JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 24, 313-325 (1968)

Multivector Calculus

DAVID HESTENES

Physics Department, Arizona State University, Tempe, Arizona 85281

Submitted by Richard Bellman

INTRODUCTION

The object of this paper is to show how differential and integral calculus
in many dimensionscan be greatly simplified by using Clifford algebra. Here
the necessarynotations, definitions, and fundamental theoremsare developed
to make the calculus ready to be used. Those features of Clifford algebra
which are neededfor this task are describedwithout pro0f.l
The discussionof differentiation and integration omits without comment
many important problems in analysis, becausethey are in no way affected
by the specialfeaturesof the approach advancedhere. The object throughout
is to show how Clifford algebracan be usedto advantage.

1. ALGEBRA

The notion of a vector asa directed number can be made preciseby intro-
ducing rules for addition and multiplication of vectors which have a geometric
interpretation. The rules governing vector addition and scalar2multiplication
are too familiar to require comment. By these operations an n-dimensional
linear space GY,,, here called arithmetic n-space,can be generated from n
linearly independent vectors. Appropriate rules governing multiplication
of vectors can be arrived at by requiring that the product of any nonzero
vector with itself be a positive scalar. The “square” of a vector a is written
u2 = 1a 12> 0, U-1)
where 1a 1 is a positive scalar called the modulus(or magnitude)of a, and
equality holds if and only if a = 0. With the exception of the commutative
rule for multiplication, all the rules of scalaralgebracan be applied to vectors
without contradicting (1.1). Then, by multiplication and addition, a Clifford

1 Further discussion of Clifford algebra together with applications to physics can be

found in my book “Space-Time Algebra,” Gordon and Breach, New York, 1966.
2 In this paper “scalar” always means “real number.”

313
314 HESTESES

Algebra ./Yn , here called the multivector algebra of &, , can be generated
from the vectors in L& . To emphasizetheir relation to vectors, the elements
of 4% are called muZtivectors.
Far from being a defect assomemight be inclined to think, the absenceof
universal commutativity for vector products is a great advantage. For the
“degree of commutativity” in a product is a measure of the relative directiom
of directed numbers. This is easily seen by decomposingthe product of
vectors a and b into a sum of commutative and anticommutative parts.

ab=a*b+aAb, (1.2)

where a -6 and a A b can be regarded as new kinds of multiplication, re-

spectively called inner and outer products, and defined by the equations

a * b = 4 (ab + bu) = 6 . a (1.3)

a A 6 = i$ (a6 - bu) = - 6 A Q. (1.4)

By virtue of (l.l), a * 6 is a scalar and may be interpreted as the usual

“Euclidean” scalar product of vectors in G& . Subject to this interpretation,
(1.2) showsthat two vectors are collinear if and only if they commute, and
they are orthogonal if and only if they anticommute. The products a . b and
a A b were invented and given a geometricalinterpretation by H. Grassmann
more than one hundred years ago. Through (1.2) they imbue the noncom-
mutative product ub and all of JY~ with geometricalsignificance.
A multivector which can be factored into a product of Korthogonal vectors
is called a simple k-vector. Since Korthogonal vectors alsospana k-dimensional’
subspaceof 02;,, it is apparent that to every simplek-vector there corresponds
a unique k-dimensionalsubspaceof C&,. In fact, every simple k-vector can be
interpreted geometrically as an oriented volume of some k-dimensional
subspaceof G&.
Any linear combination of simple k-vectors is called simply a k-vector.
The terms “l-vectors” and “O-vector” are synonyms for “vector” and
“scalar,” respectively. An n-vector of &,, is often called a pseudoscalar.
The product aA, of a vector a with a k-vector A, consists of a (k - l)-
vector plus a (k + I)-vector, denoted by a * A, and a A A, respectively, so

aA,=a-A,+ahAk. (1.5)

This is a straightforward generalization of (1.2), to which it reducesif k = 1.

