σ ϕ(2σ+1) , L1 T σ ϕ(2σ+1) τ = μ1 .
2i 2i

δ1,l (−1)σ−1
Then by (4.45), c(0) = s∗ ! μ1 . By Lemma 4.5, we have

L1 T j ϕ(2j+1) , L1 T 2σ−j ϕ(s−2j−1) τ = (−1)σ−j μ1 for 1 ≤ j ≤ 2σ − 1.

T j+1 ϕ(2j+2) , T 2σ−j ϕ(s−2j−2) τ = (−1)σ−j−1 μ2 for 1 ≤ j ≤ 2σ − 2.

By the above equations, (4.47) is reduced to

 
2σ−1
2σ
  2σ 
2σ−2
σ−1 σ−j
(−1) μ1 = (−1) μ1 + 2iδ1,l (−1)σ−j−1 μ2
j=1
j j=1
j
 2σ−1
   2σ 
2σ−2 
2σ
= (−1)σ−j + (−1)σ−j−1 μ1 .
j=1
j j=1
j
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 33

This implies (−1)σ−1 μ1 = (−1)σ−1 2σμ1 , or μ1 = 2σμ1 . Thus we conclude μ1 = c(0) = 0.

Hence in all cases, it holds that c(0) = 0. Once we know c(0) = 0, as before, we use
an induction argument that all c(2j) (z) = 0 for all 0 ≤ j ≤ s∗ . Suppose we have proved
c(2j) (z) = 0 for 0 ≤ j < k for some 1 ≤ k ≤ s∗ and consider the case j = k. By
s∗ −k ∗∗
assumption, Dzα Dw (gp )(0) ≡ 0 for any |α| = 2k − 1, and any p ∈ Hln close to 0.
∗
s∗ −k
As before, by (II.a) we have Lα T s −k f (p) = Owt,p (1). Consequently, Dzα Dw gp (0) =
∗
Lα T s −k g(p) + Owt,p (2). By (4.30) and Proposition 4.2, we have
∗
−k
 s∗ −k−1 ˜∗

Lα T s g(p) + T g(p) − 2iDzα Dw fp (0), a(p) l,l ,n + Owt,p (2) ≡ 0.

Using (4.34), we collect the Owt,p (1)-term in the above equation to obtain
∗
(s∗ − k)!Dzα c(2k) (p̂) − 2iLα T s −k−1
ϕ(s−3) (p), T ϕ(3) τ = 0

Furthermore we collect holomorphic terms in p̂ to see Dzα c(2k) (p̂) = 0. As α is arbitrary,

we have c(2k) (z) ≡ 0. By induction, c(2j) (z) ≡ 0 for all j, and thus g (s) (z, w) ≡ 0. This
finishes the proof of Lemma 4.10. 

Lemma 4.14. ϕ(s1 ) , ϕ(s2 ) τ ≡ 0 for all s1 , s2 ≥ 2, s1 + s2 = s.

Proof of Lemma 4.14. When s is odd, it follows from Lemma 4.7 that
 
∂ |α1 |+β1 (s1 ) ∂ |α2 |+β2 (s2 )
ϕ , α2 β2 ϕ (0) ≡ 0,
∂z α1 ∂wβ1 ∂z ∂w τ

for any (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . This implies ϕ(s1 ) , ϕ(s2 ) τ (0) ≡ 0. When s is even
and s = 2s∗ = 4σ for some σ ≥ 2, it follows from Lemma 4.9 and the proof of Lemma 4.10
(see (4.39) and recall we proved c(0) = 0) that
 
∂ |α1 |+β1 (s1 ) ∂ |α2 |+β2 (s2 )
ϕ , α2 β2 ϕ (0) ≡ 0,
∂z α1 ∂wβ1 ∂z ∂w τ

for any (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 . This again implies ϕ(s1 ) , ϕ(s2 ) τ (0) ≡ 0.
Similarly, if s is even and s = 2s∗ = 4σ+2, then the conclusion follows from Lemma 4.9
and the proof of Lemma 4.10 (see (4.45) and recall we proved c(0) = 0). This establishes
Lemma 4.14. 

Proposition 4.3 follows from Lemma 4.10 and Lemma 4.14 (except that it remains to
prove Lemma 4.6 and Lemma 4.7 for the case s = 5; this will be done in §5). 

