Professional Documents
Culture Documents
net/publication/345990039
CITATIONS READS
0 12
2 authors, including:
Matvei Libine
Indiana University Bloomington
31 PUBLICATIONS 223 CITATIONS
SEE PROFILE
All content following this page was uploaded by Matvei Libine on 22 February 2021.
Abstract
We extend constructions of classical Clifford analysis to the case of indefinite non-
degenerate quadratic forms. We define (p, q)-left- and right-monogenic functions by
means of Dirac operators that factor a certain wave operator. We prove two different
versions of Cauchy’s integral formulas for these functions. The two formulas arise from
dealing with singularities in distinct ways, and are inspired by the methods of [L, FL].
These results indicate the merit of these methods for dealing with singularities.
1 Introduction
Many results of complex analysis have analogues in quaternionic analysis. In particular, there
are analogues of complex holomorphic functions called (left- and right-) regular functions.
Cauchy’s integral formula for complex holomorphic functions f (z)
1 f (z)
I
f (z0 ) = dz
2πi z − z0
extends to the quaternionic setting, and its analogues for (left- and right-) regular functions
are usually referred to as Cauchy-Fueter formulas. For modern introductions to quaternionic
analysis see, for example, [S, CSSS].
Complex numbers C and quaternions H are special cases of Clifford algebras. (For an
elementary introduction to Clifford algebras see, for example, [G].) There is a further ex-
tension of complex and quaternionic analysis called Clifford analysis. For Clifford algebras
associated to positive definite quadratic forms on real vector spaces, Clifford analysis is very
similar to complex and quaternionic analysis (see, for example, [BDS, DSS, GM] and ref-
erences therein). Furthermore, J. Ryan has initiated the study of Clifford analysis in the
setting of complex Clifford algebras [R1, R2].
If a Clifford algebra is associated to a quadratic form that is not positive definite, the
function that should serve as the reproducing kernel in the Cauchy type formula has singular-
ities that always intersect the contour of integration, thus rendering a potential reproducing
integral formula meaningless. For this reason, most developments of real Clifford analysis
∗
Department of Mathematics, Indiana University, Rawles Hall, 831 East 3rd St, Bloomington, IN 47405
†
Undergraduate student at Cornell University, Ithaca, NY 14850
1
so far have been in the setting of Clifford algebras associated to positive definite quadratic
forms.
We develop Clifford analysis in the setting of Clifford algebras associated to indefinite
non-degenerate quadratic forms by treating the relevant Clifford algebras as real subalgebras
of complex Clifford algebras and, in a certain sense, restricting the complex Dirac operators
and functions to the appropriate real subspaces. This is done in complete parallel with
K. Imaeda’s approach in [I], where he first extends classical quaternionic analysis to the
algebra of biquaternions HC = H ⊗R C and then “restricts” to the Minkowski space M by
realizing M as a real form of HC .
Similarly, the works [L, FL] extend classical quaternionic analysis to split quaternions
HR (also known as coquaternions). The algebra HR is isomorphic to the algebra of real 2 × 2
matrices and is another example of a Clifford algebra, this time associated to an indefinite
non-degenerate quadratic form on R2 . The issue of singularities intersecting the contour
of integration is resolved in two different ways, leading to two types of integral formulas.
The first method uses biquaternions HC , which contain the split quaternions HR as a real
subalgebra. By considering the holomorphic extension of a regular function from HR into
HC and deforming the contour of integration so it no longer intersects the singularities,
one obtains the first version of Cauchy’s formulas for regular functions in HR . The second
method involves inserting a purely imaginary term iεkX − X0 k2 into the denominator of the
reproducing kernel to prevent singularities, then showing that the limit as ε → 0 exists and
produces another version of Cauchy’s formulas for regular functions in HR .
Methods for dealing with non-positive definite signature setting developed in [L, FL]
were used in [RCW] and [P]. Thus [RCW] extends the results of classical Clifford analysis to
the setting of what the authors call split-quaternionic Hermitian Clifford analysis. And [P]
extends analysis over octonions to split octonions. Both papers have to deal with exactly the
same issue of the set of singularities intersecting the contour of integration, and this issue is
resolved in the same way as in [L, FL]. In this paper we use methods of [L, FL] to extend
constructions of classical Clifford analysis to the case of indefinite non-degenerate quadratic
forms. In a larger context, our results suggest that the methods of handling singularities
developed in [L, FL] are quite general.
We begin by recalling in Section 2 the construction of the universal Clifford algebras.
These constructions are done in both the real and complex cases, and we regard the real
Clifford algebra as a subalgebra of the complex Clifford algebra. Following this, in Section
3 we define Dirac operators and introduce left- and right-monogenic functions as functions
annihilated by Dirac operators. Both real and complex cases are considered. The complex
case was mostly developed by J. Ryan [R1, R2], and the real case is, in a certain sense,
the restriction of the complex operators and functions to the real subspace. In Section 4 we
introduce Clifford-valued differential forms Dn z, Dp,q x and establish some of their properties.
These forms appear in the statements of Cauchy’s formulas for monogenic functions. We
also state the classical Cauchy’s integral formulas for left- and right-monogenic functions in
the positive definite case (Theorem 16). In Section 5 we state and prove the first version
of Cauchy’s integral formulas for left- and right-monogenic functions (Theorem 20). This
version requires having holomorphic extension of the monogenic functions to the complex
space and utilizes a deformation of the contour of integration to avoid the set of singularities.
In Section 6 we prove the second version of Cauchy’s integral formulas for left- and right-
2
monogenic functions (Theorem 23). This version involves inserting a purely imaginary term
iεkX − X0 k2 into the denominator of the reproducing kernel to prevent singularities, then
taking limit as ε → 0+ . We first establish Theorem 23 up to a constant coefficient (26).
Then we pin down the value of the coefficient using the first version of integral formulas
(Theorem 20), which is already established (Lemma 33).
This research was made possible by the Indiana University, Bloomington, Math REU
(research experiences for undergraduates) program, funded by NSF Award #1757857. We
would like to thank Prof. Chris Connell for organizing and facilitating this wonderful pro-
gram. We would also like to thank Ms. Mandie McCarthy for her administrative work, the
various professors who gave talks, and the other REU students for their company.