In general, the product of a simpler r-vector A, with a simple s-vector B,
is more complicated than (1.Q but the (r + s)-vector part of the product
A& is important enough to be given a symbol: A,. A B, , and a name:
outer product of A,. with B, .
 MULTIVECTOR CALCULUS 315

Any multivector A can be expressed as the sum of k-vectors A, , where

k = 0, 1, 2 ,..., a. Thus,

A = i A,. (1.6)
k=O

Equation (1.5) can now be generalized by introducing the definitions

a-A=xa*A,, aAA=~ahA, (l-7)

so that
aA=a-A+ahA. (1.8)
The reverse (or adjoint) of A, denoted by A+, can be obtained by expressing
the simple k-vector components as products of vectors and reversing the order
of multiplication. It follows that

A+ = i (- l)W-1) A,.
(l-9)
k=O

The scalar part of the product A+B is called the scalar product of multivectors
A and B, and is written (A+B), , the subscript zero denoting O-vector (or
scalar) part. This scalar product is symmetric and positive definite, the latter
property being due to (1.1).

(A+B), = i (A$B,), = i (BIA,), = (B+A)o (1.10)

k=O k=O

(A+A), = f (Ask), = f A,$Ak 2 0. (1.11)

k=O k-0

The modulus (or norm) of A is defined by the equation

I A I = [(AtA),]* = rktoI A, I?]’ b 0.

We have 1A 1 = 0 if and only if A = 0. A multivector with unit modulus

is said to be unitary. Any multivector A can be expressed as a scalar multiple
of a unitary multivector A.
A=IAIA. (1.13)
316 HESTENES

The scalar product has many important properties. For instance, every
scalar determinant of rank R can be expressedas the scalar product of two
K-vectors, and all the properties of determinants follow automatically from
properties of multivectors algebra.
There are two unitary pseudoscalarsin An corresponding to the two
possibleorientations of an n-dimensionalunit volume in G&,. Denote by i the
pseudoscalarrepresenting the unit volume with positive orientation. The
product iA is called the dual of A. In &‘a, the “cross product” a x b
introduced by J. Willard Gibbs is the dual of the bivector a A b. When
proper account is taken of the usualsign conventions, this duality is expressed
by the equation
a A b = i(a x b). (1.14)
With this definition, the entire vector algebraof Gibbs is seenasa subalgebra
of&Y,.

2. GEOMETRY

Intuitive geometricalnotions such as“continuous,” “straight,” “distance,”

and “dimension” require a speciallanguagefor precise expression.To meet
this need, multivector algebrais cultivated here.
The points of EUCI’d
a can n-space &n can be put into one-to-one correspond-
ence with the vectors of G&,. So it is convenient to usea vector x asa name
for the point to which it corresponds.A vector used asa name for a point is
called the coordinateof the point.
The correspondencebetween ~5’~and GZ&gives much more than names.
Points in c!?~are named for the purpose of describing the properties of
geometric objects (point sets)in &n . The multivector algebra&n provides a
grammarand vocabulary designedto simplify such descriptions. For instance,
the distance between two points x1 and xa in &n is simply the modulus
I x2 - Xl I Y and trigonometrical relations for discrete point sets in c?~are
readily computed with multivector algebra. It is worth remarking that &n
can alsobe usedto describenon-Euclidean geometriesif only an appropriate
change is made in the definition of scalar product.
This paper is concerned with continuous surfaces in c?~. Let 9’ be a
smooth K-dimensionalsurface in ~9~. A multivector function (or multivector
field) on 9 is a mapping of 9’ into &Yn . Let x1 , x2 , x2 ,... be a sequenceof
points in 9’ converging a point x in 9. The unit vector

xi - x
n(x) = lim (2.1)
t-m 1xi - x 1
 MULTIVECTOR CALCULUS 317

is said to be tangent to Y at X. The surface Y is k-dimensional if and only

if there are k linearly independent vectors tangent to 9 at each point of 9.
By multiplication and addition, these vectors generate a multivector algebra
Jz’~(x) which, of course, is a subalgebra of ~4’~ , A multivector function on Y
with values in JH~(x) at x is said to be tangent to 9’ at X. A multivector
function is said to be tangent to Y if it is tangent at every point of Y.3
A surface can be characterized by the multivector functions tangent to it.
If a smooth k-dimensional surface Y is orientable, there are exactly two
unitary k-vector functions tangent to 9, each corresponding to one of the
two orientations which can be given to 9. So tangent to each point x of an
oriented k-dimensional surface there is a unique unitary k-vector V(X)
characterizing the orientation of Y at x. Call V(X) the tangent of Y at x.
Call the dual of v(x) the normal of Y at x.

