MULTIVECTOR DIFFERENTIAL

CALCULUS
Eckhard M.S. Hitzer
Department of Mechanical Engineering,
Fukui University,
3-9-1 Bunkyo, 910-0024 Fukui, Japan,*

(Received: September 25, 2002; Accepted: November 03, 2002)

Abstract. Universal geometric calculus simplifies and unifies the structure and no-
tation of mathematics for all of science and engineering, and for technological appli-
cations. This paper treats the fundamentals of the multivector differential calculus
part of geometric calculus. The multivector differential is introduced, followed by the
multivector derivative and the adjoint of multivector functions. The basic rules of
multivector differentiation are derived explicitly, as well as a variety of basic multi-
vector derivatives. Finally factorization, which relates functions of vector variables
and multivector variables is discussed, and the concepts of both simplicial variables
and derivatives are explained. Everything is proven explicitly in a very elementary
level step by step approach. The paper is thus intended to serve as reference material,
providing a number of details, which are usually skipped in more advanced discus-
sions of the subject matter. The arrangement of the material closely follows chapter
2 of [3].

But the gate to life is narrow and the way that leads to it is hard, and
there are few people who find it. ... I assure you that unless you change
and become like children, you will never enter the Kingdom of heaven.
Jesus Christ [1]

... for geometry, you know, is the gate of science, and the gate is so
low and small that one can only enter it as a little child.
William K. Clifford [2]

" e-mail: hitzer©mech.fukui-u.ac.jp, homepage: http://sinai.mech.fukui-u.ac.jp/

Advances in Applied Clifford Algebras 12 No. 2, 135-182 (2002)
136Multivector Differential Calculus E. M.S. Hitzer

1, Introduction

The work' is presented as a series of definitions (D. or Def.), propositions

(P.) and Remarks. All propositions (except P. 76) are explicitly proved in a
sometimes elaborate step by step procedure. Such a degree of explicitness seems
to be lacking in the literature on the subject 2 which I have come across so far.
This motivated me to undertake this study. It is only assumed that the readers
are familiar with geometric (multivector) algebra, as presented in chapter 1 of
[3], the introductory sections of [6], [7] or in numerous other publications on
geometric (Clifford) algebra.
The presentation closely follows the arrangement in chapter 2 of [3]. It is
in fact the major purpose of this work to help non-experts work thoroughly
through such a text. There is therefore no great claim to originality, apart from
the hope that it will assist and motivate non-experts and new-comers to delve
into the material and become confident of comprehensively understanding and
mastering it. The multivector derivative is to some extent just a generalization
of the vector derivative. I encourage non-experts and new-comers therefore to
study [6], because it contains a similar study restricted to the vector derivative
of multivector functions 3 .

In the next section we will start with introducing the multivector differ-
ential of multivector functions, i.e. functions with arguments and ranges
in the universal geometric algebra 9(I) with unit pseudoscalar I. This will
be followed by sections defining the multivector derivative and the adjoint of
multivector functions. Both the multivector differential and the adjoint can be
understood as two linear approximations associated pointwise to each contin-
uously differential multivector function.
Conventional scalar differential calculus does not distinguish the three con-
cepts of multivector derivative, differential and adjoint, because in the scalar
case this distinction becomes trivial.
Standard definitions of continuity and scalar differentiability apply to multi-
vector-valued functions, because the scalar product determines a unique "dis-
tance" (A — B] between two elements A, B E 9(I).

1 First presented in the Geometry session of the 6th Int. Clifford Conference,
Cookeville, 23 May 2002.
2 Compare e.g. chapter 2 of [3], the section on multivector calculus of [4] or [5].
3 In this context compare also section 2-8 of [8] on directional derivatives, section
1I-2 of [9] on vector derivatives and differentials, section V-20 of [10] on differentiation
with its applications throughout [10], chapter 2 of [11] on geometric calculus, and
section 5 of [12] on the vector derivative.
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 137

2. Multivector Differential
Definition 1. (differential, A-derivative). For a multivector function F =
F(X) defined on the geometric algebra C9(I):
F : X E 9(I) , F(X) E c(I), I = (I) n , ( 2.1)

and for a multivector A and the projection P(A) = (A• I) I the A-derivative
(or differential) is defined by:
F(X +TP(A)) — F(X)
A * 0xF(X) - 8 F(X +TPA lim (2.2)

* signifies the scalar product defined in [31, chap. 1, p. 13, (1.44). By its defi-
nition the A- derivative is a linear function of A denoted in various ways as:
F=F(A) =F(X,A) =FA(X) =A*8XF(X) = A* 8F. (2.3)

Remark 2. Note the important convention, that inner (•), outer (A) and
scalar (*) products have priority over geometric products. If X is restricted
to a certain subspace of 9, then the projection P in definition 1 becomes the
projection into that subspace. This is e.g. the case at the beginning of section
7.
Proposition 3.
F(A) = F(P(A)). (2.4)
Proof 3.
E(A)
Dell ar F(X + TP(A))J,=o
Def.1
= 8T F(X +TP(P(A)))IT =0 F(P(A)). (2.5)
Proposition 4 (grade invariance).
A * 8(F) r = (A * 8F) r = ( F(A)) r . ( 2.6)

Proof 4.
r D=.l
A * a (F) 8i (F (X + TP (A))) r I T =o
(F(X +'rP(A))) r — (F(X)) r
= lim
r-0 T
T s ca lar / F(X + rrP(A)) — F(X) \
lim
'r-+0 T r
Def.1 Def.1
(A * 8F)r (F(A)) r . (2.7)
138 Multivector Differential Calculus E. M.S. Hitzer

Proposition 5.

F(A + B) = F(A) + F(B). (2.8)

Proof 5.

F(A + B) Def.1
li o F(X + TP(A+T B)) — F(X)
linear ity of P lim F(X +TP(A) +TP(B)) — F(X)
= lim
r—^0 T r—O

f F(X +TP(A) +TP(B)) — F(X +TP(B)) + F(X +TP(B)) — F(X)

T

X r - X +TP(B), Def. 1 lim F(Xr +TP(A)) — F(X)

+ F(B)
r—^0 T

line .o Xr= X Def. 1

, Def.l
A * aXF(X) + F(B) F(A) + F(B). (2.9)

An alternative rigorous proof of P. 5 is the following: I first define a function

eF(X, A, r), which is according to Def. 1 continuous at r = 0

—( F(X+rP(A))—F(X)
r — F( XA ) T 0
6 F(X,A,T) = I (2.10)
0 T=0.

We hence have

F(X + TP(A)) (2 10) F(X) +TF(X, A) + TEF(X, A, T) (2.11)

and

linearity of P
F(X +TP(A+B)) F(X +TP(A)+TP(B))

(2.=1)
F(X +TP(A)) +TF(X +TP(A),B) +TEF(X +TP(A), B, r)
(2.11)
F(X)+TF(X,A)+TrF(X,A,T)
+TF(X +TP(A), B) +TEF(X +TP(A), B,T ). (2.12)
Advances in Applied Clifford Algebras 12, No. 2 (2002) 139

We can now calculate F(X, A + B) according to Def. 1

F(X + i-P(A+ B)) - F(X) (2.)
F(X, A + B)
{F(X, A) + eF(X, A, i ) + F(X + rP(A), B) + rF(X + rP(A), B, r)}
-

= lim{F(X, A) + EF(X, A,i ) + F(X, B) + rF(X, B, r)}

-

20 F(XA) +E(X,B).(2.13)
Proposition 6. For constant scalar A

F(AA) = AF(A). (2,14)

Proof 6.
Def.1 F(X + rAP(A)) - F(X)
F(AA) = lim
- r
AOO F(X + rAP(A)) - F(X)
- lirnA
,r—.O 'FA
F(X + riP(A)) - F(X) D.1
- AIim AF(A), (115)

in case of A = 0 we simply have

Def.1 F(X)F(X)
F(OA) = F(0) = lim = 0 = OF(A). (2.16)
r
Proposition 7 (sum rule). For two multivector functions F, G on

(2.17)

Proof 7.

A * 5x(F + C)
Def.1
a, [F(X+ TP(A)) + C(X + i-P(A))] (2.18)
DI
= ôr F(X + rP(A)) + &r G(X + rP(A)) ' A * 8F + A * aG.
Proposition S (product rule). For two multivector functions F, C on

A*0x(FG) = (A*0xF)C+F(A*axG). (2.19)

140Multivector Differential Calculus E. M.S. Hitzer

Proof 8.

