Complez Variables

, 1995

Vol. 27, pp. 17{46

Quaternionic -hyperholomorphic functions,

singular integral operators and boundary
value problems. I. -hyperholomorphic
function theory
M.V. Shapiro

Departmento de Mathematicas, ESFM del I.P.N., Mexico City, Mexico

N.L. Vasilevski

Departmento de Mathematicas, CINVESTAV del I.P.N., Mexico City, Mexico

Quaternion-valued functions of 3 and 4 real variables are studied which form exact analogues of the class of all
holomorphic (or anti-holomorphic) functions and which are called hyperholomorphic. The set of such analogues
is parametrizable by the group O3 (R ) or O3+ (R ) depending on the number of variables. Connections between
various classes of hyperholomorphic functions are established, and some basic facts of the function theory are
given. Properties of the singular integral operator with the quaternionic Cauchy kernel are treated and commented
in detail.

Among various ways to extend one-dimensional complex analysis to the multidimensional, in our
opinion Cli ord analysis (i.e. a study of functions with values in a Cli ord algebra) is the most important and adequate on. The exact structural analogy between Cli ord analysis and one-dimensional
complex analysis together with successful applications to a number of problems in physics have caused an
intensi ed development of Cli ord analysis. A series of fundamental and (in certain cases) nal results
were obtained (see, e.g. BDS], DSS], GS]).
Some classical questions in complex one-dimensional theory { in particular the theory of boundary
value problems and of singular integral operator { have not been developed su ciently for Cli ord
analysis.
This paper is devoted to a systematic study of such questions.
The skew{ eld of quaternions H gives an example of a non-commutative Cli ord algebra with minimal
dimension. It serves as a very convenient model of general Cli ord constructions. For this reason H valued functions are considered below, although many assertions allow quite evident generalizations for
Cli ord algebra-valued functions.
We study H -valued functions of three and of four real variables. We de ne the function classes
(parametrizable by the group O3 (R ) or by O3+(R ) depending on the number of variables) which are
exact analogues of the class of holomorphic (or anti-holomorphic) functions. We clarify the connections
among these classes as well as the corresponding analogues of the Cauchy integral{type integral and of
the singular integration operator with Cauchy kernel.
It should be mentioned that we do not attempt to give a complete review of papers on this theme.
Some bibliographical remarks will be made at appropriate places.
1

For a reader who is already acquainted with quaternionic or Cli ord analysis, most of the results
formulated in Sections 2 { 3 found familiar. However, for the sake of convenience and of completeness,we
have listed them; in most cases their proofs are omitted.

1 Some basic notations and de nitions

1.1

R 4y

The following notations will be used.

{ the four-dimensional Euclidean space with a xed orthonormalized (canonical) basis
st

:= fi0 ; i1 ; i2 ; i3 g :

{ the non-commutative skew{ eld of real quaternions which is built from R 4y in such a way that i0 is its
unity and i1 ; i2 ; i3 are the imaginary units. If y =

3
X

k=0

yk ik is a quaternion then ~y = (y0 ; y1 ; y2 ; y3 )

is the corresponding vector from R 4y .

H (C ) := H R C { the corresponding algebra of complex quaternions.
H m := H
: : : H (m times, m 2 f3; 4g):
Rm
x { the m-dimensional Euclidean space with a xed orthonormal basis which does not necessarily
coincide with st (when m = 4) or with its vector part (when m = 3).
{ a connected domain in R mx ; its topological boundary @ =: is supposed to be an (m 1)dimensional compact smooth (or Liapounov) surface.
B m (x0 ; R ){closed ball in R m
x centered at a point x0 and with radius R > 0.
Sm 1 := @ B (0; 1) { the unit sphere in R m
x.
int X { the interior of a set X R mx .
h ; im = h ; iR m { the scalar product in R m .
0 X { the function x 2 X 7 ! 0 2 R .
1 X { the function x 2 X 7 ! 1 2 R .
jk { the Kronecker symbol (Kronecker delta).
R p q (respectively C p q ){the set of all p q -matrices with real (respectively complex) components.
Ep { the unit p p -matrix.
3
X

Let f = fk ik : ! H , and f~ = (f0 ; f1; f2 ; f3) is the corresponding R 4 -valued function. For
k=0
p 2 N f0g we de ne as usual
1.2

C p( ; R 4 ) := f~j fk 2 C p( ; R ) ;
n
o
C p( ; H ) := f j f~ 2 C p( ; R 4 ) :
2

For an H -valued function a we denote M a (respectively aM ) the right (respectively left) multiplication
operator by a. We shall use the operator of the quaternionic conjugation Z : f 7 ! f and the partial
di erentiation operator @k = @xk = @x@ k :
Let := f 4 m ; : : : ; 3 g 2 H m ;
D by the formulas
1.3

4 m

:= f

; : : : ; 3 g: We de ne on C 1 ( ; H ) the operators D and

0 3
1
3
X
X
k
k @ f;
D f ] := @
M @k A f ] =
k
k=4 m
k=4 m
0 3
1
3
X
X
k
@
A
@k f k :
D f ] :=
M @k f ] =
k=4 m

k=4 m

Let us put the question: what are the conditions on for the equalities

D D= D D=D D =D D =
to be true, where

f ] :=

3
X

k=4 m

fk ] ik ,

that (1.1) is valid if and only if

(1.1)

being the m-dimensional Laplace operator? It turns out

j k+ k

= 2 jk
(1.2)
for any j; k. The su ciency of (1.2) has been proved in VS1] and N1], moreover in N1] necessity has
been proved also. The equalities (1.2) are equivalent to

h ~|; ~k iR 4 = jk ;

(1.3)

which means the orthonormality of the set ~ . Any set of quaternions with the property (1.2) we shall
call \structural set".
It is evident that and are structural sets simultaneously .
If is a structural set then we shall call D and D respectively the left and the right-CauchyRiemann operator (C.- R.O.). We have

D = ZD Z; D = Z DZ;

(1.4)

so it is su cient to study the \left" (or the \right") case only.

We have also
3
D f g] = D f ] g + P k f @k g;

D f g] =

k=4 m
k+f

P3 @ f g
k

k=4 m

D g]:

(1.5)

It follows from (1.5) in particular that the left-(respectively right) C.-R.O. is linear in the right (respectively left) H -module C 1 ( ; H ).

