GENERALIZATIONS OF COMPLEX ANALYSIS BANACH CENTER PUBLICATIONS, VOLUME 37
INSTITUTE OF MATHEMATICS POLISH ACADEMY OF SCIENCES
WARSZAWA 1996
PAIRS OF CLIFFORD ALGEBRAS OF THE HURWITZ TYPE
W I E S l A W K R ´ O L I K O W S K I
Institute of Mathematics, Polish Academy of Sciences, L´ od´ z Branch Narutowicza 56, 90-136 L´ od´ z, Poland
Abstract. For a given Hurwitz pair [S(Q
S), V (Q
V), ◦] the existence of a bilinear mapping
? : C(Q
S) × C(Q
V) → C(Q
V) (where C(Q
S) and C(Q
V) denote the Clifford algebras of the quadratic forms Q
Sand Q
V, respectively) generated by the Hurwitz multiplication “◦” is proved and the counterpart of the Hurwitz condition on the Clifford algebra level is found. Moreover, a necessary and sufficient condition for “?” to be generated by the Hurwitz multiplication is shown.
1. Introduction. The general Hurwitz problem was studied e.g. by Lawrynowicz and Rembieli´ nski [2-4]. They introduced the notions of “Hurwitz pairs” and “pseudo- Hurwitz pairs” and gave their systematic classification according to the relationship with real Clifford algebras. In the present work we show the existence of a bilinear mapping
?: C(Q
S) × C(Q
V) → C(Q
V), where (S, V , ◦) is a given Hurwitz pair which makes the following diagram:
(1)
S × V V V
C(Q
S) × C(Q
V) C(Q
V)
◦ (Hurwitz multiplication)
//
iS×iV
iV?//
commutative.
Moreover, we prove that if such a mapping exists and satisfies the following “algebraic Hurwitz condition”: N (x
S? y
V) = N (x
S)N (y
V) for any x
S∈ Γ
Sand y
V∈ Γ
V, where Γ denotes the Clifford group of the Clifford algebra C(Q) and N is a spinor norm, then ? is generated by the Hurwitz multiplication, i.e. ?
|S×V= ◦. An example of a mapping ? which does not satisfy the N -norm condition is given. Since in the meantime the detailed proofs have appeared in [1], they are only sketched here.
1991 Mathematics Subject Classification: Primary 15A66, Secondary 30G35.
Research supported by the KBN grant PB 2 1140 91 01.
The paper is in final form and no version of it will be published elsewhere.
[327]
328
W. KR ´OLIKOWSKI2. Product of Clifford algebras generated by the Hurwitz multiplication.
Let (S, V , ◦) be a Hurwitz pair. Suppose that the vector spaces S and V are equipped with non-degenerate quadratic forms Q
Sand Q
V, respectively. We will only consider the elliptic and hyperbolic cases (see, e.g. [2-4]). In S and V we choose some bases (
α) and (e
j) with α = 1, . . . , p = dim S; j = 1, . . . , n = dim V . Assume that p ≤ n.
Let C(Q
S) (resp. C(Q
V)) denote the Clifford algebra of (S, Q
S) (resp. (V, Q
V)).
There are canonical injections i
S: S → C(Q
S) and i
V: V → C(Q
V). Then we get the diagram (2). It would be interesting to complete the diagram (2) by the suitable mapping C(Q
S) × C(Q
V) → C(Q
V). Define the following mapping ? : C(Q
S) × C(Q
V) → C(Q
V) by:
(2)
1
S? y
V:= y
V,
(
i1. . .
ir) ? (e
j1. . . e
jk) :=
e
jk. . . e
jr+1(
ir◦ e
jr) . . . (
i1◦ e
j1), r < k, (
ir◦ e
jr) . . . (
i1◦ e
j1), r = k,
ir◦ [
ir−1◦ [. . . ◦ [
ik+1◦[(
ik◦ e
jk) . . . (
i1◦ e
j1)] . . .], r > k, (
i1. . .
ir) ? 1
V:= k
i1k . . . k
irk1
Vfor 1 ≤ r ≤ p, 1 ≤ i
1< . . . < i
r≤ p; 1 ≤ k ≤ n, 1 ≤ j
1< . . . < j
k≤ n. Then, the required mapping ? : C(Q
S) × C(Q
V) → C(Q
V) is defined by the bilinear extension of (2).