Def•1 1 0 F(X +TP(A))G(X +TP(A)) — F(X)G(X)

A * 8X(FG)
F(X +TP(A))G(X +TP(A)) — F(X)G(X +TP(A))
= lim
T—^0

+F(X)G(X +TP(A)) — F(X)G(X)

T
( F(X +TP(A)) — F(X)
=li rn G(X +TP(A))+
G(X +TP(A)) — G(X)
+F(X)
T

Def.1
(A * axF) limo G(X + TP(A)) + F (A * 8XG)
_ (A*0 F)G+F(A*aXG). (2.20)

Proposition 9 (constant function). For B independent of X:

A*axB=0. (2.21)

Proof 9. Def. 1 gives for F = B:

A*aXB= lim B—B = 0. (2.22)

T-0 'r

Proposition 10 (identity).
A * axX = P(A). (2.23)

Proof 10. For F(X) = X:

A * axX
D = li o X +TP(A) — X = P(A). (2.24)

Proposition 11. For B independent of X:

A * 0x(X * B) = P(A) * B. (2.25)
Advances in Applied Clifford Algebras 12, No. 2 (2002) 141

Proof 11.
Def. of scalar prod P. = 8
A * ax (x * B) .A * ôx(XB)a (A * ÔxXB)o (2.26)
P.
((A * 8xX)B + XA * 8B)o '°((P(A)B + 0) = P(A) * B.
Proposition 12 (Taylor expansion).

F(X + P(A)) V 1 (A * ax)kF = exp(A * 8x)F. (2.27)

Proof 12.
dG(0) Def.1
G(r) F(X + rP(A)) dr = 8 F(X + 7P(A))j r0 = A *
(2.28)

d2C(0)
Def. of G = ôr (Or F(X+rP(A)))L_ o(2.29)
dr 2
Def.1 D ef.1
A8 (3F(X+rP(A)))l 0(A*13x)2F,

In general: dkG(0) = (A * ax )kF. (2.30)

dr"
Taylor expansion of C:

dG(0) 1 d 2 C(0) 1 dcG(0)

G(1)=C(0+1)=G(0)+ dr +...=V1 di-k
(2.31)
dr2
k0

C(1) = F(X + P(A)) = V (A * ax )kF = exp(A * 8x)F. (2.32)

Proposition 13 (chain rule).

f : X E GM - E g(I') with 1' = (I') (2.33)

The composite function F(X) = G(f(X)) has the differential
A * OF = A * OC(f(X)) = f(A) * 9C (2.34)
142Multivector Differential Calculus E. M.S. Hitzer

Proof 13. The Taylor expansion (P.12) of f gives:

f(X+rrP(A))= f(X)+(TA*8x)f+ 1 (rA*8x) 2 f +...

Def.1
.f (X) + .f ( rA) + 2 (rA * 8x) 2 f + .. .

P=6 f(X)+rf(A)+ Z (A*ax) 2 f +... (2.35)

Therefore
Def.1
A * 8xF = A * axG(f (X )) ar G(f (X + TP(A))) I r =o
Taylor ar
G(f(X) + Tf(A))I r =o
range of f_f in (I') a
r G(f (X) + r P(f (A))) I
-
r=o
Def.1
f(A) * ax G(X')I x=f(x) . (2.36)

Remark 14. Proof 13 employs the Taylor expansion (P. 12). Yet there are
continuously differentiable functions without a Taylor expansion. But the chain
rule P. 13 will still apply as long as a function f has the linear approximation
f (X + TP(A)) ,= f (X) + r f (A) for sufficiently small values of 'r. 4
I therefore want to give another proof for the chain rule based on the lin-
earization of multivector functions with the help of the differential. One such
linearization is already given in (2.10) and (2.11). In the same way I define
f(x+rP(A))—f(x)
— f (X, A) T 0, (2.37)
^f (X> A>T) = 1 0 T=0,

Hence we have the linearization of f : X E 9 -- f (X) E 9

(2.37)
f (X + TP(A)) f(X) + T f (X, A) +ref (X, A, r). (2.38)

I further define
^ c(I' B T ) _ I
G(Y+rP(B)) — G(Y)

0 r=0,
T _ G(Y B) T 0
(2.39)

4 I owe this remark to Dr. H. Ishi, of Yokohama City University.

Advances in Applied Clifford Algebras 12, No. 2 (2002) 143

and hence
(239)
G(Y +TP(B)) G(Y) +TG(Y, B) +TSC(Y, B, T). (2.40)

According to Def. 1 both functions e f and eG are continuous at 7 = 0. Consid-

ering that f (X) E g D = 1 f E G ( ) e f E 9 (the same is valid for a function
f on a linear subspace of G), we can now rewrite F(X) = G(f (X)) as

G(f (X +TP(A))) (2.38) G(f (X) +T{ f (X, A) + e f(X, A, r)})

f '^f E G(f(X)+TP(f(X,A)+ef(X,A,rr)))
(2.40)
G(Y = f (X)) + rrG(Y = 1(X), B = f (X, A) + E1(X, A, r))
+'rec(f (x), f (X, A) + e1(X, A, rr), rr). (2.41)

Subtracting G(f (X)), deviding by rr and taking the lim r _.., o we get according
to Def. 1 the differential of the composite function F(X) = G(f (X))

G(f (X +TP(A))) — G(f (X))

lim = lim
r—+0 T r— O

{G (f(X), f(X,A)+e f (X,A,rr)) + 6 G(f(x),f(X,A)+ef(X,A,rr),T)}

(2.37),(2.36)
G(f (X), f (X, A)). (2.42)

This concludes the alternative proof for the differential of composite functions
(chain rule).

Definition 15 (second differential).

FAB = FAB (X) ; B * aA * 8F. (2.43)

The over-dot notation indicates, that 8 acts only on F.

Proposition 16 (integrability). Let F(X+ rP(A)+crP(B)) be a twice conti-

nously differentiable function in the vicinity of X, i.e. for all values of 0 < 7
and 0 < a <a 1 . Then

FAB(X) = FBA(X). (2.44)

 144 Multivector Differential Calculus E. M.S. Hitzer

Remark 17. To motivate the slightly more elaborate proof of P. 16 which

follows, I present this "handwaving" argument:

FAB(X)
DeL15
B * 8A * aF
Def.1
B * a (a F(X + r P(A))
T=O
Def.1 9
,,T F(X + TP(A) + o P(B)) I T= , =0(2.45)
F(X+rP(A)+oP(B))—F(X+TP(B)) _ F(X+rP(A)—F(x))
= lim limT T = lim
v-, 0 T-+0 o. o--.0

F(X +TP(A) + aP(B)) - F(X +TP(B)) - F(X +TP(A)) + F(x)

l im
T-O Qr

The last expression is symmetric under 5 the exchange (P(A),T) — (P(B), a).
Hence
FAB(X) = B * aA * aE(X) = A * 8B * aF(X) = FBA(X). (2.46)

Proof 16. According to the fundamental theorem of calculus ([151, 216.0

and [16], p. 140, Satz 3) we have, for the interval [0, T 1 ], continously differen-
tiable functions g(r)

9(T1) - g( 0 ) = f T 1 9T9(71)d^. (2.47)

0

We now use the in T and a twice continuously differentiable function

f (T, a) - F(X +TP(A) + aP(B)) (2.48)

to define
M(ri , Q i ) = f (Ti , a i ) - f (ri , 0) - f (0, a i ) + f (0, 0). (2.49)
Following the notation of (2.45) we can therefore write
M(Ti, o'i)
FAB = lim ( lim 1 = alarf I T_a-o = 8 81- f (0, 0), (2.50)
v1-0 Tj-.O a1T1

FBA = lim lim

M(ri,ai) 1 = (2.51)
8T0Q.f I T=^=o = art9 f ( 0 , 0).
Tl-.0 vl-^0U1Ti

5 Remark 17 is "handwaving", because the implied interchange of the limits in a

and T is mathematically nontrivial. I am very much indebted to Dr. H. Ishi (Yoko-
hama City University, Japan) for discussing P. 16 with me and for suggesting the
proof of it given below.
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 145

Using the fundamental theorem of calculus twice the function M(rrl, o ) can -

be reexpressed on the one hand by

(2.47) 1 al
M(ri,Ql) (9af(Tl, )dS — ^^ 1 a, f( 0 , b)dS
0 0

_ 1[af (Tl, ) — f(o, e)l de (2.47) / °1 fTl aa^f(77, (2.52)

0 0 0

and on the other hand by

M(ri , o1) = f (r1, oi) — f (O Ql) — f (Tl 0 ) + f ( 0 , 0 )

(2.47) ITl
8Tf( 77, o, 1) d 'I —
1Tl "Tf('l ,O ) d = ,f(Thai)— 8Tf( 77 ,0 )) d '/
o 0 0

(2.47) [1 /mil a0af ( 7, ^)d^dr^.

