Université Grenoble Alpes

ANNALES DE
L’INSTITUT FOURIER

Bernhard Lamel & Duong Ngoc Son

The CR Ahlfors derivative and a new invariant for spherically
equivalent CR maps
Tome 71, no 5 (2021), p. 2137-2167.
<http://aif.centre-mersenne.org/item/AIF_2021__71_5_2137_0>

© Association des Annales de l’institut Fourier, 2021,

Certains droits réservés.
Cet article est mis à disposition selon les termes de la licence
Creative Commons attribution – pas de modification 3.0 France.
http://creativecommons.org/licenses/by-nd/3.0/fr/

Les Annales de l’institut Fourier sont membres du

Centre Mersenne pour l’édition scientifique ouverte
www.centre-mersenne.org
Ann. Inst. Fourier, Grenoble
71, 5 (2021) 2137-2167

THE CR AHLFORS DERIVATIVE AND A NEW

INVARIANT FOR SPHERICALLY EQUIVALENT CR
MAPS

by Bernhard LAMEL & Duong Ngoc SON (*)

Abstract. — We study a CR analogue of the Ahlfors derivative for conformal

immersions of Stowe that generalizes the CR Schwarzian derivative studied earlier
by the second-named author. This notion possesses several important properties
similar to those of the conformal counterpart and provides a new invariant for
spherically equivalent CR maps from strictly pseudoconvex CR manifolds into a
sphere. The invariant is computable and distinguishes many well-known sphere
maps. In particular, it vanishes precisely when the map is spherically equivalent to
the linear embedding of spheres.
Résumé. — Nous étudions un analogue CR de la dérivée au sens d’Ahlfors pour
les immersions conformes de Stowe, qui généralise la dérivée schwarzienne CR étu-
diée antérieurement par le second auteur. Cette notion possède plusieurs propriétés
importantes similaires à celles de son homologue conforme et fournit un nouvel
invariant pour les applications CR, sphériquement équivalentes, de variétés CR
strictement pseudoconvexes à valeurs dans la sphère. Cet invariant est calculable
et permet de distinguer beaucoup d’applications CR sphériques entre elles. En par-
ticulier, il s’annule précisément quand l’application est sphériquement équivalente
au plongement linéaire entre sphères.

1. Introduction

The main purpose of this paper is to extend the notion of the CR

Schwarzian derivative for CR diffeomorphisms [21] to the case of CR im-
mersions. For conformal immersions of Riemannian manifolds, the Ahlfors
derivative of Stowe [23] generalizes the Schwarzian derivative of Osgood–
Stowe [19] in a similar way and goes back to Ahlfors [1]. Precisely, we

Keywords: Sphere maps, Ahlfors derivative, CR maps.

2020 Mathematics Subject Classification: 32V05, 32H35, 53A30.
(*) Duong Ngoc Son was supported by the Austrian Science Fund, FWF-Projekt M
2472-N35. Bernhard Lamel was supported by the Austrian Science Fund, FWF-Projekt
I3472.
2138 Bernhard LAMEL & Duong Ngoc SON

shall define, for each CR immersion f : (M, θ) → (N, η) between pseudo-

hermitian manifolds, a tensor denoted by Aθ (f ). This tensor reduces to
the Schwarzian tensor introduced in [21] for CR diffeomorphisms in the
equidimensional case. We shall call this tensor the CR Ahlfors derivative
(or tensor). We refer the reader to Stowe’s paper for further discussions
regarding the history and motivations in the conformal case; however, it
turns out that the CR setting has some special properties not present in
the conformal setting, which we shall point out as we go.
The tensor which we are going to construct satisfies a “chain rule” de-
F G
scribed as follows: For a chain of CR immersions (M, θ) −−→ (N, η) −−→
(P, ζ), it holds that

(1.1) A(G ◦ F ) = A(F ) + F ∗ A(G).

It was shown in [21] that if (M, θ) = (N, η) is the sphere with its stan-
dard pseudohermitian structure, then A(F ) (which reduces to the CR
Schwarzian derivative as already noted) vanishes identically if F is a CR
automorphism of the sphere. Therefore, the chain rule (1.1) implies that A
is an invariant for spherically equivalent CR maps into spheres of higher di-
mensions. This invariant property is a main motivation for us to extend the
notion of the CR Schwarzian derivative to the case of higher dimensional
targets. We shall in fact apply the Ahlfors derivative to study equivalence
of sphere maps, a problem which has been studied extensively; we can
mention only several papers [3, 4, 6] and refer the readers to numerous
references therein.
To construct the CR Ahlfors derivative, we shall follow Stowe’s con-
struction for the conformal case. First, we define a notion of CR sec-
ond fundamental form for the “isopseudohermitian” immersions and the
(1, 0)-mean curvature vector (this step was not needed in the equidimen-
sional case). Precisely, let (N, η) be a pseudohermitian manifold and let
ι : (M, θ) ,→ (N, η) be a pseudohermitian submanifold of N . This means
the standard inclusion ι is CR and θ = ι∗ η. We denote by ∇ and ∇ e the
Tanaka–Webster connections on (M, θ) and (N, η), respectively, introduced
by Tanaka and Webster [24]. For any two vector fields X, Y ∈ Γ(CT M )
extended to smooth sections X, e Ye of CT N , we define the pseudohermitian
second fundamental form by the Gauß formula, namely,

(1.2) II(X, Y ) = IIN

M (X, Y ) := ∇X
e Ye − ∇X Y,
e
This notion was previously studied by many authors, see, e.g., [26] for the
codimension one case and [8, 9] for the case of pseudohermitian immersions

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2139

(i.e., when the Reeb field of η is tangent to ι(M ).) For our applications, we
shall make no assumption on the Reeb field of the target.
Due to the presence of the torsion, II is not necessarily symmetric and
thus we also consider the symmetrized second fundamental form, i.e.,
1

(1.3) Sym II(X, Y ) = II(X, Y ) + II(Y, X) .
2
In most situations, we shall consider the second fundamental form II as a
tensor on the “horizontal” space T 0, 1 M ⊕ T 1, 0 M (the “good directions”)
where it behaves quite well. In particular, we define the (1, 0)-mean curva-
ture vector to be the trace of II on the horizontal subspace:
n
X
(1.4) H := II (Zᾱ , Zα )
α=1

where {Zα : α = 1, 2, . . . , n} is an orthonormal frame of T 1, 0 M . The trace

of Sym II is denoted by µ, so that µ = <H. Let us point out that the
consideration here is similar to [22] in which we consider the case of CR
immersions into a Kähler manifold. Moreover, when the target is the stan-
dard sphere, the second fundamental form (1.2) is closely related to the
one for CR immersions into complex euclidean space.
N
Similarly to [23], we define the tensor ν = νM as a symmetric real tensor
1, 0 0, 1
on T M ⊕ T M via the formula
(1.5) ν(X, Y ) = 2 hSym II(X, Y ), µi − hX, Y i|µ|2 .
Moreover, we define, for each smooth function u on M ,
1 2
(1.6) Hθ (u) = Sym ∇∇u − ∂b u ⊗ ∂b u − ∂¯b u ⊗ ∂¯b u + ∂¯b u Lθ .
2
Here, Lθ (Z, W ) := −id θ(Z, W )(Z, W ∈ T 1, 0 M ) is the Levi form. We refer
the reader to Section 2.2, in particular (2.17), for the (standard) notation
used here. We can now introduce the CR analogue of the Ahlfors derivative
as follows.
Definition 1.1 (cf. [23]). — Let (M 2n+1 , θ) and (N 2d+1 , η) be strictly
pseudoconvex pseudohermitian manifolds and let F : M → N be a CR
immersion. Let u be the smooth function on M such that F ∗ η = eu θ. We
define the CR Ahlfors derivative (or CR Ahlfors tensor) of F to be
  1 1
(1.7) A(F ) := Hθ (u) + F ∗ νFN(M ) + F ∗ (Jη Lη ) − Jθ Lθ .
2 2
where Jθ = Rθ /(n(n + 1)) and Jη = Rη /(d(d + 1)) are the normalized
Webster scalar curvatures on M and N , respectively, and Lη and Lθ are
the corresponding Levi forms.