3. INTEGRATION

Let f be a multivector function defined on a smoothr-dimensional surface

V. Define the directed integral off over V by the formula

h(x) f (x) = bz f do,(x) f (xi). (3.1)

S/f =I, 2=1

This differs from the usual definition of a Riemann integral only in one
important respect. Both dv and dv(x,) are directed volume elements.The
magnitudes 1dv 1 and 1dv / are to be understood as the usual Riemunn
measures of volume. The direction of a volume element at x is characterized
by the unitary simple r-vector v(x) tangent to V at the point x. This can be
expressedby writing

dv(xJ = I Llv(x,) 1V(Xi) (3.2a)

dv(x) = I dv(x) I v(x). (3.2b)
So it is clear that the directed integral off is equivalent to the Riemunn integral
of vf-

Therefore, the details to the limiting processin (3.1) can be handled by the
techniques of Riemann integration theory.

3 For brevity, the effects of discontinuities in a surface are not discussed here.
318 HESTENES

The directed integral (3.1) is a nontrivial generalization of the Riemann

integral which makes essential use of multivector algebra. The significance
of this generalization can be seen in complex variable theory, for the integral
with respect to a complex variable is a l-dimensional directed integral, and it
is this feature which makes the theory so powerful. In another paper it will
be shown that complex variable theory can be regarded as a special case of the
multivector calculus developed here.
At a deeper level, “directed integration” is founded on a generalization
of measure theory which uses “directed measure” instead of “scalar measure.”
A “directed measure” associates a direction and a dimension as well as a
magnitude to a set. Thus, it may be said that the directed integral (3.1) makes
use of “directed Riemann measure” rather than the usual “scalar Riemann
measure.”
The volume 1Y 1 of the surface Y is

IV1 = jydvd= j ldvl.

Y
One can also associate a “directed volume” with the surface 71r given by the
integral sY dv. From the definition (3. l), it follows that the directed volume
of any closedsurfacevanishes.This is expressedby the equation

$ dv =O, (3.5)
where $ indicates that the integral is over a closed surface. Clearly, (3.5)
obtains becauseon a closed surface “directed volume elements” occur in
pairs with opposite orientations which cancel when added.
Becausemultivector multiplication is noncommutative, (3.1) is not the
most general form for a directed integral. The appropriate generalization is

(3.6)

where, of course,f and g are multivector functions defined on Y.

4. DIFFERENTIATION

Let A be a nonzero simpler-vector in .Mn , andf somemultivector function

defined on &n . The “(left) derivative off with respectto A (evaluated) at x”
is denoted by Vf(x) and defined as follows:

(4.1)
 MULTIVECTOR CALCULUS 319

where
(1) a = 1A I-IA.
(2) A smooth open r-dimensional surface Y has been chosen which
passes through x “tangent to A,” i.e., so that A = D(X), the tangent of Y
at x.
(3) The integral off is taken over the boundary, aY, of Y. The orienta-
tion of the (r - 1)-vector da = 1da 1a describing a volume element of 8Y
is chosen so that the vector n(x’) = D+(x’) a(x’) is the outward normal at a
point x’ of ar.
(4) The limit is taken by shrinking Y and its volume 1Y” 1 to zero at the
point x.
A discussion of the extent to which the choice of Y and the process of
shrinking is arbitrary is too involved to give here.
To get at the significance of the operator V, let us look at some special
cases and then ascertain its general properties.
On the basis of (4.1) no meaning can be attached to the derivative with
respect to a scalar, so the simplest example is the derivative with respect to
some vector n. In this case, 7” is an oriented curve with arc length s = 1Y I
passing through the point x with tangent proportional to tl. The boundary
of Y consists of the end points, xi and xp , of the curve. To evaluate the
integral over W’-, appeal must be made to the definition of the integral (3.1),
which shows that limiting process is unnecessary since the surface in question
consists of only two points. A point is a O-dimensional surface, so its volume
element is a O-vector. Use of Riemann measure requires that the volume
element at x2 have unit wieght, i.e., da(x,) = 1. The volume element at xi
must have opposite orientation to be consistent with (3.5), so du(x,) = - 1.
Thus

s av duf = 44fW + 4%>f(Xl>

= fW - f(xA (4.2)
and (4.1) can be written

(4.3)

