Michael Grossman, Robert Katz - Non-Newtonian Calculus (1972 - 2006, Kepler Press) PDF

Non-Newtonian Calculus
Michael arossman
(Lowell Technolo~ical Institute)
Robert Katz
1972
Lee Press
Pigeon Cove, Massachusetts 01966
0 Copyright 1972 by Michael Grossman and Robert Katz
All rights reserved
Manufactured in the United States of America
F o r e wo r d
We are pleased to present here the

results of our investigations into the
non-Newtonian calculi, which are simple
extensions of the classical calculus de-
veloped by Newton and Leibniz three cen-
turies ago.
Michael Grossman
Robert Katz
101 Granite Street

Rockport, Massachusetts
December 14, 1971
''for each successive class of
phenomena, a new calculus or a new
geometry, as the case might be,
which might prove not wholly inade-
quate to the subtlety of nature.''
Quoted, without citation,

by H.J.S. Smith; Nature,
Volume 8 (1873), page 450.
P R E F A C E
In August of 1970 we constructed a comprehensive family
of calculi, which includes the classical calculus as well as
an infinite subfamily of non-Newtonian calculi that we had
constructed three years earlier.
Each calculus possesses the following:
a distinctive method of measuring changes in
function arguments;
a distinctive method of measuring changes in ·
function values;
four operators: a gradient (i.e., an average
rate of change), a derivative, a natural
average, and an integral;
a characteristic class of functions having a
conetant derivative;
a Basic Theorem involving the gradient, deriv-
ative, and natural average;
a Basic Problem whose solution motivates a sim-
ple definition of the integral in terms of
the natural average;
and two Fundamental Theorems which reveal that
the derivative and integral are 'inversely'
related in an appropriate sense.
A popular method of creating a new mathematical system

is to vary the axioms of a known system. Although it may be
~ssible to axiomatize the classical calculus, we did not and
shall not pursue that course. Instead, in Chapter 1 we for-

mulate the classical calculus in a novel manner that leads
naturally to the subsequent construction of the non-Newtonian
calculi.
In Chapters 2 - 4, we construct three specific non-New-
tonian calculi: the geometric, anaqeometric, and bigeometric.
Chapter 5 is devoted to arithmetics (slightly specialized com-
plete ordered fields), which are used in subsequent chapters •
•
In Chapter 6 we simultaneously construct all the calculi, in-
dicate the unifornt relationships between the corresponding
operators of any two calculi, and provide suitable graphic
v
V1• P r e f a c e
interpretations. Chapters 7 and 8 contain brief developments
of the quadratic, anaquadratic, biquadratic, harmonic, anabar
monic, and biharmonic calculi. (The har1nonic calculus is not
to be confused with harn~nic analysis.) Chapter 9 includes
a variety of heuristic guides for selecting gradients, deriv-
atives, averages, and integrals. In Chapter 10 we discuss
certain generalized spaces, vectors, and least squares method:
which were suggested by our work with the non-Newtonian calcu·
li; we introduce a trend concept that may be considered as a
kind of global derivative; and we indicate some connections
between the non-Newtonian calculi and calculus in Banach
spaces. Sundry digressions and comments have been placed in
the Notes at the rear of the book, for example, in Note 2 we
explain how a simple algebraic identity suggested the possi-
bility of constructing the geometric calculus.
Since this book is intended for a wide audience, includ-
ing students, engineers, scientists, as well as mathematician:
we have presented many details that would not appear in a re-
search report and we have excluded all proofs. (All stated
results can be proved in a straightforward way.) It is as-
sumed, of course, that the reader has a working knowledge of
the rudiments of classical calculus.
Finally we wish to express our gratitude to Professor
Dirk J. Struik for his interest and encouragement, and to
Messrs. Charles Rockland, Charles K. Wilkinson, and David J.
Liben for their helpful comments and assistance. However,
we alone are responsible for the form and content of this
book.
Criticism and suggestions are cordially invited.
M.G.
R.K.
C 0 N T E N T S
Preliminaries, 1
Chapter 1: THE CLASSICAL CALCULUS
1.1 Introduction, 3
1.2 The Classical Gradient, 3
1.3 The Classical Derivative, 4
1.4 The Arithmetic Average, 5
1.5 The Basic Theorem of Classical Calculus, 6
1.6 The Classical Integral, 7
1.7 The Fundamental Theorems of Classical Calculus, 8
Chapter 2: THE GEOMETRIC CALCULUS
2.1 Introduction, 9
2.2 The Geometric Gradient, 10
2.3 The Geometric Derivative, 11
2.4 The Geometric Average, 12
2.5 The Basic Theorem of Geometric Calculus, 13
2.6 The Geometric Integral, 14
2.7 The Fundamental Theorems of Geometric Calculus, 16
2.8 Relationships to the Classical Calculus, 16
Chapter 3: THE ANAGEOMETRIC CALCULUS
3.1 Introduction, 18
3.2 The Anaqeometric Gradient, 18
3.3 The Anaqeometric Derivative, 19
3.4 The Anaqeometric Average, 20
3.5 The Basic Theorem of Anaqeometric Calculus, 21
3.6 The Anaqeometric Integral, 22
3.7 The Fundamental Theorems of Anageometric Calculus, 23
Chapter 4: THE BIGEOMETRIC CALCULUS
4.2 The Bigeometric Gradient, 25
4.3 The Biqeometric Derivative, 26
4.4 The Bigeometric Average, 27
4.5 The Basic Theorem of Bigeometric Calculus, 28
4.6 The Bigeometric Integral, 29
4.7 The Fundamental Theorems of Bigeometric Calculus, 30
Chapter 5: SYSTEMS OF ARITHMETIC
5.2 Arithmetics, 32
5.3 a-Arithmetic, 33
5.4 Geometric Arithmetic, 36
Chapter 6: THE •-CALCULUS
6.2 The •-Gradient, 39
6.3 The •-Derivative, 40
6.4 The •-Average, 41
• •
V11
• • •
V111 C o n t e n t s
6.5 The Basic Theorem of •-Calculus, 42

