Outer Pro duct Factorizati on in Cli ord
Algebra
Hongb o Li, Yihong Wu
Institute of Systems Science
Academia Sinica
Beijing 100080, P.R.China
hli, [email protected]
Abstract
Cli ord algebra plays an imp ortant role in mathematics and physics,
and has various applications in geometric reasoning, computer vision
and rob otics. When applying Cli ord algebra to geometric problems,
an imp ortanttechnique is parametric representation of geometric en-
tities, such as planes and spheres in Euclidean and spherical spaces,
which o ccur in the form of homogeneous multivectors. Computing a
parametric representation of a geometric entity is equivalenttofac-
toring a homogeneous multivector into an outer pro duct of vectors.
Such a factorization is called outer pro duct factorization.
When no signature constraint is imp osed on the vectors whose
outer pro duct equals the homogeneous multivector, a classical result
can b e found in the b o ok of Ho dge and Pedo e 1953, where a su-
cient and necessary condition for the factorability is given and called
quadratic Pluc ker relations p-relations. However, the p-relations are
generally algebraically dep endent and contain redundancy. In this
pap er we construct a Ritt-Wu basis of the p-relations, which serves
asamuch simpli ed criterion on the factorability. When there are
signature constraints on the vectors, we prop ose an algorithm that
can judge whether the constraints are satis able, and if so, pro duce a
required factorization.
1 Intro duction
Cli ord algebra is an imp ortant to ol in mo dern mathematics and physics.
Because of its invariant representation for geometric computation, Cli ord
algebra is gaining wider and wider recognition in elds like geometric reason-
ing, computer vision and rob otics. This short pap er contributes to solving
an often encountered problem in applying Cli ord algebra, the problem of
outer pro duct factorization.
n
Let K b e a eld whose characteristic 6= 2, and let V be a K -vector space of
n n
dimension n. The Grassmann algebra V generated by V is a graded K -
vector space, whose grades range from 0 to n.Themultiplication is denoted
by\^", called the outer pro duct. An element in the Grassmann algebra
is called a multivector, and an r -graded element is called an r -vector. Any
r -vector is called a homogeneous multivector.
n
Let fe ;e ;:::;e g b e a basis of V . It generates a basis fe ^ ^ e j 1
1 2 n i i
r
1
r n
i < ::: < i ng for the space V of r -vectors. Let A be an r -vector,
1 r r
then
X
: 1 e ^ ^ e A = a
i i r i :::i
r r
1 1
1i <:::
r
1
The list a j 1 i < ::: < i n is called the Pluc ker co ordinates of
i :::i 1 r
r
1
A . In this pap er, weallowany order among the suxes by requiring that
r
a be anti-symmetric with resp ect to its suxes.
i :::i
r
1
The outer pro duct of r vectors is called an r -extensor. For an r -extensor
n
A = a ^ ^ a , where the a's are vectors, a vector x 2V is in the
r 1 r
subspace A spanned bythea's if and only if x ^ A = 0. Therefore wecan
r r
use A to represent the space A . Since an extensor represents a subspace,
r r
an imp ortant question is how to judge whether or not a homogeneous multi-
vector is an extensor, and how to factor an extensor. The factorization will
b e unique up to a sp ecial linear transformation in the subspace.
In the b o ok of Ho dge and Pedo e 1953, there are two theorems that answer
the ab ove question:
Theorem 1.1. An i-vector A is an extensor if and only if its Pluc ker co-
r
ordinates a =a j 1 l ;:::;l n satisfy the following quadratic
l :::l 1 r
r
1
Pluc ker relations p-relations: for any1 i < ::: < i n and any
1 r 1
1 j <::: 1 r +1 r +1 X F a: 1 a a =0: 2 i :::i ;j :::j i :::i j j :::j j :::j 1 r 1 1 r +1 1 r 1 1 r +1 1 +1 =1 Theorem 1.2. Let A be an r -extensor with Pluc ker co ordinates a j r i :::i r 1 T 1 i ;:::;i n, where a 6=0. Let b =b ;:::;b , where 1 r l :::l j l 1 l n r 1 j j b = a : Then l k l :::l kl :::l r j 1 j 1 j +1 1 r A =a b ^ ^ b : 3 r l :::l 1 r r 1 In the b o ok of Iversen 1992, where the Grassmann algebra is replaced by a nondegenerate Cli ord algebra, Theorem 1.1 is reformulated elegantly as follows, where the dot denotes the inner pro duct in the Cli ord algebra: Theorem 1.3. An r -vector A is an extensor if and only if for anyr 1- r vector X , r 1 A ^ A X =0: 4 r r r 1 The set of p-relations 2 forms a Grobner basis for a convenient monomial order. Furthermore, it has a structure of Ho dge algebra see DeConcini, Eisenbud and Pro cesi, 1982. However, the p-relations are generally alge- braically dep endent, and contain redundancy when used as a criterion on the factorability of a homogeneous multivector. In this pap er, we construct a Ritt-Wu basis Wu, 1978, i. e., an irreducible characteristic set, of 2. The numb er of relations in the basis is much smaller than that in 2. The basis not only simpli es the criterion on the factora- bility, but also yields an ecientway to reduce a p olynomial of Pluc ker co ordinates by the Pluc ker relations. n n When V is equipp ed with an inner pro duct, the Cli ord algebra C V n n generated by V cf. Crumeyrolle, 1990 is linearly isomorphic to V . When K = R, a nonzero vector x is said to b e p ositive, or negative, or null, if x x>0, or < 0, or = 0, resp ectively. The sign of x x is called the signature of x.LetA = a ^ ^a , where the a's are mutually orthogonal r 1 r vectors and p of which are p ositive, q of which are negative. The triplet p; q ; r p q is called the signature of A . r The problem of outer pro duct factorization in Cli ord algebra is, given an r -extensor, factor it into the outer pro duct of p p ositivevectors, q negative ones and r p q null ones, if the factorization is p ossible. An algorithm is prop osed in this pap er to solve the problem. 2 Factorization in Grassmann Algebra Let A b e a nonzero r -vector with Pluc ker co ordinates a j 1 l ;:::;l r l :::l 1 r r 1 T n. Assume that A = a ^ ^ a ,wherea =a :::a with resp ect to r 1 r i i1 in n the basis fe ;:::;e g of V .Then 1 n a a 1i 1i r 1 . . . . . . a = : 5 . i :::i . . r 1 a a ri ri r 1 Since A is nonzero, at least one of its Pluc ker co ordinates is nonzero. We r assume that a 6= 0, otherwise we simply change the suxes of the basis 1:::r fe ;:::;e g to achieve this. Denote a = a , a = a . First _ 1 n 0 1:::r 1:::i 1j i+1:::r j i we derive the classical p-relations. The r -extensor A can b e represented bytherowvectors of the r n matrix r 0 1 0 1 T a a a 11 1n 1 B C B C . . . . B C B C . . . . A = = : 6 . . . . @ A @ A T a a a r 1 rn r Since a 6= 0, the matrix 0 0 1 a a 11 1r B C . . . B C . . . A = 7 . 0 . . @ A a a r 1 rr