POLONICI MATHEMATICI LXIV.2 (1996)
Fueter regular mappings and harmonicity by Wies law Kr´ olikowski ( L´od´z)
Abstract. It is shown that Fueter regular functions appear in connection with the Eells condition for harmonicity. New conditions for mappings from 4-dimensional confor- mally flat manifolds to be harmonic are obtained.
1. Introduction. There does not exist a “quaternionic analysis” in the same sense as complex analysis. This term stands for a theory investigating properties of regular functions in the sense of Fueter. Let us recall their definition.
Definition 1.1. A function f : Ω → H (H the quaternions) is said to be regular in a domain Ω ⊂ H if f is differentiable in the usual sense as a mapping of Ω, considered as an open set in R
4, with values in R
4and
D
+f :=
14(∂
0+ i
1∂
1+ i
2∂
2+ i
3∂
3)(f
0+ i
1f
1+ i
2f
2+ i
3f
3) = 0 in Ω, where i
1, i
2, i
3are the quaternionic units.
These functions are the most natural analog of holomorphic functions.
But because of the non-commutativity of quaternions one cannot define a quaternionic manifold with Fueter regular functions as transition functions.
Nevertheless, one does define a “quaternionic manifold” omitting com- pletely the problem of transition functions. This definition is connected with the so-called “Berger list” [1]. Berger proved that the holonomy group of every irreducible Riemannian manifold which is not a symmetric space is a subgroup of one of the following:
SO(n), U (n/2), Sp(n/4) × Sp(1), Sp(n/4), G
2(n = 7), Spin(7) (n = 8), Spin(9) (n = 16), where n denotes the dimension of the manifold in question.
1991 Mathematics Subject Classification: 53C10, 53C40, 58E20.
Key words and phrases : quaternions, Fueter regular functions, harmonic mappings.
[97]
The manifolds whose holonomy group is contained in SO(n) are ori- ented Riemannian manifolds. One can prove only general theorems about the topology of such manifolds.
The manifolds with holonomy group in U (n/2) are complex K¨ ahler man- ifolds. They play a very important role in complex analysis.
The next group from the Berger list constitutes a basis of the following definition:
Definition 1.2. A 4n-dimensional Riemannian manifold is called quater- nionic (precisely, quaternionic-K¨ ahler ) if its holonomy group is a subgroup of Sp(n) × Sp(1) (see e.g. [2]).
Analysis on such manifolds is another kind of “quaternionic analysis”.
At first sight it has nothing to do with Fueter regular functions.
In this paper it is shown that Fueter regular functions do appear on quaternionic manifolds. It turned out, to our surprise, that the Eells condi- tion for harmonicity [5] is, for some kind of 4-dimensional manifolds, equiv- alent to the existence of an antiregular function in the sense of Fueter. It is the most significant result of the present paper. This once more points to the importance of Fueter regular functions. In particular, rewriting the Eells condition for harmonicity in quaternionic variables we obtain new results on harmonic mappings from 4-dimensional conformally flat manifolds.
In particular, we get a very interesting new characterization of harmonic mappings from the 4-dimensional torus.
Some of these problems were discussed at the University “La Sapienza”
in Rome. The author wants to express his hearty thanks to Prof. Stefano Marchiafava for numerous discussions and encouragement in this work.
The manifolds considered in the paper are assumed to be connected, compact, orientable and without boundary.
2. Elements of quaternionic analysis. Denote by H the field of quaternions. Let V be a real m-dimensional vector space and V
∗the dual space to V . Consider the following vector spaces over H:
V
H:= V ⊗
RH — quaternionification of V (a right H-vector space), V
H:= H ⊗
RV — quaternionic conjugation of V (a left H-vector space),
(V
H)
∗:= H ⊗
RV
∗— dual of V
H(a left H-vector space), (V
H)
∗:= V
∗⊗
RH — dual of V
H(a right H-vector space).
We use the fact that qv = vq, v ∈ V
H, q ∈ H.
Let (1, i, j, k) be a fixed standard basis of H with the well known prop-
erties: i
2= j
2= k
2= −1, ijk = −1. Consider three involutions of H,
denoted by σ
1, σ
2and σ
3, given by automorphisms corresponding to i, j, k,
respectively, which satisfy the following conditions:
σ
12= σ
22= σ
32= id, σ
1σ
2= σ
2σ
1= σ
3, σ
2σ
3= σ
3σ
2= σ
1, σ
3σ
1= σ
1σ
3= σ
2. Using quaternionic units we can write
σ
1(q) = −iqi, σ
2(q) = −jqj, σ
3(q) = −kqk, q ∈ H.
Set
q
1:= σ
1(q), q
2:= σ
2(q), q
3:= σ
3(q), q ∈ H.
Explicitly, if q = x
0+ ix
1+ jx
2+ kx
3, then q
1= x
0+ ix
1− jx
2− kx
3, q
2= x
0− ix
1+ jx
2− kx
3, q
3= x
0− ix
1− jx
2+ kx
3.
Right multiplication by i, j and k determines a triple (I
1, I
2, I
3) of com- plex structures on R
4m∼ = H
msatisfying the following conditions:
I
12= I
22= I
32= − Id, I
1I
2I
3= − Id, where Id stands for the identity mapping of R
4m.
Any two such triples (I
1, I
2, I
3) and (I
1′, I
2′, I
3′) are related by a transfor- mation
I
h′=
3
X
k=1
c
hkI
k, h = 1, 2, 3, with (c
hk) ∈ SO(3).
Let (M, g) be a Riemannian manifold.