TOME 71 (2021), FASCICULE 5

2140 Bernhard LAMEL & Duong Ngoc SON

As mentioned above, the Ahlfors tensor A generalizes the CR Schwarzian

tensor for CR diffeomorphisms in [21] in the same spirit the conformal
Ahlfors generalizes the Schwarzian of Osgood–Stowe. We shall explain this
in the next section.
As briefly discussed, our first result of the paper is the following chain
rule.
Theorem 1.2. — For CR immersions F : (M, θ) → (N, η) and G : (N, η)
→ (P, ζ), we have
(1.8) A(G ◦ F ) = A(F ) + F ∗ A(G).
We point out that although this theorem is analogous to [23, Theorem 1],
the chain rule in CR case is in fact simpler than its conformal counterpart:
the excess term  does not appear in the CR case.
Specializing this chain rule to the case of CR maps into the sphere,
we obtain a new tensorial invariant for spherically equivalent classes of
such maps. This invariant property is a consequence of the fact that the
Schwarzian tensor of a CR automorphism of the sphere with its standard
pseudohermitian structure vanishes identically [21]. Recall that two CR
0
maps F and G from M into S2N +1 are said to be (left) spherically equiv-
0
alent if there exists a CR automorphism φ of S2N +1 such that G = φ ◦ F .
When M = S2N +1 is also a sphere, we can use the CR automorphisms
of M to define a weaker version of spherical equivalence. Namely, we say
that F and G are spherically equivalent if there exist CR automorphisms γ
0
(of S2N +1 ) and φ (of S2N +1 ) such that G ◦ γ = φ ◦ F .
In the following, the unit spheres are always equipped with their standard
pseudohermitian structures.
Corollary 1.3. — Let (M, θ) be a strictly pseudoconvex pseudoher-
mitian manifold.
(i) Suppose that F : M → S2N +1 is a CR immersion and φ : S2N +1 →
0
S2N +1 (N 0 > N ) is a totally geodesic embedding, then
(1.9) A(F ) = A(φ ◦ F ).
In particular, if F and G are left spherically equivalent CR maps
from M into S2N +1 , then
(1.10) A(F ) = A(G).
2n+1
(ii) Suppose that G : (S , Θ) → (M, θ) is a CR immersion and γ : N
→ S2n+1 is a totally geodesic embedding, then
(1.11) γ ∗ A(G) = A(G ◦ γ).

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2141

Remark 1.4. — In Corollary 1.3 (ii), if N admits a totally geodesic em-

bedding into a sphere, then it is necessarily CR spherical. This follows
from [9] for the case dimR N > 5 and [22] for the case dimR N = 3.
In view of Corollary 1.3, an interesting question that arises is whether
the Ahlfors derivative distinguishes the spherically equivalent classes of
sphere maps. Although we can check that this is the case for maps between
spheres of “low” codimension, we do not know the answer to this question
in the full generality. However, we prove that the CR Ahlfors distinguishes
the totally geodesic CR map: an arbitrary CR map into the sphere with
vanishing CR Ahlfors derivative must be a totally geodesic embedding.
Theorem 1.5. — Let F : (M, θ) → (S2d+1 , Θ) be a CR immersion. If
A(F ) = 0, then M is CR spherical and F is spherically equivalent to the
linear mapping.
It is also natural to ask under which conditions the Ahlfors derivative is
a nonzero functional multiple of the Levi metric. We shall discuss this ques-
tion after analyzing several examples in the last section; see Question 7.6.
The paper is organized as follows. In Section 2, we study the geometry
of the CR second fundamental form for CR immersions. In Section 3, we
prove Theorem 1.2. In Section 4, we study the maps with vanishing CR
Ahlfors derivatives. We study the case when the source is of dimension
three in Section 5. In Section 6, we provide an explicit formula for the
Ahlfors derivative which is used to analyze various examples in Section 7.

Acknowledgment

The authors would like to thank an anonymous referee for very careful
reading of the manuscript and pointing out many, many typographical
errors that we were not aware of.

2. Immersions of CR manifolds and the second

fundamental form

2.1. The second fundamental form

Let ι : (M, θ) ,→ (N, η) be a pseudohermitian submanifold, i.e., ι is CR

and θ = ι∗ η, where ι is the inclusion. In this case, ι is “isopseudohermitian”

TOME 71 (2021), FASCICULE 5

2142 Bernhard LAMEL & Duong Ngoc SON

in the sense of [8]. This notion is more general than that of “pseudohermi-
tian immersions,” as the latter requires that the Reeb field of η is tangent
to M along M . In the latter case, the pair (θ, η) is admissible in the sense
of [9].
For any two vector fields X, Y ∈ Γ(CT M ) extended to smooth sections
X,
e Ye of CT N , we recall that the second fundamental form is defined by
II(X, Y ) = ∇e Ye − ∇X Y (see (1.2)) where ∇
X
e
e and ∇ are the Tanaka–
Webster connection on (N, η) and (M, θ), respectively. We summarize the
basic properties of II as follows (cf. [22] which treats a similar situation),
where T and Te denotes the Reeb field of (M, θ) and (N, η), respectively.
Proposition 2.1. — The second fundamental form II is well-defined,
tensorial, and satisfies the following properties for all (1, 0)-vectors Z and
W:
(2.1) II(Z, W ) = II(Z, W ),
(2.2) II(Z, W ) = II(Z, W ),
(2.3) II(Z, W ) = II(W, Z),
(2.4) II(Z, W ) = II(W , Z) − ihZ, W iθ (T − Te),
(2.5) II(Z, T ) = ∇
e Z (T − Te),

(2.6) II(T, Z) = ∇ e
eZ + [Z, T − T ] + τeZ − τ Z.
T −T
e

e (T , Z) is the pseudohermitian torsion of ∇ and similarly

Here τeZ := T∇ e e
for τ . Moreover, II is symmetric if and only if ι is pseudohermitian (i.e.
Te = ι∗ T ).
Proof. — That II is well-defined and tensorial follows from standard ar-
guments. Equations (2.1) and (2.2) follow from the reality of the Tanaka–
Webster connection. Equation (2.3) follows from the equation T∇ (Z, W )
= 0 for (1, 0)-vectors Z and W on M and similarly for N . Proof of (2.4)
uses the fact that T∇ (Z, W ) = ihZ, W iT . Precisely, by [24]
(2.7) ∇Z W − ∇W Z − [Z, W ] = T(Z, W ) = ihZ, W iT,
and similarly for ∇e and thus (2.4) follows.
To prove (2.5), observe that ∇T = 0 and ∇ e Te = 0 [24], and hence
II(Z, T ) = ∇Z T − ∇Z T = ∇Z (T − T ), as desired. The proof of (2.6) also
e e e
follows from direct calculations. We omit the details.
Assume that Te = ι∗ T , then ι is called a pseudohermitian immersion [8]
and the pair (θ, η) is said to be an admissible pair [9]. In this case, it follows
from (2.5) that II(Z, T ) = 0. On the other hand, from (2.6), II(T, Z) =

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2143

τeZ − τ Z and hence both sides vanish by type consideration. Similarly,

from (2.4), II(Z, W ) = II(W , Z) and thus both sides vanish. Thus, II is
symmetric.
Conversely, if II is symmetric, then it follows from (2.4) that Te = ι∗ T .
The proof is complete. 
Definition 2.2. — Let ι : (M, θ) ,→ (N, η) be a pseudohermitian sub-
manifold, ι∗ η = θ. The (1, 0)-mean curvature vector of M in N is the
(1, 0)-vector defined by
n
1X
(2.8) H := II (Zᾱ , Zα ) ,
n α=1
where {Zα : α = 1, 2, . . . , n} is an orthonormal frame of T 1, 0 M and Zᾱ
= Z α . We also define µ = µN M to be the trace of Sym II, i.e. µ = <H.

Let ι : (M, θ) ,→ (N, η) be a pseudohermitian submanifold, ι∗ η = θ. Then

ι∗ sends T 1, 0 M into T 1,0 N . We can define the normal bundle N 1,0 M as a
subundle of T 1, 0 N as usual. Here the orthogonality only depends on the
CR structure, but not on the pseudohermitian structure. We also define
N 0, 1 M similarly.
Proposition 2.3. — Let ι : (M, θ) ,→ (N, η) be a pseudohermitian sub-
manifold, ι∗ η = θ. If T and Te are the Reeb fields corresponding to θ and
η, respectively, then for all tangent vectors Z, W in T 1, 0 M ,
(2.9) H − H = i(T − Te),
(2.10) II(Z, W ) = hZ, W iH,
(2.11) τeZ − τ Z = II(T, Z) − II(Z, T ) = −i∇
e Z H,

(2.12) II(T, Z) = −i∇

e Z H.

Moreover, H ∈ N 1, 0 M and II(Z, W ) ∈ N 1, 0 M .

Proof. — By direct calculation using (2.4), one has
n
1X
H= II (Zᾱ , Zα )
n α=1
n
(2.13) 1 X 
= II (Zα , Zᾱ ) + i hZα , Zᾱ iθ (T − Te)
n α=1
= H + i(T − Te).
Combining with (2.4),

we have II(Z, W ) − II(W , Z) = −ihZ, W i(T − Te)
= hZ, W i H − H . Taking the (1,0) and (0,1) parts, we obtain (2.10).

TOME 71 (2021), FASCICULE 5

2144 Bernhard LAMEL & Duong Ngoc SON

To show that H ∈ N 1, 0 M , observe that dθ = d(ι∗ η) = ι∗ (dη). Thus, for

every X ∈ T 1, 0 M ,

(2.14) 0 = dθ(T, X) = dη(T, X).

Therefore, if X is tangent to M ,

(2.15) hH, Xi = hH − H, Xi = −ihT − Te, Xi = 0.

This implies that H ∈ N 0, 1 M and H ∈ N 1, 0 M . Finally, for Z, W ∈ T 0, 1 M

and X ∈ T 1, 0 M , it holds that

(2.16) II(Z, W ), X = − W , II(Z, X) = −hZ, XihW , Hi = 0.

This implies that II(Z, W ) ∈ N 0,1 M and II(Z, W ) ∈ N 1,0 M . The proof of
Proposition 2.3 is complete. 