6.6 The •-Integral, 43
6.7 The Fundamental Theorems of •-Calculus, 44
6.8 Relationships to the Classical Calculus,44
6.9 Relationships Between Any Two Calculi, 45
6.10 Applications of Geometric Arithmetic, 47
6.11 Graphical Interpretations, 49
Chapter 7: THE QUADRATIC FAMILY OF CALCULI
7.1 The Quadratic Arithmetic, 52
7.2 The Quadratic Calculus, 53
7.3 The Anaquadratic Calculus, 55
7.4 The Biquadratic Calculus, 57
Chapter 8: THE HARMONIC FAMILY OF CALCULI
8.1 The Harmonic Arithmetic, 59
8.2 The Harmonic Calculus, 60
8.3 The Anaharmonic Calculus, 62
8.4 The Biharmonic Calculus, 64
Chapter 9: HEURISTICS
9.2 Choosing Gradients and Derivatives, 66
9.3 Choosing Integrals, 68
9.4 Choosing Averages, 70
9.5 Constants and Scientific Concepts, 72
Chapter 10: COLLATERAL ISSUES

10.2 •-Space, 75
10.3 •-Vectors, 78
10.4 The •-Method of Least Squares, 78
10.5 Trends, 81
10.6 Calculus in Banach Spaces, 82
10.7 Conclusion, 82
NOTES, 84
LIST OF SYMBOLS, 89
INDEX, 91
PRELIMINARIES
Since complex numbers are not used in this book, the

word "number" will mean real number. Except for f, g, and
h, lower-case Roman letters are used for referring to num-
bers. The letter R denotes the set of all numbers, and the
symbol R+ denotes the set of all positive numbers.
If r < s, then the interval [r,s] is the set of all
numbers x such that r < x < s. (We do not use open inter-
- -
vals in this book.) The interior of [r,s] is obtained by
deleting r and s therefrom. The classical extent of [r,s]
is s - r. (Although the number s - r is usually called the
length of [r,s], we use the term "classical extent" in order
to achieve a unifornt terminology for all the calculi.)
By a point we mean an ordered pair of numbers' by a
function we mean a set of points, each distinct two of which
have distinct first members. We use the letters f, g, h,
and occasionally others, for referring to functions, and we
usually identify a function with its graph, thereby avoid-
ing undesirable circumlocution.
The domain of a function is the set of all its argu-
ments (first members of the points); the range of a function
is the set of all its values (second members of the points).
A function is said to be on its domain, to be onto its range,
and to be defined at each of its arguments.
If every two distinct points of a function f have dis-
tinct second members, then f is one-to-one and its inverse,
£- 1 , is the one-to-one function consisting of all points
(y,x) for which (x,y) is a point of f. Please bear in mind
that f-l is not the reciprocal of f.
A positive function is a function whose values are all
•
positive; a discrete function is a function that has only a

finite number of arguments.
A linear function is a function of the foJ:Iu mx + c,
where m and c are constants and x is unrestricted in R. An
especially important linear function is the identity func-
tion, I, for which I(x) = x. (All linear functions are on
R.) The function exp assigns to each number x the number ex,
where e is the base of the natural logarithm function, ln,
1
2 Preliminaries
which is the inverse of exp.
In the sequel we shall introduce some new teruas and
symbols, which, however, we have endeavored to keep to a
• •
m1n1mum.
Chapter 1
THE CLASSICAL CALCULUS
1.1 INTRODUCTION
In this chapter we present some of the basic ideas of
the classical calculus in a novel manner that leads naturally
to the subsequent construction of the non-Newtonian calculi.
In the classical calculus, chanqes in function arguments
and values are measured by differences.
1.2 THE CLASSICAL GRADIENT

The classical change of a function f over [r,s] (or from
r to s) is the number f(s) - f(r).
A classically-unifonn function is a function that is on
R, is continuous, and has the same classical change over any
two intervals of equal classical extent. Clearly every con-
stant function on R is classically-uniform.
Since the classically-uniform functions turn out to be
the linear functions, we shall henceforth use the term "lin-
ear function."
It is characteristic of a linear function that for each
arithmetic progression of arguments, the corresponding se-
quence of values is also an arithmetic progression.
The classical slope of a linear function is its classi-
cal change over any interval of classical extent 1. Of
course, the classical slope of the function mx + c turns out
to be m.
The classical gradient of a function f over [r,s] is
the classical slope of the linear function containing the
points (r,f(r)) and (s,f(s)), and turns out to be
f(s) - f(r)
•
s - r
There are several reasons why we decided to use the
British texm "qradient" instead of "average rate of change"
or ''rate of change": the first term is the shortest; the
second leaves us wonderinq of what it is the average; the
third is often taken to mean "instantaneous rate of change."
3
4 The Classical Calculus
There will be no confusion with the gradient concept in vec-
tor analysis, since vector analysis is not treated in this
book.
Clearly the classical gradient is independent of the
origins used for measuring the magnitudes that the function
arguments and values may represent.
1.3 THE CLASSICAL DERIVATIVE
Throughout this section, f is assumed to be a function