2 Hyperholomorphic functions
2.1

Let us x a structural set and a domain . We introduce the sets

M( ) := ker D = nf 2 C 1 ( ; H )j D f ] = 0 o;
M ( ) := ker D = f 2 C 1 ( ; H )j D f ] = 0 :

(2.1)

We shall call elements of these sets left-(respectively, right-) - hyperholomorphic in a domain H -valued
functions of m- dimensional variable, shortering this name if misunderstanding is excluded. The sets (2.1)
will be called brie y the classes of the hyperholomorphy sometimes. Any class of left-(respectively, right)-hyperholomorphic functions is a right (respectively, left) H -submodule of the H -module C 1( ; H ). The
equalities (1.1) mean that the coordinate functions of hyperholomorphic functions of class C 2 ( ; H ) are
necessarily harmonic functions in .
2.2

Let f0 :

! R be a harmonic function of m variables and put g(x) := D f0 ](x) =

3
X

k=4 m

@k f0 :

Then D g] = D D f0 ] = f0 ] = 0 ; similarly D g] = 0 . We have shown that the classes of

hyperholomorphy are wide enough and that

M ( ) := M( ) \ M ( ) 6= ;:
It is clear that various structural sets can generate the same class of hyperholomorphy as in shown by
trivial example M( ) = M( ). This theme will be discussed in detail in Section 4.
2.3

Let f be a complex-valued function of a complex variable z = x + iy. Then as it is well known

@f dz;
dz
+
df = @f
@z
@z

(2.2)

where df is the di erential of f . This equality is one of the starting points to introduce the notion of a
holomorphic function. Now we will show that analogously one can introduce the notion of hyperholomorphic function for m = 4.
First we introduce the H -valued di erential forms
(2)

;x

:= sgn

1 dx2

^ dx3

2 dx1

^ dx3 + 3 dx1 ^ dx2 ;

(2.3)

sgn (dx1 ^ dx2 ^ dx3 1 dx0 ^ dx2 ^ dx3 +

(2.4)
+ 2 dx0 ^ dx1 ^ dx3 3 dx0 ^ dx1 ^ dx2 );
where 1 ; 2 ; 3 are purely imaginary quaternions, sgn := 1 or sgn := 1 depending on the fact:
the bases f ~1 ; ~2 ; ~3 g and fi~1 ; i~2 ; i~3 g in R 3 have the same orientation or not. Then one can check the
equality
d (2);xf = (3);x D f ] + (3);x D f ];
(2.5)
(3)

y;x

:=

where f 2 C 1 ( ; H ) and d is the exterior di erentiation operator of a di erential form. The equality
(2.5) is an exact analog of (2.2).
Further let us denote by 0f (x0 ) { call the left- -derivative of f at a point x0 { the quaternion satisfying
the equality
(2.6)
d( (2);x f ) = (3);x 0f (x0 ):
4

Thus the function f is left- -hyperholomorphic if it has a left- -derivative in and in this case
0f (x) =

D f ](x):

(2.7)

Apparently for the rst time -hyperholomorphic functions (with = (i0 ; i1 ; i2 ; i3 ) for m = 4 and
= (i1 ; i2 ; i3 ) for m = 3) were investigated by Gr. C. Moisil M1], M2] (see M3, pp. 23{25, 80-85]
also), who used a quaternionic language only in the four-dimensional case (m = 4) ( M2], while for
m = 3 he used a matrix one M3]). R. Fueter in F1], F2] and in other papers and V.S. Vinogradov
Vi1] complemented essentially the results from M2]. Note the papers Su], D], Sp1], Sp2] also. An
extension of these ideas for Cli ord algebra-valued functions can be found in BDS], DSS], R1], GM],
R2], Sp3], L] for example. V. Iftimie I] investigated the case = f"0 ip0 ; : : : ; "m ipm g, where "0 ; : : : ; "m
are real numbers.
For H -valued functions a general de nition of the class M was introduced independently in VS1]
and N1] (see also N2] and VS2], VS3]), for a corresponding extension to Cli ord analysis see Sh1],
Sh2].
2.4

3 Quaternionic Cauchy integral-type integrals (C.i.t.i) and some of

their properties
Hyperholomorphic functions have the exact analogues of the main theorems of one-dimensional
complex analysis. Now we give some of them which we need for further purposes.
If (m;x 1) is de ned by (2.3) or (2.4) and if we denote as usual by dxb(k) the di erential form dx4 m ^
: : : ^ dx3 with the multiple dxk omited then
3.1

(m 1)

;x

3
X

= sgn

k=4 m

( 1)k k dxbk :

(3.1)

If dsm 1 is the di erential form of the (m 1)-dimensional volume in R m then we have

(m 1)

;x

j = dsm 1 :

This fact allows us to call (m;x 1) the H -valued di erential form of the (m 1)-dimensional volume in
R m corresponding to .
Let ff; gg C 1 ( ; H ). Then direct computation leads to
d(g (m;x 1) f ) =
(3.2)
= ( 1)m 1 sgn

D g] f + g D f ] dx;

where dx is the di erential form of the m-dimensional volume in R m .

The equality (3.2) is the base for a series of essential constructions further on. Thus the general
Stokes formula gives an integral form of (3.2):

g(x)

(m 1)

;x

f (x) =

= ( 1)m 1 sgn

(3.3)

D g](x) f (x) + g(x) D f ](x) dx:

One may call it the quaternionic Stokes' formula. The general (that is for two functions) Cauchy
integral theorem follows in turn from (3.3): if f is a left- -hyperholomorphic in and g is right- hyperholomorphic in then
Z
(3.4)
g(x) (m;x 1) f (x) = 0:
@

3.2

Now we introduce the Cauchy kernel K generated by a structural set . Let

K (x) := (m 2)1jSm 1 j D m](x) =
3
X
1
k k
x;
=

jSm 1 j jxjm k=4

where

: x 2 R m nf0g 7 ! jxjm1 2 ;
(
= 3;
m
1
jS j = 42 ;2 ; ifif m
m = 4;
m

is the (m 1)-dimensional volume of the unit sphere Sm 1 in R m . Then obviously:

1) K 2 C 1 (R m nf0g; H )
2) K 2 M (R m nf0g)
3) Let z 2 R m ; and let K ;z (x) := K (z x): Then

K ;z 2 M (R m nfzg):
The second property is a mere corollary of the harmonicity of m .
3.3 Let x0 be xed and let " > 0 be small and let
" := nB m (x0 ; "). Using formula (3.3) for " and
the Cauchy kernel properties we obtain by letting " ! 0 the analog of the well-known Borel-Pompeiu
formula : if ff; gg C 1 ( ; H ), then
( 1)m 1 sgn

K(

x)

(m 1)

f ( ) + g( )

(m 1)

K(

x)

K (y x) D f ](y) + D g](y) K (y x) dy =
=

(3.5)

f (x) + g(x); if x 2 ;
0;
if x 2 R m n :
6

If g 2 M ( ) and f 2 M( ) then (3.5) gives us the Cauchy integral formula for a function which is a
sum of a left and a right- -hyperholomorphic functions. Assuming additionally that f = 0 or g = 0 we
obtain the Cauchy integral formula for a right and for a left- -hyperholomorphic functions respectively:

and

1)m 1 sgn

1)m 1 sgn

Z
@

g( )

(m 1)

K(

x) = g(0x;); x 2x R2m n; ;

K(
(m 1)

x)

f ( ) = f (0x; ); x 2x R2mn; :

(3.6)
(3.7)

One can consider the upper lines of these formulas as the representations of a right and of a left- hyperholomorphic in functions by their respective right and left Cauchy integrals.
If g = f = 1 in (3.6) and (3.7) then
( 1)m 1 sgn

=(
3.4

1)m 1 sgn

(m 1)

K (" x)

K(

Now let f 2 C (@ ; H ). Then for x 62 @ we de ne

(m 1)

=
(3.8)

x) = 10;; x 2x R2mn; :

x 7 ! K f ](x) := ( 1)m 1 sgn K (

@R
x 7 ! K f ](x) := ( 1)m 1 sgn f ( )
@

(m

(m 1)

f ( );
1)
K ( x):

x)