7 (2.53)
o o

It follows from (2.52) and (2.53) that

M(ri, a'1) (2.52) [01 1 T

l aTa,f (
(2.53) f Tl / al
77, S)djdS 80 ,9Tf ( 17 , S) d 6 &7 .
0 0 0 0
(2.54)

The assumption of the double continous differentiability of f means that

85f(i) and ao a-r f (17, ^) are both continuous for

OG77<-ri 3 O<^<Q l . (2.55)

This in turn has the consequence that

m(rj, o 1 ) as max{o<1<T,, O<{<ol} l avarf (^), S) — warf ( 0 , 0 )I 0

for (7-11 a) --> (0, 0), (2.56)

and

n(ri ol ) as max{o<T7<T 1 ,0<^<0l} I 8T 80f (77, S) — 0Ta0f ( 0, 0)I —p0

for (711 o l ) --* (0, 0). (2.57)
146 Multivector Differential Calculus E. M.S. Hitzer

We therefore have
` M(T1,a1)
—.9Taa.f(0,0)!
Q1T1

°i Tl 1 °3 Ti
(2.52) 1
_
aTaQ.f ( 7 t, 6)di7d6 — — r t^Tay.f ( 0 , 0 )di1d6I
'IT1 0 0 CrIT1 0 0
1a^ Ti

< ff Tl a (aT ,9a.f ( 71, 6) — 0T 8of 0 0 d?)d6 ( , )1

(2.57) 1 (°' 7 TL n(r1, o'i)di d6

r r """"
011 T1 ° o

aI Ti
= n( 1,u1) ° °
7 = n(Tl,orl). (2.58)
171Ti

Because of (2.57) we then get

lim I M(Tl, 1)
_ 8T8rf(0, 0)t (2.57)
0. (2.59)
{Ti,oi)--i(°>°) a1T1

Likewise we have
M(ri,171)
_.9daTf(0, 0)I
I a1T1
Ti
(2.53) 1 I °' IT'

C 1
-9,ar (77,^a) d r
d 1 ^ ai,a-rf( ,0 )4j
'"`?
0-17-1 0 0 O'1T1 0 0
1 fTl- fai ' 0
UOTf('/,S) -8Tf(0,0)I ddd1I
O'1 "rl r0 0
(2.56) 1 [O1
m fTi,17i )d6d71
'l i 0 0
rTl r a l
= m(Ti,al) t07 ° dsdq =m(ri,ci). (2.60)
(TiTI

Because of (2,56) we then get

M(Ti, ^1) _ 8 8 (2.56)
lim T.f ( 0 , 0)I 0. (2.61)
('rx,ai) -.(0,0) j o'i ri

Equations (2.54), (2.59) and (2.61) show that

8T87.f ( 0 , 0 ) O6 8Tf (0, 0). (2.62)
Finally (2.62) results with (2.50) and (2.51) in the proof of P. 16:
FAB(X) = FBA(X). (2.63)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 147

3. Multivector Derivative
Definition 18. The brackets r indicate the selection of the r-grade part
(... )

of the multivector expression enclosed by them.

AT = (A)r, (A) = (A)o = A0. (3.1)

Definition 19 (derivative). The derivatives of a multivector function F(X)

by its argument X

axF(X) = 8F, (3.2)

with derivative operator ax, is assumed
(i) to have the algebraic properties of a multivector in 9(I), with I the unit
pseudoscalar.

(ii) that as in definition 1: A * aX with A E 9(I) equals the differential

A *axF= FA= YAT *OXF= A*OT F= 7Ar *a r F. (3.3)

r =0 r =0 r =0
[Compare [3], p. 13, (1.46) and (1.47a).]

Remark 20. Property (ii) in Def. 19 expresses that the derivative operator
acts like a map from the space (usually C9 or a linear subspace of G) of multi-
vectors A to the space of (differentials of) multivector functions F:

OXF:AE9— >A *8XFEg. (3.4)

In order to avoid easily occuring confusions, I want to point out that Remark
2 clearly implies that
A *BxF= (A *ax)F
A * ( axF). (3.5)

It is important to note that based on directed integration, which uses a "di-

rected Riemann measure", one can define the multivector derivative by a limit pro-
cess applied to an integral over the boundary of a smooth open r-dimensional surface
"tangent" to (in this definition r-vector) X. In the limit process the volume of the
r-dimensional surface is shrunk to zero. (Compare [5], section 4.) For an analogous
definition of the vector derivative see section 5 of [12].
148Multivector Differential Calculus E. M.S. Hitzer

Proposition 21 (algebraic properties of 8x).

for ai = const.
ax=P(ax)=^a aJ*8x = Y'ai*axa , (3.6)
J
with P the multivector projection into G(d 1 A a2 A ... A d 0 ). aj is a simple
blade basis? of 9(I):

aj = dd1 A 2 n ... n a"3 (3.7)

with ik = k or 0 and elimination of elements with ik = 0. Hence jl < j2 <

... < n J stands for the combined index

J = (j1,j2,...,jn). (3.8)
For all 3k = 0 we define

aj = a(o,o,... ,o) = 1 . (3.9)

The a J are the corresponding reciprocal blades:

a j __a7 ^A...Adi 2 Ad^ 1 . (3.10)

The vectors with upper index are reciprocal vectors with

a"k a",, = 8, (the Kronecker delta.) (3.11)

For further details of the notation employed here compare [31, pages 30 and
31. The scalar differential operator aj * ax after the last equality of (3.6) is
not to be applied to a j , but rather to any multivector function to which the
multivector derivative operator ax on the left hand side is intended to be
applied.
Proof 21. See the definition of the algebraic properties of 8x in definition
19(i). For the last equality in (3.6) take into account that aJ*8x is algebraically
scalar and [3], chapter 1-1, (1.11).
Proposition 22 (constant scalar factor). Another algebraic property of the
multivector derivative ax is that we have for constant scalar factors A:

ax(AF) = \ BxF. (3.12)

The vectors k' k = 1, ... , n in (3.7) are not necessarily orthonormal, but can
be orthogonalized by a procedure similar to the Schmidt orthogonalization {[3], pp.
27, 28 (3.1)-(3.4)}.
Advances in Applied Clifford Algebras 12, No. 2 (2002) 149

Proof 22.

ax (AF) i3 = Y" a Ja j * ax (AF) 116 ^ a J A aj * aX F

J J
[3], chap. 1 1 , (1.11)
AY, a J aj * axF i3=6) A OXF (3.13)
J

Proposition 23.
n

aX = Y' a(x) with ,.

r=0

(ax)r = (a') r (a j ) r *1X. (3.14)

alx),. is thus the derivative with respect to a variable (X) r E gr(I).

Proof 23.

a,, * aX [3], p. 13 (1.46) ^ (a (ibidem)

J)r * ax ^(aJ) r * (aX) r(3.15)
r=O r=0

(ax) r is the r-blade part of ax, which has the algebraic properties of a multi-
vector (Def. 19, P. 21). The sum > J aJa j * ax is therefore naturally performed
in two steps:

1. Sum up over all index sets J = (jl, j2, ... , jn) with r non-zero members.

2. Sum up over all r = 0 ... n:

ax =77(a J )r(aj)r *ax. (3.16)

r=0 J

Proposition 24.
n
a = ' ar . (3.17)
r=0
150 Multivector Differential Calculus E. M.S. Hitzer

Proof 24.
n n n
aX P_23 a(X)r P. 23 Def. 18
7(8X)r 7 ar. (3.18)
r =O r =0 r -0
Proposition 25.

aX = BAA * 8X. (3.19)

Proof 25.

(AAA * ax R21 a aj *
DA (A * &x)
J
P. 11 aJP(aj) P. 21
* aX =Va j ar * aX aX. (3.20)
J J

Proposition 26 (derivative from differential).

aF - axF(X) = 8AF(X, A) = aAFA(X) - OF. (3.21)
a means the derivative with respect to the differential argument A of F.
Proof 26.
OxF(X) P. 25 aAA * 8XF D = 1
8AF(X, A). (3.22)

4. Adjoint of a Multivector Function

Definition 27 (adjoint). The adjoint of a multivector function F is
F = F(A') - a(F * A'), (4.1)
or explicitly
Def. 1
8A [{A * axF(X)} * A'] aA [F(X, A) * A'] . (4.2)

Proposition 28.
F(A') = ax(F * A'). (4.3)
or explicitly
(X, A') = ax (F(X) * A'). (4.4)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 151

Proof 28.
A*ax
aig . scalar
F(x A' ) Def. 27 aA 1L(A * ax)F(X)I * A'] aA(A * ax ) [F(X) * A']
P. 25
ax(F(X) * A'). (4.5)
Proposition 29 (common definition of adjoint).
B * F(A) = F(B) * A. (4.6)

Proof 29.
B*8X
algebraic scalar (B * axF) * A Del.
B * F(A) P. 2 8 B * 8xF * A 1 F(B) * A.
(4.7)
Proposition 30.

P(F(A)) = F(A). (4.8)

Proof 30.

P(F(A)) P'= $ P(ax (F * A)) = P(ax)(F A)

P' z'
ax(F * A) P'=s F(A).
scalar
(4.9)

Proposition 31.
((A))r = 8(F * A) = Of F * A. (4.10)

Proof 31.
Del. Def. 18 a
( (A}}r (^ F * A} r = (aB} r F(X, B) * A r (F * A), (4.11)
scalar

(F, (A))r P. 28 (ax(F(X) * A)) * A) Del. 18

r = (ax)r(F(X) aF(X) * A.
scalar
(4.12)
 152 Multivector Differential Calculus E. M.S. Hitzer

Proposition 32 (linearity of adjoint).

F(A + B) = F(A) + F(B), (4.13)

F(AA) = )F(A), if A = (A) scalar. (4.14)

Proof 32.

linearity of
scalar product Ox (F * A + F * B)
F(A + B) P. ' 8X (F * (A + B)}
R21

J
P. 5, distributivity
' r a J aJ * ox (F * A) +Ya J aJ * ox (F * B)
J J
P. al 9x (F * A) + ax (F * B) P ' 2 8 F(A) +(B), (4.15)

where I mean the distributivity of geometric multiplication with respect to

addition as in [31 p. 3, (1.4), (1.5).