2 Clifford Algebras
In this section we recall basics of Clifford Algebras and establish notations. For an elementary
introduction to Clifford algebras see, for example, [G].
Let V be an n-dimensional real vector space, and Q a quadratic form on V . The form Q
uniquely extends to a bilinear form on V . Diagonalizing Q, we have that there exist integers
p, q and orthogonal basis {e1 , . . . , en } of (V, Q), such that
1
1 ≤ j ≤ p;
Q(ej ) = −1 p + 1 ≤ j ≤ p + q;
0 p + q < j ≤ n.
By Sylvester’s Law of Inertia, we the numbers p and q are independent of basis chosen. The
ordered pair (p, q) is the signature of (V, Q). From this point on we restrict our attention to
non-degenerate quadratic forms Q, in which case p + q = n. If Q is such a form with q = 0
or p = 0, then it is positive or negative definite quadratic form respectively.
Recall the standard construction of the universal Clifford algebra associated to (V, Q) as
a quotient of the tensor algebra. We start with the tensor algebra over V ,
O
V = R ⊕ V ⊕ (V ⊗ V ) ⊕ (V ⊗ V ⊗ V ) ⊕ . . . ,
consider a set of elements of the tensor algebra
O
S = {v ⊗ v + Q(v) : v ∈ V } ⊂ V,
N
and let (S) denote the ideal of V generated by the elements of S. Then the universal
Clifford algebra associated to (V, Q) can be defined as a quotient
O
AQ = V /(S). (1)
We note the sign convention chosen for elements of the ideal is not standard in the literature,
resulting in a possible interchange of p and q. Let e0 ∈ AQ denote the multiplicative identity
of the algebra. If v1 and v2 are orthogonal elements of (V, Q),
v1 ⊗ v2 + v2 ⊗ v1 = (v1 + v2 ) ⊗ (v1 + v2 ) − v1 ⊗ v1 − v2 ⊗ v2
= −Q(v1 + v2 ) + Q(v1 ) + Q(v2 ) = 0. (2)
3
Thus AQ is a finite-dimensional algebra over R generated by e1 , . . . , en , and let Ap,q denote
the algebra AQ constructed from (V, Q), where Q has signature (p, q). We fix an orthogonal
basis
{e1 , e2 , . . . , ep , ẽp+1, . . . , ẽp+q }
of V such that Q(ej ) = 1 and Q(ẽj ) = −1 for all applicable j. In the positive definite case
the basis of V is given by {e1 , e2 , . . . , en }, and we let An = An,0. We have the following
relations for 1 ≤ i, j ≤ n with i 6= j,
B = {i1 , i2 , . . . , ik } ⊆ {1, . . . , n}
or, more precisely, eB is the image of this tensor product in AQ . If B is empty, we set e∅ be
the identity element e0 . These elements
eB : B ⊆ {1, 2, . . . , n}
form a vector space basis of Ap,q over R. We will be especially concerned with the space
R ⊕ V , which we identify with the vector subspace Rp+q+1 ⊂ Ap,q spanned by
n+1
Pnparticular, in the positive definite case any X ∈ R
In ⊂ An can be expressed as X =
j=0 xj ej .
We can also construct a complex universal Clifford algebra from a quadratic space over
C. Let V C be an n-dimensional complex vector space with quadratic form Q. Diagonalizing
Q, we have that there exists an orthogonal basis {e1 , . . . , en } of V C and integer p such that
Q(ej ) = 1 if 1 ≤ j ≤ p and Q(ej ) = 0 otherwise. The form Q is non-degenerate if and only
if p = n, and we only consider this case. We perform a similar construction to (1): form the
tensor algebra over V C
O
V C = C ⊕ V C ⊕ (V C ⊗ V C ) ⊕ (V C ⊗ V C ⊗ V C ) ⊕ . . . ,
consider a set O
S C = {v ⊗ v + Q(v) : v ∈ V C } ⊂ V C,
4
and let (S C ) denote the ideal generated by S C . Then the complex universal Clifford algebra
associated to (V C , Q) is a quotient
O
ACQ = V C /(S C ).
We also use notation ACn for this algebra. The analogues of (2) and (3)-(4) remain valid for
ACn , and the products eB defined as in (5) form a basis of ACn .
We can also consider ACn as a real algebra generated by
(These conjugations can be extended to all of ACp+q , but for the purposes of this paper
we do not need this.) The Clifford conjugation fixes the C-span of e0 identified with C,
and the complex conjugation fixes the R-span of e0 , . . . , en identified with Rn+1 , and these
conjugations can be viewed as reflecting over the respective subspaces.
These two conjugations commute and lead to two useful quadratic forms on Cn+1 ⊂ ACn .
The first such form is n
X
+ +
N(Z) = ZZ = Z Z = zj2
j=0
5
then all Z ∈ Cn+1 \ NnC are invertible with inverse given by Z −1 = N(Z)−1 Z + . The other
quadratic form is real valued:
n
21 1 X
kZk = (Z Z̄ + + Z̄Z + ) = (Z̄ + Z + Z + Z̄) = |zj |2 .
2 2 j=0
As was mentioned earlier, we consider Ap,q as a real subalgebra of ACp+q . We describe the
restrictions of these two conjugations and two quadratic forms to Rp+q+1 ⊂ Ap,q ⊂ ACp+q :
p p+q p p+q
X X X X
+
X = x0 e0 + xj ej + x̃j ẽj 7→ X = x0 e0 − xj ej − x̃j ẽj ,
j=1 j=p+1 j=1 j=p+1
p p+q p p+q
X X X X
X = x0 e0 + xj ej + x̃j ẽj 7→ X̄ = x0 e0 + xj ej − x̃j ẽj , (8)
j=1 j=p+1 j=1 j=p+1
p p+q
X X
+ +
N(X) = XX = X X = x2j − x̃2j ,
j=0 j=p+1
p q
1
2 + + 1 + +
X
2
X
kXk = (X X̄ + X̄X ) = (X̄ X + X X̄) = xj + x̃2j . (9)
2 2 j=0 j=p+1
then all X ∈ Rp+q+1 \ Np,q are invertible with inverse given by X −1 = N(X)−1 X + . Finally,
we note that in the positive definite case Nn,0 = {0}.