2.2. Change of contact forms

Let ι : (M, θ) ,→ (N, η) be a pseudohermitian submanifold, ι∗ η = θ. The

total differential du of a smooth function u can be decomposed into (1, 0),
(0, 1), and the transverse parts as follows:

(2.17) du = ∂b u + ∂¯b u + (T η u) η.

This decomposition depends on the choice of pseudohermitian structure η.

We then define

(2.18) grad1, 0 u = uγ Zγ , grad0, 1 u = grad1, 0 ū.

If ι : M ,→ (N, η) and θ = ι∗ η, then we have

 ⊥
(2.19) grad1, 0 1, 0 1, 0
N u = gradN, M u + gradN, M u ,

and similarly for grad0, 1 u. Here the orthogonal complements in T 1, 0 N and

T 0, 1 N are defined using the Levi metric corresponding to any pseudoher-
mitian structure on N .

Proposition 2.4. — Let ι : (M, θ) ,→ (N, η) be a pseudohermitian sub-

manifold, ι∗ η = θ. Suppose that P is the manifold N with ηe = eu η.
Put θe := eu◦ι θ. Let eII be the second fundamental form of the inclusion
ι : (M, θ)
e ,→ (P, ηe). Then

(2.20) II(Z, W ) = II(Z, W ),

e Z, W ∈ T 1, 0 M,

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2145

II(Z, W )
(2.21) e
 ⊥
= II(Z, W ) − hZ, W iη grad1, 0
N, M u , Z ∈ T 0, 1 M, W ∈ T 1, 0 M,
and
 ⊥
1, 0
(2.22) eu µP
M = µN
M − < gradN, M u .

Proof. — The first two identities follow from Lee’s formulas for the pseu-
doconformal change of the metrics [14]. Precisely, on M , we have
(2.23) ∇
e Z W = ∇Z W + Z(u)W + W (u)Z,

and
(2.24) e W = ∇ W − hW, Ziθ grad1, 0 u.
∇ Z Z N

Similar formulas hold on M and hence (2.20) and (2.21) follow immediately.
The last identity (2.22) also follows by taking the trace of (2.21) and its
conjugate. 
Remark 2.5. — In view of (2.20), II(Z, W ), where Z, W are (1, 0)-vectors,
is called the CR second fundamental form of the CR immersion. It can be
computed by any pair of pseudohermitian structures θ = ι∗ η, not necessary
admissible. This notion has been extensively used in the study of the CR
immersions [8, 9, 10].

2.3. The Gauß and Weingarten equations

Proposition 2.6 (Pseudohermitian Weingarten Equation). — If N is

a section of N 1, 0 M ⊕ N 0, 1 M , then
D E
(2.25) ∇e X N, Y = − N, II(X, Y )

for all sections X, Y of T 1, 0 M ⊕ T 0, 1 M .

In the following, we shall use the following convention for the curvature
operator:
(2.26) R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X, Y ] Z.
Then for X, Y, Z tangent to M ,

(2.27) R(X,
e Y )Z = R(X, Y )Z + II(X, ∇Y Z) − II (Y, ∇X Z)
+∇
e X (II(Y, Z)) − ∇
e Y (II(X, Z)) − II ([X, Y ], Z) .

Here, R and R
e are the curvature on (M, θ) and (N, η) respectively.

TOME 71 (2021), FASCICULE 5

2146 Bernhard LAMEL & Duong Ngoc SON

Proposition 2.7 (Pseudohermitian Gauß equation). — Let ι : (M, θ)

,→ (N, η) be a pseudohermitian submanifold, ι∗ η = θ. Then the Gauß
equation holds, i.e.,
D E
(2.28) R(X,
e Y )Z, W = R(X, Y )Z, W + II(X, Z), II(Y , W )
− |H|2 hY , ZihX, W i + hX, Y ihZ, W i .

Moreover,
(2.29) he
τ Z, W i = hτ Z, W i + i II(Z, W ), H .

Remark 2.8. — For the special case of pseudohermitian immersions, this

is in [9, Eq. (5.3) or Proposition 5.1] as in this case, H = 0 ; see also [22]
for similar equations.
It is sometimes helpful to write the Gauß equations using index notations,
that is
a b̄ 2


(2.30) R αβ̄γ σ̄ = Rαβ̄γ σ̄ + ωαγ ωβ̄ σ̄ gab̄ − |H| gγ β̄ gασ̄ + gαβ̄ gγ σ̄ ,
e

and
a
(2.31) A
eαβ = Aαβ + iωαβ H b̄ gab̄ .
a
Here ωαβ is the components of the second fundamental form in a local
a
frame, i.e. II(Zα , Zβ ) = ωαβ Za , H = H b̄ Zb̄ , Aαβ = hτ Zα , Zβ i, and so on.
These hold for all isopseudohermitian immersions.

Proof of Proposition 2.7. — The proof is analogous to the case of CR

immersions into Kähler manifolds considered in [22]. Namely, for X, Z ∈
T 1,0 M and Y , W ∈ T 0,1 M ,
D E

R(X,
e Y )Z, W = R(X, Y )Z, W − II [X, Y ], Z , W
D
E D E
+ ∇ e X II(Y , Z) , W − ∇ e (II(X, Z)) , W
Y


(2.32) = R(X, Y )Z, W − II [X, Y ], Z , W
− II(Y , Z), II(X, W ) + II(X, Z), II(Y , W )
= R(X, Y )Z, W + II(X, Z), II(Y , W )
− hY , ZihX, W i|H|2 − II [X, Y ], Z , W .

On the other hand, by the defining properties of the Tanaka–Webster

connection [24, Proposition 3.1], we have
(2.33) [X, Y ] = ∇X Y − ∇Y X − ihX, Y iT.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2147

Therefore, using (2.11),

(2.34) II [X, Y ], Z , W
= −ihX, Y i II(T, Z), W = hX, Y ihZ, W i|H|2 .
Putting this into (2.32) we obtain (2.28). Equation (2.29) follows easily
from (2.11) and we omit the detail. 

3. The chain rule: Proof of Theorem 1.2

Let ι : (M, θ) ,→ (N, η) be a pseudohermitian submanifold, ι∗ η = θ. Re-

N
call that the tensor ν = νM introduced in the introduction is the symmetric
real tensor on T M ⊕ T 0, 1 M defined by
1, 0

(3.1) ν(X, Y ) = 2 hSym II(X, Y ), µi − hX, Y i|µ|2 ,

where µ is the trace of Sym II, i.e., µ = <H. The components of ν in the
“horizontal” directions are given in a local frame by
a
(3.2) ναβ = νᾱβ̄ = ωαβ H b̄ gab̄ , ναβ̄ = νβ̄α = 12 |H|2 gαβ̄ .
Here the Greek indices α and β run from 1 to n := dimCR M , the lowercase
indices a and b run from n + 1 to dimCR N , and g is the Levi metric.

Proposition 3.1. — Let ι : (M, θ) ,→ (N, η) be a pseudohermitian sub-

manifold, ι∗ η = θ. Suppose P is the manifold N with the pseudohermi-
tian structure ηe = eu η and put θe = eu◦ι θ. Then for any sections X, Y of
T 1, 0 M ⊕ T 0, 1 M ,
P N
(3.3) νM (X, Y ) = νM (X, Y ) − 2 hSym II(X, Y ), ξiθ + hX, Y iθ |ξ|2θ

where ξ = <(grad1, 0 ⊥
N u) .

Proof. — Observe that

P
 ⊥
1, 0
(3.4) IIM (X, Y ) = IIP
e
M (X, Y ) − hX, Y iθ gradN u ,

and thus, with ξ = <(grad1, 0 ⊥

N u) ,

(3.5) II(X, Y ) = Sym II(X, Y ) − hX, Y iθ ξ.

Sym e
Taking the trace, we obtain
(3.6) eu µP N
M = µM − ξ.

TOME 71 (2021), FASCICULE 5

2148 Bernhard LAMEL & Duong Ngoc SON

Therefore,

P
(3.7) νM (X, Y )
D E 2
II(X, Y ), µP
= 2 Sym e M − hX, Y ie
θ
µP
M e
θ
eθ
2
= 2 Sym II(X, Y ) − hX, Y iθ ξ, µN
M −ξ θ
− hX, Y iθ µN
M −ξ θ
N
= νM (X, Y ) − 2 hSym II(X, Y ), ξiθ + hX, Y iθ |ξ|2θ .

The proof of Proposition 3.1 is complete. 

Recall that the operator H(u) on a pseudohermitian manifold (M, θ) is
defined, for horizontal vectors, by
2
(3.8) Hθ (u) = Sym ∇∇u − ∂b u ⊗ ∂b u − ∂¯b u ⊗ ∂¯b u + 1
2 ∂¯b u Lθ .

The tensor Hθ (u) is closely related to the CR Schwarzian tensor [21]. In

the notations of [21],
 
1 1  2

(3.9) Bθ u = Hθ (u) + ∆b u − n ∂¯b u Lθ
2 2n
is the traceless part of H. Here Lθ (X, Y ) = hX, Y i is the Levi metric and
we use the convention that ∆b is a nonnegative operator. We can sometimes
write u for u ◦ ι as a function on M .