P.28 scalar product

F(AA) OF * ( AA) ax(AF * A)
P. 2i
Y'a J aJ*Ox(AF*A) P=6 'a J Aaj*8x(F*A)
J J
[3] P. 4_(1.11) P. 28, P. 21 AF(A).
A Y a J aj * &x(F * A) (4.16)
J

5. Differentiating Sums and Products, Changing Variables

Proposition 33 (sum rule).

Or (F + G) = OF + OrG. (5.1)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 153

Proof 33.

i (F + G)
P. 23 r (a J )r(aj)T * ax (F + G)
J

P= 7 'T(a J )r {(aj)r * BxF+ (aj) r * 8xG}

J
distributivity 'T (a J )T(a J)jw * axF+T (a
J )r(aj)T * 8xG
J J
P. 23
aF + a G, (5.2)
where I again mean the distributivity of geometric multiplication with respect
to addition as in [31 p. 3, (1.4), (1.5).
Proposition 34 (product rule).

6 (FG) = i FG + 8FG. (5.3)

Proof 34.
. 23 ^(a J
3(FG) )r(aj)T * ax(FG)
J

P. 8(a){ ((aj)r * BxF)G + F( (aj)r * (fix )GI

algebraic scalar

= (a J )T 1((aJ)- * ax)FG + ((aj) r * ax )FG

J
distributivity 7 (a
J )- ((a j)T * cx )FG + 7 (a J )- ((a j) T * 6x )FG
J J

P'23 i FG + 8FG, (5.4)

where I again mean the distributivity of geometric multiplication with respect
to addition as in [3] p. 3, (1.4), (1.5).
Definition 35 (product rule variation). Left and right side derivation is indi-
cated by:

F G=FibG+Fc9G. (5.5)
154Multivector Differential Calculus E. M.S. Hitzer

Remark 36. Expanding the variation (5.5) of the product rule explicitely in
the aj blade basis (3.7) of the geometric algebra 9 shows how (5.3) and (5.5)
are algebraically different (compare Proof 34):

F c G D.35 FAG + Ft7^ G

P. 23 ^`((
aj)iz * ax)F(aJ)FG+'F(aJ)r((aj)r * ax)G (5.6)
J J

Proposition 37 (change of variables). For F(X) = G(f(X)), i.e. f : X —}

X'—f(X)

0rF = AG(i) = (1(A')),.G, (5.7)

i.e. 5r = (8x)r = (Aax'))T = (1( 8')) ,.. (5.8)

Proof 37.
P.26
F(X) (9A (A * (,9x)r)G(f(X))
[31 P. 13, (1.46)
8Ar (Ai * (ox)r)G(f (X ))
Def. 1 8T
G(f (X +TP(Ar)))IO
P. 12 aA
r (or G(.f (X) + T (Ar * (aX )r)f (X)) IT=O
Def. 1
8Ar ({A* ( 3x)rf(X)} *ax) G(X')Ix,=f(x)
Def. 1 8
r (f (AT) * 8x,) G(X'))X,=f(x)
P. 31
(.f ( 8X' ))r G(X ' ) I x,=f(x) . (5.9)

Remark 38. Remark 14 also applies to proof 37. I conclude from proof 37
that formulas (2.26a+b) in [3], p. 56 are slightly incorrect. There the grade
selector (... ),. is applied to the argument of f, but in order to be correct it
should be applied to f itself, i.e. appear around f as in P. 37.
Advances in Applied Clifford Algebras 12, No. 2 (2002) 155

Proposition 39 (using the full derivative 8x).

Sum rule: 8(F + G) = OF + 8G, (5.10)
Constant scalar factor A: 8(AF) = ADDF, (5.11)
Product rule: 8(FG) = 8FG + 8FG, (5.12)
with variation: F'c3G = FaG + FOG. (5.13)
For F(X)=G(f(X)), i.e. f :X --fX'= f(X)
Chain rule: OF = OG(f) = f (a)G, (5.14)
i.e. a = j(axe) = f (a'). (5.15)

Proof 39. One just needs to take the sum over all grades r : E^_ a on both
left and right hand sides of P. 33, P. 34, Def. 35 and P. 37, as well as take into
account P. 23 and that

'f {r3'} _ (f (c )),., compare [6], (13). (5.16)

r=0
For (5.11) compare P. 22.
Remark 40. The above results obtain, if F = F(X) is defined for X = Xv =
(X) r E 9'^'(I). The results can be adapted to functions on any linear subspace
of 9(I). The differential F = F(X, A) (Def. 1) and the adjoint F = F(X, A')
(Def. 27) are the only two linear functions of A (and A' respectively) that can
be formed from F using the derivative 8 = 8x and the scalar product (compare
the definitions Def. 1 and Def. 27.)

6. The Scalar Case

Proposition 41 (scalar di ferential calculus). For a multivector function of a
f

scalar variable X = X(r) = f (r) :

Multivector derivative 8 X = (,g,)X= dX , (6.1)

dX
Differential X (A) = A * aX = A0 X = A
T , (6.2)

Adjoint X(A) = 8 .X * A = d—
1 Y ^ * A, (6.3)

Chain rule = f(S)F(X) _ * axF(X). (6.4)

 156 Multivector Differential Calculus E. M.S. Hitzer

The second expression of equation (2.27d) in [3] seems to be sligthly wrong

compared to (6.4). In the special case of a scalar function X = x(r) = (x( ,r)),
only the scalar part a = (A) of A contributes to the adjoint:

X(a) = ads, (6.5)

and the chain rule for such a scalar function has the form:
dF _ dx dF
( 6.6)
dT dT dx '

Proof 41. For the multivector derivative (6.1):

Because rr is scalar, (ar ) r X = 0 for r tr 0, =. a = a = dTfor the differential
T

(6.2) :

De£. scalar product

X (A) 'A * aX = (A * a')x (A)a X A =( `4) 8r X. (6.7)
For the adjoint (6.3) compare P. 28.
For the chain rule (6.4) compare P. 37 and (6.3) with A = ax.
We further have from (6.3) for scalar x(r) = f(r) : dT = ( L ) !

T(A) = dx * A scalar product dx * (A) a __(A) dx a

(6.8)
- dr dT d'r
and from (6.4)

dF = f(ax)F(x) = ( x
d ) * 9F(x) = — d F(x). (6.9)
dr r- dr dx
Remark 42. In scalar differential calculus differentials (6.2) and adjoints (6.5)
are identical for ) a. The distinction with the derivative (6.1) is trivial (i.e.
only the scalar factor )). The single concept of derivative in elementary differ-
ential calculus is therefore now generalized to three distinct, related concepts
of multivector derivative, differential and adjoint.

7. Basic Multivector Derivatives

In the following I assume

(i) X = F(X) to be the identity function on some linear subspace of G(I) of

dimension d,
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 157

(ii) A = Psubpace(A) to be the projection into the above mentioned d—dimen-

sional subspace of 9(I),

(iii) that singularities at X = 0 are to be excluded.

Proposition 43.

A*axx=axx*A=.A. (7.1)

Remark 44. To help avoid confusion I refer to Remarks 2 and 20 in order to

clarify that the computation of (7.1) implies the following brackets:

(A * ax)X = 8x (X * A) _ 4. (7.2)

Proof 43.

F(X) = X,
D 1 P.10
A * 9XX = F(A) P=3 F(P(A)) =F(4) 4, (7.3)
aXX*A Def.19
a J aj *0XX*A=Y'a J a j *A=4, (7.4)
^dusing (7.3) ^d

where Jd is the index subset for the multivector base of the assumed linear
d—dimensional subspace of C9(I) of X.

Proposition 45.

A*axX =dxX-*A=4, (7.5)

where the tilde operation (") indicates reversion as defined in [3], pp. 5,6,
(1.17) to (1.20).

Proof 45.
X)P_43
A*0xX—(A (7.6)
algebraic scalar

1 P_43aX X * A
^3]p.13,(1.48)6XX*A
(7.7)
 158 Multivector Differential Calculus E. M.S. Hitzer

Proposition 46.

8xX = d. (7.8)

Proof 46.
axX Def. 19
V' a ' a j * 8xX P. 43 7 a'aj =' 6 j j = d, (7.9)
Jd Jd Jd

where Jd is the index subset for the multivector base of the assumed linear
d— dimensional subspace of 9(I) of X.

Proposition 47.

9xIXI2 = 2X. (7.10)

Proof 47.

axIX12 [3] p. 13, (1.49) Def. 19, P. 21

8Y(XX) 7a 1 a j * 8 x (XX)
id

P. 8, P. 4
distributivity
Y a aj *ax(XX) +'V'a aj*8x(XX) (7.11)
Jd id

Def. 19, P. 21
ax (XX) + ax (XX) = ax (X * X) + ax (x * X)
[3] p. 13_ (1.47a) P. 43_P. 45
x( * X) + (fix (X * X) X+= 2X,

where I mean the distributivity of geometric multiplication with respect to

addition as in [3] p. 3, (1.4), (1.5). Jd is the index subset for the multivector
base of the assumed linear d— dimensional subspace of 9(I) of X.