6
We introduce linear differential operators on Cn+1
n n
∂ X ∂ ∂ X ∂
∇+
C = e0 + ej and ∇C = e0 − ej
∂z0 j=1 ∂zj ∂z0 j=1 ∂zj
which may be applied to functions on the left and on the right. Let C be the complex
Laplacian
n
X ∂2
C = 2
.
j=0
∂z j
Then
∇C ∇+ +
C f = ∇C ∇C f = C f and g∇C ∇+ +
C = g∇C ∇C = C g, (10)
where f is a holomorphic function on Cn+1 with values in ACn or a left ACn -module, and g is
a holomorphic function on Cn+1 with values in ACn or a right ACn -module.
Definition 1. Let U ⊆ Cn+1 be an open set, and MnC a left ACn -module. A holomorphic
function f : U → MnC is called complex left-monogenic if
n
∂f X ∂f
∇+
C f = e0 + ej =0
∂z0 j=1 ∂zj
at all points in U.
Similarly, let M̃nC be a right ACn -module, then a holomorphic function g : U → M̃nC is
called complex right-monogenic if
n
∂g X ∂g
g∇+
C = e0 + ej = 0
∂z0 j=1
∂zj
at all points in U.
The factorization (10) leads to a natural method of constructing complex left- and right-
monogenic functions. If ϕ : U → MnC is complex harmonic (i.e., C ϕ = 0), with U and
MnC as in the definition, then ∇C f is complex left-monogenic. Similarly, if ϕ̃ : U → M̃nC is
complex harmonic, ϕ̃∇C is complex right-monogenic.
We restrict these definitions to the subalgebra Ap,q ⊂ ACp+q . Thus we introduce linear
differential operators
p p+q
∂ X ∂ X ∂
∇+
p,q = e0 + ej − ẽj and
∂x0 j=1 ∂xj j=p+1 ∂ x̃j
p p+q
∂ X ∂ X ∂
∇p,q = e0 − ej + ẽj
∂x0 j=1 ∂xj j=p+1 ∂ x̃j
which may be applied to functions on the left and on the right. Let p,q be a wave operator
on Rp+q+1 ,
p p+q
X ∂2 X ∂2
p,q = − .
j=0
∂x2j j=p+1 ∂ x̃2j
7
Then
∇p,q ∇+ +
p,q f = ∇p,q ∇p,q f = p,q f and g∇p,q ∇+ +
p,q = g∇p,q ∇p,q = p,q g, (11)
where f is a C 2 function on Rp+q+1 with values in Ap,q or a left Ap,q -module, and g is a C 2
function on Rp+q+1 with values in Ap,q or a right Ap,q -module.
Definition 2. Let U ⊂ Rp+q+1 be an open set, and Mp,q a left Ap,q -module. A C 1 function
f : U → Mp,q is called (p, q)-left-monogenic if
p p+q
∂f X ∂f X ∂f
∇+
p,q f = e0 + ej − ẽj =0
∂x0 j=1 ∂xj j=p+1 ∂ x̃j
at all points in U.
Similarly, let M̃p,q be a right Ap,q -module, then a C 1 function g : U → M̃p,q is called
(p, q)-right-monogenic if
p p+q
∂g X ∂g X ∂g
g∇+
p,q = e0 + ej − ẽj = 0
∂x0 j=1
∂xj j=p+1
∂ x̃j
at all points in U.
The factorization (11) leads to a natural method of constructing (p, q)-left and right-
monogenic functions. If ϕ : U → Mp,q satisfies p,q ϕ = 0, with U and Mp,q as in the
definition, then ∇p,q f is (p, q)-left-monogenic. Similarly, if ϕ̃ : U → M̃p,q satisfies p,q ϕ̃ = 0,
ϕ̃∇p,q is (p, q)-right-monogenic.
The following case is an important example of such a construction, in which we consider
a Green’s function on Cp+q+1 ⊂ AC p+q and a suitable restriction to R
p+q+1
⊂ Ap,q .
Definition 3. For odd n ≥ 2, we define the following function Hn (Z): Cn+1 \ NnC → C by
1 1
Hn (Z) = n−1 = Pn n−1 .
N(Z) 2 ( j=0 zj2 ) 2
We note that if n is even, this may not be well defined. In this case, we introduce a
domain
Cn+1
G = Cn+1 \ {Z ∈ Cn+1 ; N(Z) ∈ R, N(Z) ≤ 0}
1
and choose a branch of N(Z) 2 : Cn+1
G → C with values in the right half-plane {w ∈ C; Re w >
1 1−n
0}. Then Hn (Z) = N(Z) 2 is a well defined function Cn+1
G → C. From now on we
n+1
restrict the domain of Hn (Z) to CG regardless of the parity of n, as the same proofs hold
in either case. Observe that on this domain Hn (Z) is complex harmonic, and, since Hn (Z) is
scalar-valued, ∇C Hn (Z) = Hn (Z)∇C is complex left- and right-monogenic. Thus, we obtain
a function on Cn+1
G with values in ACn that is complex left- and right-monogenic:
Lemma 4. An ACn -valued function Gn (Z) defined on Cn+1
G as
Pn
1 z0 − j=1 zj ej Z+
Gn (Z) = ∇C Hn (Z) = Pn n+1 = n+1
1−n ( j=0 zj2 ) 2 N(Z) 2
is complex left- and right-monogenic.
8
Restricting this function to Rp+q+1 ⊂ Ap,q as was done with Dirac operators earlier,
we obtain the following solution to the wave equation p,q ϕ = 0 and the corresponding
(p, q)-left- and (p, q)-right-monogenic functions. Let
p+q+1
RG = Rp+q+1 \ {X ∈ Rn+1 ; N(X) ≤ 0} = Rp+q+1 ∩ Cp+q+1
G .