defined at least on an interval containing the number a in
its interior.
If the following limit exists, we denote it by [Of] (a),
call it the classical derivative of f at a, and say that f
is classically differentiable at a:
f (x) - f (a)
lim •
x-ta x - a
The classical derivative of f, denoted by Of, is the

function that assigns to each number t the number [Of] (t) ,
if it exists.
The operator D is additive and homogeneous:
O(f + g) = Df + Og,
D(c·f) = c•Df, c constant.
(We are intentionally avoiding the term "linear operator.")
In the classical calculus a distinctive role is played
by the linear functions, for only they possess classical de-
rivatives that are constant on R.
Next we define a familiar concept in a special way that
will perntit a suitable generalization in Section 6. 3.
The classical tangent to a function f at the point
(a,f(a)) is the unique linear function g, if it exists,
which possesses the following two properties.
1. The li~ear function g contains (a,f(a)).
2. For each linear ·function h containing (a,f(a))
and distinct from q, there is a positive number
p such that for every number x in [a - p, a + p]
but distinct from a,
1.4 The Arithmetic Average 5
lg(x) - f(x) I < lh(x) - f(x) I .

Mr. Charles Rockland of Princeton University has proved
that [Df] (a) exists if and only if f has a classical tangent
at (a,f(a)), and that if [Df] (a) does exist, it equals the
classical slope of that classical tangent.
We say that two functions are classically tangent at a
common point if and only if they have the same classical tan-
gent there.
1.4 THE ARITHMETIC AVERAGE

The arithmetic average of n numbers v 1 , ••. ,vn is the
number (v 1 + ••• + vn)/n.
In our treatment of classical calculus an important role
is played by the arithmetic average of a continuous function
on an interval, the definition of which will be simplified by
introducing the concept of an arithmetic partition of an in-
terval.
An arithmetic partition of an interval [r,s] is any
arithmetic progression whose first and last terms are r and
s, respectively. An arithmetic partition is n-fold if it
has n terms.
The arithmetic average of a continuous function f on
[r,s] is denoted by
M8 f
r
and defined to be the limit of the convergent sequence whose
nth term is the arithmetic average of f(a 1 ), ••• ,f(an)' where
a 1 , ••• ,a is then-fold arithmetic partition of [r,s].
n s
The operator Mr is additive and homogeneous:
The arithmetic average of a linear function on [r,s] is

equal to the arithmetic average of its values at r and s, and
is equal to its value at the arithmetic average of r and s.
The arithmetic average is the only operator that has the
following three properties.
1. For any interval [r,s] and any constant function
h(x) =bon [r,s],
8
Mr h = b.
2. For any interval [r,s] and any functions f and g
that are continuous on [r,s], if f(x) < g(x) for
-
every number x in [r,s], then
Msf < Mrsq.
r -
3. For any numbers r,s,t such that r < s < t, and
any function f continuous on [r,t],
- (t - r)• M~f.
1.5 THE BASIC THEOREM OF CLASSICAL CALCULUS

Our discussion of the Basic Theorem of Classical Cal-
culus begins with its discrete analogue, which is a propo-
sition that concerns discrete functions and appropriately
conveys the spirit of the theorem.
The Discrete Analogue of the
Basic Theorem of Classical Calculus
Let h be a discrete function whose arguments
a 1 , ••• ,an are an arithmetic partition of [r,s].
Then the arithmetic average of the following
n - 1 classical gradients is equal to the
classical gradient of hover [r,s]:
I i = l, ••• ,n-1.
The Basic Theorem of Classical Calculus

If Dh is continuous on [r,s], then its arithmetic
average on [r,s] equals the classical gradient of
hover [r,s], that is
h(s) - h(r)
- ------.
s - r
In view of the preceding theorem we say that the arith-
1.6 The Classical Integral 7
metic average fits naturally into the scheme of classical

calculus.
The Basic Theorem of Classical Calculus provides an
immediate solution to the following problem.
The Basic Problem of Classical Calculus
Suppose that the value of a function h is known at
an argument r, and suppose that f, the classical
derivative of h, is continuous and known at each
number in [r,s]. Find h(s).
Solution
By the Basic Theorem of Classical Calculus we have
h(s) - h(r)
= M;(Dh) =
s - r
------.
Solving for h(s), we get
s
h ( s) = h ( r) + ( s - r ) • Mr f •
The number (s- r)•M;f in the foregoing solution arises

with sufficient frequency to warrant a special name, which
is introduced in the next section and is justified by the en-
suing discussion.
1.6 THE CLASSICAL INTEGRAL