(3.9)

Those functions will be called respectively the left and the right Cauchy integral-type integral (C.i.t.i.)
of the function f and the maps

K : f 7 ! K f ]; K : f 7 ! K f ]
are called C.i.t.i. operators.
When f is a restriction to @ of a function from M( ) then of course the left C.i.t.i of f turns into
its left Cauchy integral. Analogously for the right case.
It follows immediately from the Leibnitz rule that both C.i.t.i. of f belong to C 1 (R m n@ ; H ) and
that
Z
Dx K f ]](x) = ( 1)m 1 sgn
Dx K ;x ]( ) (m; 1) f ( ) =
= ( 1)m 1 sgn

D K ](

x)

(m 1)

f ( ) = 0;

analogously D K f ]](x) = 0. Consequently the left (the right) C.i.t.i. of a continuous function f is a
left-(right-)- -hyperholomorphic function in R m n@ :
7

3.5 Now we proceed to study boundary properties of the C.i.t.i. of a xed function f . As usual it is
convenient to start from the case of a Holderian density f . Let for 0 <
1;

H ( ; H ) := ff j jf (t1) f (t2 )j Lf jt1 t2 j ; 8ft1 t2 g

Lf = constg;

and let f 2 H ( ; H ): We show that the integral K f ](t) is de ned in the sense of the principal value
for t 2 . To simplify the statement we assume m = 3 and sgn = 1. For " > 0 we put
:= \ B 3 (t; "); +" := +ZnB 3 (t; ");
+
t) (2); f ( ):
" := @ " nB 3 (t; "); At (") := K (
"

"

We have: At (") = K (

(2)

t)

"

(f ( ) f (t))+

0
Z
+B
@ K(
"

Let us transform N :

0
N =@

R
3
" (@ B (t;")n + )

1
t) (2); C
A f (t) =: M + N:

1
0
t) (2); A f (t) @

@ B 3 (t;")n +

1
t) (2); A f (t) =: N1 + N2 :

According to (3.8) for any " > 0 N1 = f (t): Changing the variables in N2 to spherical coordinates we
obtain
Z 1
2
N2 =
4 "2 w " sin 1 w d 1 ^ d 2 f (t) =
"
Z
1
= 4 sin 1 d 1 ^ d 2 f (t);
"

where
:= f( 1 ; 2 )j 1 2 0; 2 ]; 2 2 0; '( 1 ; ")]g ;
w := 1 sin 1 cos 2 + 2 sin 1 sin 2 + 3 cos 1 :
"

It follows from the smoothness of the surface that with a decreasing of " > 0 part @ B 3 (t; ")n of the
sphere @ B 3 (t; ") will di er from hemisphere less and less. This means
when " ! 0. Hence

'( 1 ; ") ! 2 ;

"!

:= 0; 2 ]

0; 2 ]

1 Z sin d ^ d f (t) = 1 f (t):

=
lim
N
1 1
2
"!0 2 4
2
8

Thus N = 21 f (t). As for M , from the estimate

jK (

t)

(2)

const j (2) j j
j tj2 ;
const ds

(f ( ) f (t))j

j tj2

tj

we conclude that the integral M exists as an improper integral. Hence "lim

A (") exists.
!0 t
Thus we have nally: if t 2 then
R
K f ](t) := 12 f (t)+( 1)m 1 sgn K ( t) (m; 1) (f ( ) f (t));
R
K f ](t) := 21 f (t)+( 1)m 1 sgn (f ( ) f (t)) (m; 1) K ( t):
It is clear that (3.10) is true also for a non-connected surface . Moreover
boundary , but t 62 then.
Note also that if f = 1 then it follows from (3.10) that

( 1)m 1 sgn

= ( 1)m sgn

K(

t)

(m 1)

(3.10)

may be a surface with a

=
(3.11)

(m 1)

t) = 21 :

K(

In addition to the operators K and K de ned in (3.11) it is convenient to consider the operators

S := 2 K ; S := 2 K :

(3.12)

We call them, like in the complex case, the operators of (respectively left or right) singular integration
with the quaternionic Cauchy kernel along a surface .
It was established earlier that C.i.t.i. of a Holderian function is de ned in the whole R mx . Let us
consider now the behavior of the functions K f ] and K f ] when x ! t; x 2 =: + or x 2 :=
R mn ; t 2 .
We have
Z
K f ](x) = ( 1)m 1 sgn K ( x) (m; 1) (f ( ) f (t)) +

3.6

+(

1)m 1 sgn

The second term here equals f (t) if x 2

8
< f (t)+(
=:
( 1)m

K(

x)

(m 1)

and equals zero if x 2

f (t):
so that

K f ](x) =

1)m 1 sgn
1 sgn

RK

x) (m;

x) (m;

1) (f ( )

1) (f ( )

f (t)); x2 +;

f (t)); x2 :

(3.13)

These are analogous formulas for K f ](x) as well. The integrals B (x) and B (x) in both formulas exist
also in the improper sense when x = t. Let us denote the corresponding values B (t) and B (t). One
can show (using arguments similar to the ones in 3.5) that lim B (x) =: B (t) and lim B (x) =:
3x!t
3x!t
B (t). Moreover
B (t) = 12 f (t) + K f ](t);
(3.14)
B (t) = 21 f (t) + K f ](t):
This means in particular that f B; B g C (R m ; H ). Finally combining (3.13) and (3.14) we obtain the
Sokhotski-Plemelj formulas
lim K f ](x) =: K f ] (t) = 21 f (t) + K f ](t);
3x!t
(3.15)
lim K f ](x) =: K f ] (t) = 21 f (t) + K f ](t);
3x!t

where K f ](t) and K f ](t) were de ned by (3.14).

There are very useful equivalent formulas:
K f ] (t) = 12 f (t) + 21 S f ](t);
(3.16)
K f ] (t) = 12 f (t) + 21 S f ](t):
3.7 In 3.4 wee supposed that f 2 C (@ ; H ), hence the C.i.t.i. of f could be understood as the Riemann
integral. Now let f 2 Lp (@ ; H ). Then the C.i.t.i. of f has to be understood as the Lebesgue integral
preserving the results from 3.4{3.6 with the corresponding changes. For instance the limit values of
K f ] and K f ] exist almost everywhere (with respect to Lebesgue surface measure) on @ , and the
Sokhotski-Plemelj formulae are true almost everywhere only. Note that the questions related to these
discussed above have been examined in B], Vi2] for instance.
The Sokhotski-Plemelj formulae allow (just as in the one-dimensional complex analysis) to solve
some simple but essential boundary value problems. Let f 2 H (@ ; H ) be given; than we would like to
nd functions
1 2 M( ) \ H ( );
2 2 M ( ) \ H ( );
such that boundary values statisfy
++ +
(3.17)
1
2
1
2 = f:
The content of the previous pages yields that one of the solutions to this problem is given by the C.i.t.i.
of f :
1
1
(3.18)
1 = 2 K f ]; 2 = 2 K f ]:
Actually, the formulae (3.18) give the solution of such two \jump problems" for left-and for right- hyperholomorphic functions separately:
+
(3.19)
1
1 = f;
+
(3.20)
2
2 = f:
One can show that the Liouville theorem is true for hyperholomorphic functions. It thus follows that if
we are interested in solutions of (3.19) and (3.20) vanishing at the in nity then the solution (3.18) is the
only one.