The right side of (4.3) is recognized as the average of left and right derivatives
with respect to arc length.
It is desirable to eschew such expressions as df/ds for “derivative with
respect to arc length” and aflax for “partial derivatives with respect to the
(scalar) coordinate x,” becausethey contain irrelevances. Different as they
appear, they refer to one and the samelimiting process.The derivative at x
320 HESTENES

depends only on the direction at x along which the limit is taken and not
on any particular curves passing through the point. The essentials are
expressed by the notation 0, f.
The operator 0, may sometimes be awkward to use because the li on
the right of (4.3) d oes not commute with other multivectors. In such cases
the “scalar differential operator” a, = 6V, may be recommended. Never-
theless, V, is more fundamental than a, because of its generalization by (4.1).
A comparably simple generalization of a, does not exist.
The derivative with respect to a pseudoscalar is expecially important. The
same result is obtained for all pseudoscalars, so it is convenient to drop the
subscript and write V.
Call V the gradient operator, to agree with common parlance when V
operateson a scalar. The gradient is a “vector differential operator,” so,
by virtue of (1.8),
Vf=V*fj-VAf. (4.4)
Call V *f the divergence off and V Af the curl off, to agree with the
terminology of vector and tensor analysis.For the specialcaseof a vector field
on &a, (1.4) can be used to get
VAf=iV Xf. (4.5)
Another familiar differential operator easily obtained from V is the laplacian
Vs. In fact, every differential operator on G?~
can be expressedassomeoperator
function of V.
One can think of VA as the gradient operator for the subspaceof &n
determined by A. However, if A is a function of x, the subspacewill depend
on the point at which the derivative is evaluated. Let it be understood that,
unlessotherwise specified, VAj has the value V,(,)f(x) at x, that is, the deri-
vative off at x is taken with respect to the value of A at x.
The general properties of the operator V, follow from the definition (4.1).

v-, = v, (4.6)

vet = VA for positive scalar A (4.7)

VAB = V, + V, if AB=AAB (4-Y

VAVB = - VBVA if AB=AAB and V,B V,A = 0

(4.9)

V,(f +gj = V*f + v‘4g (4.10)

v/dfg) = (vAf )g if g constant. (4.11)

 MULTIVECTOR CALCULUS 321

The operator equations (4.6) and (4.7) express the fact that VA depends
only on the direction of A and not on the orientation or magnitude of A.
Equations (4.8) and (4.9) sh ow how “gradient operators for orthogonal
subspaces of c+‘~” are related, and they determine how “laplacians” for ortho-
gonal subspaces “combine”:
(VA=)2 = VA2 + VB2. (4.12)
Equations (4.10) and (4.11) hardly need comment.
The convention that VA differentiates only to the right can be awkward
because of the noncommutivity of multiplication. If the convention is retained,
it is convenient to have a mark which indicates differentiation both to the left
and right when desired. Accordingly, the definition

This definition admits a simple form for the “Leibnitz rule” for differentiating
a product:
g 74f = (gY4)f + dLf ). (4.14)

On the right, only the function inside the parenthesis is to be differentiated.

The proof of (4.14) uses the identity

(4.15)

where f = f(x) is the value f at the point where the derivative is to be

taken, and f’ = f(d) is the value off at a point x’ on i3Y; likewise for the
other quantities. The last term on the right of (4.15) is identically zero
because of (3.5). In the limit, the next to the last term on the right of (4.15)
vanishes and the remaining terms give (4.14).
The relation of VA to the gradient is shown by the following:
V = A-IAV = A-l(A . V + A A V)
v = VA -/- vi, (4.16)
VA = A-IA .V (4.17)
ViA = A-lA A V. (4.18)

409/24b6
322 HESTENES

5. THE FUNDAMENTAL THEOREM OF CALCULUS

Letf be a multivector function defined on an oriented r-dimensional surface

Y in 8%with tangent v. Call Vf the tangential derivative off on Y. These
things being understood, the fundamental theorem can be stated as follows:
The integral of the tangential derivative off over V is equal to the integral
off over the boundary of V. As an equation,