4.3. Proofs of Theorem 4.1 and Theorem 1.1

Proof of Theorem 4.1. As was mentioned, we prove only the implication from (1) to (2).
This easily follows from the following claim.
34 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Claim 4.15. Let F be as in the assumption in Theorem 4.1. Assume F is CR transversal

near 0 and satisfies the normalization in Lemma 3.1. If F is of geometric rank zero near
0, then

f (t−1) ≡ 0, g (t) ≡ 0, ϕ(s1 ) , ϕ(s2 ) τ ≡ 0, ∀s1 , s2 ≥ 2, s1 + s2 = t ≥ 4. (4.48)

We prove the claim by induction on t. First note by Lemma 3.1 and the geometric
zero condition, (4.48) holds for t = 4. Now assume Claim 4.15 holds for all 4 ≤ t < s
with s ≥ 5, and consider the case t = s. But Claim 4.15 (with 4 ≤ t < s) applies also to
Fp∗∗ . We thus have (4.11) holds for 4 ≤ t < s. Then by Proposition 4.3, (4.48) holds for
t = s. Hence by induction, Claim 4.15 holds. 

Proof of Theorem 1.1. Composing F with automorphisms, we assume that p =

[1, 0, · · · , 0, 1] ∈ ∂Bln and F (p) = [1, 0, · · · , 0, 1] ∈ ∂BlN . Denote by Ψn the Cayley
transformation from Sln to Bln , and ΦN the Cayley transformation from Sl,l N N
 ,n to Bl
−1
(see §3). Then G := ΦN ◦ F ◦ Ψn is well-defined in a small neighborhood of 0 ∈ Hl . N

Note F is side-preserving, CR transversal, and of geometric rank zero if and only if G is

so. Then Theorem 1.1 follows from Theorem 4.1 and Theorem 2.1. 

5. Completion of the proof

In this section, we complete the proof of Theorem 1.1 by giving a proof of Lemma 4.6
and Lemma 4.7 in the case s = 5. More precisely, we prove the following proposi-
tion.

Proposition 5.1. Let s1 ≥ 2, s2 ≥ 2 and s1 +s2 = 5. Assume (α1 , β1 ) ∈ Is1 , (α2 , β2 ) ∈ Is2 .
Then

Lα1 T β1 ϕ(s1 ) , Lα2 T β2 ϕ(s2 ) τ (p) ≡ 0.

Proof of Proposition 5.1. Note A, B̄ τ = B, Ā τ for two vectors A, B and thus we can
always assume s1 ≤ s2 . Since s1 + s2 = s = 5, this implies s1 = 2, s2 = 3. We will verify
the conclusion by a direct computation case by case.
(1). We first consider the case when Lα1 T β1 = L2i for some 1 ≤ i ≤ n − 1.
(1.a). If Lα2 T β2 = Li T , then we have

−δj,l 2 (2)
L2i ϕ(2) , Li T ϕ(3) τ = L2i ϕ(2) , T Li ϕ(3) τ = Li ϕ , (L̄j Lj − Lj L̄j )Li ϕ(3) τ .
2i

Here we have used identity (3.3) and let j

= i. Furthermore, by Lemma 4.4 it equals
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 35

−δj,l 2 (2) δj,l

Li ϕ , L̄j Lj Li ϕ(3) τ = Lj L2i ϕ(3) , Lj Li ϕ(2) τ
2i 2i
−δj,l −δj,l (5.1)
= Li Lj ϕ(2) , Lj L̄i Li ϕ(3) τ = Li Lj ϕ(2) , Lj 2iδi,l T ϕ(3) τ
2i 2i
=δj,l δi,l Li Lj ϕ(2) , Lj T ϕ(3) τ .

This equals zero by case (2.a).

(1.b). If Lα2 T β2 = Lj T with 1 ≤ j
= i ≤ n − 1, we have by (3.3),

−δj,l 2 (2)
L2i ϕ(2) , Lj T ϕ(3) τ = Li ϕ , (L̄j Lj − Lj L̄j )Lj ϕ(3) τ
2i
(5.2)
−δj,l 2 (2) δj,l 2 (2)
= Li ϕ , L̄j L2j ϕ(3) τ + L ϕ , Lj L̄j Lj ϕ(3) τ
2i 2i i
Furthermore, by Lemma 4.4 and (3.3) respectively, we have

L2i ϕ(2) , L̄j L2j ϕ(3) τ = −Lj L2i ϕ(3) , L2j ϕ(2) τ = Lj Li ϕ(2) , L̄i L2j ϕ(3) τ = 0;

L2i ϕ(2) , Lj L̄j Lj ϕ(3) τ = L2i ϕ(2) , Lj 2iδj,l T ϕ(3) τ = −2iδj,l L2i ϕ(2) , Lj T ϕ(3) τ .

Then by (5.2), L2i ϕ(2) , Lj T ϕ(3) τ = −L2i ϕ(2) , Lj T ϕ(3) τ , and thus L2i ϕ(2) , Lj T ϕ(3) τ =
0.
(1.c). If Lα2 T β2 = L3i , by Lemma 4.4 and (4.22) we have

L2i ϕ(2) , L3i ϕ(3) τ = −L̄i L2i ϕ(3) , L2i ϕ(2) τ

= −L2i L̄i ϕ(3) , L2i ϕ(2) τ − 4iδi,l T Li ϕ(3) , L2i ϕ(2) τ .