1
Gp,q (X) = Gp+q (X) = ∇p,q Hp,q (X)
p+q+1
RG 1−p−q
x0 − pj=1 xj ej − p+q
P P
j=p+1 x̃j ẽj X+
= P p+q+1 = p+q+1 .
p 2
Pp+q 2 N(X)
j=0 xj − j=p+1 x̃j
2 2
Consequently, p,q Hp,q (X) = 0 and Gp,q is (p, q)-left- and right-monogenic.
4 Differential Forms
In this section we introduce the Clifford-valued differential n-forms Dn z and Dp,q x that will
appear in the statement of Cauchy’s integral formulas. We also prove several properties of
these forms. At the end of the section we state the classical Cauchy’s integral formulas for
monogenic functions. The complex case originally was developed in [R1].
Definition 6. Let dVC be the n + 1 complex holomorphic form on Cn+1 :
where
dẑj = dz0 ∧ dz1 ∧ · · · ∧ dzj−1 ∧ dzj+1 ∧ . . . dzn .
9
Lemma 8. Let U ⊂ Cn+1 be an open set and let f : U → ACn , g : U → M̃nC be holomorphic
functions, where M̃nC is a right ACn -module. Then,
d(gDn zf ) = (g∇+ +
C )f dVC + g(∇C f )dVC and d(gDn z) = (g∇+
C )dVC . (15)
Similarly, if f : U → MnC and g : U → ACn are holomorphic functions, where MnC is a left
ACn -module,
d(gDn zf ) = (g∇+ +
C )f dVC + g(∇C f )dVC and d(Dn zf ) = (∇+
C f )dVC .
This corollary may be used to give an alternative definition of complex left- and right-
monogenic functions.
Recall that the generators of the real universal Clifford algebra Ap,q and complex universal
Clifford algebra ACp+q are related by (6). Therefore, restricting to Rp+q+1 , we have
dzj (xj ej ) = xj = dxj (xj ej ), hence dzj = dxj if 0 ≤ j ≤ p;
Rp+q+1
dzj (x̃j ẽj ) = ix̃j = idx̃j (x̃j ẽj ), hence dzj = idx̃j if p + 1 ≤ j ≤ p + q.
Rp+q+1
Definition 11. Let Dp,q x be the restriction of Dp+q z to Rp+q+1, so that we have, for all
(X0 , X1 , . . . , Xp+q ) ∈ Rp+q+1 ,
X0 , Dp,q x(X1 , X2 , . . . , Xp+q ) = dVp,q (X0 , X1 , . . . Xp+q ).
Explicitly,
p
X p+q
X
Dp,q x = i q j
(−1) ej dx̂j − j ˆ
(−1) ẽj dx̃j , (17)
j=0 j=p+1
where
We have the following real analogue of Lemma 8; its proof is the same as in the complex
case.
10
Lemma 12. Let U ⊂ Rp+q+1 be an open set and let f : U → Ap,q , g : U → M̃p,q be C 1
functions, where M̃p,q is a right Ap,q -module. Then,
d(gDp,q xf ) = (g∇+ +
p,q )f dVp,q + g(∇p,q f )dVp,q and d(gDp,q x) = (g∇+
p,q )dVp,q .
Similarly, if f : U → Mp,q and g : U → Ap,q are C 1 functions, where Mp,q is a left Ap,q -
module,
d(gDp,q xf ) = (g∇+ +
p,q )f dVp,q + g(∇p,q f )dVp,q and d(Dp,q xf ) = (∇+
p,q f )dVp,q .
Corollary 13. A C 1 function f on U is (p, q)-left-monogenic if and only if the n-form Dp,q xf
is closed on U. Similarly, a C 1 function g on U is (p, q)-right-monogenic if and only if the
n-form gDp,q x is closed on U.
Again, this corollary may serve as an alternative definition of (p, q)-left- and (p, q)-right-
monogenic functions.
We orient Rp+q+1 ⊂ Ap,q so that
is a positively oriented basis. For an open subset U ⊂ Rp+q+1 with piecewise smooth bound-
ary ∂U, at each smooth boundary point X ∈ ∂U, let
be an outward pointing normal unit vector to U at X (with respect to (9)). Then we can
orient the boundary ∂U so that a basis {v1 , . . . , vp+q } of the tangent space TX U is positively
oriented if and only if {nX , v1 , . . . , vp+q } is a positively oriented basis of Rp+q+1 . Let dSp,q
be the contraction of nX with dVp,q , so that
Lemma 14. With n̄X denoting the complex conjugate of nX (recall (8)), the restriction
Dp,q x = n̄X dSp,q .
∂U
Proof. By definition, nX is orthogonal to the tangent space TX ∂U. Thus, for every vector
X = x0 e0 + x1 e1 + · · · + xp ep + x̃p+1 ẽp+1 + · · · + x̃p+q ẽp+q in TX ∂U, we have
p p+q
X X
nj xj + ñj x̃j = 0,
j=0 j=p+1
11
and hence
n0 dx0 + · · · + np dxp + ñp+1 dx̃p+1 + · · · + ñp+q dx̃p+q = 0.
Without loss of generality we may suppose n0 6= 0, then we can isolate dx0 :
1
dx0 = − (n1 dx1 + n2 dx2 + · · · + np dxp + ñp+1 dx̃p+1 + · · · + ñp+q dx̃p+q ). (19)
n0
We introduce the symbol dx̌j denoting wedge products of all but dx0 , dxj :
Then
Substituting (19) into the expression (18) for dSp,q and using these notations,
p n2 p+q ñ2
j j
X X
q j j
dSp,q = i n0 dx̂0 + (−1) − dxj ∧ dx̌j ) + (−1) − dx̃j ∧ dx̌j
j=1
n0 j=p+1
n 0
p p+q
iq X 2 iq
X
2
= n + ñ dx̂0 = dx̂0 ,
n0 j=0 j j=p+1 j n0
since nX is a unit vector. We perform a similar computation for Dp,q x substituting (19) into
(17) to get
p
X p+q
X
q j
(−1) ẽj dx̃ˆj
j
Dp,q x =i (−1) ej dx̂j −
∂U j=0 j=p+1
p n p+q ñ
j j
X X
= iq n0 dx̂0 + j
(−1) − ej dxj ∧ dx̌j − (−1) − j
ẽj dx̃j ∧ dx̃ˇj
j=1
n0 j=p+1
n 0
p p+q
iq X iq
X
= nj ej − ñj ẽj dx̂0 = n̄X dx̂0 = n̄X dSp,q .
n0 j=0 j=p+1
n0
12
These two expressions are the same in the positive definite case (when q = 0).