3.8

10

Let fa; b; c; dg

and consider boundary problems with the next boundary conditions:

+ a+b ++
1
2
1 c+d
2 = f;
+ a+
1
1 c = f;
+
b 2 + d 2 = f:
Obviously one can directly reduce these problems to (3.17), (3.19), (3.20).
H

Let be a smooth surface (without boundary) in R m bounding a nite domain

domain . Let f 2 H ( ; H ). Next assertions follow directly from 3.6.

3.9

Assertion A. In order that a function f be the boundary value (that is a restriction on

fe from

and an in nite
) of a function

M( + ) \ H ( + ; H )

the following necessary and su cient conditions should be satis ed:

1 f (t) + K f ](t) = 0
2

(3.21)

8t 2

Assertion B. In order that a function f be the boundary value of a function fe from

M( ) \ H ( ; H ); fe(1) = 0;

the following necessary and su cient conditions should be satis ed:

1 f (t) + K f ](t) = 0
2

(3.22)
8t 2
To formulate corresponding assertions for f 2 Lp ( ; H ) the following de nition is useful: if X is any
space consisting of H -valued functions de ned on then A( ; X ) denote the classes of functions fe

such that:
1. fe 2 M( )
2. when 3 x ! t 2 there exist

lim fe (x) =: fe (t)

(along, non-tangential paths at any rate) everywhere or almost everywhere on generating function
from X .
3. fe is represented by a C.i.t.i. with a density from X .
If X = Lp( ; H ) then we shall write Ap( ).

Assertion C. In order that the values of function f 2 Lp( ; H ) coincide almost everywhere on

with

limit values of fe 2 Ap( + ) the following necessary and su cient condition should be realized almost

everywhere on :

1 f (t) + K f ](t) = 0:
2
It is clear how one can reformulate Assertion C for the case
case.

and the right{ {hyperholomorphic

Let us make here several remarks on Lp( ; H ); p > 1. First of all there are two equivalent de nitions
of Lp as a set:

3.10

11

8
9
Z
3
< X
=
1. Lp ( ; H ) := :f = fk ik j jf (t)jp dsm 1 (t) < +1; :
k=0
)
( X
3
k
k
2. Lp ( ; H ) := f = f ik j f 2 Lp ( ; R ); k 2 f0; 1; 2; 3g :
k=0

It is clear that one can introduce a structure of both left and right H -modulus in Lp ( ; H ). We shall
make more porecise what we mean in concrete situations. In both cases Lp ( ; H ) is a Banach module.
Let Lp provided with a right structure. Then classes Ap( ) introduced in 3.9 generate sub-modules
of the H -module Lp ( ; H ). One can show that for + = B m (0; 1) the de nition of the class Ap( + )
is a natural hyperholomorphic analog of the complex Hardy space Hp (B m ) de nition. That is why we
shall denote it as Hp (B m ; H ). Analogously for Hp (B m ): As usual we identify an element of Hp(B m ; H )
with a function from Lp ( ; H ) generated by its limit values and so Hp(B m ; H ) becomes a sub-module of
Lp( ; H ) just in this sense.
3.11 In the sequel we shall frequently use the notion of a right (left) Hilbert H {module. For convenience
we only recall some properties of such modules and we refer the reader to Ist] for forther information,
even in more general cases.
The inner product on a Hilbert H { module H will be denoted by h ; i; for a (non-empty) subset H0
of H , its orthogonal complement is denoted by H0?.
The Riesz theorem tells us that a continuous linear functional in a right (respectively left) Hilbert
H -module H acts by the rule (a) = hg ; ai (respectively (a) = ha; g i) for 8a 2 H where g 2 H is
uniquely de ned and j j = jg j. Correspondingly an antilinear functional in a right (or left) Hilbert
H -module H acts by the rule (a) = ha; g i (or (a) = hg ; ai) for 8a 2 H where g is the same as above.
The set of all continuous linear functions in a right (respectively left) Hilbert H -module H is a left
(respectively right) Hilbert H -module H with the scalar product

h ; iH := hg ; g iH :
It also follows from the Riesz theorem that for any right (respectively left) Hilbert H -module H there
exists an antilinear invertible operator B : H ! H with the property

hx; yiH = hBx; ByiH ;

which means that B is an isometric bijection and so B is an isomorphism in this sense.
Let H be a right or a left Hilbert H -module, and let S be a linear operator in H . Then the adjoint
operator S acts in a (left or right correspondingly) H -module H by the rule

S ](x) := (S x]) 8 2 H ; 8x 2 H:
By the Riesz theorem the last equality takes the form: for the right case

hB 1S ]; xi = hB
and for the left case

]; S x]i;

hx; B 1 S ]i = hS x]; B

]i:
Let H be a right H -module. An operator S is called self-adjoint if

hB

]; xi;

]; S x]i = hSB
12

that is SB 1 = B 1 S and this means

=S:
The same holds for the left case. The notion of a self-adjoint operator can be de ned in another but
equivalent way. Let H be a right H -module, denote

B
then

BSB

] =: g 2 H; B 1 S ] =: gS

2 H;

hg ; S x]i = hgS ]; xi:

Consequently a (left) linear operator S generates a map

S 0 : g 2 H ! gS

2H

which is linear (that is right-linear) in H . The operators S 0 and S are connected by the evident equality

S = BS 0 B 1 :
It is clear that S is self-adjoint in H in the sense of the above de nition if and only if

S 0 = S:
3.12

Let us consider L2 ( ; H ) as a right H {module. Then

Z
hf; gi := f (t) g(t) dsm

de nes a scalar product on it turning L2 ( ; H ) into a right Hilbert H -module. In the left L2 ( ; H ) a
scalar product is given by
Z
hf; gi := f (t) g(t) dsm 1 :
The hyperholomorphic Hardy spaces H2 ( ; H ) and H2 ( ; H ) introduced in 3.10 form Hilbert submodules
in the corresponding L2 ( ; H ).
Let us note that if A : L2 ( ; H ) ! L2 ( ; H ) is a self-adjoint operator in the sense of 3.11 then we
have
Z
Z
A f ](t) g(t) dsm 1 = f (t) A g](t)dsm 1
or

A f ](t) g(t) dsm

f (t) A g](t) dsm 1 :

4 Dependence of some main objects on a structural set

Now we return to the discussion concerning the connections between the classes of hyperholomorphy
and the structural sets generating them.
Let be a structural set and let h 2 H ; jhj = 1. As the multiplication operator (both from the
left and from the right) by h is an orthogonal transformation of R 4 the sets y := h and ye := h
4.1

13

are structural also. In this case we shall call y and

equivalent.
4.2

We have

Thus
In exactly the same way, from

3
X

@k ; D h = M h D

(4.1)

M( ) = h M( ); M ( ) = M h ( ):

(4.2)

D=

k=4 m

(respectively ye and ) left-(respectively right-)