The classical of a continuous function f on
inteqra~
s
[r,s] is the number (s - r)•Mrf' and is denoted by ls f •
r
We set lr f
r
= 0.
The classical integral is a weighted arithmetic average,

since
lr
sf
= M8
r
[ ( s - r) • f 1•
Furthermore, ls f equals the limit of the convergent sequence

r
whose nth terua is the sum kn•f(a 1 ) + ••• + k 0 •f(an_ 1 ), where
a 1 , ••• ,an is then-fold arithmetic partition of [r,s], and kn

is the common value of
s
The operator l
r
is additive and homogeneous:
[sr (c• f) =c ·1r f,

8
c constant.
The classical integral is the only operator that has

the following three properties.
1. For any interval [r,s] and any constant function
h(x) =bon [r,s],
1r
8
h = (s - r)•b.
2. For any interval [r,s] and any functions f and g

that are continuous on [r,s], if f(x) < q(x) for
-
3. For any numbers r, s, t such that r < s < t, and

t
lsf + l t f -
r s r
f. 1
1. 7 THE FUNDAMENTAL THEOREMS OF CLASSICAL CALCULUS
The classical derivative and integral are 'inversely'

related in the sense indicated by the following two theorems.
First Fundamental Theorem
Iff is continuous on [r,s], and
q(x) = [xf, for every number x in [r,s],

r
then
Dg = f, on [r,s].
Second Fundamental Theorem
If Dh is continuous on [r,s], then
ls
r
(Db) = h(s) - h(r).
Chapter 2
THE GEOMETRIC CALCULUS
2.1 INTRODUCTION
During the Renaissance many scholars, including Galileo,

discussed the following problem:
Two estimates, 10 and 1000, are proposed as
the value of a horse. Which estimate, if any,
deviates more from the true value of 100?
The scholars who maintained that the deviations should
be measured by differenqes concluded that the estimate of 10
was closer to the true value. However, Galileo eventually
maintained that the deviations should be measured by ratios,
and he concluded that the two estimates deviated equally from
the true value.
Now let us consider a more sophisticated problem. At
time r, a man invests f(r) dollars with a promoter who guar-
antees that at a certain subsequent time s, the value of the
investment would be f(s) dollars. In event that the investor
should desire to withdraw at any other time t, it was agreed
that the value of the investment increases continuously and
uniforutly. The problem is this: How much, f(t), would the
investor be entitled to at time t? We shall give two reason-
able solutions; there is no unique solution.
Solution 1. Since it was agreed that the value of the invest-
ment increases uniformly, we may reasonably assume that it in-
creases by equal amounts in equal times. Since it was agreed,
furthermore, that the value of the investment increases con-
tinuously, it can be proved that the growth must be linear
and, in fact, that
f(s) - f(r)
f(t) = f(r) + (t - r) •
s - r
Notice that the expression within the brackets represents the
classical gradient off over [r,s].
Solution 2. Since it was agreed that the value of the invest-
ment increases uniformly, we may reasonably assume that it in-
9
10 The Geometric Calculus
creases by equal percents in equal times. Since, furthermore,
the value of the investment increases continuously, it can be
proved that the growth must be exponential and, in fact, that
1 t-r
f (s) s-r
f(t) = f(r)• •
f (r)
We shall see that the expression within the brackets is of

fundamental importance in geometric calculus.
In the geometric calculus, changes in function arguments
and values are measured by differences and ratios, respec-
tively, and the operators are applied only to positive func-
tions. Therefore, every function in this chapter is positive,
except for the natural logarithm function, ln, which is used
for special purposes. Lest the reader be disappointed by
such a restriction, we hasten to say that a geometric-type
calculus for negative (valued) functions is discussed briefly
in Section 6.10.
Heuristic guides for selecting appropriate gradients,
derivatives, averages, and integrals are discussed in Chapter
9.
2.2 THE GEOMETRIC G~DIENT
The geometric change of a positive function f over [r,s]

is the number f(s)/f(r).
A geometrically-unifor!n function is a positive function
that is on R, is continuous, and has the same geometric change
over any two intervals of equal classical extent. Clearly
every positive constant function on R is geometrically-unifornt.
The geometrically-uniforut functions are those expressible
in the form exp(mx +c), where m and care constants and xis
unrestricted in R. If p is a positive constant, then the func-
tions pmx+c are also geometrically-uniform, for they are ex-
pressible in the form indicated.
It is characteristic of a geometrically-uniform function
that for each arithmetic progression of arguments, the corre-
sponding sequence of values is a geometric progression. That
fact provides one reason for the name "geometric calculus";
another reason will appear in Section 2.4.
2.3 The Geometric Derivative 11
The geometric slope of a geometrically-uniform function

is its geometric change over any interval of classical extent
1. For example, the geometric slopes of the functions
exp(mx + c) and pmx+c turn out to be e m and pm , respect1vely.
.
If one plots a geometrically-uniform function h on semi-

log paper that is logarithmically scaled on the y-axis, the
result is a straight line whose classical slope equals the
natural logarithm of the geometric slope of h. Other aspects
of graphical interpretation will be discussed more fully in
Section 6.11.
The geometric gradient of a positive function f over
[r,s] is the geometric slope of the geometrically-uniform
function containing (r,f(r)) and (s,f(s)), and turns out to
be
1
f(s) s-r
•
f (r)
Clearly the geometric gradient is independent of the

origin and unit used for measuring the magnitudes that the
arguments and values may represent, respectively.
The foregoing expression for the geometric gradient
yields the indeter1ninate form 1 00 when r = s, in contrast to
the indeterntinate forua 0/0 yielded by the expression for the
classical gradient.
2.3 THE GEOMETRIC DERIVATIVE