Definition 2.1 [2]. An almost quaternionic structure is defined as a covering {U
i} of M with two almost complex structures I
iand J
ion each U
isuch that I
iJ
i= −J
iI
iand the 3-dimensional vector spaces of endomor- phisms generated by I
i, J
iand K
i:= I
iJ
i,
End
Ui:= {αI
i+ βJ
i+ γK
i: α, β, γ ∈ R}
are the same on all of M .
A Riemannian metric g is quaternionic-Hermitian if g is Hermitian for each I and J. A manifold (M, g) equipped with an almost quaternionic struc- ture with g quaternionic-Hermitian is called almost-quaternionic-Hermitian.
Definition 2.2. The standard enhanced quaternionic structure of H
mis the 3-dimensional subspace Q
0of the space End
RH
mspanned by (any)
triple (I
1, I
2, I
3) as above, called an admissible hypercomplex base of Q
0(we
will also write (I
1, I
2, I
3) ∈ Q
0).
Definition 2.3. Let (M
4m, g) and (N
4n, h) be two almost-quaternionic- Hermitian manifolds and Φ : (M
4m, g) → (N
4n, h) a smooth map. Then Φ is called Q-holomorphic if for every p ∈ M
4mand each hypercomplex base (I
1′, I
2′, I
3′) ∈ Q
Mpthere exists a hypercomplex base (I
1, I
2, I
3) ∈ Q
NΦ(p)such that
I
a(Φ
∗)
|p= (Φ
∗)
|pI
a′for a = 1, 2, 3.
Let Φ : (M
4m, g) → (N
4n, h) be a C
∞-mapping between two almost- quaternionic-Hermitian manifolds of dimensions 4m and 4n, respectively.
We can regard the quaternionic extension d
HΦ of the differential dΦ as a section of the bundle (Φ
−1T
HN )⊗
H(T
HM )
∗over M .
Let p ∈ M . It is well known (see e.g. [3]) that T
pHM := T
pM ⊗
RH can be decomposed in the following way:
T
pHM := U
pH⊕ τ
1U
pH⊕ τ
2U
pH⊕ τ
3U
pH, where τ
1, τ
2, τ
3are the semi-involutions defined on T
pHM as
τ
1= id ⊗σ
1, τ
2= id ⊗σ
2, τ
3= id ⊗σ
3and σ
1, σ
2, σ
3are the involutions of the algebra of quaternions H defined above.
Take p ∈ M and let U be a small open neighbourhood of p. On U , by Definition 2.1, there are almost complex structures I, J and K := IJ. Using I, J and K we can define U
pHby
U
pH:= {X + iIX + jJX + kKX : X ∈ T
pM }.
Then we get
τ
1U
pH= {X + iIX − jJX − kKX : X ∈ T
pM }, τ
2U
pH= {X − iIX + jJX − kKX : X ∈ T
pM }, τ
3U
pH= {X − iIX − jJX + kKX : X ∈ T
pM }.
R e m a r k 2.1. There exist elements X
1, . . . , X
mof T
pM such that the system (X
1, . . . , X
m, IX
1, . . . , IX
m, JX
1, . . . , JX
m, KX
1, . . . , KX
m) forms an orthonormal basis for T
pHM with respect to the metric g.
P r o o f. It is analogous to that in the complex case [8].
Let (x
i0, x
i1, x
i2, x
i3), i = 1, . . . , m, be local real coordinates at the point p.
We can introduce the quaternionic coordinates (q
1, . . . , q
m) as follows:
q
k:= x
k0+ ix
k1+ jx
k2+ kx
k3, k = 1, . . . , m.
If the almost quaternionic structure (I, J, K) is integrable then we can
assume that I, J and K are given by
I
∂
∂x
k0 |p= − ∂
∂x
k1 |p, J
∂
∂x
k0 |p= − ∂
∂x
k2 |p, K
∂
∂x
k0 |p= − ∂
∂x
k3 |p, I
∂
∂x
k1 |p= ∂
∂x
k0 |p, J
∂
∂x
k1 |p= ∂
∂x
k3 |p, K
∂
∂x
k1 |p= − ∂
∂x
k2 |p, I
∂
∂x
k2 |p= − ∂
∂x
k3 |p, J
∂
∂x
k2 |p= ∂
∂x
k0 |p, K
∂
∂x
k2 |p= ∂
∂x
k1 |p, I
∂
∂x
k3 |p= ∂
∂x
k2 |p, J
∂
∂x
k3 |p= − ∂
∂x
k1 |p, K
∂
∂x
k3 |p= ∂
∂x
k0 |p. It follows that
(2.1)
∂
∂q
k |p= 1 4
∂
∂x
k0 |p− i ∂
∂x
k1 |p− j ∂
∂x
k2 |p− k ∂
∂x
k3 |p,
∂
∂q
k |p1= 1 4
∂
∂x
k0 |p− i ∂
∂x
k1 |p+ j ∂
∂x
k2 |p+ k ∂
∂x
k3 |p,
∂
∂q
k |p2= 1 4
∂
∂x
k0 |p+ i ∂
∂x
k1 |p− j ∂
∂x
k2 |p+ k ∂
∂x
k3 |p,
∂
∂q
k |p3= 1 4
∂
∂x
k0 |p+ i ∂
∂x
k1 |p+ j ∂
∂x
k2 |p− k ∂
∂x
k3 |p.