Proposition 48.

A * axX k = 4IXk-i + X4X '- 2 +, ... + X k -1 1. (7.12)

 Advances in Applied Clifford Algebras 12, No. 2 (2002) 159

Proof 48.

FX,
De 1 P8
A * 5X" A * ôxF' + FF 2 + ... + F''
P. 43,F X XXk2
4X' + + ... + Xk .
( 7.13)
Proposition 49.
axlX = kjX 2 X. (7.14)

Proof 49.

f(X) 1X1 2 ,

8xiX = ax(iX1 2 ) 7 '8 kI

128
x(lI2 * 8,) a k2l n=f(X)=1X12 = ( 8xjX12)8 2
Ic=f(X)=IXI2

=' 2XlXI = kiXI k _ 2 X. (7.15)

Proposition 50,

8x log Xl=—. (7.16)

Proof 50.

f(X) lxi,
ôxlogJXI = f(8n) log
P.37— P.28
= ô3c(Pl *3) logai f(x)x}
1 P.491
= X
(8XIXI) XI (7.17)

Proposition 51.

A*8x(lXi'X) IXIkI4+k AJX\ . ( 7.18)

160Multivector Differential Calculus E. M.S. Hitzer

Proof 51.

A * ax (IXlkx) P=8 (A * 8x IXVk)X + IX}kA * aX x (7.19)

Def. 19(ii), P. 49, P. 43
A * (kI
, xi k_2 X)X + Ixlk = IXIk 41 + k AIX12
* XX 1

scalar

Proposition 52.

Xx1 .
ox(IXI k x) = IXIk / d+k (7.20)

Proof 52.
aX(jXj k X) P. 34, Rem. 40 (3X'X^
k )X + X lXj k
scalar
P. 49 (kjX1 k_2X )X + lXI k aXx
P. 46 kIXIk XX + jXjk =1xjk a + xx 1 (7.21)
/
IXI IX

Proposition 53. For X = X defined on the whole of g(I) and i3 = aX

A*8X=ax*A=P(A). (7.22)

Proof 53. In Proposition 43, the assumed "subspace" becomes now all of
9(I), therefore 4=P(A).

Proposition 54. For X = F_r^ a X;. defined on the whole of 9(I) and a = ax
A * ar X = 6f ± * A = P(A). (7.23)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 161

Proof 54.
[3], P• 13, (1.45a) 53
A * af X Ar * l9XX P. P(A?) (7.24)
and
23
C^TX * A P. r(aJ)r (aJ)r * 8XX *A (7.24) Y-(aJ)rP((aj)r) * A

[3], p. 13, (1.45a) Y (a )r/a

J)r * Af = P(A T ). (7.25)
J

Proposition 55.

afx=aXr=arx = (n} . (7.26)

r

Proof 55.
arX P. 23 (aJ\r(a J)r * axX [3], P. 13, (1.45a) (a)r (a
J)r * C7(y) ,X

P. 54 7
(a J )rP((aj)r) = 7 (a J )r(aj)r = (r , (7.27)
J J

where ( r 1 is the dimension of the r-vector subspace of 9(I).

Br Xf is in this subspace and we therefore have according to P. 23 that

arXr P. 23 P_43
^(aJ)r(aJ)r * 8TXT N-(a J ) r (aj) r = ( r \ , (7.28)
J J

and
l9X V(aJ)8(aj)8 * 3gXT
s s J

P. 4Y(CL ((aJ)s * ôSX) r P. 54 Y V (a J )s ( P ((aJ)s))r

s J $ J

_ (a J )s ((aj)s)r = V (a J )r(aJ)r = ( rr (7.29)

$ J ^^
J
=5 ,.(aj)s
 162 Multivector Differential Calculus E. M.S. Hitzer

Proposition 56.

,qX = (r 1 = 2 n . (7.30)
r

Proof 56.
t7X P. 23 rX P. 55
(r =2 g . (7.31)
r r

Proposition 57. For A = P(A), .K = 2 (r + s — jr — sI) and 1= 0 if j > i

(7.32)
as(XAr)m = (AT&)mX = (r 1 ( s — r ^ b +s-2kAr>
with8r+-2k the Kronecker symbol for non-negative integers.
Proof 57. (This step by step proof extends over the next two pages.) I will
now assume A r to be a simple r-blade A r = a1Q2 ... nr i.e. a geometric product
of r orthogonal unit vectors. We now select an orthonormal basis of 9(I) which
includes d l , d2 , ... , car . The first step in our proof is to show the formula
c?s(XAr),n. = t95(X5Ar),n.. (7.33)

The left hand side of (7.33) yields

P
(aJ)s l (aJ) s * (05X A r ^
.

as(XAr)m
J m

P. 54 V'(aJ)s
((aJ)sA )m ,. , (7.34)
J
and the right hand side of (7.33) gives

8s(X5Ar)m P =?3
(a) '( J)s
J
s` Ar l
m
P. 54 ^(aJ}s
/ag t * (aJ) S A r \
J m
P. 54 7.34)
C'om(a')s ((aj)s * (&&XAr)m. as(XAr)m (7.35)
)J
Advances in Applied Clifford Algebras 12, No. 2 (2002) 163

where to obtain the second equality we first set in P. 54 r = s and A = (aj)s :

(ai)s * asX = a 9 X * (aj)s. (7.36)

Second, by applying the s-grade selector () S on both sides of (7.36) we get

the necessary relationship

(aj)s * agXg = 8 5 X * (ai)s . (7.37)

alg.scalar

This completes the proof of (7.33). Writing the vector factors (a j ),, (aj),g and
A r in the last expression of (7.34) explicitely, we obtain from (7.33) that

as( XA r)m = as(XsAr)m = Y " (a J )s ((ai)sAr)m

_ a's...a32a3'(ajiaj2...aj,dla"2...ar m,(7.38)

s vectors r vectors
r + s vectors

where all vectors ai l , a32 , ... , ad s are taken from the aforementioned orthonor-
mal basis of C9(I) and j s < n.

Selecting the m-grade part (i.e. the m = r + s — 2k-grade part) from a prod-
uct of r + s vectors means, that because of the orthogonality (all s vectors
d 1 , ai l , ... , ad s , and all r vectors aa1 i a2, ... , ar are orthogonal unit vectors,
and stem from a complete basis of orthogonal unit vectors of g 1 (I)) precisely k
vectors of the two sets ail , adz , ... , ak a , and al, a"2 i ... , a"r must be in common.
Otherwise the m-grade condition for each specific m is not fulfilled and the
()-part is zero. For all non-zero m-grade parts we can drop the (),,,,-bracket,
use the reciprocal frame relationship of P. 21, and end up with A r = a1a2 ... ar .
How many such terms occur in the sum E^ l< <j y ...? This question can be
answered by considering that there are (r 1 different choices of k vectors
from the set of r vectors di, a2 i ... , air . Then there are n — r vectors left from
which we can still freely choose s — k vectors. The freedom of choice is reduced
to s — k vectors, because the set of s vectors dj 7 , 2 , ... , aj. must have k vec-
tors in common with the r vectors al, a2, .•. , aar , as just explained. There are
164 Multivector Differential Calculus E. M.S. Hitzer

(n—r\
different ways to choose s — k vectors from n — r vectors. Hence
s— k

aps...ithdiI(Ztj Qd 2 ...a^A 2.. .ar),n

(r\ (n-r\
(( 7 .39 )
s — k 5r+s-2kAr.
= k
The result (7.39) extends by linearity from the case of simple A r blades to
general grade-homogeneous multivectors A. :

85 (X 9r)" = as(XSAr)r _ ( k ( s — k (7.40)

1 S +s-2kAr.
The last part of our proof of P. 57 will be to show that the same result as
in eq. (7.40) holds true for (A r ar ) m X. As before I again first assume A r to
be a simple r-blade A r = a l a2 ... ar of orthonormal vectors dj, 2,••• , d,.. We
then have
(Aras)mX P. 23 Y (alaa2 ... ar(aj)s)m,,. (a J ) 8 * aX X
J
P_53 r (aia2
... ar(aj)s). (a J )s
J

_ A 2...dr dj7 a.12...di.)Ina7 '...Q7z aJi

h <... <^s
_ (r (n—r m
—k s — k 6r+s-2k A r , (7.41)

where I use in the last step the same argument as for the derivation of (7.39),
an important part of which was the reciprocal frame relationship of P. 21.
Equation (7.41) can again be extended by linearity from simple r-blades A r to
general grade-homogeneous multivectors A. This together with (7.40) com-
pletes the full proof of P. 57:
as(XA ^ )m = ( ATas)mX = ( k ^ (s — k (7.42)
1 b +s-2kAT.