Dh = D M h ;

hD =

D hM

(4.3)

it follows that

M ( ) = Mh ( )h; M( ) = h hM( ):
(4.4)
Thus a class of hyperholomorphy is not changed under:
1. multiplication of left-(respectively right-) -hyperholomorphic functions from the right (respectively
from the left) by constants from H ,
2. replacing a structural set by an equivalent one (see (4.2)),
3. simultaneous multiplication \from the wrong side" by h 2 S3 H of all functions of a class and of
a structural set (see (4.4)).
We note also that from (1.4) we have
M( ) = Z (M ( )); M ( ) = Z ( M( )):
Now let a 2 H nf0g. Since all classes of hyperholomorphy are R -linear spaces, for h := ja1j a we have:
if f 2 M( ) then af 62 M( ) in general but af = jaj hf 2 hM( ): So by multiplying a left- hyperholomorphic function by quaternion from the left we get a left-hyperholomorphic function again
but from another class generally saying. The same holds in the right case.
We shall describe now how an a ne change of variables (with an orthogonal matrix) a ects on a C.R.O. with xed . This will allow us to point out the connection between arbitrary -hyperholomorphy
and the standard hyperholomorphy.
Let
Rm
x ; and let x = (y) := a y + b where
4.3

a = (akq ) 2 Om (R ); b 2 R m; y 2

1(

) =: :

Let us denote by W the operator of the change of variable: W f ](y) = g = (f )(y). As one can easily
see we have
0 3
1
3
X
X
k a A @ q g(y):
@
(W Dx ) f ](y) =
y
qk
q=4 m k=4 m

It turns out that the set ' := f'4 m ; : : : ; '3 g, with

'q :=

3
X

k=4 m

14

aqk ;

(4.5)

is structural. Consequently the C.-R.O. Dx turns into C.-R.O. 'Dy :

W Dx = 'Dy W
(4.6)
and
f 2 M( ) () f 2 ' M( ):
(4.7)
4.4 As was established in 4.1 equivalent structural sets de ne the same class of hyperholomorphy. Later
on we shall speak sometimes about \structural 3-tuples (4-tuples)" instead of \structural sets" when we
must distinguish between m = 3 and m = 4. Firstly suppose m = 3. We show that among all equivalent
structural 3-tuples there is one which, in the space R 4y corresponding to H , lies in the hyperplane

e 3 := n(0; y1 ; y2 ; y3)j (y1 ; y2 ; y3) 2 R 3o :

Ry

Really, for any xed structural 3-tuple there are exactly two quaternions 0 such that sets of vectors
f ~0 ; ~1 ; ~2 ; ~3 g form orthonormalized bases in R 4y . Let 0 be chosen in such a way that the basis
f ~0 ; ~1 ~2 ; ~3 g is co-oriented with the canonical basis in R 4y and denote h := 0. Then y0 := h 0 = i0 ,
whence a 3-tuple y := h which is left equivalent to a 3-tuple lies in the hyperplane Re 3y :

Analogously if := h then a 3-tuple which is right equivalent to lies in Re 3y .

It then follows that for m = 3 a class of hyperholomorphy is de ned in fact by a certain orthonormal
basis in R 3 (but not by three orthonormal vectors in R 4 ).
It is clear that at the same time we can conclude that for m = 4 a class of hyperholomorphy is de ned
by a structural 4-tuple consisting of the unit and the orthonormalized basis in Re 3 , i.e., it is de ned just
by this basis.
However it is important to note that for m = 4 (in contrast to m = 3) a basis in R 3 is not obliged to
have positive orientation. Thus for m = 3 the set of classes of hyperholomorphy is parametrized by the
elements of the group O3+ (R ) while for m = 4 by the elements of the group O3 (R ).
In the sequel the structural set is chosen in accordance with the procedure described. In the
opposite case a special remark is made explicitly.
Now we point out the way for establishing connections between di erent classes of hyperholomorphy
with the help of a change of variables. Let = f 1 ; 2 ; 3 g be a structural 3-tuple and let
k 3
j k;j =1 :

a :=

(4.8)

Then a 2 0+3 (R ). Consider the change of variable x = a y with matrix (4.8). Formula (4.5) gives

'q = P k qk = iq ;
k=1
q 2 f1; 2; 3g:
3

(4.9)

For m = 4 let = fi0 ; 1 ; 2 ; 3 g, where f 1 ; 2 ; 3 g Re 3 , and where,in general the matrix a in (4.8)
belongs to O3 (R ) (not to O3+ (R )). Then the change of variables x = ae y, where

1 0 ;
0 a

ae =
gives 'e = fi0 ; i1 ; i2 ; i3 g =
15

st ;

(4.10)

and the standard set st corresponds to the standard embedding of R m into H : for m = 3 R 3
is identi ed with Re 3 corresponding coordinate axes coinciding; for m = 4 just the same thing. Besides
inequitable arrangement of coordinates in H occasioned by the existence of the quaternionic multiplication
is taking into account.
Summarazing we thus have by a linear change of variables that

f 2 M( ) () f
f 2 M ( ) () f

2 stM( 1( ));
2 Mst( 1 ( )):

(4.11)

Every structural set, including the standard one, de nes C.i.t.i. and singular integration operator. In
this section we show the connection between these important objects for the standard and for arbitrary
structural sets.
Let us start with the case m = 3. Let be a structural 3-tuple and let 0 be (as in 4.4) such
quaternion that a base f ~0 ; : : : ; ~3 g is equally oriented with the canonical base in R 4y . As above h = 0 ,
'k = h k ; a is the matrix (4.8), and a is the corresponding linear change of variables.
Direct computation shows that
4.5

Dx = hM W a1 stDy W a ;
(2)
(2)
(2)
;ay = h st;y ;
;x =

(4.12)

K (ay) = Kst (y) h;

where y 2 . Hence we have for C.i.t.i. and for the singular integration operator that

K@
S@

= W a1
= W a1

stK
@
stS
@

W a;
W a:

(4.13)
(4.14)

In the case m = 4 the matrix a is from O3 (R ) in general. Thus

where ae is the matrix (4.10).

S@ = sgn ( ) W ea1 stS @ W ea ;

5 Properties of the operator of the singular integration

The operator S introduced in 3.5 plays an exceptionally important role in various questions. That
is why we shall study its properties more systematically and thoroughly.
Let us rst note that the operator S can be extended onto sets of H (C )-valued functions. Moreover
in many cases it is more natural to consider it in the complexi ed spaces Lp ( ; H (C )) := Lp ( ; H ) R C
and
H ( ; H (C )) := H ( ; H ) R C ; p > 1; 0 < < 1:
Let X denotes any of these functional sets where X is considered as a right Banach module. Then
L(X ) will denote the set of all (right) linear bounded operators acting in X , and T (X ) is the set of all
compact in X operators.
5.1

16

5.2 From the Sokhotski-Plemelj formulas the following property may be easily derived: If
is a smooth
hypersurface without boundary in R m , then S belongs to any of the sets L(Lp ( ; H ), L(Lp( ; H (C )),
L(H ( ; H )), L(H ( ; H (C )) for p 2 (1; +1), 2 (0; 1). Morewover S is an involution, i.e.