Then the system {∂/∂q
1|p, . . . , ∂/∂q
m|p} forms a basis for U
pHand {∂/∂q
a1 |p, . . . , ∂/∂q
am|p} are bases for τ
aU
pH, a = 1, 2, 3. It is also clear that {dq
1|p, . . . , dq
m|p} and {dq
a1 |p, . . . , dq
am|p} are bases for (U
pH)
∗and (τ
aU
pH)
∗, a = 1, 2, 3, respectively.
3. Main theorems. Let V be a real vector space of dimension 4n. A quaternionic structure in V corresponds to the following decomposition (see Sect. 2):
V
H= U
H⊕ τ
1U
H⊕ τ
2U
H⊕ τ
3U
H. Consider the tensor products
V
H⊗
H(V
H)
∗, V
H⊗
HV
H, (V
H)
∗⊗
H(V
H)
∗, . . .
Each of them can be decomposed into the direct sum of 16 parts with respect to the given quaternionic structure in V . In particular, to every real covariant 2-tensor S on V , S ∈ V
∗⊗
RV
∗, there corresponds a 2-tensor S
H∈ (V
H)
∗⊗
H(V
H)
∗which can be decomposed into 4 parts, namely real tensors S
0H, S
1H, S
2Hand S
3H(of genus 0, 1, 2, 3, respectively) in the following way:
S
0H∈ (U
H)
∗⊗
H(U
H)
∗+ (τ
1U
H)
∗⊗
H(τ
1U
H)
∗+ (τ
2U
H)
∗⊗
H(τ
2U
H)
∗+ (τ
3U
H)
∗⊗
H(τ
3U
H)
∗,
S
1H∈ (U
H)
∗⊗
H(τ
1U
H)
∗+ (τ
1U
H)
∗⊗
H(U
H)
∗+ (τ
2U
H)
∗⊗
H(τ
3U
H)
∗+ (τ
3U
H)
∗⊗
H(τ
2U
H)
∗,
S
2H∈ (U
H)
∗⊗
H(τ
2U
H)
∗+ (τ
2U
H)
∗⊗
H(U
H)
∗+ (τ
3U
H)
∗⊗
H(τ
1U
H)
∗+ (τ
1U
H)
∗⊗
H(τ
3U
H)
∗,
S
3H∈ (U
H)
∗⊗
H(τ
3U
H)
∗+ (τ
3U
H)
∗⊗
H(U
H)
∗+ (τ
1U
H)
∗⊗
H(τ
2U
H)
∗+ (τ
2U
H)
∗⊗
H(τ
1U
H)
∗.
In each line the tensors are mutally biconjugate.
Assume that
dim
RV = 4.
Let x ∈ V and (∂/∂x
0|x, ∂/∂x
1|x, ∂/∂x
2|x, ∂/∂x
3|x) be a base of the tangent space T
xV (this base is good for every x ∈ V and because T
xV ∼ = V , we can treat it as a base for V ) and let (dx
0, dx
1, dx
2, dx
3) be the dual base.
Consider a real symmetric 2-tensor on V : S = dx
iS
ijdx
j.
The quaternionic decomposition of S
Hin pure components looks as follows:
S
H= dqs
00dq + dq
1s
11dq
1+ dq
2s
22dq
2+ dq
3s
33dq
3+ dqs
01dq
1+ dq
1s
10dq + dq
2s
23dq
3+ dq
3s
32dq
2+ dqs
02dq
2+ dq
2s
20dq + dq
1s
13dq
3+ dq
3s
31dq
1+ dqs
03dq
3+ dq
3s
30dq + dq
1s
12dq
2+ dq
2s
21dq
1,
where
s
00= s
11= s
22= s
33∈ R,
s
10= (s
01)
1, s
23= τ
2s
01, s
32= τ
2s
01, s
23= (s
32)
1, s
20= (s
02)
2, s
13= τ
1s
02, s
31= τ
1s
02, s
21= (s
12)
3, s
30= (s
03)
3, s
12= τ
3s
03, s
21= τ
3s
03, s
31= (s
13)
2.
The relationship between the quaternionic components s
ijand the real com- ponents S
mnis
1
4
S
00= s
00+ Re(s
01+ s
02+ s
03),
1
4
S
11= s
00+ Re(s
01− s
02− s
03),
1
4
S
22= s
00+ Re(−s
01+ s
02− s
03),
1
4
S
33= s
00+ Re(−s
01− s
02+ s
03),
1
4
S
01= − Re[(s
02+ s
03)i] = − Re[(±s
01+ s
02+ s
03)i],
1
4
S
02= − Re[(s
01+ s
03)j] = − Re[(s
01±s
02+ s
03)j],
1
4
S
03= − Re[(s
01+ s
02)k] = − Re[(s
01+ s
02±s
03)k],
1
4
S
12= − Re[(s
01− s
02)k] = − Re[(s
01− s
02±s
03)k],
1
4
S
13= − Re[−(s
01− s
03)j] = − Re[(−s
01±s
02+ s
03)j],
1
4
S
23= − Re[−(s
02− s
03)i] = − Re[(±s
01− s
02+ s
03)i].
Writing ±s
0kwe indicate that we are free to choose the sign due to the fact that Re(s
01i) = Re(s
02j) = Re(s
03k) = 0.
On the other hand, we have s
00=
161(S
00+ S
11+ S
22+ S
33),
s
01=
161(S
00+ S
11− S
22− S
33) +
18(S
02− S
13)j +
18(S
03+ S
12)k, s
02=
161(S
00− S
11+ S
22− S
33) +
18(S
01+ S
23)i +
18(S
03− S
12)k, s
03=
161(S
00− S
11− S
22+ S
33) +
18(S
01− S
23)i +
18(S
02+ S
13)j.