Proposition 58. For A = P(A), K = 2 (r + s — jr — sI) and = 0 if j > i

asATXs = Y(a J )SA-(aj)s = rrAr, (7.43)

J
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 165

with
K
rs = ^(-1)rs-k ^ r (7.44)
s - r l.
k=0

Proof 58.
(3], p. 6, (1.20a) (-1)m(,m_l)/2(ArX5)m
(ArXs)m
m(m-1)/2+r( r -1) / 2 +s ( s -1) / 2 (X 5 A r m .
= (-1) ) (7.45)

Therefore
K K
(3], p. 10, (1.36)
ArXs = ^(ArXs),r—sI+2k = 7 *(ATXs)r+s-2k
k=0 k=0
K
(7.45) 2k)(r+s - 2k - 1)/2+r(r - 1)/2+s(s - 1)/2.
V (XsAr)r+s
-
- 2k ( - 1)(r+s
k=0
(7.46)
The two different summations in line one of (7.46) correspond to counting up
as in [3], p. 10, (1.36):
r-sI,Ir-s[+2jr-sj+4,...
... ,jr-s[+2K= Ir-sI+r+s - ft-sj =r+s, (7.47)

and counting down as in [31, p. 58, after (2.38c):

r+s,r+s-2,r+s-4,...
... , r + s — 2K = r + s — (r + s) + Ir — sl = Ir — sl. (7.48)

The exponent of (-1) in (7.46) can further be simplified to

(-1)(r+s-2k)(r+s-2k-1)/2+r( r-1)/2+s(s-1)/2

l (1) 2 (r 2 +s 2 +4k Z +2rs-4ks-4kr—r—s+2k Fr 2 —r+s 2 —s)

= ( - 1) (2r 2 +2s 2 -2r-2s+2rs+2k) = ( - 1)(r 2 +s 2 —r—s+rs+k)

= ( -1 )r(r-1)+s (s-1)+rs—k+2k = (-1) rs-k.

(7.49)
166Multivector Differential Calculus E. M.S. Hitzer

Equation (7.46) simplifies therefore to

K
AT X5 = 7(XsAr)r+s-2k (-1) r5—k . (7.50)
k=O
The s-grade derivative of (7.50) gives
K
0BATXa = V 5g(XsAr)r+s-2k (-1) rs—k
k=O

(7.40) 7(-1)rs—k ^ 1 (s — k 1 A r = rrA f(7.51)

k=o
=r,

To prove the first identity in P. 58 we write

a3AfXs P_23
Y'( aJ)s (aj)s * a s AX s = Y ' (a J )5Af ((aj)s * asXs)
'V^
JJ
scalar
P. 43 Y ( a5)5AfP aj)s) _ 7(a J )5Ar(aJ)s. (7.52)
s-dim. ((

)j subspace j

Proposition 59.
n
85 AX = BAX s = Y r A,.., (7.53)
r=0

with r;=F, o(-1)rs—k(r^ I —r and K=2(r+s—Jr—si).

sk
Proof 59. We will first proof that for an arbitrary but fixed grade r
as A T X = a s A r Xs = BA r Xs . (7.54)
(For r = 0 we end up with a scalar multiple of P. 55.)
asAfX P. 23 Y(a)5
(aj)s * as ArX = (aJ)5Af (aj) s * 9sX
scalar
P_54 P.58
V(aJ)sAT(aj)s = 05- AX. (7.55)
J
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 167

The second identity in (7.54) can be shown as follows:

n
P. 23 ['^ C^(aJ)t (aJ)t
aArXs * (8 ArXg
t )=0 )J
scalar

[3) P. 13, (1.46) )

7(aJ)t (aJ)t * at ArXe
t=0 J ^^
scalar
n
_ 7 Y(a J )tAr (aJ)t * afXs P=4 7 Y(a J )tAr ((aj)t * atX)s
t=0 J `'^`^^ t J
P. 54 77( a J )tAr 77 (a J )tAr
(P((aj)t)) s = ((aj)t) s
t J t J
= 6 ts(aj)t
P. 58
_ Y(aJ)5Ar(aj), 9g A f X s . (7.56)
J

Performing the sum over all grades r in (7.54) we get

n n
P. 5s
as AX = aAX s = S` asArXs 7 r9A r , (7.57)
r=0 r=0

where I used the fact that according to Def. 18 and [6] (13):

n
A= AT . (7.58)

Proposition 60.

a5XAAr=ArA85X= (n S r1 A r . (7.59)

Proof 60.

(7A0) as ( X
asX A Ar =.9s ( XA r)max. grade sAT)r+s =max. grade
P. 57
= 09 (XAT) r +s (A r8 )r+sX = A r Aas X, (7.60)
 168Multivector Differential Calculus E. M.B. Hitzer

(7.60) a P__57 7 r 1 (n—r\ Sm—_r+s A

e^ X
s r st/ XA
A r^r+s k A_ s—k r+s-2k r

k=0(r\ln s r1 AT= tnr1

s Ar. (7.61)

Proposition 61.

r s X.Ap=A f—i9§ X— f (r\ A

^n s r^
Ar ifo<r<s
s—r
(7.62)

Proof 61.
( 7.40) 8s
55X ' Ar = i9. (XA' )min_ grade = (XsAi)jr — sj =min. grade
(Aras ) jr —SI X = Ar 8 X
= as (X Ar)j r _ S) P.57 (7.63)

(7.63) p. 57 (r \ (n — r s m
t9s (XA f ) r — s
•

C' S X• Ar _s - 2k r

jr^ (n—r
^A-6r —s
r r+s-2k if0<s<r
k s— k
_ k=s
(r^ jn—r\A SS
if0<r<s
I k s —k
—r
r r+s-2k
t k=r

— J f rl

(
r
rsns r01A;.if0<s<r

(
=1
_
s—]c\ArifO<r<s
(7.64)

=2

Proposition 62. For simple, X-independent A r the expansion

85X5 = 85(X5Ar)Ar 1 = t s Xs A A r Ar 1 + 85(X5 A A r ) r+s -2 A'

+ ... + 3s (X A Ar) 17_s1+2 Ar' + 85 X3 - A^. Ar' (7.65)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 169

is termwise equivalent to (assuming 0 < s < r)

( n 1_( r t! n— r 1 I r t n_ r 1
s 0 s + 1 s-1

s r 1 \ (n 1 r\ -1- ( \r (n ^ . (7.66)

Proof 62. For simple A r :

X9 [3], p. 3, (1.3), [ 6], (30)
(Xs-A r )Ar 1 . (7.67)

The expansion of XgA r is done according to (31, p. 10, (1.36).

The correspondence of the first and last term are shown by P. 60 and P. 61
respectively. In general each term in the right hand side expansion (7.65) has
the form
j 1 P. 57, (7.40) (r \ f n — r
as\XsAr)r+,,-2kAr A1
k s — k Arr

— (rl (n—r^
(7.68)
k s—k

with k = 0,... ,K and K = 2 (r + s — Ir — sl). The left hand side of (7.65) is

- 9sXs ". ss (n 1
(7.69)
s

The resulting binomial coefficient identity is known as theorem of addition:

[141, p. 105, (2.4).

8. Factorization
Factorization relates functions of multivector variables with corresponding
functions of (several) lower grade content multivector variables. In the sim-
plest case the latter functions will just be functions of several vector variables.
Proposition 63. For two multivector variables A, B:

8A G(AAB)=OA(A.AB)*8UGU(AAB)=5AG(AAB,A.AB), (8.1)
170Multivector Differential Calculus E. M.S. Hitzer

with
P6
auGU(AAB) aXG(X)Ix=AnB (8.2)

Proof 63. For f (A) = A A B and F = G(f (A)):

P. '
f(ax') 8A(AAB)*ax'. (8.3)
Therefore
P. 37_P. 39
OAF = aAG(.f (A)) f (oX')G(X ' ) I X'= P(A)=AAB
D=.
(8=3) aA(AAB) *Bx,G(X')Ix'=AAB 18AG(X',AAB)Ix'_AAB
= 0AG(AAB,AAB). (8.4)
Proposition 64.
0 0
aBaAG(A A B) = aB 8A(AA B) * aU Gu(A A B)
0 0

+ 8B (AA B) * av 6A(A A B) * 8v Guv (A A B), (8.5)

with

Guv = V * d U * a G(A A B). (8.6)

Proof 64.
P' 63 8B
8B OA G(A A B) 10A (A A B) * auG u (A A B)1
[3], p. 13 (1.44), P. 4, Def. 19(ii), P. 34 0
aA(AA B) * auGU(A A B)
00

+ aB OA (A A B)* (a U G U (A A B)). (8.7)

The second term on the right hand side of (8.7) becomes
P_s3
8B 0A(A A B)* (OUGu(A A B))
0 0
BBaA(A A B)*{(AA B)* 9x , auGu(X')I X' =AAB }
Def. 19(ii), P. 26, Def. 15
= avBuGw(AnB)
=aB I (AA B) * av l dA I (A AB) * 8v Guv (A AB). (8.8)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 171

Proposition 65. For a multivector function G defined on gr(I), i.e.