S 2 = I:

(5.1)

This assertion is also preserved for Lp if

n
o
= R m 1 := (0; x5 m ; : : : ; x3 )

5.3

Denote

P := 21 (I

In accordance with 5.2

Rm

S ):

(5.2)

P 2 = P ; P + P = P P + = 0;
so that P are mutually orthogonal projections both on Lp and on H . Denote
P (Lp ) =: Lp ; P (H ) =: H :

(5.4)

From 3.9 it follows that, for instance Lp ( ; H ) is the space of traces on

(see 3.9). Analogously for H .
Thus
Lp ( ; H ) = L+p ( ; H ) Lp ( ; H );
H ( ; H ) = H + ( ; H ) H ( ; H ):
5.4

(5.3)

of functions from Ap( )

(5.5)

Let a 2 H ( ; H (C )) and consider the composition

M a S : f 2 Lp( : H (C )) 7 ! S f ] a:
We have:

= 2 K (

M a S f ](t) = S f ](t) a(t) =

t)

f ( ) (a(t) a( )) + 2 K (

t)

f ( )a ( ) =

=: T f ](t) + S f a](t) = T f ](t) + ( S M a ) f ](t):

The kernel of operator T has a weak singularity, whence T 2 T (Lp ( ; H (C ))). Also it follows from
the above that T 2 T (Lp ( ; H )) for a 2 H ( ; H ). These results can be easily extended to the case
of continuous functions using the density of H in C . Moreover one can show that if a 2 H then
T 2 T (H ).
Thus we have shown that the commutator

M a ; S ] := M a S

S Ma

is compact in the considered spaces.

It is clear that the same holds for the commutators of M a with the projections P .

17

Consider the algebra R generated by the operators: S; M a and T , where a 2 C ( ; H (C )) in the

case Lp and a 2 H ( ; H (C )) in the case H and T 2 T . It follows from 5.4 that an arbitrary operator
A from R is respresentable in the form
A = Ma P+ + Mb P + T =
= M c + M d S + T:
We emphasize in particular that the right (not left) multiplication is essential, and that for the operator
a M the result will not be true. This is the e ect of left- -hyperholomorphic functions forming a right
but not a left H -module.
5.5

We now consider the important situation when = Sm 1 or = R m 1 and the operators act on
L2 ( ; H (C )). We begin with the hyperplane R m 1 , in which case we need the well-known Riesz operators
Rk de ned by
Z k tk
(5.6)
Rk f ](t) := m2 1
j tjm f ( )d ;
5.6

jS

Let m = 4. Then

= R3

m 1

k 2 f1; : : : ; m 1g:

=f

g R 4. We have
Z 1 X
3
2
k
S f ](t) = 3 j tj4
( k tk )f ( )d =
jS j 3
k=1
R
Z k tk
3
X
k 1
=
2
4 f ( )d ;
j
t
j
k=1
3
(0; x1 ; x2 ; x3 )

hence

SR 3 =

3
X

k=1

3
X

Rk =

Rk :

k=1

(5.7)

Now let m = 3, = R 2 = f(x1 ; x2 ; 0)g R 3 :

Then we have analogously
Z
3
X
k k k 3
(
t ) f ( )d =
S f ](t) = 22 j 1 tj3

jS j

=
=
and hence

1 Z

2
2

j
2
R
1 Z

k=1

tj3

t1 ) +

1 3( 1

t1 f ( )d
tj3

2 3( 2

t2) f ( )d =

1 Z 2 t2 f ( )d ;
2 2 j tj3
R

(5:70 )
m 1 ; H (C )):
5.7 Let us now consider the direct and the inverse Fourier transforms of functions from L2 (R
R e ihx; i f (x) dx;
F f ]( ) := (2 ) (m 1)=2
m 1
RR
(5.8)
1
(
m
1)
=
2
eihx; i f (x) dx:
F f ]( ) := (2 )

SR 2 = 2 R1

1R

18

m 1

2:

From the properties of Fourier transform of complex-valued functions it follows that the expressions (5.8)
are well{de ned and that F and F 1 are isometric isomorphisms of L2 . Moreover for the Riesz operators
(5.6) we have
k
(5.9)
Rk = F 1 i j j F
or
k
FRk F 1 = i j j ;
(5.10)
where = ( 1 ; : : : ; m 1 ). From this and 5.6 we obtain

3
SR 3 = iF 1 P k j kj F;
k=1 1
1 2 F:
1
2
SR 2 = iF
jj
jj

(5.11)

As is well known the function

sym (Rk ) : 2 Sm 2 7 ! i k 2 C
is called the symbol of the operator Rk . Therefore we introduce the symbol of the operator SR m 1 by

P3 k k ;
sym ( SR 3 ) : 2 S2 7 ! i
k=1
sym ( SR 2 ) : 2 S 1 7 ! i 2 1 1

5.8

(5.12)

Now let = Sm 1 . Recall that (see Section 3)

S f ](t) = jSm2 1 j

m 1
S

P3

j tjm k=4 m

( k tk ):

P3 ( 1)p+m pdb f ( ):

(5.13)

p=4 m

For the sphere Sm 1 we have that ( 1)p+m d bp = p dsm 1 , whence (5.13) takes the form

S f ](t) =

jSm 1 j

m 1

3
X

tjm k=4

k
m

( k tk )
(5:130 )

3
X

pf (

p=4 m

Transforming (5.130 ) we obtain that

S f ](t) = (

X
4 m k<p 3

p2

1)m

jSm 1 j

)dsm 1 ( ):

3
X

j
m 1 k=4 m

tk ( 1)k dbk f ( )+
tjm

( 1)m Z ( k tk )( 1)p d bp ( p tp)( 1)k d bk f ( ) =

m 1
j tjm

jS

m 1

19

=: ( 1)m I0 + ( 1)m

X
4 m k<m 3

k p
I

kp:

Notice that the following formula has been used for the computations: if is a structural set, fak ; bp g R
then
P3 a k P3 b p =
p
k
p=4 m
k=4 m
(5.14)
P3 a b +
P (a b a b ) k p :
=
p k
k k
k p
k=4 m

4 m k<p 3

In order to de ne the symbol of S it will be convinient for us to use the stereographic projection of the
sphere Sm 1 onto R m 1 and the change of variables operators generated by this projection. Let us recall
that that the stereographic map is given by

w : t = (t4 m; : : : ; t3 ) 2 Sm 1 7 ! x = (x1 ; : : : ; xm 1 ) =

4 m
2
=: 1t t3 ; : : : ; 1 t t3 2 R m 1 ;

and that thecorresponding change of variables operators are given by

m 1
2
2
f (w 1 (x));
2
jxj + 1
!m 1
jw(t)j2 + 1 2 g(w(t)):
2

W f ](x) :=
W 1 g](t) :=

The operator W is an isometric isomorphism of L2 (S m 1 ) onto L2 (R m 1 ). Let us consider the operator

W S W 1 acting on L2 (R m 1 ; H (C )). Making the change of variables = w 1 (y) in the integral
W S W 1 f ], direct computations give:
2m jy xjm
j tjm =
;
(jyj2 + 1) m2 (jxj2 + 1) m2
3

2
2
t3 = 2 (jxj2 +jyj1) (jjxyjj2 + 1) ;
m 1
2
db3 = 2 (1 +(1jyj2j)ymj ) dy;

while for k 2 f4 m; : : : ; 2g
k

tk = 2 y

k+m 3

(jxj2 + 1) xk+m 3 (jyj2 + 1) ;