Hereafter we will consider a real Riemannian manifold M which is locally conformally flat with dim
RM = 4 (e.g. the sphere S
4∼ = HP
1or the torus T
4∼ = H/Z
4, see e.g. [2]). Then we can assume that in a neighbourhood of every point p of M there exists a system of local coordinates (x
0, x
1, x
2, x
3) such that the metric g is expressed by
g = g
0R[(dx
0)
2+ (dx
1)
2+ (dx
2)
2+ (dx
3)
2],
where g
R0is a real positive C
∞-function defined about p. Consider the quater- nionic coordinate q := x
0+ ix
1+ jx
2+ kx
3associated with the given system of real coordinates. Then the quaternionic expression of g is
(3.1) g = g
H0[dq ⊗ dq + dq
1⊗ dq
1+ dq
2⊗ dq
2+ dq
3⊗ dq
3].
Comparing the expressions for g in real and quaternionic coordinates we get:
Proposition 3.1. g
R0= 4g
H0.
P r o o f. This follows by straightforward calculations.
Definition 3.1 [5]. Let Φ : (M, g) → (N, h) be a smooth map between Riemannian manifolds. The stress-energy tensor of Φ is the symmetric 2- tensor on M given by
S(Φ) := e · g − Φ
∗h, where e = e(Φ) denotes the energy density of Φ:
e(Φ) :=
12|dΦ|
2=
12g
ijh
αβΦ
αiΦ
βj, and (Φ
αi) = (∂Φ
α/∂x
i) is a local representation of dΦ.
We will compute the quaternionic components of S(Φ).
I. Computation of Φ
∗h. The real expression of the metric h on N is h = X
α,β
h
αβdy
α⊗ dy
β(h
αβ= h
βα),
where (y
α) is a local system of real coordinates defined in an open neigh- bourhood of Φ(p), with p a fixed point in M . If (x
j) is a system of local coordinates about p then
Φ
∗h = dx
iΦ
αih
αβΦ
βjdx
j.
In order to pass to quaternionic coordinates we have to consider the ex- tension of the metric h to the quaternionified tangent bundle of N , T
HN :=
T N ⊗
RH. If X
H, Y
H∈ T
pHN then
X
H= (X
1, . . . , X
n), Y
H= (Y
1, . . . , Y
n), where 4n = dim
RN and
X
α= X
0α+ iX
1α+ jX
2α+ kX
3α, Y
β= Y
0β+ iY
1β+ jY
2β+ kY
3β. Then we set
hX
H, Y
Hi = hX
H, Y
Hi
h:=
n
X
α,β=1
(X
0α+ iX
1α+ jX
2α+ kX
3α)h
αβ(Y
0β+ iY
1β+ jY
2β+ kY
3β).
Taking into account (2.1) we can write Φ
∗h =
dq ∂Φ
α∂q + dq
1∂Φ
α∂q
1+ dq
2∂Φ
α∂q
2+ dq
3∂Φ
α∂q
3h
αβ× ∂Φ
β∂q dq + ∂Φ
β∂q
1dq
1+ ∂Φ
β∂q
2dq
2+ ∂Φ
β∂q
3dq
3. Finally, we get
Φ
∗h =hΦ
q¯, Φ
qi(dqdq + dq
1dq
1+ dq
2dq
2+ dq
3dq
3)
+ [dqhΦ
q¯, Φ
q1idq
1+ dq
1hΦ
q¯, Φ
q1i
1dq + dq
2hΦ
q¯, Φ
q1i
2dq
3+ dq
3hΦ
q¯, Φ
q1i
3dq
2]
+ [dqhΦ
q¯, Φ
q2idq
2+ dq
2hΦ
q¯, Φ
q2i
2dq + dq
1hΦ
q¯, Φ
q2i
1dq
3+ dq
3hΦ
q¯, Φ
q2i
3dq
1]
+ [dqhΦ
q¯, Φ
q3idq
3+ dq
3hΦ
q¯, Φ
q3i
3dq + dq
1hΦ
q¯, Φ
q3i
1dq
2+ dq
2hΦ
q¯, Φ
q3i
2dq
1].
II. Computation of e(Φ) · g. By the definition we have
e(Φ) =
12|dΦ|
2=
12g
ijh
αβΦ
αiΦ
βj=
12g
0H(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2).
Then
e(Φ) · g =
18(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2)(dqdq + dq
1dq
1+ dq
2dq
2+ dq
3dq
3)
= 2hΦ
q¯, Φ
qi(dqdq + dq
1dq
1+ dq
2dq
2+ dq
3dq
3).
Now, we can write an explicit expression for the stress-energy tensor of Φ:
S(Φ) = hΦ
q, Φ
qi(dqdq + dq
1dq
1+ dq
2dq
2+ dq
3dq
3) (3.2)
− [dqhΦ
q¯, Φ
q1idq
1+ dq
1hΦ
q¯, Φ
q1i
1dq + dq
2hΦ
q¯, Φ
q1i
2dq
3+ dq
3hΦ
q¯, Φ
q1i
3dq
2]
− [dqhΦ
q¯, Φ
q2idq
2+ dq
2hΦ
q¯, Φ
q2i
2dq + dq
1hΦ
q¯, Φ
q2i
1dq
3+ dq
3hΦ
q¯, Φ
q2i
3dq
1]
− [dqhΦ
q¯, Φ
q3idq
3+ dq
3hΦ
q¯, Φ
q3i
3dq + dq
1hΦ
q¯, Φ
q3i
1dq
2+ dq
2hΦ
q¯, Φ
q3i
2dq
1].