Throughout this section, f is assumed to be a positive
function defined at least on an interval containing the num-
ber a in its interior.
If the followinq limit exists and is positive, we de-
note it by [0£] (a), call it the geometric derivative off at
a, and say that f is geometrically differentiable at a:
1
f (x)
x-a
lim • •
X.fa f(a)
It is possible for tRe preceding limit to be 0 (this
happens, for example, if f(x) = exp(-x11 3 ) and a= 0). Our
reason for requiring [0f] (a) to be positive will be given in
Section 6.10.
It can be proved that [Of] (a) and [0f] (a) coexist; that
is, if either exists then so does the other. Moreover, if
they do exist, then [5f] (a) equals exp{[Df] (a)/f(a)} and
equals the geometric slope of the unique geometrically-uni-
form function that is classically tangent to fat (a,f(a)).
(Incidentally, the expression [Df] (a)/f(a) represents the
so-called logarithmic derivative of f at a, which is not a
non-Newtonian derivative according to our use of the term.)
The geometric derivative of f, denoted by Of, is the
function that assigns to each number t the number [~f] (t) ,
if it exists.
If f is geometrically-uniform, then 0£ has a constant
value equal to the geometric slope of f. Indeed, only geo-
metrically-uniform functions have geometric derivatives that
are constant on R. In particular, if f is a positive con-
stant function on R, then Of is everywhere equal to 1.
~ ,.,., ~ ,.,.,
Since D(l + 1) ~ D(l) + D(l), the operator D is not
""'-~
additive; however D is multiplicative:

~ ~ l'ttJ
D(f•g) = Df·Dg.
Furthermore,
Recalling that the function exp equals its classical de-

rivative, we point out that the function exp[exp(x)] equals
its geometric derivative.
In Note 1 we define the relative gradient, which is in-
timately related to the geometric gradient, is often called
the compound growth rate,and gives rise to the relative de-
rivative.
2.4 THE GEOMETRIC AVERAGE

The geometric calculus was so named principally because
the geometric average fits naturally thereinto, a fact that
will be explained in the next section.
The geometric average of n positive numbers v 1 , ••• ,vn
is the positive number
2.5 The Basic Theorem of Geometric Calculus 13
The geometric average of a continuous positive function
f on [r,s] is denoted by Msf and defined to be the positive
r
limit of the convergent sequence whose nth term is the geo-
metric average of f(a 1 ), ••. ,f(an)' where a 1 , •.. ,an is the
n-fold arithmetic partition of [r,s].
~s
The operator M is multiplicative:
r
~s
~
r
(f·g) = Mr f
~s ~s
· M g.
r
Also,
M~(fc) = (M~f)c, c constant.
The geometric average of a geometrically-uniform func-
tion on [r,s] is equal to the geometric average of its val-
ues at r and s, and is equal to its value at the arithmetic
average of r and s.
The geometric average is the only operator that has the
following three properties.
1. For any interval [r,s] and any positive constant func-
tion h(x) =bon [r,s],
~h = b.
2. For any interval [r,s] and any positive functions f
and g that are continuous on [r,s], if f(x) ....
< g(x)
for every number x in [r,s], then
M~£ ~ M~g.
3. For any numbers r, s, t such that r < s < t, and any

positive function f continuous on [r,t],
rMs£1s-r • rMt£1t-s = rMrt£1t-r.
r s
2.5 THE BASIC THEOREM OF GEOMETRIC CALCULUS
Our investigation of non-Newtonian calculi began with

an observation of the simple identity indicated in the fol-
lowing proposition. (In Note 2 we explain how this identity
provided the original insight into the constructibility of
the geometric calculus.)
The Discrete Analosue of the
Basic Theorem of Geometric Calculus
Let h be a discrete positive function whose arguments
a 1 , ••• ,an are an arithmetic partition of [r,s].
Then the geometric average of the following n-1
geometric gradients is equal to the geometric
gradient of h over [r,s]:
1
, 1. = 1 , ••• ,n- 1 .
h (a.)
1
The Basic Theorem of Geometric Calculus

~
If Dh is continuous on [r,s], then its geometric

average on [r,s] equals the geometric gradient
of hover [r,s], that is
1
h (s) s-r
•
h (r)
In view of the preceding theorem we say that the geo-

metric average fits naturally into the scheme of geometric
calculus.
The Basic Theorem of Geometric Calculus provides an
immediate solution to the following problem.
The Basic Problem of Geometric Calculus
Suppose that the value of a positive function h
is known at an argument r, and suppose that £,
the geometric derivative of h, is continuous and
known at each number in [r,s]. Find h(s).
Solution
h (s) = h (r)·[M8 f] s-r.
r
The number [M 8 £] 8 -r in the foregoing solution arises
r
with sufficient frequency to warrant a special name, which
is introduced in the next section.
2.6 THE GEOMETRIC INTEGRAL
The geometric integral of a continuous positive func-

tion f on [r,s] is the positive number [M8 £]s-r and is de-
r
2.6 The Geometric Integral 15
noted by l~s
r
f• We set
r
f = 1.
The geometric integral is a weighted geometric average,
•
s1nce s
f =
,.....,s
FurtherJnore, 1f r
equals the positive limit of the conver-
gent sequence whose nth term is the product

k k k
[f (a 1 ) 1 n • [f (a 2 ) 1 n • • · [f (an-l) 1 n,
where a 1 ,a 2 , ••• ,an is then-fold arithmetic partition of
[r,s], and kn is the common value of a 2 - a 1 , a 3 - a 2 , ••• ,
8 n- an-1·
~
The operator l s is multiplicative:

r
~s ~s ~s
1 (f·g) = I
r r
f • 1 9·
r
Also,
c
, c constant.
The geometric integral is the only operator that has