Proposition 3.2 (cf. [21]). — Let u, v : M → R be smooth functions

on M . Then

(3.10) Hθ (u + v) = Hθ (u) + Hθ̂ (v)

where θb = eu θ.

We point out that the equality of the traceless parts of both sides was
proved in [21].
Proof. — We need to verify (3.10) for each pair of (1, 0)-vectors and for
each pair of one (1, 0)- and one (0, 1)-vector. First, for any vector Z, W of
type (1, 0), we have from (2.22)

(3.11) Hθ̂ (v)(Z, W )

1 2
= Sym ∇v(Z,
b W ) + ∂¯b v θ̂ hZ, W iθ̂
2
 
1 2
= Sym ∇v(Z, W ) + < ∂¯b u, ∂b v + ∂¯b v θ hZ, W iθ .
2

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2149

Thus,

(3.12) Hθ (v)(Z, W ) + Hθ̂ (v)(Z, W )

1 2
= Sym ∇u(Z, W ) + ∂¯b u hZ, W iθ
2 
¯ 1 ¯ 2
+ Sym ∇v(Z, W ) + < ∂b u, ∂b v + ∂b v θ hZ, W iθ
2
1 2
= Sym ∇(u + v)(Z, W ) + |∂b u + ∂b v|θ hZ, W iθ
2
= Hθ (u + v).

This verifies (3.10) for any pair of vectors of mixed type. The identity for
vectors of pure type is exactly the same as [21]. We omit the details. 

Proposition 3.3. — If (M, θ) ⊂ (N, η) is a pseudohermitian submani-

fold, X, Y ∈ Γ(T 1, 0 ⊕ T 0, 1 M ), and u ∈ C 2 (N ), then

(3.13) Hθ (u|M )(X, Y ) − Hη (u)(X, Y )

1  ⊥ 2
= (du) (Sym II(X, Y )) − grad1, 0
N, M u hX, Y iθ .
2
Proof. — On M , we have

(3.14) ∇2 (u|M )(X, Y ) = X(Y (u)) − d(u|M )(∇X Y ),

e 2 u(X, Y ) on N . Thus,
and similarly for ∇

(3.15) ∇2 u(X, Y ) − ∇
e 2 u(X, Y ) = (du)(II(X, Y )).

Consequently,

(3.16) Hθ (u|M )(X, Y ) − Hη (u)(X, Y )

1 1 2 2

= (du) (II(X, Y ) + II(Y, X)) + |∂b (u|M )| − |∂b u| hX, Y iθ
2 2
1  ⊥ 2
= (du) (Sym II(X, Y )) − grad1, 0
N, M u hX, Y iθ . 
2
Proposition 3.4. — For any tower of pseudohermitian submanifolds
M ⊂ N ⊂ (P, η), it holds that

(3.17) µP N P
M = µM + µN ,
P N

∗ P
(3.18) νM − νM = ιN
M νN .

TOME 71 (2021), FASCICULE 5

2150 Bernhard LAMEL & Duong Ngoc SON

Proof. — For Z ∈ T 1,0 M and W ∈ T 0,1 M , we have that

P
hZ, W iHM = IIP
M (Z, W )

(3.19) = IIP N
N (Z, W ) + IIM (Z, W )
P N


= hZ, W i HN + HM .

P N P
Taking Z = W 6= 0, we immediately obtain HM = HM + HN and
hence (3.17) follows. Plugging this into the definition of ν, we obtain (3.18).
The proof is complete. 

Proof of Theorem 1.2. — The idea of the proof is essentially the same
as in [23] and based on the calculations above. The formula in the CR case
turns out to be simpler than its conformal counterpart because of Proposi-
tion 3.4 above. Indeed, assume that F : (M, θ) → (N, η) and G : (N, η) →
(P, ζ). The chain rule for A is equivalent to the analogous assertions for
A0 where
 
(3.20) A0 : F 7→ Hθ (u) + F ∗ νFN(M ) .

In the equidimensional case, this and (1.8) reduce to an analogous state-

ment for the CR Schwarzian that was proved in [21].
Next, we assume that P and N have the same dimension. In this case,
we can suppose that P is the manifold N with a pseudohermitian structure
ζ = ev η and G is the identity map. Since eu θ = F ∗ η, by Proposition 3.2,

Hθ (u + v ◦ F ) = Hθ (u) + Heu θ (v ◦ F )
= Hθ (u) + F ∗ H(F (M ), ι∗ η) (v).

Therefore,

(3.21) A0 (G ◦ F ) − A0 (F ) − F ∗ A0 (G)
(N, ev η)
 
= Hθ (u + v ◦ F ) + F ∗ νF (M )
  
(N, η)
− Hθ (u) + F ∗ νF (M ) − F ∗ (Hη (v))
(N, ev η)
 
(N, η)
= F ∗ H(F (M ), ι∗ η) (v) − Hη (v) + νF (M ) − νF (M )
= 0.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2151

To conclude the proof, we consider the general case when G is a CR

immersion, G∗ ζ = ev η. We have

(3.22) A0 (G ◦ F ) − A0 (F ) − F ∗ (A0 (G))

 
= Hθ (u + v ◦ F ) + (G ◦ F )∗ νFP (G(M ))
     
− Hθ (u) + F ∗ νFN(M ) − F ∗ Hη (v) + G∗ νG(N P
)
   
∗ P P ∗
= (G ◦ F ) νG(F (M )) − νG(N ) − F νFN(M )
= 0.
At the last step, we have used Proposition 3.4. The proof of Theorem 1.2
is complete. 

4. CR maps with vanishing Ahlfors derivative

It is well-known that the Schwarzian derivative of a conformal diffeomor-

phism measures the change of the traceless component of the Ricci tensor,
see [19]. The same is true for the CR analogue of the Schwarzian [21].
The case of immersions is a bit different due to the presence of the sec-
ond fundamental form. We point out that in the CR case a certain part of
the Ahlfors derivative actually measures how the pseudohermitian torsion
changes when going from the original structure to the pull-back.
We denote Aαβ (F ) = A(F )(Zα , Zβ ) the “holomorphic” components and
Aαβ̄ (F ) = A(F )(Zα , Zβ̄ ) the “mixed type” components of the Ahlfors de-
rivative. As usual, hαβ̄ denotes the Levi metric on (M, θ), i.e. hαβ̄
= hZα , Zβ̄ i = −idθ(Zα , Zβ ), and similarly, gAB̄ denotes the one for (N, η).
Proposition 4.1. — Let F : (M, θ) → (N, η) be a CR immersion. Let
Aαβ be the pseudohermitian torsion of (M, θ) and A eAB the pseudohermi-
tian torsion of (N, η). Also let Rαβ̄ be the Ricci tensor of the Tanaka–
Webster connection associated to θ. In an adapted coframe the following
hold:
(i) The “holomorphic” components of the Ahlfors tensor are
 
(4.1) Aαβ (F ) = −i Aeαβ − Aαβ .

(ii) The tracefree components of mixed type are

(4.2) tf Aαβ̄ (F ) = tf Rαβ̄ − tf R
b ,
αβ̄

b is the Ricci tensor associated to θb = F ∗ η.

where R αβ̄

TOME 71 (2021), FASCICULE 5

2152 Bernhard LAMEL & Duong Ngoc SON

(iii) If (N, η) is the sphere with its standard pseudohermitian structure,

then
1 2
(4.3) traceθ A(F ) := hαβ̄ Aαβ̄ (F ) = eu II2, 0 ,
2(n + 1)
where |II2, 0 |2 is the squared norm of the CR second fundamental
form.

Remark 4.2. — This proposition exhibits some new features of the CR

analogue of the Ahlfors tensor derivative compared to its conformal coun-
terpart. Observe that (i) and (ii) generalize similar formulas for the CR
Schwarzian in [21] while (iii) says that the Ahlfors tensor is traceless if and
only if F has vanishing CR second fundamental form. Moreover, part (iii)
also holds when (N, η) has the Tanaka–Webster curvature tensor of the
form
RAB̄C D̄ = c (gAB̄ gC D̄ + gAD̄ gC B̄ ) , c ∈ R.
In particular, it holds when (N, θ) is the Heisenberg hypersurface with a
pseudohermitian structure of vanishing curvature and torsion.
b → (N, η) as
Proof. — We identify M with F (M ) and consider ι : (M, θ)
∗ u
the inclusion of a submanifold and thus θ := ι η = e θ for some smooth
b
function u. Let Abαβ be the pseudohermitian torsion of θb on M . By the
Gauß equation, we have
(4.4) A bαβ + i ω a H b̄ g .
eαβ = A
αβ ab̄

bαβ − i(uα, β − uα uβ ), by [14]. Here the indices

On the other hand, Aαβ = A
preceded by a comma indicate covariant derivatives. Therefore,
a
Aαβ (F ) = uα, β − uα uβ + ωαβ H b̄ gab̄
(4.5) = uα, β − uα uβ − i(A
eαβ − A
bαβ )

= −i(A
eαβ − Aαβ ).

Thus, Proposition 4.1 (i) is proved.