Note Li T ϕ(3) , L2i ϕ(2) τ = 0 by (1.a). Hence we conclude L2i ϕ(2) , L3i ϕ(3) τ ≡ 0.
(1.d). If Lα2 T β2 = Lj Lk Ll where j
= i and k, l may be equal to i, then

L2i ϕ(2) , Lj Lk Ll ϕ(3) τ = −L̄j L2i ϕ(3) , Lk Ll ϕ(2) τ = −L2i L̄j ϕ(3) , Lk Ll ϕ(2) τ = 0.

(2). It remains to consider the case where Lα1 T β1 = Li Lj for some 1 ≤ i

= j ≤ n − 1.
(2.a). If Lα2 T β2 = Lj T , then by (3.3),

Li Lj ϕ(2) , Lj T ϕ(3) τ = Li Lj ϕ(2) , T Lj ϕ(3) τ

−δj,l
= Li Lj ϕ(2) , (L̄j Lj − Lj L̄j )Lj ϕ(3) τ
2i
−δj,l δi,l
= Li Lj ϕ(2) , L̄j L2j ϕ(3) τ + Li Lj ϕ(2) , Lj L̄j Lj ϕ(3) τ
2i 2i
Note by Lemma 4.4 and (3.3) respectively,

Li Lj ϕ(2) , L̄j L2j ϕ(3) τ = −Li L2j ϕ(3) , L2j ϕ(2) τ = −L2j ϕ(3) , L̄i L2j ϕ(2) τ = 0;
36 X. Huang et al. / Advances in Mathematics 374 (2020) 107388

Li Lj ϕ(2) , Lj L̄j Lj ϕ(3) τ = Li Lj ϕ(2) , Lj (Lj L̄j + 2iδj,l T )ϕ(3) τ

= −2iδj,l Li Lj ϕ(2) , Lj T ϕ(3) τ .

Then we have Li Lj ϕ(2) , Lj T ϕ(3) τ = −Li Lj ϕ(2) , Lj T ϕ(3) τ . Thus

Li Lj ϕ(2) , Lj T ϕ(3) τ = 0.

(2.b). If Lα2 T β2 = Lk T with k

= i, j, a similar argument as in (2.a) yields

Li Lj ϕ(2) , Lk T ϕ(3) τ = 0.

(2.c). If Lα2 T β2 = L3j , then by Lemma 4.4 and (3.3),

Li Lj ϕ(2) , L3j ϕ(3) τ = −L̄j Lj Li ϕ(3) , L2j ϕ(2) τ = −2iδj,l T Li ϕ(3) , L2j ϕ(2) τ .

This equals 0 by (1.b).

(2.d). If Lα2 T β2 = L2i Lj , then by Lemma 4.4 and (3.3),

Li Lj ϕ(2) , L2i Lj ϕ(3) τ = −L̄i Li Lj ϕ(3) , Li Lj ϕ(2) τ = −2iδi,l T Lj ϕ(3) , Li Lj ϕ(2) τ .

This equals 0 by (2.a).

(2.e). If Lα2 T β2 = Lk Lμ Lν with k
= i, j and μ, ν may equal i, j, Lemma 4.4 implies

Li Lj ϕ(2) , Lk Lμ Lν ϕ(3) τ = −L̄k Li Lj ϕ(2) , Lμ Lν ϕ(3) τ

= −Li Lj L̄k ϕ(2) , Lμ Lν ϕ(3) τ = 0.

This proves Proposition 5.1. 

References

[1] M.S. Baouendi, P. Ebenfelt, X. Huang, Super-rigidity for CR embeddings of real hypersurfaces into
hyperquadrics, Adv. Math. 219 (5) (2008) 1427–1445.
[2] M.S. Baouendi, P. Ebenfelt, X. Huang, Holomorphic mappings between hyperquadrics with small
signature difference, Am. J. Math. 133 (6) (2011) 1633–1661.
[3] M.S. Baouendi, P. Ebenfelt, L.P. Rothschild, Transversality of holomorphic mappings between real
hypersurfaces in different dimensions, Commun. Anal. Geom. 15 (3) (2007) 589–611.
[4] M.S. Baouendi, X. Huang, Super-rigidity for holomorphic mappings between hyperquadrics with
positive signature, J. Differ. Geom. 69 (2) (2005) 379–398.
[5] S.S. Chern, J. Moser, Real hypersurfaces in complex manifolds, Acta Math. 133 (1974) 219–271.
[6] J.P. D’Angelo, Several Complex Variables and the Geometry of Real Hypersurfaces, Studies in
Advanced Mathematics, CRC Press, Boca Raton, FL, ISBN 0-8493-8272-6, 1993, xiv+272 pp.
[7] J.P. D’Angelo, M. Xiao, Symmetries in CR complexity theory, Adv. Math. 313 (2017) 590–627.
[8] P. Ebenfelt, Partial rigidity of degenerate CR embeddings into spheres, Adv. Math. 239 (2013)
72–96.
[9] P. Ebenfelt, X. Huang, D. Zaitsev, Rigidity of CR-immersions into spheres, Commun. Anal. Geom.
12 (3) (2004) 631–670.
 X. Huang et al. / Advances in Mathematics 374 (2020) 107388 37