1. For any interval [r,s] and any positive constant

function h(x) =bon [r,s],
J~s
r
h = bs-r.
2. For any interval [r,s] and any positive functions
f and q that are continuous on [r,s], if f(x) < g(x)
-
<
-
3. For any numbers r,s,t such that r < s < t, and any
positive function f continuous on [r,t],
~s ,._,t ,...,t
1r 1s
f • f = Ir £.
2.7 THE FUNDAMENTAL THEOREMS OF GEOMETRIC CALCULUS
The geometric derivative and integral are 'inversely'

If f is positive and continuous on [r,s], and
,._.,X
g(x) = 1 r
£, for every number x in [r,s],
then
~
Dg = f, on [r,s].
~
If Dh
.1s continuous on [r,s], then
1 IIIWS
r
(Db) = h ( s) /h ( r) •
Just as the Second Fundamental Theorem of Classical Cal-

culus is useful for evaluating classical integrals, so is the
Second Fundamental Theorem of Geometric Calculus useful for
evaluating geometric integrals. For example, let f(x) =
IIIW
exp(l/x) and h(x) = x, for x > 0. Then f = Dh, and so
IIIWS 5
13
f = (0h) = h(5)/h(3) = 5/3.
2.8 RELATIONSHIPS TO THE CLASSICAL CALCULUS

Our presentation of the geometric calculus was indepen-
dent of the classical calculus, for the operators of the for-
mer were not defined in terms of the operators of the latter.
However, the corresponding operators of the two calculi are
uniformly related.
Let f be a positive function and set f(x) = ln(f(x)). -
s- -
Let Grf be the classical gradient off over [r,s], and let
G~f be the geometric gradient off over [r,s]. Then we have
the following uniforut relationships.
(1) G~f = expjG~fJ
(2) (a) = exp I [Of] (a)J
[0£]
(3) M~f = exp~M~f}
(4) Js f
r
= exp{/s
r
f}
2.8 Relationships to the Classical Calculus 17
,...,
Remark. For (2) we assume that [Df] (a) exists; for (3) and
(4), that f is continuous on [r,s].
The preceding observations clearly indicate that for
each theorem in classical calculus there is a corresponding
theorem in geometric calculus, and conversely. For example,
we have the following mean value theorem of geometric calcu-
lus.
If a positive function f is continuous on [r,s]
and geometrically differentiable everywhere be-
tween r and s, then between r and s there is a
number at which the geometric derivative of f
equals the geometric gradient off over [r,s].
The relationship of the nth-geometric derivative to the

nth-classical derivative follows the familiar pattern:
(i5n£] (a) = exp{ [Dnf] (a)},
-
where f(x) = ln(f(x)).
The second geometric derivative of the function exp(-x 2 )
2
turns out to be the constant l/e , a fact that may provide
additional insight into that important function.
Chapter 3
THE ANAGEOMETRIC CALCULUS
3.1 INTRODUCTION
In the anageometric calculus, changes in function argu-
ments and values are measured by ratios and differences, re-
spectively, and the operators are applied only to functions
whose arguments are positive. However, an anageometric-type
calculus for functions with negative arguments is discussed
briefly in Section 6.10.
A positive interval is an interval [r,s] for which
0 < r < s. The geometric extent of a positive interval
[r,s] is the number s/r.
We were interested to learn that in his Principia, New-
ton expressed Galilee's law of descent of freely falling bod-
ies as follows: "When a body is falling, ••• the spaces de-
scribed in proportional times are as ••• the squares of the
times." (Scholiwn to Corollary VI in "Axioms, or Laws of
Motion.") That is, in time intervals of equal geometric ex-
tent, the distances traversed are proportional to the squares
of the times elapsed.
derivatives, averages, and integrals are discussed in Chap-
ter 9.
3.2 THE ANAGEOMETRIC GRADIENT
An anageometrically-unifor!" function is a function that

is on R+' is continuous, and has the same classical change
over any two positive intervals of equal geometric extent.
Clearly every constant function on R+ is anageometrically-
uniforiu.
The anageometrically-uniform functions are those express-
ible in the form ln(cxm), where c and mare constants, c > 0,
and x is unrestricted in R+.
It is characteristic of an anageometrically-uniform func-
tion that for each geometric progression of arguments, the
corresponding sequence of values is an arithmetic progression.
The anaqeometric slope of an anageometrically-uniform
18
3.3 The Anageometric Derivative 19
function is its classical change over any positive interval