From the definition, the mixed type components are given by
1

(4.6) Aαβ̄ (F ) = uα, β̄ + uβ̄, α
2
1 ¯ 2 2

+ ∂b u + eu HF (M ) ◦ F + eu Jη ◦ F − Jθ hαβ̄ .
2
Therefore, using [14], we obtain
1
1
(4.7) tf Aαβ̄ (F ) = uα, β̄ + uβ̄, α + ∆b u hαβ̄ = tf Rαβ̄ − tf R
b ,
αβ̄
2 2n

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2153

where R b is the Ricci curvature of θb = eu θ = F ∗ η.

αβ̄
Suppose that (N, η) is a sphere with the standard pseudohermitian struc-
ture, then the Gauß equation implies that
2 1 2
(4.8) Jη ◦ F = Jθ̂ − HF (M ) ◦F + II0, 2 ◦ F.
n(n + 1) F (M )
On the other hand,
1 2
(4.9) ∆b u − ∂¯b u .
eu Jθ̂ = Jθ +
n
Putting them together, we obtain that
2 2 1
(4.10) eu Jη ◦ F − Jθ + ∂¯b u + eu HF (M ) ◦F − ∆b u
n
1 2
= eu II2, 0
F (M ) ◦ F.
n(n + 1)
It thus holds that
1 2
(4.11) traceθ A(F ) = hαβ̄ Aαβ̄ (F ) = eu II2, 0 ,
2(n + 1)
and the proof of Proposition 4.1 is complete. 

Definition 4.3. — Let ι : (M, θ) ,→ (N, η) be a pseudohermitian sub-

manifold, ι∗ η = θ. We say that ι (or M if the embedding is understood) is
umbilic at p if IIN
M (p) is a multiple of the Levi metric, i.e., II(Z, W ) = 0 for
all (1, 0) vectors Z and W at p.

We remark that the pseudohermitian total umbilicity does not depend

on the choice of pseudohermitian structures. In fact, if ι is pseudoher-
mitian umbilic, then it is CR totally geodesic for any admissible pair of
contact forms. This holds because the CR second fundamental form (as
defined in [9]) associated to any admissible pair of pseudohermitian struc-
tures coincides with the “holomorphic” components of the pseudohermitian
second fundamental form.

Corollary 4.4. — Let (M, θ) be a pseudohermitian manifold and as-

sume that F : (M, θ) → (S2N +1 , Θ) a CR immersion into a sphere. Then
trace A(F ) > 0, and equality holds at a point p ∈ M iff (M, θ) is um-
bilic at p. If trace A(F ) = 0 identically, then M is CR spherical and F is
spherically equivalent to the linear embedding.

Proof. — That trace A(F ) > 0 follows directly from (4.11). The equal-
ity occurs precisely when the CR second fundamental form vanishes. This
implies that M must be CR spherical by [9] for n > 2 and [22] for the case

TOME 71 (2021), FASCICULE 5

2154 Bernhard LAMEL & Duong Ngoc SON

n = 1. Moreover, F is equivalent to the linear embedding by [13] for the

case dim M > 5 and [22] for the three-dimensional case. 

5. Maps from three-dimensional CR manifolds into

spheres

Suppose that (N, η) = (S, Θ) where Θ is the standard pseudohermitian

structure on the sphere S := S2N +1 . If F = φ ◦ G for some CR automor-
phism φ of the target sphere, then A(F ) = A(G) by Theorem 1.2 and, in
particular, Aθαβ̄ (F ) = Aθαβ̄ (G). This last equality implies that uα, β̄ = vhαβ̄
for some function v. When n > 2, using Lee’s characterization of CR-
pluriharmonicity [15], we can infer from the last condition that u must be
CR-pluriharmonic. This argument does not work in the case n = 1. We
shall deduce the CR-pluriharmonicity from a simpler argument which also
works in the case n = 1 as follows.

Proposition 5.1. — Let M be a strictly pseudoconvex CR manifold

and assume that F and G are CR immersions from M into the sphere S.
Let θ = F ∗ Θ and θ0 = G∗ Θ, where Θ is the standard pseudohermitian
structure on S. Put θ0 = eu θ. If F = φ ◦ G, for some CR automorphism φ
of S, then u is CR-pluriharmonic.

Proof. — If G = φ ◦ F for some CR automorphism φ of S, then this

automorphism extends to a biholomorphic automorphism of the ball B
which we will also denote by φ. Let φ(0) = a and let ρ = kzk2 − 1 be the
defining function for S. Then by [20, Theorem 2.2.2],
!
2 1 − |a|2
(5.1) (ρ ◦ φ)(z) = |φ(z)| − 1 = 2 · ρ(z).
|1 − z · ā|

Observe that Θ = −ι∗ (i∂ρ) and hence φ∗ Θ = −ι∗ (i∂(ρ ◦ φ)) = eϕ Θ, where
2
ϕ(z) = log 1 − |a|2 − log |1 − z · ā | , z ∈ S.


(5.2)

In particular, ϕ is CR-pluriharmonic and so is u = ϕ ◦ F . 

In view of this proposition, we can construct another invariant which
plays the role of the traceless part of A(F ) as follows. By [15, Proposi-
tion 3.4], u is CR-pluriharmonic if and only if
 
(5.3) P u := u1̄, 1̄ 1 + iA11 u1 θ1 = 0.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2155

Here A11 is the component of the pseudohermitian torsion. This suggests

the following notion: For each CR immersion F from the 3-sphere into a
sphere, define the (1, 0)-form A1 (F ) by

(5.4) A1 (F ) = P (u),

where as before u is determined by the relation F ∗ Θ = eu θ.

We formulate the discussion above as follows.

Proposition 5.2. — Let (M, θ) be a three-dimensional strictly pseu-

doconvex pseudohermitian manifold and assume that F and G are CR
immersions of M into the sphere S. Then A1 is invariant with respect to
spherical equivalences in the following sense:
(1) If φ is a CR automorphism of the target, then

(5.5) A1 (φ ◦ F ) = A1 (F ).

(2) If M is the 3-sphere and γ is a CR automorphism of M , then

A1 (F ◦ γ) = γ ∗ eψ A1 (F )


(5.6)

where ψ is determined by the relation γ ∗ θ = eψ θ.

Proof. — We write F ∗ Θ = eu θ and G∗ Θ = ev θ. Suppose that G = φ ◦ F ,

then Proposition 5.1 implies that u − v is CR pluriharmonic and hence

(5.7) A1 (φ ◦ F ) = P (v) = P (u) = A1 (F ),

as desired.
Suppose that M is a 3-sphere, γ is a CR automorphism of M , and θ is
the standard pseudohermitian structure on M . Suppose that G = F ◦ γ,
then

(5.8) v = u ◦ γ + ψ.

Under the change of contact form θe = eψ θ, the operator P changes as

follows: We write P θ for the operator P associated to θ and similarly for
θ.
e Then

(5.9) eψ P e
θ
(u) = P θ (u) + 2i P θ (u), ∂¯b ψ θ.

Since P (ψ) = 0, we have that

−1 ∗
 
(5.10) P θ (v) = P θ (u ◦ γ) = γ ∗ P (γ ) θ (u) = eψ γ ∗ P θ (u)


mod θ.

This completes the proof of Proposition 5.2. 

TOME 71 (2021), FASCICULE 5

2156 Bernhard LAMEL & Duong Ngoc SON

6. Explicit calculations in local coordinates

6.1. The Ahlfors tensor A in general dimensions

The purpose of this section is to give an explicit formula for the her-
mitian part of the Ahlfors tensor. The formula will be explicit in terms
of the defining functions of the source and target and will be simplified
when the target is the sphere. Let M ⊂ Cn+1 be a strictly pseudoconvex
real hypersurface defined by ρ = 0. It is well-known that there exists a
unique (1, 0) vector field ξ satisfying the following two conditions (see, e.g.
[12, 18]):
(6.1) ¯ = ir∂ρ,
ξ c i∂ ∂ρ ¯ ∂ρ(ξ) = 1.
The function r = r[ρ] := ρj k̄ ξ j ξ k̄ is called the transverse curvature of M
and ρ [12]. When ρ is strictly plurisubharmonic, then r−1 = |∂ρ|2 in the
Kähler metric with potential ρ.
Proposition 6.1. — Let M be a strictly pseudoconvex real hypersur-
face in Cn+1 and F : M → N ⊂ Cd+1 a CR immersion. Suppose that
F extends holomorphically to a neighborhood of M . Let ρe be a strictly
¯ ρ = ρe ◦ F ,
plurisubharmonic defining function for N and let η = i∂ρ,
¯ ∗
θ = i∂ρ, so that F η = θ. Then
2
(6.2) HF (M ) ◦ F = r[ρ] − r[e
ρ] ◦ F.
ρ] ◦ F 6 r[ρ] on M . The equality holds if and only if (θ, η)
In particular, r[e
is an admissible pair for the CR immersion F .
Proof. — As already noted, F ∗ η = θ. Also note that the Reeb vector
field is T = i(ξ − ξ). ¯ We shall compute F∗ ξ as follows. In local coordinates
(z1 , . . . , zn+1 ) we write ξ = ξ j ∂j (using summation convention). By direct
calculations (see, e.g., [16, 18]) we obtain
(6.3) ξ j = rρj = rρj k̄ ρk̄ ,
where ρj = ∂ρ/∂zj , ρj k̄ = ∂ 2 ρ/∂ z̄k ∂zj , and ρj k̄ is the inverse transpose
matrix of ρj k̄ . Thus, for p ∈ M and q = F (p) ∈ M ⊂ N , and in local
coordinates z 0A ,
(6.4) F∗ (ξp ) = FjA (p) ξpj ∂A .
On the other hand, the Reeb vector field on N is given by Te = i(ξ 0 − ξ 0 ),
with ξ 0 = ξ 0A ∂A . Thus, by Proposition 2.3,
 
(6.5) HF (M ) F (p) = FjA (p)ξpj − ξ 0A ∂A .