According to the above decomposition we can define four tensors S
0, S
1, S
2, S
3so that every square bracket [·] corresponds to one component of the decomposition of S(Φ) in these tensors: S(Φ) = S
0+ S
1+ S
2+ S
3.
Proposition 3.2. (Φ is conformal) ⇔ (S
1= S
2= S
3= 0).
P r o o f. Notice that the equations
hΦ
q¯, Φ
q1i = hΦ
q¯, Φ
q2i = hΦ
q¯, Φ
q3i = 0,
which express the vanishing of the components in S
1, S
2and S
3, are equiv- alent to the conditions
|Φ
0|
2= |Φ
1|
2= |Φ
2|
2= |Φ
3|
2,
hΦ
0, Φ
ii = 0, i = 1, 2, 3; hΦ
i, Φ
ji = 0, i 6= j, i, j 6= 0, which just express the conformality of Φ.
R e m a r k 3.1. (Φ is conformal) ⇔ (S(Φ) is pure of genus 0).
Corollary 3.1. If Φ is locally regular , i.e. Φ
q¯= 0, then S(Φ) = 0.
Corollary 3.2. S(Φ) = 0 if and only if Φ = const.
P r o o f. By (3.2), S(Φ) = 0 if and only if
hΦ
q¯, Φ
qi = hΦ
q¯, Φ
q1i = hΦ
q¯, Φ
q2i = hΦ
q¯, Φ
q3i = 0.
By straightforward calculations the above equalities are equivalent to
|Φ
0|
2= |Φ
1|
2= |Φ
2|
2= |Φ
3|
2= 0 and hΦ
i, Φ
ki = 0, i, k = 0, 1, 2, 3.
But this is possible if and only if Φ is a constant.
Definition 3.2. A 2-tensor S defined on an almost quaternionic manifold (M
4, g) with dim
RM
4= 4 and standard enhanced quaternionic structure Q
0is Hermitian if for any p ∈ M
4we have
S(I
αX, I
αY ) = S(X, Y ) for α = 1, 2, 3, where (I
1, I
2, I
3) ∈ Q
0|pand X, Y ∈ T
pM
4.
R e m a r k 3.2. (S is pure of genus 0) ⇔ (S is Hermitian).
R e m a r k 3.3. (Φ is conformal) ⇔ (S(Φ) is Hermitian).
Proposition 3.3. Let Φ : (M
4, g) → (N
4n, h) be a smooth mapping between two almost-quaternionic-Hermitian manifolds. Assume that (M
4, g) is locally conformally flat. If Φ is Q-holomorphic then S(Φ) is Hermitian.
P r o o f. By the definition of S(Φ) it is enough to show that Φ
∗h is Her- mitian on (M
4, g). Indeed,
Φ
∗h(I
αX, I
αY ) = h(dΦ(I
αX), dΦ(I
αY )) = h(I
α′(dΦ(X), I
α′(dΦ(Y )))
= h(dΦ(X), dΦ(Y )) = Φ
∗h(X, Y ).
Proposition 3.4. Let Φ : (M
4, g) → (N
4n, h) be a smooth mapping between almost quaternionic Hermitian manifolds. Assume that (M
4, g) is locally conformally flat. If Φ is Q-holomorphic then Φ is harmonic if and only if it is homothetic.
P r o o f. Note that
(3.3) e(Φ) = 2g
0HhΦ
q¯, Φ
qi.
If Φ is Q-holomorphic then Φ is conformal. By Proposition 3.2 and (3.2) we get
(3.4) S(Φ) = g
H0hΦ
q¯, Φ
qig.
On the other hand, conformality of Φ means that Φ
∗h = µg for some con- tinuous and non-negative function µ defined on M . By (3.3) and (3.4) we obtain µ = g
H0hΦ
q¯, Φ
qi. Thus
(div S(Φ) = 0) ⇔ [hd(g
0HhΦ
q¯, Φ
qi), gi = 0] ⇔ [d(g
0HhΦ
q¯, Φ
qi) = 0]
⇔ (dµ = 0) ⇔ (µ = const) ⇔ (Φ is homothetic).
If Φ is homothetic then Φ
∗h = µ
0g, where µ
0= const. So, we have S = (e − µ
0)g and div S = hde, gi. But on the other hand, e = 0, and so Φ is harmonic.
Recall that if S is a real 2-tensor on a (real) Riemannian manifold (M, g), then one defines the divergence of S (see e.g. [5]) in the local coordinates (x
i) by
(div
RS)
i= (div S)
i:= g
jk∇
∂jS
ki.
Definition 3.3. We define a quaternionic divergence of the quaternionic 2-tensor s
Hby
(div
Hs
H)
γ:= g
Hαβ∇
∂αs
βγ, where α, β, γ stand for q, q
1, q
2, q
3.
Proposition 3.5. If the metric g is locally conformally flat then
(3.5)
64 Re[(div
Hs
H)
q] = (div
RS)
0, 64 Re[i(div
Hs
H)
1] = (div
RS)
1, 64 Re[j(div
Hs
H)
2] = (div
RS)
2, 64 Re[k(div
Hs
H)
3] = (div
RS)
3, P r o o f. This follows by straightforward computations.
Let us quote a very important result of Eells [5]:
Theorem 3.1. Suppose that Φ : (M, g) → (N, h) is a smooth mapping between smooth Riemannian manifolds. If Φ is harmonic then S(Φ) is con- servative (i.e. div
RS(Φ) = 0). If Φ is a differentiable submersion almost everywhere and div
RS(Φ) = 0, then Φ is harmonic.