of geometric extent e. (In Note 3 and Section 6.10 we ex-
plain why e was chosen here.) The anageometric slope of the
function ln(cxm) turns out to be m.
If one plots an anageometrically-~niforna function h on
semi-log paper that is logarithmically scaled on the x-axis,
the result is a straight line whose classical slope equals
the anageometric slope of h. Other aspects of graphical in-
terpretation will be discussed more fully in Section 6.11.
The anageometric gradient of a function f over a posi-

tive interval [r,s] is the anageometric slope of the anageo-
metrically-uniforna function containing (r,f(r)) and (s,f(s)),
and turns out to be
f(s) - f(r)
•
ln(s) - ln(r)
It is easy to verify that the anageometric gradient is
independent of the unit and origin used for measuring the
magnitudes which the arguments and values may represent, re-
spectively.
The foregoing expression for the anageometric gradient
yields the indeterminate form 0/0 when r = s.
3.3 THE ANAGEOMETRIC DERIVATIVE

defined at least on a positive interval containing the num-
ber a in its interior.
If the following limit exists, we denote it by lR£1 (a),
call it the anageometric derivative of f at a, and say that
f is anageometrically differentiable at a:
f(x) - f(a)
lim •
x-ta ln(x) - ln(a)
It can be proved that [Df] (a) and [Df] (a) coexist; that
#IW
if they do exist, then

[l}f] (a) = [Df] (a)/ [D (ln)] (a) = a· [Df] (a);
and that [Df] (a) equals the anageometric slope of the unique
~
20 The Anageometric Calculus
anageometrically-uniform function which is classically tan-
gent to fat (a,f(a)).
Observe that [Ef1 (a) equals the classical derivative of
f with respect to ln at a.
The anageometric derivative of £, denoted by Rf, is the
function that assigns to each number t the number [~f) (t) ,
if it exists.
The operator ~ is additive and homogeneous:
If f is anageometrically-uniform, then Of has a constant

--
value equal to the anageometric slope of f. Indeed, only ana-
geometrically-uniform functions have anageometric derivatives
that are constant on R+.
It is worth noting that if h(x) = mx, where m is a con-
stant and x > 0, then ,.,Dh = h.
3. 4 THE ANAGEOMETRIC AVERAGE
Since changes in function arguments are measured by ra-

tios in the anageometric calculus, one should not be surprised
that the definition of the anageometric average requires the
use of partitions in which the successive points form a geo-
metric progression.
A geometric partition of a positive interval [r,s] is
any geometric progression whose first and last terms are r
and s, respectively. A geometric partition is n-fold if it
has n terms.
The anageometric a~erage of a continuous function f on
a positive interval [r,s] is denoted by M8 f and defined to
~r
be the limit of the convergent sequence whose nth term is

the arithmetic average of f(a 1 ) , ••• ,f(an), where a 1 , ••• ,an
is then-fold geometric partition of [r,s].
The arithmetic and anageometric averages are NOT iden-
tical; for example, if f(x) = x, for x > 0, then
M 8
Alfr f = (s - r)/[ln(s) - ln(r)],
whereas
Msf
r
= (s + r)/2.
3.5 The Basic Theorem of Anageometric Calculus 21
Nevertheless, the operator Ms is additive and homoge-
~r
neous:
s s
~r(c·f) = c·~rf' c constant.
The anageometric average of an anageometrically-uniform

function on [r,s] is equal to the arithmetic average of its
values at r and s, and is equal to its value at the geomet-
ric average of r and s.
The anageometric average is the only operator that has
1. For any positive interval [r,s] and any constant

s
M h = b.
~r
2. For any positive interval [r,s] and any functions

f and g that are continuous on [r,s], if f(x) < g(x)
-
Msf < Msg •
.wr - ~r
3. For any numbers r,s,t such that 0 < r < s < t, and
8
Nr f + [ln(t) - ln(s)]·Mtf
[ln(s) - ln(r)]·M NS
= [ln(t) - ln(r)]·Mtf.
IIIWr
3.5 THE BASIC THEOREM OF ANAGEOMETRIC CALCULUS
In view of the following theorem we say that the anageo-

metric average fits naturally into the scheme of anageometric
calculus. (We forego the discrete analogue.)
The Basic Theorem of Anageometric Calculus
If Rh is continuous on a positive interval [r,s],
then its anageometric average on [r,s] equals the
anageometric gradient of hover [r,s], that is,
h(s) - h(r)
Mrs (Dh) =
N _, •
ln(s) - ln(r)
The preceding theorem provides an immediate solution to
the following problem, which will motivate our definition of
the anageometric integral.
The Basic Problem of Anageometric Calculus
Suppose that the value of a function h is known
at a positive argument r, and suppose that f,
the anageometric derivative of h, is continuous
and known at each number in [r,s]. Find h(s).
Solution
h(s) = h(r) + [ln(s) - ln(r)].M 8
,_r f.
3.6 THE ANAGEOMETRIC INTEGRAL

The anageometric integral of a continuous function f on
8
a positive interval [r,s] is the number [ln(s) - ln(r)]·M
._.,r f
and is denoted by J
"'r
sf. We set / r f
~r
= 0.
The anageometric integral is a weighted anageometric

•
average, s1nce
sf-- ~;{[ln(s) - ln(r)]·f}.