We state the classical Cauchy’s integral formulas for left- and right-monogenic functions
in the positive definite case (q = 0). For details and proofs see, for example, [BDS, DSS, GM].
Recall that in this case we write An instead of An,0 .
Theorem 16. Let n ≥ 2. Let U ⊂ Rn+1 ⊂ An be an open bounded set with piecewise C 1
boundary ∂U, and let Mn be a left An -module. Suppose that an Mn -valued function f is
(n, 0)-left-monogenic function on a neighborhood of the closure U . We have:
(
ωn f (X0 ) if X0 ∈ U,
Z
Gn,0 (X − X0 )Dn,0 xf (X) =
∂U 0 if X0 ∈
/ U,
Note that the map hε depends on p and q, but we suppress this dependence in order to
avoid cumbersome notations.
In particular, when ε = 1,
N(h1 (X)) = 2ikXk2 .
hε,Z0 : Z 7→ Z0 + hε (Z − Z0 ).
13
Theorem 20. Let p+q ≥ 2. Let U ⊂ Rp+q+1 ⊂ Ap,q be an open bounded set with piecewise C 1
C
boundary ∂U, and let Mp+q be a left ACp+q -module. Suppose that an Mp+qC
-valued function f
is (p, q)-left-monogenic on a neighborhood of the closure U. Furthermore, suppose f extends
to a complex left-monogenic function f C : W C → Mp+q C
, with W C ⊂ Cp+q+1 ⊂ ACp+q an open
subset containing U. We have:
(
ωp+q f (X0 ) if X0 ∈ U,
Z
Gp+q (Z − X0 )Dp+q zf C (Z) =
(hε,X0 )∗ (∂U ) 0 if X0 ∈
/ U,
Remark 21. For all ε 6= 0 sufficiently close to 0 the contour of integration (hε,X0 )∗ (∂U) lies
inside W C and the integrand is non-singular, thus the integrals are well-defined. Moreover,
we will see that the value of the integral becomes constant when the parameter ε is sufficiently
close to 0.
Remark 22. If we require ε < 0, then we need to insert a factor of (−1)q into the right
hand sides of the above formulas. Thus,
(
(−1)q ωp+q f (X0 ) if X0 ∈ U,
Z
Gp+q (Z − X0 )Dp+q zf C (Z) =
(hε,X0 )∗ (∂U ) 0 if X0 ∈
/ U,
for all ε < 0 sufficiently close to 0. And similarly for the other formula.
Proof. We consider only the left-monogenic case, the right-monogenic case is similar. By
translation, we may assume that X0 = 0. Let M = supX∈∂U kXk. We restrict W C to be the
δ-neighborhood of U in Cp+q+1 for some sufficiently small δ > 0. Consider 0 < |ε| < δ/M.
If X ∈ Rp+q+1 \ {0}, then, by (20), N(hε (X)) is not a negative real, hence hε (X) ∈ Cp+q+1G .
p+q+1 C
Thus, we have that (hε )∗ (∂U) lies inside CG ∩W . By Lemma 8, the integrand is a closed
p+q+1 C
form in the region CG ∩ W , so the integral stays unchanged for all −δ/M < ε < 0 and
0 < ε < δ/M (but the values of the integral may be different on these two intervals). If
X0 = 0 ∈ / U , we have that (hε )∗ (U ) ⊆ Cp+q+1
G ∩ W C , the integrand is a closed form, and the
integral over ∂U is zero, thus we are done.
Now, suppose X0 = 0 ∈ U. Choose an r > 0 sufficiently small such that r < δ/2
and the ball Br = {X ∈ Rp+q+1 : kXk ≤ r} is contained in U. We orient the sphere
Sr = {X ∈ Rp+q+1 : kXk = r} as the boundary of Br . By Stokes’ Theorem and (15) we
14
have:
Z Z
C
Gp+q (Z)Dp+q zf (Z) − Gp+q (Z)Dp+q zf C (Z)
(hε )∗ (∂U ) (hε )∗ Sr
Z Z
C
= d(Gp+q (Z)Dp+q zf (Z)) = 0dVC = 0.
(hε )∗ (U \Br ) (hε )∗ (U \Br )
Hence, Z Z
C
Gp+q (Z)Dp+q zf (Z) = Gp+q (Z)Dp+q zf C (Z).
(hε )∗ (∂U ) (hε )∗ Sr
By varying the parameter ε from its original value to 1, we can continuously deform (hε )∗ Sr
into (h1 )∗ (Sr ) within Cp+q+1
G ∩ W C without changing the value of the integral (here we are
using r < δ/2), and so we have:
Z Z
C
Gp+q (Z)Dp+q zf (Z) = Gp+q (Z)Dp+q zf C (Z).
(hε )∗ (∂U ) (h1 )∗ Sr
Next, we rotate the sphere (h1 )∗ Sr using the map Z 7→ e−it Z, 0 ≤ t ≤ π/4. Since
N(e−it Z) = e−2it N(Z) and ke−it Zk = kZk, as t varies continuously from 0√to π/4, the
sphere (h1 )∗ Sr gets rotated inside Cp+q+1G ∩W C into S̃r√2 – the sphere of radius r 2 contained
p+q+1
in Span{e0 , e1 , . . . , ep+q } = R ⊂ Ap+q,0. Thus, we have by Stokes’ Theorem and the
classical integral formula for the definite case (Theorem 16):
Z Z
C
Gp+q (Z)Dp+q zf (Z) = Gp+q (Z)Dp+q zf C (Z)
(hε )∗ (∂U ) S̃r √2
Z
= Gp+q,0(X)Dp+q,0xf C (X) = ωp+q f (0).