G(X) = G((X) r ), (8.9)

and for A = (A), in P. 62, P. 63 we have B = (B) r _ s and
8AG(A A B) = B . auCu(A A B). (8.10)

Proof 65.
G on 4'(I)
G(A A B) = G((A A B)r) G(((A)s A B)r) = r-*(((A)s (B)r—s),
(8.11)
and
[3], p. 13 (1.46)
8A(AAB)* aX— aA ( A A B ) aX
=(AAB),. =(a) ,. =( A), =( B) ,.- s =(a),.
[3], p. 7 (1.25b)
OA(A (B ' aX ))
[3], p. 13 (1.46), A = (A)s 8
A(A * (B ' aX ))
P. 54 B ,
aX. (8.12)
Using (8.11) and (8.12) we finally get
P' 63
aAG(A A B) 0A(A A B) * aucu (A A B)
similar to (8.12)
= B . BUGU(A A B). (8.13)
Proposition 66. For a multivector function G defined on gr(I), i.e.

G(X) = G((X)r),
and for A = (A), in P. 62, P. 63 we have B = (B) r _ 9 and

8BaA G(AAB) _ ( r 1 avGU(AAB)

r—s
+ (-1)s(r- 5)A ' avB . au Guv(A A B). (8.14)
172 Multivector Differential Calculus E. M.S. Hitzer

Proof 66. The following three identities arise from proof 65:

B = (B)r-s, (8.15)

0A(A A B) * aX = B . aX, (8.16)

a ( A A B) *
$
[6[ =
(40)
aB (B A A) * aX(- 1 )s ( r - s ) = A - a x (- 1)s(r - s).
(8.17)

We can therefore show that

(8.16) P.61 j TrS
aB aA(AA B) * aX 0x. (8.18)
(all)r_. (B) --. tax?r
,

We finally get for

OBaAG(A A B) P. 64 aB aA(AA .B) * auGj(A A B)

0 0

+as (AA B) *avaA(AAB) *8UGUv(AAB)

similar to (8.17) similar to (8.16)
similar to (8.18) / r
ajGu (A A B)
r—s
+ (-1)A - avB . au Guv(A A B). (8.19)

Proposition 67. For a linear multivector function L = L(X) we have

L Def. 1
LU(X) = L(U), (8.20)
Luv = 0. (8.21)

Proof 67.
L Def. i LU(X) Bef. i U
* a9 L(X)
scalar
linearity for P(U) = U L(U),
L(U * 9xX) = L(P(U)) (8.22)
 Advances in Applied Clifford Algebras 12, No. 2 (2002) 173

Luv De£ 15 V * a (8.22)

x U * aXL(X) V * axL(P(U))
linearity linearity
L(V * ayP(U)) 0. (8.23)

Proposition 68. A multivector function L = L(U) is linear, if and only if

L = L(U). (8.24)

Proof 68.

(=)
L = L(X) is linear P..7 L = L(U), (8.25)

()
L = L(U) = L(aU + /3V) = L(X, caU + /3V)
P. 5_P. 6
aL(X, U) +,3L(X, V)
L _—L(U)
= caL(U) + /3L(V) L is linear. (8.26)
Proposition 69 (factorization of derivative of lin. function). If G(A A B)
L(A A B) is a linear function on 9(I), and A = (A) 5 B = (B) r _ S , ,

aBaAL(A A B) _ ( r r s 1 ar L. (8.27)

Proof 69.
P. 64 P. 67
8BaAL(A A B) aB 0A(AA B) * DULU(A A B)
`

Def. 19(ii), P. 26 o
= aB dA (AA B) * aXL(X)IX=AAB
A = (A)s, B = (B)r_s, [3], p. 13 (1.45a) o 0

aB aA(AA B)r * (aX)rL(x)I X =AAB

Proof 66, P_23, Def. 18
r—s
r ( 1 ar L = r r—s
r 1 aL (8.28)

where the last equality holds if L(X) = L((X) T ) on

 174 Multivector Differential Calculus E. M.S. Hitzer

Proposition 70 (vector derivative factor). For a linear function L:

,9B aa L(a
1 l A B) = ra ? L. (8.29)

Proof 70.

OBaa1 L( al A B) P. 69
r
r — ( —1)1
a L(a
r i A B) = ra f L. (8.30)

=r
Proposition 71. The function aa 1 L(a"1 A B) is linear in B.

Proof 71. For ca, ,Q independent of d l :

8 1 L(1 A ( aB /3C)) L linear6] (23) aa l {aL(a" 1 A B) + QL(a" 1 A C) }

[s] P. sl
aj, l {aL(d j A B)} + aal {f3L(d 1 A C)}
[6] Def. 17(i) and eq. (21), (23), Def. 19(i)
= aaa1L(dd1 A B) + Qaa 1 L( d i A C).
(8.31)

Proposition 72.

a, ...a a L( dl Ad2... A a ) =r! a L

2 1 r f

= a, A ... A a2 A a 1 L(& l A d2 ... A d,), (8.32)

with ak __ aa, .

Proof 72.
arL P. 70 1 aBa1L((al A A ar r)
r
=Br=^
= 8M(B) M is linear according to P. 71
r
aT-1 M P. 70r 1 1 8CO2M((a2 A d3 ... n dr r-i)
=cr=2
P. 70 P.10 1 a
r Z(¢r ) Z is linear according to P. 71. (8.33)
Advances in Applied Clifford Algebras 12, No. 2 (2002) 175

Hence

r! af L = 8r Z(dr ) = '.. 8r8r_1 .. .82 8 1 L(d 1 Ad2.., A dr) (8.34)

We finally consider that

8r 8r i ... 828iL(diAd2 ... Ad r )

I6 Def. 17(i)
8r 8r1 . . . a 81 L(d1 Ad2... A d)
+8r 8r j ... 82A8i L(di Ad2 ... Adr )
8r 8r .-.i ... 82A81L(diAd2 ... Adr), (8.35)

because

8r 8r ..i...ä2'8iL(diA62..,A6r)

&r 8r i(828i+8i(92)L(diAd2 ... Adr )

8r 8r 1.,.(8281 (9281 ) L( d1Ad2 ... Adr )0. (836)

relabeling interchanging
14-2 61A2=-62A1

The factor (.-1) in the argument of L caused by the interchange & j A d2 =

—62Ad1 and the relabeling (1 +- 2), can be factored out, because of the linearity
of L. Applying the same consideration to any pair of indizes i,j E {1, 2,... , r}
of
8r8r.,1...828iL(diAd2 ... Adr )

as to the pair 1,2 in (8.35) and (8.36), we see that the vector derivatives
on the left hand side are completely antisymmetric with respect to pairwise
interchanges. Hence

8r 8r _i...8281L(di Ad2 ... Ad r )8r A...AO2A81L(di Ad2...Adr ).

(8.37)

9. Simpilcial Derivatives and Variables

Definition 73 (simplicial variable, simplicial derivative). A simplicial variable
a( r ) is a simple blade of the form

a( r )alA ... AcL,..(9.1)

176 Multivector Differential Calculus E. M.S. Hitzer

The simplicial derivative 8( r ) is defined as

8( r ) - I aa l n ... n ait r(9.2)

Proposition 74 (equivalence). For linear functions L the r-vector derivative

(right hand side) is equivalent to the simplicial derivative (left hand side):

8( r )L(a( r )) = 8F L. (9.3)

Proof 74. P. 72 and Def. 73.

Proposition 75.

a(r)a(r) = afXr = ( r}• (9.4)

Proof 75.

a(r)a(r) P. 74 arX P. 55 arxf p . 55 (n \ (9.5)

r

Proposition 76.

0(r) (a( r )) 2 = (r + 1)a( r) . (9.6)

Remark 77. Though I lack the general proof so far, the examples of one, two
and three dimensions are quite instructive.
1-D

, [s] P. 69 2.
0(1) (a(1)) 2 = aa^ (ai)2 P. 55 (9.7)
Advances in Applied Clifford Algebras 12, No. 2 (2002) 177

2-D

0(2) (CL(2)) 2 = 282 A al(d1 A a2) 2 = Z02 A 81[(d1 a2) 2 — a"iaz]

_ la2 A 91(d1 • a2) 2 — 2(02d2) A ( 8 1a1)

= 2a2 A {2(d1 d2) 8 1(dl • d2)} — -2d2 A ( 2 d1)

=02(a1 •a2)Ad2+2d1 Ad2

= d1 A dd2 + d1 • d2 82Ad2 +2d1 A d2 = 3d1 A d2. (9.8)

[6], P. 67

3-D

0(3) (a(3)) 2 = 3^ 03 A a2 A al(dd1 .A d2 A d3) 2

= 3r 83 A a2 A 81(—d2l 2d3 + d (CL2 • d3) 2 + o2(CL3 • CLl) 2 + CL3(d1 • CL2) 2

— 2a1 • d2d2 • d3d3 • ai)

= 1 ( — ( 03d3) A (a2d2) A (alai) + (alai) A ( 03 A 8 2)(a2 . a3) 2

— (a2d2) A ( 03 A a1)(d1 • d3) 2 + (a3d3) A (a2 A a1)(d1 . Q'2) 2

— 2(a3 A a2 A 81)[d1 • d2a2 • d3d3 . d1])

2-D {-8d3 A d2 A d1 + 2d1 A [2(62 A d3)] — 2d2 A [2(d1 A d3)]

+ 2d3 A [2(d1 A d2)] — 2a3 A a2 A (d2d2 • d3a3 • d1 + d3d1 • d2d2 • d3)}

[6], P. 67 1
= 6 {8a 1 Aa2Aaa3+(4+4+4)al Aa2Aaa 3

—2 a3A (ad3Aad2(ad3•al)+d1 Ad3(d2 • d3)+d3Ad3(d1 •d2))}

=0
[6], P. 67 1
= 6 {20d1 A d2 A d3 —2(d1 A d3 A d2 + d2 A d1 A d3)

2
= 6al Aa2A663=4d1Ad2Aad3. (9.9)
 178 Multivector Differential Calculus E. M.S. Hitzer

Proposition 78.