Now, we can state
Theorem 3.2. Let Φ : (M
4, g) → (N
4n, h) be a smooth map between smooth Riemannian manifolds. Assume that M is locally conformally flat.
If Φ is harmonic then Φ satisfies the following system of real equations:
(3.6)
1
32
∇
∂0(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
+ ∇
∂1hΦ
0, Φ
1i + ∇
∂2hΦ
0, Φ
2i + ∇
∂3hΦ
0, Φ
3i = 0,
1
32
∇
∂1(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
− ∇
∂0hΦ
0, Φ
1i + ∇
∂2hΦ
0, Φ
2i − ∇
∂3hΦ
0, Φ
3i = 0,
1
32
∇
∂2(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
− ∇
∂0hΦ
0, Φ
1i − ∇
∂1hΦ
0, Φ
2i + ∇
∂3hΦ
0, Φ
3i = 0,
1
32
∇
∂3(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
− ∇
∂0hΦ
0, Φ
1i + ∇
∂1hΦ
0, Φ
2i − ∇
∂2hΦ
0, Φ
3i = 0,
where Φ
i:= ∂Φ/∂x
iand h·, ·i means a real scalar product. Moreover , if Φ is a differentiable submersion almost everywhere and the system (3.6) is satisfied then Φ is harmonic.
P r o o f. By the result of Eells, Theorem 3.1, and by Proposition 3.5 the condition div
RS(Φ) = 0 is equivalent to the following system of quaternionic equations:
(3.7) Re[(div
Hs
H(Φ))
0] = 0, Re[i(div
Hs
H(Φ))
1] = 0, Re[j(div
Hs
H(Φ))
2] = 0, Re[k(div
Hs
H(Φ))
3] = 0, where s
H(Φ) denotes the quaternionic stress-energy tensor of Φ.
By the assumption, the metric g has the form (3.1). Hence, the only non-zero quaternionic components of g are g
00, g
11, g
22, g
33and they equal g
0H6= 0, which is real. Then we have
(3.8)
(div
Hs
H)
0= X
α
g
αα∇
∂αs
α0, (div
Hs
H)
γ= X
α
g
αα∇
∂αs
αγ, γ = 1, 2, 3.
Substituting the quaternionic expression s
Hfor the stress-energy tensor into (3.8) we get
(div
Hs
H)
0g
H0= ∇
∂/∂qhΦ
q¯, Φ
qi − ∇
∂/∂q1hΦ
q¯, Φ
q1i
1− ∇
∂/∂q2hΦ
q¯, Φ
q2i
2− ∇
∂/∂q3hΦ
q¯, Φ
q3i
3, (div
Hs
H)
1g
H0= − ∇
∂/∂qhΦ
q¯, Φ
q1i + ∇
∂/∂q1hΦ
q¯, Φ
qi
− ∇
∂/∂q2hΦ
q¯, Φ
q3i
2− ∇
∂/∂q3hΦ
q¯, Φ
q2i
3, (div
Hs
H)
2g
H0= − ∇
∂/∂qhΦ
q¯, Φ
q2i − ∇
∂/∂q1hΦ
q¯, Φ
q3i
1+ ∇
∂/∂q2hΦ
q¯, Φ
qi − ∇
∂/∂q3hΦ
q¯, Φ
q1i
3, (div
Hs
H)
3g
H0= − ∇
∂/∂qhΦ
q¯, Φ
q3i − ∇
∂/∂q1hΦ
q¯, Φ
q2i
1− ∇
∂/∂q2hΦ
q¯, Φ
q1i
2+ ∇
∂/∂q3hΦ
q¯, Φ
qi.
Since
∇
∂/∂qihΦ
q¯, Φ
qmi
i= [∇
∂/∂qhΦ
q¯, Φ
qmi]
ifor i, m = 0, 1, 2, 3 and
Re [∇
∂/∂qhΦ
q¯, Φ
qni]
n= Re ∇
∂/∂qhΦ
q¯, Φ
qni, n = 0, 1, 2, 3,
we see that the system (3.7) is equivalent to
(3.9)
Re[∇
∂/∂qhΦ
q¯, Φ
qi − ∇
∂/∂qhΦ
q¯, Φ
q1i
− ∇
∂/∂qhΦ
q¯, Φ
q2i − ∇
∂/∂qhΦ
q¯, Φ
q3i] = 0, Re i[∇
∂/∂qhΦ
q¯, Φ
q1i − ∇
∂/∂q1hΦ
q¯, Φ
qi
+ ∇
∂/∂qhΦ
q¯, Φ
q3i + ∇
∂/∂qhΦ
q¯, Φ
q2i] = 0, Re j[∇
∂/∂qhΦ
q¯, Φ
q2i + ∇
∂/∂qhΦ
q¯, Φ
q3i
− ∇
∂/∂q2[Φ
q¯, Φ
qi + ∇
∂/∂qhΦ
q¯, Φ
q1i] = 0, Re k[∇
∂/∂qhΦ
q¯, Φ
q3i + ∇
∂/∂qhΦ
q¯, Φ
q2i
+ ∇
∂/∂qhΦ
q¯, Φ
q1i − ∇
∂/∂q3hΦ
q¯, Φ
qi] = 0.