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2157

By direct calculations,
(6.6) ρj = ρeA FjA , ρj k̄ = FjA Fk̄B̄ ρeAB̄ .
Therefore,
(6.7) FjA ξ 0B̄ ρeAB̄ = reFjA ρeA = reρj .
Consequently,
2
(6.8) HF (M ) ◦F

 0B̄ 
= ξ 0A − FjA ξ j ξ − Fk̄B̄ ξ k̄ ρeAB̄
 
= ξ 0A ξ 0B̄ ρeAB̄ − 2< FjA ξ j ξ 0B̄ ρeAB̄ + FjA Fk̄B̄ ξ j ξ k̄ ρeAB̄

= re − 2< reξ j ρj + ξ j ξ k̄ ρj k̄

= re − 2e
r+r
= r − re.
The proof of Proposition 6.1 is complete. 
We also need the following computational result.

Proposition 6.2 ([22]). — Let M be a strictly pseudoconvex real hy-

persurface defined by ρ = 0, with dρ 6= 0 along M . Let σ be a smooth
function in a neighborhood of M and ρb = eσ ρ. Then
2
(6.9) eσ rb = r + 2<(ξ)σ − ∂¯b σ .

Here, ∂¯b is the Cauchy–Riemann operator associated to θ := −i∂ρ and ξ is

the transverse vector field of ρ.

Suppose that ρ and ρe are defining functions for (M, θ) and (N, η) such
¯ and η = i∂¯ρe. Let F : M → N be a CR immersion, extended
that θ = i∂ρ
as a holomorphic immersion in a neighborhood of a point p ∈ M , and let
Q be such that
(6.10) ρe ◦ F = Qρ.
By a transversality argument, Q 6= 0 on M . Thus, we may assume that
Q > 0 on M . Then F ∗ η = eu θ with u = log Q|M .
If r[ρ] and J[ρ] are the transverse curvature of ρ and the Levi–Fefferman
determinant (see [18]), i.e.
 
det ρj k̄
 
ρ ρk̄
(6.11) J[ρ] = − det , r[ρ] = ,
ρj ρj k̄ J(ρ)

TOME 71 (2021), FASCICULE 5

2158 Bernhard LAMEL & Duong Ngoc SON

then by the Li–Luk formula for the Webster scalar curvature [17] (see
also [18, Proposition 4.1]),
(6.12) Jθ = r[ρ] + Pρ log J[ρ],
where
1  
(6.13) Pρ := ξ j ξ k̄ − ψ j k̄ ∂j ∂k̄ .
2n(n + 1)
Here, ψ j k̄ is the inverse transpose of ψj k̄ := ρj k̄ + (1 − r[ρ])ρj ρk̄ .
Then by using Propositions 6.1 and 6.2 and the formulas above, we obtain
that, in terms of the local frame Zα := ∂α − (ρα /ρw )∂w , the mixed type
components of A are given by
1
1
(6.14) Aαβ̄ (F ) = uα, β̄ + uβ̄, α + (Nρ u) hαβ̄
2 2
1  u  
+ e Peρ
log J[eρ ] ◦ F − P ρ log J[ρ] hαβ̄ ,
2
where u = log Q and Nρ := ξ + ξ. ¯
Following [17], we define the second order operator
(6.15) Dαρ β̄ = ∂β̄ ∂α −(ρα /ρw ) ∂w ∂β̄ − ρβ̄ /ρw̄ ∂w̄ ∂α + ρα ρβ̄ /|ρw |2 ∂w ∂w̄ ,

which satisfies
Dαρ β̄ ϕ = ϕZZ Zα , Zβ̄ .


(6.16)
We can now give a completely explicit formula for the mixed type compo-
nents of the Ahlfors tensor.
Proposition 6.3. — With the notations introduced above, it holds
that
1  u  
(6.17) Aαβ̄ (F ) = Dαρ β̄ log Q − e Peρ
log J[e
ρ ] ◦ F − P ρ log J[ρ] hαβ̄ ,
2
Proof. — This follows from (6.14) and a well-known formula for the
Christoffel symbols of the Tanaka–Webster connection [17]. 
We point out that this formula involves both tangential and normal
derivatives of the quotient Q and the Fefferman determinants on the source
and the target. The interesting case is when ρ and ρe are approximate
Fefferman defining functions for the source and target of order 3, i.e., when
J(ρ) = 1+o(ρ2 ) and J(e ρ) = 1+o(eρ2 ), because then the formula simplifies to
(6.18) Aαβ̄ (F ) = Dαρ β̄ log Q.
In particular, we have a simple formula when both source and target are
spheres.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2159

Corollary 6.4. — Let (M, θ) be a strictly pseudoconvex pseudoher-

mitian manifold, F and G nonconstant holomorphic maps sending M into
S2N +1 , and
(6.19) kF k2 − 1 = QF · ρ, kGk2 − 1 = QG · ρ.
where ρ is a defining function for M , θ := i∂ρ. ¯ If F = γ ◦ G for some
2N +1
γ ∈ Aut(S ), then
(6.20) ι∗ ∂ ∂¯ log (QF /QG ) = 0.
Here ι : M → Cn+1 is the inclusion.
Proof. — As explained above, in the frame Zα = ∂α − (ρα /ρw )∂w , we
have that
1
(6.21) Aαβ̄ (F ) = Dαρ β̄ log QF + Pρ log J(ρ)hαβ̄ ,
2
and similar for Aαβ̄ (G). Thus, if there exists such a γ, then by Theorem 1.2,
we have that
(6.22) Dαρ β̄ log (QF /QG ) = 0.
The last equality is clearly equivalent to (6.20), and the proof of Corol-
lary 6.4 is complete. 
Note that if F and G are nonconstant then QF and QG are nonvanishing
on M by a Hopf Lemma (a transversality result). The quotient QF defined
as above has been used extensively in the study of proper holomorphic maps
between balls or sphere maps; see, e.g., [5, 7] and the references therein.
The explicit formula can be used to deduce a necessary condition, which
is simple to check, for a map to be equivalent to a monomial map as follows.
Corollary 6.5. — Let F be a monomial map between spheres in the
standard coordinates, then trace A(F ) only depends on the moduli of the
|zj |, j = 1, 2, . . . , n + 1. Moreover, for each j, Aαβ̄ (F ) is tracefree on the
set |zj | = 1. Consequently, if F is equivalent to a monomial map, then the
umbilical locus of F (S2n+1 ), if not empty, is invariant under a torus action.
We mention a similar result characterizing the monomial maps F in
terms of the invariant group ΓF in a recent paper by D’Angelo and Xiao [7].
Precisely, they proved that F is spherically equivalent to a monomial map
iff the group ΓF contains an n-torus.
Proof. — Suppose that F is a monomial map, then kF k2 − 1 is a mul-
tivariate polynomial in |zj |2 , j = 1, 2, . . . , n + 1: There is a polynomial
f (t1 , . . . , tn+1 ) such that
kF k2 − 1 = f |z1 |2 , . . . , |zn+1 |2 .


(6.23)

TOME 71 (2021), FASCICULE 5

2160 Bernhard LAMEL & Duong Ngoc SON

Since F sends sphere into sphere, f (t) is divisible by t1 +t2 + . . . +tn+1 −1 in

the polynomial ring R[t]. Thus, the quotient g(t1 , . . . , tn+1 ) is a polynomial
in tj with real coefficients. As log QF (z) = log g(|z1 |2 , . . . , |zn+1 |2 ), we can
write at point for which w = zn+1 6= 0, that
kzk2 −1
(6.24) Aαβ̄ (F ) = Dαβ̄ log QF = φα δαβ + z̄α zβ φαβ ,
where
∂ log g
φα = ,
∂tα tj =|zj |2

and
∂2 ∂2 ∂2 ∂2
 
φαβ = − − + 2 log g(t) ,
∂tα tβ ∂tβ ∂tn+1 ∂tα ∂tn+1 ∂tn+1 tj =|zj |2

are real-valued. Since the inverse of the Levi matrix is hαβ̄ = δαβ − zα z̄β ,
we obtain that Aαβ̄ (F ) is tracefree on |w| = 1. Moreover,
X X 2
X 2 2
(6.25) trace A(F ) = φα + (φαα − φα ) |zα | − φαβ |zα | |zβ | .
α α α, β

These prove the Corollary 6.5. 