(r
g a(r +s)L(a(r +s)) = a(r) A a (s )L(a( r ) A a( s )) = a( r) a( s) L(a (r ) A a( S
))

(9.10)

for a( r ) and a (r) as in Def. 73 and L a linear function.

Proof 78.

Def. 73 a(r) A a(
a(r +s) s) (9.11)

Def. 73 1
a(r +s) r+s n ... A a s+1 A as A... A a l
(r + s)! a
1r!s! 1
_ (r+s)!(—,9r+9n...Aas+i)A(siasn...Aai)

S
_ ( +
(r r s )
s) !s! 1 0()A0(3)= (r S a(r) n 9( s ). (9.12)

Hence

(r+sl
Sa(r +s) = a(r) A 0(9), (9.13)

and

(r+s\
S
a(r+s)L(a(r+,g)) = a( r ) A a (s) L(a (r) A a(s))
= a( r ) a( s ) L(a( r) A a( 8) ). (9.14)

The last equality holds because of the symmetry of the argument a (r) A a (s) .

Proposition 79. For linear functions L:

aBaAL(A A B) = aB A aAL(A A B). (9.15)

Advances in Applied Clifford Algebras 12, No. 2 (2002) 179

Proof 79.
P_69 r arL P. 74 r 1 a( r )L( a
oB aA L(A A B) )
r-s r-s (r)

I r 1_ r 1
r—s S (r—s+s1
— S (r—s+s)L(a(r—s+s))

P. 78
= 8( r - s ) A 8( S) L(a( 9) A a(r_s)) (9.16)
P = 8- A 8
(s) L (a
(s) —
A
r B)
s

r—s A^L(A T AB—)

r—s = a B A A(
aLA-AB)
s r—s

Proposition 80. For a linear function Fr = Fr (d i , a2, ... , d r ) of r vector

variables

ak . 8a" k Fr = Fr (k = 1, ... ,r). (9.17)

Proof 80.

ak • aa k Fr = aak aa,, Fr(ai, a,... , ak, ... , ar)

P. 67, P. ss, X = a^ = U Fr(al, a2, ... , ak, ... , ar), (9.18)

for all k = 1, ... , r.

Definition 81 (skew-symmetrizer).

G r - r^ ( al A ...Aar ) • (8r A ... A 8i)Fr = a(r) - 8(r)Fr, (9.19)

for a linear function Fr of r vector variables.

Proposition 82.

Cr = Gr(di, ... ,r) (9.20)

is according to Def. 81 skew-symmetric in its arguments. If Fr is skew-symmetric,

then
Gr = Fr.. (9.21)
180Multivector Differential Calculus E. M.S. Hitzer

Proof 82.
(d j A...Adr )•(ar A ... A91)Fr
[3] P. 7 (1.25b), P. 11 (1.38) (
dl A ... A dr _1) • [ a'r • (or A ... A al)]Fr
I3I P. 11 (1.38) rr-.
(-1) r—k (a" 1 A... A a r _ 1 ) • (Or A... A ak A... A a l )
k )=1
ar . akFr(x1 ... , Xk, ... , x r )
r
F linear, P. 80
^(-1) r— k( al n ... n ar _ 1 ) (ar n ... n a k n ... n a l )
k=1
position k
Fr(,1 ... , a r ,... ,xr )
r
F skew-symmetric
'(a l n...Ad r _ 1 )• (Or A...A6 k A...Aa l )
k=1

Fr(xl ... ,Xk,...

relabeling of
:1 ,... ,sr
= r(al A ... A ar _ 1 ) (ar _1 A ... n a l )Fr (xl, ... , Xr-1, ar)
P.80
r(di A ... A C'b r _1) • (ar _ 1 A ... A al) dr . arFr(Y1 .. • , ^r-1, xr)
P.80 r'F
_... =r!dl •al...d r •ar Fr , (9.22)

where the symbols xk, ak in the above mean that xk and ak = ax k are to be
left out.

Proposition 83 (canonical form of alternating linear form). The simplicial

derivative of an alternating linear r-form a r = a r (al,... ,dr ) ([3], pp. 33 ff.)
yields an r-vector
Ar = a( r)a, (9.23)
with the property
a r =(dd1A...Adr )-AT =a( r )A , (9.24)

also called the canonical form of an alternating linear r-form c r .

Proof 83. Def. 73 and Def. 19(i) (for vector derivatives) show that the above
defined A;. must be an r-vector.
a(r) . Ar = a(r) • a(r)ar = a r . (9.25)
Advances in Applied Clifford Algebras 12, No. 2 (2002) 181

The last equality is due to P. 80, and the linearity and skew-symmetry of a,.

10. Conclusion

In order to make the treatment in the future more self-contained, it may be

useful to compile a set of important geometric algebra relationships, which
are necessary for multivector differential calculus, and often referred to in this
paper.
It may be of interest to notice that there is a MAPLE package software
implementation of the multivector derivative and multivector differential [13].
After discussing vector differential calculus [6] and multivector differential
calculus in some detail, I would like to proceed and produce a similar discussion
on directed integration of multivector functions in the future.

Acknowledgements

I first of all thank God for the joy of studying his creation:
" ... since the creation of the world God's invisible qualities - his eternal
power and divine nature - have been clearly seen, being understood from
what has been made ... "[17].
I thank my wife for encouragement, H. Ishi for corrections and improve-
ments, G. Sobczyk for some hints, and 0. Giering and J.S.R. Chisholm for
attracting me to geometry. Fukui University provided a good research environ-
ment. I thank K. Shinoda (Kyoto) for his prayerful support.

References
[1] Jesus Christ, "Gospel according to Matthew", Bible, Today's English Version.
[2] Gull S., A. Lasenby, C. Doran, Imaginary Numbers are not Real. - the Geo-
metric Algebra of Spacetime, Found. Phys. 23 (9), 1175 (1993).
[3] D. Hestenes, G. Sobczyk, "Clifford Algebra to Geometric Calculus", Kluwer,
Dordrecht, 1999.
[4] Bell I., "Maths for (Games) Programmers", Section 4 - Multivector Methods,
<http://www.iancgbell.clara.net/maths/geoalg.htm>
[5] Hestenes D., Multivector Calculus,J. Math. Anal. and Appl., 24 (2), 313-325
(1968).
<http://modelingnts.la.asu.edu/html/GeoCalc.html#Multivector>
[6] Hitzer E.M.S., Vector Differential Calculus, Mem. Fac. Eng. Fukui Univ., 50
(1) (2002).
<http: //sinai.mech.fukui-u.ac.jp/gcj/pubs.html>
182Multivector Differential Calculus E. M.S. Hitzer

[7] Hitzer E.M.S., Antisymmetric Matrices are Real Bivectors, Mega. Fac. Eng.
Fukui Univ., 49 (2), 283-298 (2001).
<http://sinai.mech.fukui-u.ac.jp/gcj/pubs.html>
[8] Hestenes D., "New Foundations for Classical Mechanics", Kluwer, Dordrecht
1999.
Hestenes D., "Space Time Calculus",
<http://modelingnts.la.asu.edu/pdf/SpaceTimeCalc.pdf>
Hestenes D., "Space Time Algebra", Gordon and Breach, New York 1966.
Hestenes D., "New Foundations for Mathematical Physics",
<http://modelingnts.la.asu.edu/html/NFMP.html>
Sobczyk G., "Simplicial Calculus with Geometric Algebra", available at:
<http://modelingnts.la.asu.edu/html/GeoCalc.htmlMultivector>
Cambridge Maple GA package,
<http://www.mrao.cam.ac.uk/clifford/software/GA/>
Bronstein I.N., K.A. Semendjajew, "Taschenbuch der Mathematik" (German),
22nd ed. , Verlag Harri Deutsch, Frankfurt am Main 1985.
[15] Ito K. (ed.), "Encyclopedic Dictionary of Mathematics", 2nd ed., Mathematical
Society of Japan, The MIT Press, Cambridge, Massachusetts, 1996.
[16] Forster 0., "Analysis 1 (Differential-und Integralrechnung einer Verander-
lichen)", Viewig, Wiesbaden & Rowohlt Taschenbuch Verlag, Reinbek bei Ham-
burg, 1976.
[17] Paul, Romans 1:20, "The Holy Bible New International Version", IBS, Col-
orado, 1973. <http://bible.gospelcom.net/>