R e m a r k. If (S, Q
S) is a Euclidean vector space then all k
ik
2> 0. In this case the Clifford algebras C(Q
S) and C(Q
V) are considered to be real. But, if (S, Q
S) is a pseudo-Euclidean vector space then there are some
i1, . . . ,
ir, 1 ≤ r ≤ p, such that k
isk
2< 0, 1 ≤ s ≤ r. This time the Clifford algebras have to be treated as complex ones.
Proposition. ? is a well defined bilinear mapping. Moreover , ?
|S×V= ◦, the Hurwitz multiplication, i.e. the diagram (1) is commutative.
Lemma. Let x
S∈ Γ
Sand y
V∈ Γ
V, where Γ
S(resp. Γ
V) denotes the Clifford group in C(Q
S) (resp. C(Q
V)) and let N
S, N
Vbe the spinor norms in C(Q
S) and C(Q
V), respectively. Then
(3) N
V(x
S? y
V) = N
S(x
S)N
V(y
V).
Theorem. Let S and V be real vector spaces equipped with non-degenerate quadratic forms Q
Sand Q
V, respectively. Denote by C
C(Q
S) (resp. C
C(Q
V)) the complex Clifford algebras of (S, Q
S) (resp. (V ,Q
V)). Suppose that there is a bilinear mapping ? : C
C(Q
S)×
C
C(Q
V) → C
C(Q
V) satisfying the condition (3). Then ? is generated by the Hurwitz multiplication, i.e. ?
|S×V= ◦, where ◦ : S × V → V is a bilinear mapping such that ks ◦ vk
V= ksk
Skvk
Vfor all s ∈ S and v ∈ V .
P r o o f. Let s ∈ S ⊂ Γ
Sand v ∈ V ⊂ Γ
V. By definition of N we have (4) N
V(s ? v) = N
S(s)N
V(v) = ksk
2Skvk
2V∈ R.
Let (e
1, . . . , e
n) be an orthogonal base in V . Suppose s ? v = a
0+
n
X
i=1
a
ie
i+
n
X
l=2
X
i1<...<il
a
li1...ile
i1. . . e
il.
PAIRS OF CLIFFORD ALGEBRAS OF THE HURWITZ TYPE
329
Then
N (s ? v) = a
20+
n
X
i=1
(a
i)
2Q
V(e
i) +
n
X
l=2
X
i1<..<il
(a
il1..il)
2Q
V(e
i1) . . . Q
V(e
il) + R(e
1, . . . , e
n), where
R(e
1, . . . , e
n) = X
b
ie
i+ X
i<j
b
ije
ie
j+ . . . + X
i1<...<im
b
i1...ime
i1. . . e
im+ be
1. . . e
n. Since N (s ? v) is a scalar then R(e
1, . . . , e
n) must vanish. The multiplication ? is bilinear so the coefficients a
0, a
iand a
il1...ilare bilinear functions in s and v. Thus N (s ? v) should be separated into two parts, first depending only on s and second only on v. Then we can write
a
20+
n
X
i=1
(a
i)
2Q
V(e
i) +
n
X
l=2
X
i1<...<il
(a
il1...il)
2Q
V(e
i1) . . . Q
V(e
il)
= ksk
2Skvk
2V= ksk
2S[c
20+
n
X
i=1
(c
i)
2Q
V(e
i) +
n
X
l=2
X
i1<...<il
(c
il1...il2Q
V(e
i1) . . . Q
V(e
il)]
Thus, the following equality has to be satisfied:
c
20+
n
X
i=1
(c
i)
2Q
V(e
i) +
n
X
l=2
X
i1<...<il
(c
il1...il)
2Q
V(e
i1) . . . Q
V(e
il) =
n
X
i=1
(v
i)
2Q
V(e
i).