6.2. The (1, 0)-form A1 in dimension three

As discussed in the last section, in dimension three, the (1, 0)-form A1

can be thought of as a replacement for the tracefree part of the Ahfors
derivative, which is trivial in in this case. It is possible to give an explicit
formula for A1 in general, but we shall focus on the case where M =
S3 is the 3-sphere with the standard pseudohermitian structure Θ. Since
the pseudohermitian torsion vanishes, we have P (u) = u1̄, 1̄ 1 (covariant
derivatives). Put
(6.26) L = w̄∂z − z̄∂w , θ1 = wdz − zdw.
Then clearly,
(6.27) P (u) = (LLL u) θ1 .
In the case of the 3-sphere, the conjugate of P reduces to the operator
characterizing the CR pluriharmonic functions introduced by Bedford [2].
Let F, G : S3 → SN be CR maps with
(6.28) kF k2 − 1 = QF · ρ, kGk2 − 1 = QG · ρ.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2161

Then
A1 (F ) = LLL log QF θ1 ,


(6.29)
and similarly for G. Therefore, if F = φ ◦ G, then



(6.30) LLL log QF = LLL log QG
Moreover, if F ◦ γ = φ ◦ G, then
LLL log QF ◦ γ = e−2ψ LLL log QG



(6.31)
where ψ is determined by γ ∗ θ = eψ θ.
Similar to Corollary 6.5, we have the following characteristic of the A1
of monomial maps. For a monomial map F from a 3-sphere into another
sphere, A1 (F ) expressed in the standard coframe (6.26), must have the
following form:
A1 (F ) = z̄ w̄p |z|2 , |w|2 /Q3F ,


(6.32)
where p is a polynomial of |z|2 and |w|2 with real coefficients. In particular,
the vanishing locus of A1 (F ), which is an invariant for equivalent sphere
maps, must contain at least two circles. The proof of this fact is similar to
that of the aforementioned corollary. We omit the details.

7. Examples and a question

We calculate the Ahlfors tensor A and the (1, 0)-form A1 for various
sphere maps that have previously appeared in the literature. For the sphere
case, the calculations are simple by Proposition 6.3. All calculations can
be done by hand, but some tedious calculations can also be done by a
computer algebra system.
Example 7.1. — In [3], D’Angelo provided a list of 13 discrete and two
1-parameter analytic families of monomial maps from S3 to S7 which in-
cludes 4 trivial extensions of maps from S3 → S5 of Faran’s list. Later,
Watanabe [25] found another map (numbered 16 in the Table 7.1). We
compute the trace of the Ahlfors derivative of each map. We then locate
the umbilical points of the images of S3 in S7 via the maps. There are four
types of umbilical loci that occur: the empty set, the whole sphere, one
circle, and the union of two circles; this is predicted in Corollary 6.5. To
simplify the notations, we put S1 = {(eit , 0) : t ∈ R} and S2 = {(0, eit ) : t
∈ R}. In Table 7.1, the expressions in the Ahlfors column are the traces of
the Ahlfors derivatives which only depend on |z|2 , and thus we put s = |z|2 .

TOME 71 (2021), FASCICULE 5

2162 Bernhard LAMEL & Duong Ngoc SON

This trace determines the “hermitian part” (or mixed-part) of the Ahlfors
derivative as n = 1. The “holomorphic” parts vanish in all cases since the
standard spheres have vanishing pseudohermitian torsions. Moreover, the
norms of the CR second fundamental forms |IICR | can be computed easily
from these results.
Observe that the Ahlfors derivatives of these 16 maps are all different.
Thus, two different maps are not left equivalent. Moreover, the umbilical
loci can be used to distinguish equivalent classes. For example, the maps
in the Faran’s list are pairwise nonequivalent, since their umbilical loci are
not congruent under the CR automorphisms of S3 .
We point out that in Table 7.1, the maps numbered 1, 3, and 5, are special
cases of the homogeneous maps. The trace of the Ahlfors derivatives are
constant while the A1 ’s vanish identically. More generally, for homogeneous
maps of degree d from S2n+1 with n > 1, the Ahlfors derivatives are nonzero
multiples of the Levi metric.

ANNALES DE L’INSTITUT FOURIER

 Table 7.1. The CR Ahlfors derivative A and the (1, 0)-form A1 of monomial sphere maps from S3 → S7 .

F trace A(F ) s = |z|2 A1 (F ) s = |z|2 Umbilic

1 (z, w, 0, 0) 0 0 S3
s 4z̄ w̄


2 z, zw, w2 , 0 (s−2)2
− (s−2) 3 S2
√
1
3 z 2 , 2zw, w2 , 0 2
0 ∅
√
3s(s−1) 2z̄ w̄(s−2)(s+1)(2s−1)
4 z 3 , 3zw, w3 , 0 − 2 S1 ∪ S2
√ √
(s −s+1)2 (s2 −s+1)3

TOME 71 (2021), FASCICULE 5

5 z 3 , 3z 2 w, 3zw2 , w3 1 0 ∅

(1−s)(s2 +4s+1) 18z̄ w̄s(s+1)
6 z 3 , z 2 w, zw, w − 2 S2
(s2 +s+1)2 (s +s+1)3
3

s3 +4s2 −5s+2 8z̄ w̄(s +3s−2)
7 z 2 , z 2 w, zw2 , w − ∅
(−s2 +2s+1)2 (s2 −2s−1)3
√ √ 2

2
7s −7s+3 2z̄ w̄(2s−1)(s2 −s+4)
8 z2 , 2z 2 w, 2zw2 , w 2 − ∅
2(−s2 +s+1) (s2 −s−1)3
√ √ 2

2
9s −6s+6 12z̄ w̄(s3 +6s−4)
9 z3 , 3z 2 w, 2zw2 , w − ∅
(−s2 +2s+2)2 (s2 −2s−2)3
√
2s2 −3s+3 12z̄ w̄
10 z, z 2 w, 2zw2 , w3 (3−2s)2
− (2s−3) 3 ∅
√ 4 3 2 2 4 3
3

3s(s −4s +12s −9) 36z̄ w̄(s −2s−1)(s −2s −6s+3)
11 z 4 , z 3 w, 3zw, w − − S1 ∪ S2
(s3 +3s2 −3s+3)2 (s3 +3s2 −3s+3)3
√ 2

3(9s4 −12s2 +4s−1) 36z̄ w̄(s +2s−1)(3s4 −6s3 −2s+1)
12 z4 , 3z 2 w, zw3 , w − S1
(3s3 −3s2 +3s+1)2 (3s3 −3s2 +3s+1)3
√ √ 5

s(s6 −7s5 +15s4 −25s3 +35s2 −27s+9)
13 z5 , 5z 3 w, 5zw2 , w A1 (F ) S2
(s4 −3s3 +4s2 −2s+1)2
√ √ 2

(t2 −1)s 2z̄ w̄(t4 −3t2 +2)
14 z, tw, 1 − t2 zw, 1 − t2 w − 2 S2
THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT

(t (1−s)+s−2)2 ((t2 −1)s−t2 +2)3

√ √
t2 (s+1)−s+1 4z̄ w̄(1−t4 )
15 z2 , 1 + t2 zw, tw2 , 1 − t2 w ∅
(t2 (1−s)+s+1)2 ((t2 −1)s−t2 −1)3
√ 3

−7s2 +6s+3 8z̄ w̄(s3 −9s+6)
16 z2 , 2zw, zw2 , w ∅
(s2 −2s+3)2 (s2 −2s+3)3
2163
2164 Bernhard LAMEL & Duong Ngoc SON

The formula for the A1 tensor of the 13rd map is too big to fit in the
table. Precisely,

A1 (F13 ) =
2s9 − 6s8 + 18s7 − 64s6 + 117s5 − 114s4 + 42s3 + 24s2 − 27s + 6
2z̄ w̄ 3 .
(s4 − 3s3 + 4s2 − 2s + 1)
Example 7.2. — Consider the following cubic map (appeared earlier
in [4])
 2
z − z 2 w zw − zw2 z + zw 2

(7.1) √ , √ , √ ,w .
2 2 2
Then the trace of the CR Ahlfors derivative is


|w|6 + 4|w|4 + 2<(w) 1 − 3|w|2 − 5|w|2 + 4
(7.2) trace A(F ) =
Q S3
3
for some polynomial Q positive on S . Observe that this trace vanishes if
and only if w = 1, i.e., the umbilical locus is a singleton. This immediately
implies that F is not equivalent to any map in Table 7.1 whose umbilical
locus is either empty or of positive dimension. In fact, by Corllary 6.5, it is
not equivalent to any monomial map regardless of the target dimension, a
fact that was first observed by D’Angelo in [4] for the target dimension 4.

Example 7.3. — Consider the following map which was discussed in [11,
Proposition 3.3],
√ √
3 2
6 2

(7.3) F (z, w) = z + 4z − 2 , z +z+1 ,
9 9
√ √ √ 
3 6 2 13
w(3z + 5), w , w(z − 1) .
12 6 12
Observe that this map is not monomial and does not send 0 to 0.
Then the trace of its CR Ahlfors derviative is
30|z|2 + 24<z + 18
(7.4) trace A(F ) = 4 .
|z| − 16|z|2 <z + 32<(z 2 ) + 272<z + 289
Since this trace does not vanish on S3 , F (S3 ) is a submanifold of S9 without
umbilical point. Moreover,


264 w̄ 1 + 4z̄ + z̄ 2
(7.5) A1 (F ) = 3 .
(17 + 8 <z − |z|2 )
Thus, F is not left equivalent to any monomial map from S3 into SN ,
N > 3.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2165

Example 7.4 (D’Angelo’s maps [3]). — For each t, put c = cos(t) and
s = sin(t) and consider the maps
Ft (z, w) = z1 , . . . , zn , cw, sz1 w, . . . , szn w, sw2 .