S̃r √2
Theorem 23. Let p+q ≥ 2. Let U ⊂ Rp+q+1 be a bounded open region with smooth boundary
∂U, and let Mp,q be a left Ap,q -module. Suppose that f is an Mp,q -valued function that is
(p, q)-left-monogenic in a neighborhood of the closure U . Then, for any X0 ∈ Rp+q+1 such
that ∂U intersects the cone {X ∈ Rp+q+1; N(X − X0 ) = 0} transversally, we have:
(
ωp+q f (X0 ) if X0 ∈ U,
Z
lim+ Gp,q,ε(X − X0 )Dp,q xf (X) =
ε→0 ∂U 0 if X0 ∈
/ U,
15
where Gp,q,ε is the modified Green’s function defined by
X+
Gp,q,ε(X) = p+q+1
N(X) + iεkXk2 2
Pp Pp+q
x0 − j=1 xj ej − j=p+1 x̃j ẽj
= P p+q+1 .
p 2
Pp+q 2
Pp 2
Pp+q 2
j=0 xj − x̃
j=p+1 j + iε( x
j=0 j + x̃
j=p+1 j ) 2
Similarly, if M̃p,q is a right Ap,q -module and g is an M̃p,q -valued function that is (p, q)-
right-monogenic in a neighborhood of U, then, for any X0 ∈ Rp+q+1 such that ∂U intersects
the cone {X ∈ Rp+q+1 ; N(X − X0 ) = 0} transversally, we have:
(
ωp+q g(X0) if X0 ∈ U,
Z
lim+ g(X)Dp,q xGp,q,ε (X − X0 ) =
ε→0 ∂U 0 if X0 ∈
/ U.
Remark 24. As in the case of the first Cauchy’s integral formula (Theorem 20), if we let
ε → 0− , we need to insert a factor of (−1)q into the right hand sides of the above formulas.
Thus, (
(−1)q ωp+q f (X0 ) if X0 ∈ U,
Z
lim Gp,q,ε(X − X0 )Dp,q xf (X) =
ε→0− ∂U 0 if X0 ∈
/ U.
And similarly for the other formula.
Proof. We change coordinates to hybrid spherical coordinates given by
Φ(ρ, ϕ1 , . . . , ϕp , θ, ψ1 , . . . , ψq−1 ) = (x0 , x1 , . . . , xp , x̃p+1 , . . . , x̃p+q ),
x0 = ρ cos θ cos ϕ1
x1 = ρ cos θ sin ϕ1 cos ϕ2
x2 = ρ cos θ sin ϕ1 sin ϕ2 cos ϕ3
..
.
xp−1 = ρ cos θ sin ϕ1 . . . cos ϕp 0 ≤ ϕ1 , ϕ2 , . . . , ϕp−1 ≤ π,
xp = ρ cos θ sin ϕ1 . . . sin ϕp 0 ≤ ψ1 , ψ2 , . . . , ψq−2 ≤ π, (21)
x̃p+1 = ρ sin θ cos ψ1 0 ≤ ϕp , ψq−1 ≤ 2π,
x̃p+2 = ρ sin θ sin ψ1 cos ψ2 0 ≤ ρ, 0 ≤ θ ≤ π2 .
..
.
x̃p+q−1 = ρ sin θ sin ψ1 . . . cos ψq−1
x̃p+q = ρ sin θ sin ψ1 . . . sin ψq−1
These are essentially the spherical coordinates for the (x0 , . . . , xp )-subspace combined with
the spherical coordinates for the (x̃p+1 , . . . , x̃p+q )-subspace. In these coordinates, we have:
p p+q
X X
N(X) = x2j − x̃2j = ρ2 cos2 θ − ρ2 sin2 θ = ρ2 cos(2θ) and
j=0 j=p+1
p p+q
2
X X
kXk = x2j + x̃2j = ρ2 cos2 θ + ρ2 sin2 θ = ρ2 .
j=0 j=p+1
16
We thus have that the null cone Np,q is the set of X in Rp+q+1 such that θ = π4 , and we
structure our argument in the vein of [L]. We use the symmetry of the change of basis matrix
with respect to p and q − 1 to calculate the determinant by means of block matrices. Let
Sn,α be the Jacobian matrix corresponding to the transformation into the standard (n + 1)-
dimensional spherical coordinates, (ρ, α1 , . . . , αn ) → (x0 , . . . , xn ) ∈ Rn+1 ,
Lemma 25. The determinant of the Jacobian matrix of the change of coordinates (21) is
given by
det(DΦ) = ρ cosp θ sinq−1 θ det(Sp,ϕ ) det(Sq−1,ψ ).
Proof. We spell out the Jacobian matrix of the change of coordinates (21) and divide it into
blocks as follows:
∂x0 ∂x0 ∂x0 ∂x0 ∂x0 ∂x0
. . . ∂ϕ . . . ∂ψ∂xq−1 0
∂ρ ∂ϕ1 ∂ϕ2 p ∂θ ∂ψ1
∂x1 ∂x1 ∂x1
. . . ∂ϕ∂x1 ∂x1 ∂x1
. . . ∂ψ∂xq−1 1
∂ρ ∂ϕ1 ∂ϕ2 p ∂θ ∂ψ1
. .. .. .. .. .. .. .. ..
.. . . . . . . . .
∂x ∂xp ∂xp ∂xp ∂xp ∂xp ∂xp
p . . . . . .
∂ρ ∂ϕ1 ∂ϕ2 ∂ϕp ∂θ ∂ψ1 ∂ψq−1
A B
DΦ = = .
Γ ∆
∂ x̃p+1 ∂ x̃p+1 ∂ x̃p+1 ∂ x̃p+1 ∂ x̃p+1 ∂ x̃p+1 ∂ x̃p+1
∂ρ ∂ϕ1 ∂ϕ2
. . . ∂ϕp ∂θ ∂ψ1
. . . ∂ψq−1
.. .. .. . . . ..
. . . . . . .
. . . . . . . . .