(jxj2 + 1) (jyj2 + 1)

m k+m 3
dbk = ( 1)k 2 (1y+ jyj2 )m dy :

Let us now compute WI0 W 1. We have

WI0 W f ](x) =
1

jSm 1 j(jxj2 + 1) m2 1
20

Z (jyj2 + 1) m2 (jxj2 + 1) m2
2m jy xjm
m 1

0 2
k+m 3 (jxj2 + 1) xk+m 3 (jyj2 + 1)
@ X 2m+1 y
yk+m
2 + 1) (jy j2 + 1)m+1
(
j
x
j
k=4 m
!
2 jxj2 ) (1 jy j2 )
m 1
(
j
y
j
m
2 (jxj2 + 1) (jyj2 + 1)m+1 (jyj2 + 1) 2 f (y)dy =
Z
2
f (y)dy
= m 1 p
2
jS j jxj + 1 m 1 (jyj2 + 1) 32 jy xjm

2 jyj2 (jxj2 + 1) 2hx; yiR m 1 (jyj2 + 1) (jyj2 jxj2 ) (1 jyj2 ) =

Z
2
p 2 f (y)dy m 2
= m 1 p
2
jS j jxj + 1 m 1 jyj + 1 jy xj
R

Hence, the operator WI0 W is compact on L2 (R m 1 ).

Now we compute WIkpW 1 for p 2. We have
1

WIkpW 1 f ](x) =

Z (jyj2 + 1) m2 (jxj2 + 1) m2
2m jy xjm
m 1

(jxj2 + 1)

m 1
2

(jxj2 + 1) xk+m 3 (jyj2 + 1) 2m yp+m

(jxj2 + 1) (jyj2 + 1)m+1

(jxj2 + 1) xp+m 3 (jyj2 + 1) 2m yk+m 3 (jyj2 + 1) m2 1 f (y)dy =

(jxj2 + 1) (jyj2 + 1)m+1
Z
f (y)dy
p4 2
xp+m 3 yk+m 3 xk+m 3 yp+m 3 :
1
2
m
jxj + 1 m 1 (jyj + 1) 2 jy xj

yp+m

= m
jS

yk+m

jSm 1 j

Thus

p+m 3
k+m 3
WIkpW = 2 xjxj2 + 1 Rk+m 3 xjxj2 + 1 Rp+m 3 + Tkp;
where Rj denotes j th Riesz operator and Tkp is a compact operator.
Now let p = 3. Computing WIk3 W 1 for k 2 f4 m; : : : ; 2g, we have
1

W Ik3 W f ](x) =
1

jSm 1 j (jxj2 + 1) m2 1

Z (jyj2 + 1) m2 (jxj2 + 1) m2
2m jy xjm
m 1

(jxj2 + 1) xk+m 3 (jyj2 + 1) ( 1) 1 jyj2

(jxj2 + 1) (jyj2 + 1)
(1 + jyj2 )m
2
2
k+m 3
m 1
21+m (jx(jjy2j+ 1)jxj(j)yj2y+ 1)m+1 (jyj2 + 1) 2 f (y)dy =
Z
f (y)dy
2
= m 1 p 2
jS j jxj + 1 m 1 (jyj2 + 1) 12 jy xjm

k+m
2m y

yk+m 3

xk+m 3

+ jyj2 xk+m
21

jxj2 yk+m

0
@ yk+m
Thus

Z
f (y)dy
= m 1 p2 2
2
jS j jxj + 1 m 1 (jyj + 1) jy xjm
R

xk+m

(1 jxj2 ) +

2
WIk3W 1 = jjxxjj2 + 11 Rk+m

xk+m 3 R
jxj2 + 1 p+m

q=1

xk+m

mX1 2

where Tk3 is a compact operator.

We obtain nally

W S W 1 = ( 1)m 2

mX1

q=1

(yq

xq xk+m 3 R + T ;
q
k3
jxj2 + 1

xp+m 3 R
jxj2 + 1 k+m

4 m k<p 2
2
X

+ ( 1)m

k=4 m
mX1 2xq xk+m 3

jxj2 + 1

q=1

1
xq ) (yq + xq )A :

jxj2 1 R
jxj2 + 1 k+m

(5.15)
3

Rq + T;

where T is a compact operator.

Now let us write down the symbol of the operator (5.15). We have
sym S := sym W S W 1 =
=(

1)m

i 2

xp+m 3
jxj2 + 1

4 m k<p 2

2
X

k=4 m
Sm 2 .

0
2 1
3 @ jxj
jxj2 + 1

k+m 3

k+m 3
mX1 2

xk+m 3
jxj2 + 1

xq xk+m
jxj2 + 1

q=1

!
p+m 3

1
qA ;

+
(5.16)

where (x; ) 2 R m 1
We again distinguish between m = 3 and m = 4. Firstly let m = 3. Recall that in this case sgn = 1,
1
2 = 3, k
k = 1. Then we have from (5.16) that
sym S = i
where

3
X

k=1

k (x;

);

((x2 )2 (x1 )2 + 1) 2 + 2x1 x2 1 ;

jxj2 + 1
((x1 )2 (x2 )2 + 1) 1 + 2x1 x2 2 ;
2 (x; ) :=
jxj2 + 1

(5.17)

1 (x; ) :=

3 (x;

) := 2 x jx1j2 +x1
2

22

(5.18)

Let us note that for all 2 S1 the vector (x; ) := ( 1 ; 2 ; 3 ) may be interpreted as a unit tangent vector
to the sphere S2 at the point t = w 1 (x).
Now consider the case m = 4 and sgn = 1. Then (5.16) gives:

sym S = jxj2 + 1

where (x; ) 2 R 3

S2 .

0
@

0
@2x2

2x1

(jxj2 1) 3 + 2x3

3
X

q=1

1
xq q A +

0
1
3
X
@2x3 1 2x1 3 + (jxj2 1) 2 2x2 xq q A +
q=1
0
11
3
X
@2x3 1 2x3 2 + (jxj2 1) 1 2x1 x1 q AA ;
q=1

Returning from the variable x 2 R 3 to the initial variable t 2 S3 we obtain

0
sym S = i @
+

0
@t1

t0

1
q+1 A +
1
2
t1 X
A
1 t3 q=0 tq q+1 +
11
2
t0 X
tq q+1 AA ;

2
X
tq
t3 3 + 1 t2 t
3 q=0

0
@t2 1 t0 3 + t3 2
0
@t1 3 t2 2 + t3 1

1 t3 q=0

where 2 S2 as before and where t already belongs to S2 .

Now let m = 4 and let sgn = 1. Then we obtain from (5.16):

sym S = jxj2 + 1

where (x; ) 2 R 3

S2 .

0
@

0
@2x2

(5.19)

2x1 2 + (jxj2 1)

2x

3
3X

1
xq q A +

q=1
1
0
3
@2x3 1 2x1 3 (jxj2 1) 2 + 2x2 X xq q A +
q=1
11
0
3
X
@2x3 2 2x2 3 jxj2 1 1 2x1 xq q AA ;
q=1

Returning again from the variable x 2 R 3 to the initial variable t 2 S3 we obtain

sym S = i
+
+

t1

t2

2
3

t2

t0 2 + t3

2
t2 X
1 t3 0 tq

q+1

!
2
t
1 X
tq q+1 +
t0 3 t3 2 + 1 t
3 0
!!
2
t
0 X
t1 3 + t3 1 1 t
t1 q+1 ;
3 0
23

(5.20)

where again 2 S2 and t 2 S2 already.