(7.6)
Then Ft maps S2n+1 into S4n+3 . It is clear that F0 = L is the linear
embedding and Fπ/2 ∼ = (W, 0) where W is the complex Whitney map from
S2n+1 into S4n+1 . We compute that
¯ = eut θ, ut = log 1 + s2 |w|2 .
Qt = 1 + s2 |w|2 , Ft∗ Θ = i∂ρ


(7.7)

In the local frame Zeα = w̄∂α − z̄α ∂w , we have that

s2 z̄α zβ
(7.8) Aαβ̄ (Ft ) = 2
(1 + s2 |w|2 )
is non-negative and of rank one except at the umbilical points. Taking the
trace, we have


s2 1 − |w|2
(7.9) trace A(Ft ) = 2 .
(1 + s2 |w|2 ) S3
Thus, Ft and Ft0 are not left equivalent for t 6= t0 . Moreover, if Ψ ∈
Aut(S2n+1 , Ft ) for t 6= 0, then Ψ(0, w) = (0, ψn+1 (0, w)). The inclusion
F (S2n+1 ) ⊂ S4n+3 is umbilical along the locus {F (0, eiy )}.

The following example appeared originally in Webster [26].

Example 7.5. — Let M ⊂ Cn be the strictly pseudoconvex real hyper-

surface defined by ρ = 0, with
(7.10) ρ = |z|2 + b(z) + b(z) − 1.
¯ and let
Put θ := i∂ρ
1
(7.11) F (z) = (z1 , . . . , zn , b(z)) .
1 − b(z)
Then F maps M into the unit sphere in Cn+1 , actually, one can compute
kF k2 − 1 = |1 − b(z)|−2 ρ. Since Dαρ β̄ log |1 − b(z)|−2 = 0, we see that

(7.12) tf Aαβ̄ (F ) = 0.
Let L and W be the linear embedding and Whitney map from Sn+1 into
S2d+1 with d > 2n − 1. Then L ◦ F and W ◦ F are inequivalent CR im-
mersions from M into S2d+1 . (In general, post composing with inequivalent
maps may still yield equivalent maps. For example, take g to be the linear
embedding of S2n−1 into S2n+1 , then L◦g = W ◦g is the linear embedding.)

TOME 71 (2021), FASCICULE 5

2166 Bernhard LAMEL & Duong Ngoc SON

To compute the trace part, we need to compute the Fefferman deter-

Pn
minant J[ρ]. In a special case b(z) = 12 k=1 zk2 , we can easily compute
J[ρ] = ρ + 2 and thus,
(7.13) Aαβ̄ (F ) = (n + 1)−1 hαβ̄ .
In particular, the imbedding image has no umbilical point.
Question 7.6. — Among the examples discussed above, we have en-
countered three families of CR maps with the property that their Ahlfors
derivatives are constant multiples of the Levi metric: linear embeddings
between spheres, the homogeneous maps, and the ones considered in Ex-
ample 7.5. Motivated by these examples, we pose the following question:
Suppose that F is a nonconstant CR map between spheres and that A(F )
is a nonzero constant (resp. functional) multiple of the Levi metric (i.e.,
Aαβ (F ) = 0 or Aαβ̄ (F ) = ghαβ̄ , g 6= 0, respectively). Does it follow that
F is spherically equivalent to a homogeneous monomial map?

BIBLIOGRAPHY

[1] L. V. Ahlfors, “Cross-ratios and Schwarzian derivatives in Rn ”, in Complex anal-

ysis, Edited by J.Hersch, A. Huber, articles dedicated to Albert Pfluger on the
occasion of his 80th birthday, Birkhäuser, 1988, p. 1-15.
[2] E. Bedford, “The Dirichlet problem for some overdetermined systems on the unit
ball in Cn ”, Pac. J. Math. 51 (1974), no. 1, p. 19-25.
[3] J. P. D’Angelo, “Proper holomorphic maps between balls of different dimensions”,
Mich. Math. J. 35 (1988), no. 1, p. 83-90.
[4] ——— , “Polynomial proper holomorphic mappings between balls II.”, Mich. Math.
J. 38 (1991), no. 1, p. 53-65.
[5] ——— , Several complex variables and the geometry of real hypersurfaces, Studies
in Advanced Mathematics, vol. 8, CRC Press, 1993.
[6] J. P. D’Angelo & J. Lebl, “Homotopy equivalence for proper holomorphic map-
pings”, Adv. Math. 286 (2016), p. 160-180.
[7] J. P. D’Angelo & M. Xiao, “Symmetries in CR complexity theory”, Adv. Math.
313 (2017), p. 590-627.
[8] S. Dragomir, “On pseudohermitian immersions between strictly pseudoconvex CR
manifolds”, Am. J. Math. 117 (1995), no. 1, p. 169-202.
[9] P. Ebenfelt, X. Huang & D. Zaitsev, “Rigidity of CR-immersions into Spheres”,
Commun. Anal. Geom. 12 (2004), no. 3, p. 631-670.
[10] P. Ebenfelt & A. Minor, “On CR embeddings of strictly pseudoconvex hypersur-
faces into spheres in low dimensions”, Trans. Am. Math. Soc. 366 (2014), no. 11,
p. 5693-5706.
[11] J. Faran, X. Huang, S. Ji & Y. Zhang, “Rational and polynomial maps between
balls”, Pure Appl. Math. Q. 6 (2010), no. 3, p. 829-842.
[12] C. R. Graham & J. M. Lee, “Smooth solutions of degenerate Laplacians on strictly
pseudoconvex domains”, Duke Math. J. 57 (1988), no. 3, p. 697-720.
[13] S. Ji & Y. Yuan, “Flatness of CR submanifolds in a sphere”, Sci. China, Math. 53
(2010), no. 3, p. 701-718.

ANNALES DE L’INSTITUT FOURIER

 THE CR AHLFORS DERIVATIVE AND A NEW INVARIANT 2167

[14] J. M. Lee, “The Fefferman metric and pseudo-Hermitian invariants”, Trans. Am.
Math. Soc. 296 (1986), no. 1, p. 411-429.
[15] ——— , “Pseudo-Einstein Structures on CR Manifolds”, Am. J. Math. 110 (1988),
no. 1, p. 157-178.
[16] J. M. Lee & R. B. Melrose, “Boundary behavior of the complex Monge–Ampère
equation”, Acta Math. 148 (1982), no. 1, p. 159-192.
[17] S.-Y. Li & H.-S. Luk, “An explicit formula for the Webster pseudo-Ricci curvature
on real hypersurfaces and its application for characterizing balls in Cn ”, Commun.
Anal. Geom. 14 (2006), no. 4, p. 673-701.
[18] S.-Y. Li & D. N. Son, “The Webster scalar curvature and sharp upper and lower
bounds for the first positive eigenvalue of the Kohn–Laplacian on real hypersur-
faces”, Acta Math. Sin., Engl. Ser. 34 (2018), no. 8, p. 1248-1258.
[19] B. Osgood & D. Stowe, “The Schwarzian derivative and conformal mapping of
Riemannian manifolds”, Duke Math. J. 67 (1992), no. 1, p. 57-99.
[20] W. Rudin, Function theory in the unit ball of Cn, Grundlehren der Mathematischen
Wissenschaften, vol. 241, Springer, 1980.
[21] D. N. Son, “The Schwarzian derivative and Möbius equation on strictly pseudo-
convex CR manifolds”, Commun. Anal. Geom. 26 (2018), no. 2, p. 237-269.
[22] ——— , “Semi-isometric CR immersions of CR manifolds into Kähler manifolds
and applications”, https://arxiv.org/abs/1901.07451v1 to appear in Annali della
Scuola Normale Superiore di Pisa, Classe di Scienze, 2019.
[23] D. Stowe, “An Ahlfors derivative for conformal immersions”, J. Geom. Anal. 25
(2015), no. 1, p. 592-615.
[24] N. Tanaka, A differential geometric study on strongly pseudoconvex manifolds,
Lectures in Mathematics. Department of Mathematics, Kyoto University, vol. 9,
Kinokuniya Company Ltd., Tokyo, 1975.
[25] C. Watanabe, “On proper monomial mappings from 2-ball B2 , to 4-ball B4 ”, Ann.
Sci. Kanazawa Univ. 29 (1992), p. 9-13.
[26] S. M. Webster, “The rigidity of C-R hypersurfaces in a sphere”, Indiana Univ.
Math. J. 28 (1979), no. 3, p. 405-416.

Manuscrit reçu le 14 octobre 2019,

révisé le 1er octobre 2020,
accepté le 28 octobre 2020.

Bernhard LAMEL
Texas A & M University Qatar,
Science Program,
Education City, Doha, Qatar
[email protected]
Duong Ngoc SON
Fakultät für Mathematik,
Universität Wien,
Oskar-Morgenstern-Platz 1,
1090 Wien, Austria
[email protected]

TOME 71 (2021), FASCICULE 5