∂ x̃p+q ∂ x̃p+q ∂ x̃p+q ∂ x̃p+q ∂ x̃p+q ∂ x̃p+q ∂ x̃p+q
∂ρ ∂ϕ1 ∂ϕ2
. . . ∂ϕp ∂θ ∂ψ1
. . . ∂ψq−1
We make the following observations about the blocks A, B, Γ and ∆. First, A is the
(p + 1) × (p + 1) matrix that is obtained from Sp,ϕ by multiplying each entry by cos θ; we
denote the first column of A by ~a. Next, B is the (p + 1) × q matrix whose first column is ~b,
described below, and all other columns are zero. Similarly, Γ is the q × (p + 1) matrix whose
first column is ~c, described below, and the rest of the columns are zero. Finally, ∆ the q × q
matrix obtained from Sq−1,ψ by multiplying the first column by ρ cos θ and multiplying the
remaining columns by sin θ; we denote the first column of ∆ by d. ~
ρ cos θ 0 0 ...
0 sin θ 0 . . .
A = cos θ · Sp,ϕ , ∆ = Sq−1,ψ · 0 ,
0 sin θ . . .
..
0 0 0 .
cos ϕ1
sin ϕ1 cos ϕ2
.. ~b = − ρ sin θ · ~a,
~a = cos θ · ,
. cos θ
sin ϕ1 sin ϕ2 . . . cos ϕp
sin ϕ1 sin ϕ2 . . . sin ϕp
17
cos ψ1
sin ψ1 cos ψ2
..
ρ cos θ
~c = sin θ · , d~ = · ~c.
. sin θ
sin ψ1 sin ψ2 . . . cos ψq−1
sin ψ1 sin ψ2 . . . sin ψq−1
Lemma 26. If A is an m × m matrix, Bis an m × n matrix, Γ is an n × m matrix, and
A B
∆ is an invertible n × n matrix, then det = det(A − B∆−1 Γ) det(∆).
Γ ∆
Proof. This standard result on block matrices is proved by factoring into triangular matrices:
A − B∆−1 Γ B∆−1
A B Im,m 0m,n
= .
Γ ∆ 0n,m In,n Γ ∆
We next calculate A − B∆−1 Γ. We first compute B −1 ∆Γ, labelling the jth row of the matrix
∆−1 by r~j (horizontal vector),
~r1 ~c · ~r1 0 . . .
~r2 ~c · ~r2 0 . . .
−1 ~ ~ ~
B(∆ Γ) = B .. ~c 0 . . . = b 0 . . . ..
~ ~
= (~c · ~r1 )b 0 . . . .
. .
~rq ~c · ~rq 0 . . .
Using the multilinearity of the determinant, and factoring out the cos θ and sin θ from the
columns of A and ∆,
1
det(DΦ) = det A − B∆−1 Γ det(∆) =
det(A) det(∆)
cos2 θ
1
= (cosp+1 θ · ρ cos θ · sinq−1 θ) det(Sp,ϕ ) det(Sq−1,ψ )
cos2 θ
= ρ cosp θ sinq−1 θ det(Sp,ϕ ) det(Sq−1,ψ ).
18
Lemma 27. We have:
kXk2 − X̄X +
∇+
p,q Gp,q,ε = iε(p + q + 1) p+q+3 ,
2 2
N(X) + iεkXk
kXk2 − X + X̄
Gp,q,ε∇+
p,q = iε(p + q + 1) p+q+3 .
2 2
N(X) + iεkXk
By Lemma 12,
In other words, the above two limits of integrals define continuous linear functionals on the
topological vector space of smooth functions1 on the interval [ π4 − θ0 , π4 − θ0 ].
Proof. In the case of p + q = 1 mod 2, this is a consequence of Lemma 17 of [L]. Otherwise,
we modify the proof to fit the fractional case. We have that (p + q + 3)/2 = n + 1/2 for some
non-negative integer n, and we use induction on n. For the base case, n = 0, we integrate
by parts,
π π
+θ0 +θ0
g(θ)dθ sin(2θ) g(θ)dθ
Z Z
4 4
1/2 = 1/2
π
4
−θ0 cos(2θ) + iε cos(2θ) + iε
π
4
−θ0 sin(2θ)
Z π +θ0
1/2 g(θ) π/4+θ0
4 21 d g(θ)
= − cos(2θ) + iε + cos(2θ) + iε dθ.
sin(2θ) π/4−θ0 π
4
−θ 0
dθ sin(2θ)
1
Note that this vector space has the standard Frechét toplogy determined by the seminorms
maxθ∈[ π4 −θ0 , π4 −θ0 ] g (k) (θ), where k = 0, 1, 2, . . . .
19
1/2
As cos(2θ) + iε is integrable for all values of ε, including ε = 0, the limits as ε → 0±
exist and depend continuously on g(θ). Now we consider the case of n > 0, in which case we
can integrate by parts,
π π
+θ0 +θ0
g(θ)dθ sin(2θ) g(θ)dθ
Z Z
4 4
n+1/2 = n+1/2
π
4
−θ0 cos(2θ) + iε cos(2θ) + iε
π
4
−θ0 sin(2θ)
1/2−n π/4+θ0 Z π +θ0
cos(2θ) + iε (2n − 1)−1
g(θ) 4 d g(θ)
= − n− 21 dθ sin(2θ) dθ,
2n − 1 sin(2θ)
π
π/4−θ0 4
−θ 0 cos(2θ) + iε
for some function g(θ). Due to the transversality of the boundary ∂U with respect to the
null cone Np,q = {θ = π4 }, the function g(θ) is smooth at least for θ lying in some interval
R π +θ0
[ π4 − θ0 , π4 + θ0 ] with θ0 ∈ (0, π4 ). We can apply Lemma 28 to π4−θ0 . . . , and the limit of the
4
integral on the remainder of the interval exists, so the limit of (23) as ε → 0 is zero.