Now, note that
Re ∇
∂/∂qhΦ
q¯, Φ
qi =
641∇
∂0(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2), Re ∇
∂/∂qhΦ
q¯, Φ
q1i
=
641∇
∂0(|Φ
0|
2+ |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
+ 2∇
∂2[hΦ
0, Φ
2i − hΦ
1, Φ
3i] + 2∇
∂3[hΦ
0, Φ
3i + hΦ
1, Φ
2i], Re ∇
∂/∂qhΦ
q¯, Φ
q2i
=
641∇
∂0(|Φ
0|
2− |Φ
1|
2+ |Φ
2|
2− |Φ
3|
2)
+ 2∇
∂1[hΦ
0, Φ
1i + hΦ
2, Φ
3i] + 2∇
∂3[hΦ
0, Φ
3i − hΦ
1, Φ
2i], Re ∇
∂/∂qhΦ
q¯, Φ
q3i
=
641∇
∂0(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2+ |Φ
3|
2)
+ 2∇
∂1[hΦ
0, Φ
1i − hΦ
2, Φ
3i] + 2∇
∂2[hΦ
0, Φ
2i + hΦ
1, Φ
3i], Re i∇
∂/∂qhΦ
q¯, Φ
q1i
=
641∇
∂1(|Φ
0|
2+ |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
− 2∇
∂3[hΦ
0, Φ
2i − hΦ
1, Φ
3i] + 2∇
∂2[hΦ
0, Φ
3i + hΦ
1, Φ
2i], Re i∇
∂/∂qhΦ
q¯, Φ
q2i
=
641∇
∂1(|Φ
0|
2− |Φ
1|
2+ |Φ
2|
2− |Φ
3|
2)
− 2∇
∂0[hΦ
0, Φ
1i + hΦ
2, Φ
3i] + 2∇
∂2[hΦ
0, Φ
3i − hΦ
1, Φ
2i], Re i∇
∂/∂qhΦ
q¯, Φ
q3i
=
641∇
∂1(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2+ |Φ
3|
2)
− 2∇
∂0[hΦ
0, Φ
1i − hΦ
2, Φ
3i] − 2∇
∂3[hΦ
0, Φ
2i + hΦ
1, Φ
3i], Re j∇
∂/∂qhΦ
q¯, Φ
q1i
=
641∇
∂2(|Φ
0|
2+ |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
− 2∇
∂0[hΦ
0, Φ
2i − hΦ
1, Φ
3i] − 2∇
∂1[hΦ
0, Φ
3i + hΦ
1, Φ
2i],
Re j∇
∂/∂qhΦ
q¯, Φ
q2i
=
641∇
∂2(|Φ
0|
2− |Φ
1|
2+ |Φ
2|
2− |Φ
3|
2)
+ 2∇
∂3[hΦ
0, Φ
1i + hΦ
2, Φ
3i] − 2∇
∂1[hΦ
0, Φ
3i − hΦ
1, Φ
2i], Re j∇
∂/∂qhΦ
q¯, Φ
q3i
=
641∇
∂2(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2+ |Φ
3|
2)
+ 2∇
∂3[hΦ
0, Φ
1i − hΦ
2, Φ
3i] − 2∇
∂0[hΦ
0, Φ
2i + hΦ
1, Φ
3i], Re k∇
∂/∂qhΦ
q¯, Φ
q1i
=
641∇
∂3(|Φ
0|
2+ |Φ
1|
2− |Φ
2|
2− |Φ
3|
2)
+ 2∇
∂1[hΦ
0, Φ
2i − hΦ
1, Φ
3i] − 2∇
∂0[hΦ
0, Φ
3i + hΦ
1, Φ
2i], Re k∇
∂/∂qhΦ
q¯, Φ
q2i
=
641∇
∂3(|Φ
0|
2− |Φ
1|
2+ |Φ
2|
2− |Φ
3|
2)
− 2∇
∂2[hΦ
0, Φ
1i − hΦ
2, Φ
3i] − 2∇
∂0[hΦ
0, Φ
3i − hΦ
1, Φ
2i], Re k∇
∂/∂qhΦ
q¯, Φ
q3i
=
641∇
∂3(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2+ |Φ
3|
2)
− 2∇
∂2[hΦ
0, Φ
1i − hΦ
2, Φ
3i] + 2∇
∂1[hΦ
0, Φ
2i + hΦ
1, Φ
3i], Re i∇
∂/∂q1hΦ
q¯, Φ
qi =
641∇
∂1(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2),
Re j∇
∂/∂q2hΦ
q¯, Φ
qi =
641∇
∂2(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2), Re k∇
∂/∂q3hΦ
q¯, Φ
qi =
641∇
∂3(|Φ
0|
2+ |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2).
Now substituting the above expressions into (3.9) we get immediately (3.6), as required.
Definition 3.4. A function f : Ω → H is said to be antiregular in the sense of Fueter in a domain Ω ⊂ H if f is differentiable (in the usual sense) as a mapping of Ω, regarded as an open set in R
4, with values in R
4and
Df :=
14(∂
0− i∂
1− j∂
2− k∂
3)(f
0+ if
1+ jf
2+ kf
3) = 0 in Ω; here ∂
k:= ∂/∂x
k, k = 0, 1, 2, 3.
R e m a r k 3.4. Let F = F
0+ iF
1+ jF
2+ kF
3: Ω → H, where Ω is an open set in H, be a differentiable mapping (i.e. each component is differentiable as a mapping R
4→ R). Then DF = 0 if and only if
∂
0F
0+ ∂
1F
1+ ∂
2F
2+ ∂
3F
3= 0,
∂
1F
0− ∂
0F
1+ ∂
2F
3− ∂
3F
2= 0,
∂
2F
0− ∂
0F
2− ∂
1F
3+ ∂
3F
1= 0,
∂
3F
0− ∂
0F
3+ ∂
1F
2− ∂
2F
1= 0.