The coefficients c
0, c
i, c
il1...ilare linear in v so, by continuity, we can write c
0(v) = c
0jv
j, c
i(v) = c
ijv
j, c
il1...il(v) = c
ilj1...ilv
j. Thus, for any 1 ≤ j, k ≤ n we get the identity
c
0jc
0k+
n
X
i=1
(c
ijc
ik− δ
jiδ
ki)Q
V(e
i) +
n
X
l=2
X
i1<...<il
c
ilj1...ilc
ilk1...ilQ
V(e
i1) . . . Q
V(e
il) ≡ 0.
Take an orthogonal transformation R ∈ O(Q
V). In a new base e
0= Re we have c
0jc
0k+
n
X
i=1
( e c
ije c
ik− δ
ijδ
ik)Q
V(Re
i) +
n
X
l=2
X
i1<...<il
e c
ilj1...ile c
ilk1...ilQ
V(Re
i1) . . . Q
V(Re
il) ≡ 0.
But Q
V(Re
i) = Q
V(e
i). Then the new coefficients e c
jand e c
ilj1...il, obtained by the changing of the base, satisfy the same identity as the previous ones. This is possible if and only if
c
0j≡ 0 for j = 1, . . . , n,
c
ijc
ik− δ
jiδ
ki≡ 0 for 1 ≤ i, j, k ≤ n,
c
ilj1...il≡ 0 for l = 2, . . . , n; 1 ≤ i
1< . . . < i
l≤ n; j = 1, . . . , n.
Thus, we get s ? v = ksk
SP
ni=1,j=1
c
ijv
je
i∈ V and ksk
2Skvk
2V= N
V(s ? v) = ks ? vk
2V, so
?
|S×Vsatisfies the Hurwitz condition, as required.
Example. We now construct a bilinear map : C
C(Q
S) × C
C(Q
V) → C
C(Q
V)
which does not satisfy the condition (4). Choose some bases (
α) and (e
j) in S and V ,
330
W. KR ´OLIKOWSKIrespectively. Define
(5)
1
S 1V := e
1. . . e
n, 1
S (ei1. . . e
ik) := e
i1d . . . e
ik,
. . . e
ik) := e
i1d . . . e
ik,
1
S (e1. . . e
n) := 1
V,
(
j1. . .
jr) (e
i1. . . e
ik) := k
j1k . . . k
jrk e
i1d . . . e
ik, (
j1. . .
jr) 1
V:= k
j1k . . . k
jrke
1. . . e
n, (
j1. . .
jr (e1. . . e
n) := k
j1k . . . k
jrk1
V, where “ b .” is defined by
e
i1d . . . e
ir:= e
j1. . . e
jswith j
1< . . . < j
sand (i
1, . . . , i
r, j
1, . . . , j
s) = (1, . . . , n).
The map : C
C(Q
S) × C
C(Q
V) → C
C(Q
V) is defined by the bilinear extension of (5).
It is easy to see that does not satisfy the condition (4). Indeed, take s ∈ S and v ∈ V . We have
s v = s
αv
iα
ei= s
αv
ik
αk
Se
1. . . b e
i. . . e
n6∈ V.
and
N
V(s v) = s
αs
βk
αk
Sk
βk
S(v
i)
2Q
V(e
1) . . . Q d
V(e
i) . . . O
V(e
n).
Suppose that N
V(s v) = N
S(s)N
V(v). Then we get X
α,β
s
αs
βk
αk
Sk
βk
S= X
α
(s
α)
2k
αk
2S, X
i
(v
i)
2Q
V(e
1) . . . Q d
V(e
i) . . . Q
V(e
n) = X
i