We introduce a constant Cp,q (it will be evaluated later in Lemma 33):
π
cosp ϑ sinq−1 ϑ
Z
2
Cp,q = lim+ p+q+1 dϑ. (24)
ε→0 0 cos(2ϑ) + iε 2
where
dΩn,α = ρ−n det(Sn,α )dα1 dα2 . . . dαn
20
represents the spherical volume element on the n-dimensional sphere. The factor of iq comes
from the normalization of the Euclidean volume element on this space, (16). Note that the
factors of r in the numerator and denominator of
rf (X)dSp,q
p+q+1
N(X) + iεr 2 2
Proof. Note that, as in the proof of previous lemma, once we convert to the hybrid spherical
coordinates (21), the factors of r cancel. After conversion to these coordinates, the integrand
depends on ψk as follows. For k = 1, . . . , q−1, the integrand is of the form cos ψk ·hk , where hk
is a function does not depend on ψk . And for k = q, it is of the form sin ψq−1 ·hq . Additionally,
we see from Lemma 25 that the Jacobian, as a function of ψk , is proportional to sinq−1−k ψk
21
p+q+1
for k = 1, . . . , q − 1. By absorbing the remaining factors of xj (cos(2θ) + iε)− 2 dSp,q into
∂hk
the hk , so we still have ∂ψ k
= 0, for all ε, we can rewrite the integral as
R
q−1−k
RSr cos ψk sin
ψk hk f (X)dθdϕ1dϕ2 . . . dϕp dψ1 dψ2 . . . dψq−1 if k = 1, . . . , q − 2,
S
cos ψq−1 hq−1 f (X)dθdϕ1dϕ2 . . . dϕp dψ1 dψ2 . . . dψq−1 if k = q − 1,
R r
Sr
sin ψq−1 hq f (X)dθdϕ1dϕ2 . . . dϕp dψ1 dψ2 . . . dψq−1 if k = q.
p p+q
2
X X
+
X X̄ = kXk − 2 xj x̃k ẽk ej .
j=0 k=p+1
p p+q Z
X X xj x̃k ẽk ej f (X) dSp,q
−2 p+q+1 .
j=0 k=p+1 Sr N(X) + iεr 2 2 r
22
We have proved Theorem 23 up to a constant coefficient. Indeed, combining (22) with
Lemmas 29 and 32, we obtain
Z
lim+ Gp,q,ε(X − X0 )Dp,q xf (X) = iq ωp ωq−1 Cp,q f (X0 ) if X0 ∈ U. (26)
ε→0 ∂U
And if X0 ∈ / U the same argument shows that the limit of integrals is zero. Then Theorem
23 follows from Lemma 33 below.
Lemma 33.
ωp+q
Cp,q = (−i)q .
ωp ωq−1
Proof. We prove this by applying both integral formulas to the constant function f (X) = 1
on the sphere Sp,q = {X ∈ Rp+q+1 : kXk = 1}. When we apply the first integral formula
(Theorem 20), we obtain, for all ε > 0 sufficiently close to zero,
Z + Dp+q z
Z
p+q+1 = ωp+q .
(hε )∗ (Sp,q ) N(Z) 2
We calculate the pullback of Dp+q z and rewrite this as an integral over the sphere Sp,q . Note
that hε is a linear transformation and that
(
∗ (1 + iε)dzj if 0 ≤ j ≤ p,
(hε ) dzj =
(1 − iε)dzj if p + 1 <≤ j ≤ p + q.
Using the expansion of Dp+q z in the standard basis (14), we find:
p p+q
X X
(hε )∗ Dp+q z = (−1)j ej (1 + iε)p (1 − iε)q dẑj + (−1)j ej (1 + iε)p+1(1 − iε)q−1 dẑj
j=0 j=p+1
= Dp+q z + εD̃p+q,εz,
where the form D̃p+q,ε z depends on ε polynomially and does not depend on Z. Let D̃p,q,ε x
be the restriction of D̃p+q,ε z to Rp+q+1 . Recall the conjugation (8). If X ∈ Rp+q+1 and
Z = hε (X), then, using (20),
(hε )∗ Z + = (X + iεX̄)+ = X + + iεX̄ + ,
(hε )∗ N(Z) = (1 − ε2 )N(X) + iεkXk2 = (1 − ε2 )N(X) + iε.
We have:
Z + Dp+q z (X + + iεX̄ + )(Dp,q x + εD̃p,q,εx)
Z Z
p+q+1 = p+q+1
(hε )∗ (Sp,q ) N(Z) 2 Sp,q (1 − ε2 )N(X) + iε 2
X + Dp,q x εX̄ + D̃p,q,ε x + iεX̄ + (Dp,q x + εD̃p,q,εx)
Z Z
= p+q+1 + p+q+1 .
Sp,q (1 − ε2 )N(X) + iε 2 Sp,q (1 − ε2 )N(X) + iε 2
As ε → 0+ , the first integral approaches iq ωp ωq−1 Cp,q by the already established formula
(26). And the second integral approaches zero because we can factor out ε and proceed in
the same manner as we proved Lemma 29 using the hybrid coordinates (21) and Lemma
28.
23
References
[BDS] F. Brackx, R. Delanghe, F. Sommen, Clifford Analysis, Pitman, London, 1982.
[FL] I. Frenkel, M. Libine, Split quaternionic analysis and the separation of the series
for SL(2, R) and SL(2, C)/SL(2, R), Advances in Math 228 (2011), 678-763; also
arXiv:1009.2532.
[G] D. Garling, Clifford Algebras: An Introduction, London Math. Society Student Texts,
Cambridge Univ. Press, 2011.
[GM] J. Gilbert, M. Murray, Clifford Algebras and Dirac Operators in Harmonic Analysis,
Cambridge Univ. Press, Cambridge, UK, 1991.
[P] B. Prather, Split Octonionic Cauchy Integral Formula, Adv. Appl. Clifford Algebras
29 (2019), issue 5.
[R1] J. Ryan, Complexified Clifford analysis, Complex Variables, Theory and Appl. 1
(1982), 119-149.
[R2] J. Ryan, Cells of harmonicity and generalized Cauchy integral formulae, Proc. Lon-
don Math. Soc. (3) 60 (1990), no. 2, 295-318.
[S] A. Sudbery, Quaternionic analysis, Math. Proc. Camb. Phil. Soc. 85 (1979), 199-225.
24