It will be convenient for further purposes to introduce the following notations. Let = (i0 ; 1 ; 2 ; 3 )
be an arbitrary structural set (i.e. sgn may be equal both +1 and 1), and let t = (t0 ; t1 ; t2 ; t3 ) 2 S3 .
De ne for each t 2 S3 and 2 S2 the vector (t; ) = ( 0 ; 1 ; 2 ; 3 ) 2 R 4 by

t0

0 (t;

) :=

1 (t;

) :=

1 t3

1 t3

t1

3 (t;

);

3 (t;

);

) := 3 1 t2 t 3 (t; );
3
3 (t; ) := t0 1 + t1 2 + t2 3 :
Direct computations show that for every xed t 2 S3 and 2 S2 the vector (t; ) is orthogonal to the
vector t and j j = 1. Moreover for any xed t 2 S3 the point is an arbitrary point of S2 , consequently
the set f (t; )g coincides with the set of all unit tangent vectors to the sphere S3 at the point t. For
t 2 S3 and h(t; ) 2 R 4 we introduce
2 (t;

:=
:=

3
X

tk ;

k:

k=0
3
X

k=0

Then as it can be easily seen the formulas (5.19){(5.20) may be written in the form
sym S = i t
(t; ):
(5.21)
5.9 We nish this section with the proof of the following property: the operators P
(introduced in
5.3) are orthoprojections on L2 ( ; H ), i.e. P = P , where = R m 1 or = Sm 1 , and L2 ( ; H ) is
understood as a right H -module. For the de nition of a self-adjoint operator we refer to 3.11{3.12.
It is su cient to show that S = S . For the sake of simplicity we restrict ourselves to the case
= R 3 . We have:

h S f ]; gi =

Z
R

f ](t) g(t)dt =

3 Z
X

k=1

3 Z
X

k=1

Rk f ](t) g(t)dt =
k

Rk f ](t)

g(t)dt =

1
0
Z
Z
X BB k tk
CC k
f
(
)
d
=
A g(t)dt =
@
4
j
t
j
k=1 3
3
Rt R
Z k tk
3 Z
X
k
=
f ( ) d j tj4 g(t)dt =
k=1 3
3
R

24

3 Z
X

k=1

f( )

Rk g]( )d =

= hf; S g]i;
whence S = S .
References

B] A.V. Bitzadze, Boundary value problems for elliptic equations of the second order. Nauka, 1966
(Russian).
BDS] F. Brackx, R. Delanghe, F. Sommen, Cli ord Analysis. Boston, Pitman, 1982.
D] C.A. Deavours. The quaternionic calculus.Amer. Math. Monthly, 80 (1973). 995{1008.
DSS] R. Delanghe, F. Sommen, V. Soucek, Cli ord Analysis and Spinor Valued Functions, Dordrecht,
Kluwer Acad. Publ., 1992.
F1] R. Fueter, Analytische Funktionen einer Quaternionen{variablen. Comm. Math. Helv., 4 (1932).
9{20.
F2] R. Fueter, Die Funktionentheorie der Di erentialgleichungen u = 0 und
variablen. Comm. Math. Helv., 7 (1934{35). 307{330.

u = 0 mit vier reelen

GM] J. Gilbert, M. Murray. Cli ord Algebras and Dirac Operators in Harmonic Analysis. Cambridge
Univ. Press. 1990.
GS] K. Gurlebeck, W. Sprossig. Quaternionic Analysis and Elliptic Boundary Value Problems. Math.
Research, v. 56, 1989.
I] V. Iftimie. Functions hypercomplexes.Bull. Math. Soc. Sci. Math. RSR, 9, no. 4 (1965). 279{332.
Ist] ????????????
L] P. Lounesto. Spinor valued regular functions. Contemp. Math., 11 (1982). 155{175.
M1] G.C. Moisil. Sur les quaternions monogenes. Bull. Sci. Math. Paris (2), 55 (1931). 169{194.
M2] G. Moisil, N. Theodoresco. Functions holomorphes dans l'espace. Mathematica, V (1931). 142{
159.
N1] K. N^ono. On the quaternion linearization of Laplacian . Bull. Fukuoka Univ. Educ. Nat. Sci.,
35 (1985). 5{10.
N2] K. N^ono. Hyperholomorphic functions of a quaternion variable. Bull. Fukuoka Univ. Educ. Mat.
Sci., 32 (1982). 21{37.
R1] J. Ryan. Complexi ed Cli ord analysis. Complex Variables, 1 (1982). 119{179.
25

R2] J. Ryan. Cli ord analysis with generalized elliptic and quasi elliptic functions. Appl. Anal., 13,
no. 3 (1982). 151{171.
Sh1] M.V. Shapiro. On some boundary value problem for functions with values in Cli ord algebra.
Mat. Vesnik, Beograd, 40 (1988). 321{326.
Sh2] M.V. Shapiro. On analogues of the Riemann boundary value problem for a class of hyperholomorphic functions. \Integral equations and boundary value problems". World Scienti c, 1991, p.
184{188.
Sp1] W. Sprossig. Stetigkeitseigenschaften des Nablaoperators. Wiss. Z. Techn. Hochsch. Karl-MarxStadt, 20. no. 4 (1978). 517{521.
Sp2] W. Sprossig. Berechnung von Thermospannungen uber die Auswertung singularer Integrale in
Raumlichen Gebieten. Wiss. Schift. Techn. Hochsch. Karl-Marx-Stadt, 7 (1979). 73{85.
Sp3] W. Sprossig. Analoga zur funktionentheoritischen Satzen in R m -Anwendungen auf mehrdimensionale singulare Integralgleichungen. Beitr. Anal. 12 (1978). 113{126.
Su] A. Sudbery. Quaternionic analysis. Math. Proc. Cambridge Phil. Soc.

85, no. 2 (1971). 199{225.

Vi1] V. Vinogradov. On analogue of the Cauchy-Riemann system in four-dimensional space. Doklady

AN SSSR, 154, no. 1 (1954). 16{19 (Russian).
Vi2] V. Vinogradov. On an elliptic system which has no Fredholm boundary value problems. Doklady
AN SSSR, 199, no. 5 (1971). 1008{1010 (Russian).
VS1] N.L. Vasilevski, M.V. Shapiro. On an analogue of monogenity in the sense of Moisil-Teodoresco
and some applications in the theory of boundary value problems. Reports of Enlarged Session of
Seminars of the I.N. Vekua Institute of App. Math. Tbilisi, 1 (1985). 63{66 (Russian).
VS2] N.L. Vasilevski, M.V. Shapiro. Quaternionic -monogenic functions, singular operators with the
quaternionic Cauchy kernel and analogues of the Riemann problem. Dep. UkrNINTI no. 629{Uk{
87, 1987, 68p. (Russian).
VS3] N.L. Vasilevski, M.V. Shapiro. Holomorphy, hyperholomorphy, Toeplitz operators. Uspekhi Mat.
Nauk (Russian Math. Surveys) 44, no. 4 (1989). 196{197 (Russian).

26