P r o o f. This follows by straightforward computations.
Let us recall that the properties of antiregular functions are analogous to those of regular functions in the sense of Fueter (see e.g. [6, 7, 13]).
Theorem 3.3. Let Φ : (M
4, g) → (N
4n, h) be a smooth map between smooth Riemannian manifolds. Assume that M is locally conformally flat.
If Φ is harmonic then the function
F
Φ= F
0+ iF
1+ jF
2+ kF
3:=
321(|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2) (3.10)
+ ihΦ
0, Φ
1i + jhΦ
0, Φ
2i + khΦ
0, Φ
3i is antiregular in the sense of Fueter. Moreover , if Φ is a differentiable sub- mersion almost everywhere and F
Φis antiregular in the sense of Fueter , then Φ is harmonic.
P r o o f. Note that the system (3.7) can be written in the following very suggestive and condensed form:
(3.11) ∇
DF
Φ= 0,
where D =
14(∂
0−i∂
1−j∂
2−k∂
3) and F
Φis the quaternionic-valued function defined by (3.10). Since F
Φis scalar, (3.11) is equivalent to D·F
Φ= 0, which proves the theorem.
The above result is rather unexpected and it emphasizes the importance of the class of regular functions in the sense of Fueter.
Theorem 3.4. On the torus T
4:= R
4/Λ, where Λ is a lattice, consider a global linear system of coordinates q = x
0+ ix
1+ jx
2+ kx
3. Then any harmonic map Φ : T
4→ (N
4n, h) satisfies
(3.12) |Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2= a,
hΦ
0, Φ
1i = b
1, hΦ
0, Φ
2i = b
2, hΦ
0, Φ
3i = b
3, for suitable constants a, b
1, b
2, b
3∈ R.
P r o o f. If Φ is harmonic then F
Φis antiregular in the sense of Fueter.
Any antiregular function satisfies the maximum principle. Since the torus T
4is compact, F
Φhas to be constant. Then, by the definition of F
Φ, we get (3.12), as required.
Corollary 3.3. If Φ : (T
4, g) → (N
4n, h) is harmonic and non-constant then either Φ
0or (Φ
1, Φ
2, Φ
3) has no zero on T
4.
P r o o f. Indeed, otherwise at a point p
1∈ T
4where Φ
0(p
1) = 0, by Theorem 3.4, we would have
−|Φ
1|
2− |Φ
2|
2− |Φ
3|
2= a ≤ 0 and b
1= b
2= b
3= 0,
and at a point p
2∈ T
4where (Φ
1(p
2), Φ
2(p
2), Φ
3(p
2)) = 0 we would get
|Φ
0|
2= a ≥ 0, b
1= b
2= b
3= 0.
Then we would have
a = b
1= b
2= b
3= 0.
This means that
(3.13) |Φ
0|
2= |Φ
1|
2+ |Φ
2|
2+ |Φ
3|
2, hΦ
0, Φ
1i = hΦ
0, Φ
2i = hΦ
0, Φ
3i = 0 at all points of T
4, with Φ
i:= ∂Φ/∂x
i, i = 0, 1, 2, 3.
But any smooth map Φ : (T
4, g) → (N
4n, h) satisfying (3.13) has to be constant. Indeed, (3.13) must hold with x
0replaced by any one of the variables x
1, x
2, x
3. Thus
hΦ
i, Φ
ki = 0, i 6= k, i, k = 0, 1, 2, 3, and
(3.14)
|Φ
0|
2− |Φ
1|
2− |Φ
2|
2− |Φ
3|
2= 0,
|Φ
1|
2− |Φ
0|
2− |Φ
2|
2− |Φ
3|
2= 0,
|Φ
2|
2− |Φ
0|
2− |Φ
1|
2− |Φ
3|
2= 0,
|Φ
3|
2− |Φ
0|
2− |Φ
1|
2− |Φ
2|
2= 0.
Note that the determinant of the system (3.14) is
det
1 −1 −1 −1
−1 1 −1 −1
−1 −1 1 −1
−1 −1 −1 1
6= 0.
Thus, Φ would be constant, which contradicts our assumption.
Let S
4denote the 4-dimensional sphere.
Corollary 3.4. If Φ : (T
4, g) → (S
4, h) is a C
∞-map such that (Φ
0, Φ
1, Φ
2, Φ
3) 6= 0 on T
4then deg(Φ) = 0.
P r o o f. The tangent bundle T S
4is of rank 4, so we can complexify it.
Denote by T S
C4the complexification of T S
4. By the definition the bundle Φ
−1(T S
C4) is also a complex vector bundle (now of rank 2).
Denote by c
2(Φ
−1(T S
C4)) the second Chern class of Φ
−1(T S
C4). Since the bundle Φ
−1(T S
C4) is trivial we have
c
2(Φ
−1(T S
C4)) = 0.
On the other hand (see [4, 11]),
c
2(T S
C4) = 3
4π
2v
Sg4,
where v
gis the volume form of S
4. Then we get 0 =
\
T4
c
2(Φ
−1T S
C4) =
\
T4
Φ
∗[c
2(T S
C4)] = 3 4π
2\
T4
Φ
∗(v
Sg4)
= 3
4π
2deg(Φ) Vol(S
4) = 2 deg(Φ)
because by the definition of the topological degree we have
\
T4