[10] P. Ebenfelt, X. Huang, D. Zaitsev, The equivalence problem and rigidity for hypersurfaces embedded
into hyperquadrics, Am. J. Math. 127 (1) (2005) 169–191.
[11] F. Forstnerič, Extending proper holomorphic mappings of positive codimension, Invent. Math. 95
(1989) 31–62.
[12] Y. Gao, S. Ng, On rational proper mappings among generalized complex balls, Asian J. Math. 22
(2018) 355–380.
[13] D. Grundmeier, J. Lebl, L. Vivas, Bounding the rank of Hermitian forms and rigidity for CR
mappings of hyperquadrics, Math. Ann. 358 (2014) 1059–1089.
[14] X. Huang, On a linearity problem of proper holomorphic maps between balls in complex spaces of
different dimensions, J. Differ. Geom. 51 (1999) 13–33.
[15] X. Huang, On a semi-rigidity property for holomorphic maps, in: A Special Issue Dedicated to Y. T.
Siu on the Occasion of his 60th Birthday, Asian J. Math. 7 (4) (2003) 463–492.
[16] S. Ji, Algebraicity of real analytic hypersurfaces with maximal rank, Am. J. Math. 124 (6) (2002)
1083–1102.
[17] S. Kim, Holomorphic maps between closed SU (l + 1, m) orbits in Grassmannian, Math. Res. Lett.
(2020).
[18] S. Kim, D. Zaitsev, Rigidity of CR maps between Shilov boundaries of bounded symmetric domains,
Invent. Math. 193 (2) (2013) 409–437.
[19] S. Kim, D. Zaitsev, Rigidity of proper holomorphic maps between bounded symmetric domains,
Math. Ann. 362 (1–2) (2015) 639–677.
[20] I. Kossovskiy, B. Lamel, On the analyticity of CR-diffeomorphisms, Am. J. Math. 140 (1) (2018)
139–188.
[21] B. Lamel, N. Mir, Convergence of formal CR mappings into strongly pseudoconvex Cauchy-Riemann
manifolds, Invent. Math. 210 (3) (2017) 963–985.
[22] B. Lamel, N. Mir, Convergence and divergence of formal CR mappings, Acta Math. 220 (2) (2018)
367–406.
[23] N. Mok, Extension of germs of holomorphic isometries up to normalizing constants with respect to
the Bergman metric, J. Eur. Math. Soc. 14 (5) (2012) 1617–1656.
[24] N. Mok, Holomorphic isometries of the complex unit ball into irreducible bounded symmetric do-
mains, Proc. Am. Math. Soc. 144 (10) (2016) 4515–4525.
[25] N. Mok, S. Ng, Germs of measure-preserving holomorphic maps from bounded symmetric domains
to their Cartesian products, J. Reine Angew. Math. 669 (2012) 47–73.
[26] S. Ng, Proper holomorphic mappings on flag domains of SU (p, q)−type on projective spaces, Mich.
Math. J. 62 (2013) 769–777.
[27] S. Ng, Holomorphic double fibration and classical domains, Int. Math. Res. Not. 2015 (2) (2015)
291–324.
[28] M. Umehara, Diastases and real analytic functions on complex manifolds, J. Math. Soc. Jpn. 40 (3)
(1988) 519–539.
[29] S.M. Webster, The rigidity of C-R hypersurfaces in a sphere, Indiana Univ. Math. J. 28 (3) (1979)
405–416.
[30] M. Xiao, Y. Yuan, Complexity of holomorphic maps from the complex unit ball to classical domains,
Asian J. Math. 22 (2018) 729–760.
[31] M. Xiao, Y. Yuan, Holomorphic maps from the complex unit ball to type IV classical domains, J.
Math. Pures Appl. 133 (2020) 139–166.
[32] Y. Yuan, Y. Zhang, Rigidity for local holomorphic isometric embeddings from Bn into BN1 × · · · ×
BNm , J. Differ. Geom. 90 (2) (2012) 329–349.
[33] Y. Yuan, Y. Zhang, CR submanifolds with vanishing second fundamental forms, Geom. Dedic. 183
(2016) 169–180.