1, dv V,f = I,, daf-

Note that this formula is independent of the dimensionof Y and of the space
in which Y’- is imbedded. A major motivation for the formulation of integra-
tion and differentiation in this paper hasbeen to achieve assimple and general
a statementof the fundamental theorem aspossible.For instance,the “l-vec-
tor property” of V, is appropriate becauseit relates integrals over 7cr and
?W-surfaces which differ by one dimension. Furthermore, the definition of
the derivative (4.1) has been made as similar to (5.1) as possible.
Various special cases of the fundamental theorem are called Green’s
theorem, Gauss’theorem, Stoke’stheorem, etc. But the general theorem is so
basic that it deservesa name which describesits scope.
A proof of the fundamental theorem is obtained by establishingthe follow-
ing sequenceof equations

(5.2)

The analytical details of the proof do not depend on the dimension of Y;

they differ in no essentialway from detailsin the proofs of specialcasesof the
theorem. Such proofs have been given on many occasions,though seldom
with the utmost generality, so no further comment is needed here.4
To illustrate the felicity and generality of (5.1), the integration theorems
of Gibbs’ “vector calculus” in Es can easily be derived.
If Y’Zsis a 3-dimensionalregion in 8s ?then one can write

dv = 1 dv 1 i, V, = V, da = in 1 da 1 ,

’ For a careful discussion of the problems involved, see M. R. Hestenes, Duke

Math. J., 8,300 (1941); A. B. Carson in “Contributions to the Calculus of Variations,”
p. 457. Univ. of Chicago Press, Chicago, Ill., 1938-1941.
 MULTIVECTOR CALCULUS 323

where n is the outward normal to Ya . Since the pseudoscalar i is constant, it

can be “factored out,” and (5.1) becomes

If f = C+J
is a scalar function, then (5.3) becomes

If f = E is a vector field, then (1.2), (1.14), (4.4), (4.5) can be used, and O-vec-
tor and 2-vector parts on each side of the equation can be equated separately
to get

If Ya is a 2dimensional surface, then one can write

dv = - in I da I , da = dx

V, = (in)-1 (in) * V = - ini(n A V) = in(n X V),

where dx is the differential of the coordinate x of a point on aYa , and n is the

“right-handed normal” to the surface Ya . So (5.1) becomes

[ysldaln xVf=$dccf. (5.4)

If f = v is a scalar, then

If f = E is a vector, then, as before, O-vector and 2-vector parts in (5.4)

can be separately equated to get

s, 1da 1n . (V x E) = $ & . E

j,, 1da I (n x V) x E = $ dx x E.
324 HESTENES

If v1 is a curve in 6s with endpoints a and 6, then one can write

dv = dx, dvV, = dx . V.

So (5.1) becomesthe familiar formula

j,ldxevf = j*df=f(b) -f(a). (5.5)

a

In spite of the easewith which the formulas of vector analysiscan be derived,

it is even easierto use (5.1) asit is or sometimesthe specialforms (5.3), (5.4),
or (5.5).
For each choice of a particular function f, (5.1) yields a formula relating
integrals over Y to integrals over 8% For instance, if v is a k-vector and x
is the coordinate of a point in c?~, then

V,x = k. (5.6)

So, if -Y is an k-dimensional surface, then

IY dv = $ $ dax. (5.7)

Becauseof (3.5), th is integral is independent of the choice of origin. If Y’”

is a flat surface, then its tangent v is constant, so

If W is an (r - 1)-dimensionalspherewith radius R and area 1a%‘-1, then

(5.8) reduces to

(5.9)

Many other useful consequencesof (5.7) can be easily found.

Actually, becauseof the noncommutivity of multiplication, (5.1) is not the
most generalform of the fundamental theorem. The necessarygeneralization
can be written

j,g dv o,f = javg daf9 (5.10)

where it is understood that dv is not differentiated by 0, . More explicitly,

since
o,v = (- l>,-1 v 8,) (5.11)
 MULTIVECTOR CALCULUS 325

(5.10) can be written

j,g dv v,f + (-- I>,l I, k OS)dvf = javg daf, (5.12)

The fundamentals have been set down. A complete geometric calculus

of multivector functions is now waiting to be worked out along lines similar
to the calculus of real and complex functions.