/41Wr
Furthermore, ls f
~r
equals the limit of the convergent se-
quence whose nth term is the sum

[ln(kn)]·f(a 1 ) + ••• + [ln(kn)]·f(an-l)'
where a 1 , ••• ,an is then-fold geometric partition of [r,s],
and kn is the common value of a 2 /a 1 , a 3/a 2 , ••• , an/an-l·
The reader who is familiar with Stieltjes integrals
will notice that the preceding result justifies the asser-
tion that
is the Stieltjes integral off with respect to ln on [r,s].
The operator / s is additive and homogeneous:

41Wr
3.7 Fundamental Theorems of Anageometric Calculus 23
/
-v r
8
(c· f) =c ·ls
-..,r
f, c constant.
The anageometric integral is the only operator that has

1. For any positive interval [r,s] and any constant
8
[ h = [ln(s) - ln(r)]·b.
~r
2. For any positive interval [r,s] and any functions

f and 9 that are continuous on [r,s], if f(x) < g(x)
-
any function f continuous on [r, t] ,
[sf + ltf - l t f.
"'r --s ,...,r
3 • 7 THE FUNDAMENTAL THEOREMS OF ANAGEOMETRIC CALCULUS
The anageometric derivative and integral are 'inversely'

Iff is continuous on a positive interval [r,s], and
g(x) = /xf, for every number x in [r,s],
-vr
then
99 = f, on [r,s].

If gh is continuous on a positive interval [r,s], then
1
.-,r
8
( Bh) = h ( s) - h ( r) •
The preceding theorem is useful, of course, for evalu-

ating anageometric integrals.
G5 f be the anageometric gradient of f over a posi-

Let Nr
.. X -
tive interval [r1sl1 and set f(x) = f(e) 1 r = ln(r) 1 s=
ln(s), and a= ln(a). (Of course, G~f
r
is the classical gra-
dient of f- over [r,s] .) Then we have the following uniform
relationships between the corresponding operators of the ana-
geometric and classical calculi.
(1) 5
G f
-
= Gsf
.-vr r
(2) [Of]
#1111
(a) -- ro£1 (a)
8 M!l
(3) M f =
~r f
-
(4) /sf= /_sr
~r r
Remark. For (2) 1 we assume that [Df] (a) exists; for (3) and
. N
(4) 1 that f is continuous on [r,s].

theorem in anageometric calculus, and conversely.
Chapter 4
THE BIGEOMETRIC CALCULUS
4.1 INTRODUCTION
In the bigeometric calculus, changes in function argu-

ments and values are measured by ratios, and the operators
are applied only to functions with positive arguments and
positive values. However, bigeometric-type calculi for oth-
er functions can be constructed (Section 6.10).
derivatives, averages, and integrals are discussed in Chap-
ter 9.
4.2 THE BIGEOMETRIC GRADIENT
A bigeometrically-uniform function is a positive func-

tion that is on R+, is continuous, and has the same geomet-
ric change over any two positive intervals of equal geomet-
ric extent. Clearly ~very positive constant function on R+
is bigeometrically-uniform.
The bigeometrically-uniform functions are those express-
ible in the form cxm, where c and mare constants, c > 0, and
x is unrestricted in R+.
It is characteristic of a bigeometrically-uniform func-
tion that for each geometric progression of arguments, the
corresponding sequence of values is also a geometric progres-
•
s1on.
The bigeometric slope of a bigeometrically-uniform func-
tion is its geometric change over any positive interval of
geometric extent e. (Our reason for choosing e here is sim-
ilar to the reason for using e in defining anageometric slope,
as explained in Note 3 and Section 6.10.) The bigeometric
slope of the function cxm turns out to be em.
If one plots a bigeometrically-uniform function h on
log-log paper, the result is a straight line whose classical
slope equals the natural logarithm of the bigeometric slope
of h. Other aspects of graphical interpretation will be dis-
cussed more fully in Section 6.11.
The bigeometric gradient of a positive function f over a
25
26 The Bigeometric Calculus
positive interval [r,s] is the bigeometric slope of the bi-
geometrically-uniform function containing (r,f(r)) and
(s,f(s)), and turns out to be
f(s) 1/[1n(s)-1n(r)]
•
f (r)
It is easy to verify that the bigeometric gradient is

independent of the units used for measuring the magnitudes
which the arguments and values may represent.
The foregoing expression for the bigeometric gradient
yields the indeterminate form 1~ when r = s.
4.3 THE BIGEOMETRIC DERIVATIVE

Throughout this section, f is assumed to be a positive
function defined at least on a positive interval containing
the number a in its interior.
If the following limit exists and is positive, we de-
note it by [~£](a), call it the bigeometric derivative off
at a, and say that f is bigeometrically differentiable at a:
f(x) 1/[1n(x)-1n(a)]
lim •
x-ta f (a)
Our reason for requiring [p£1 (a) to be positive will be in-

dicated in Section 6.10.
It can be proved that [Df] (a) and lD£1 (a) coexist; that
if they do exist, then
[~f] (a) =
exp{a. [Df] (a) /f (a)};
and that [~f] (a) equals the bigeometric slope of the unique
bigeometrically-uniform function which is classically tan-

gent to f a t (a,f(a)).
The bigeometric derivative of £, denoted by ~£, is the
function that assigns to each number t the number [~£] (t) ,
if it exists.
The operator B is multiplicative:
o(f·g> = i5£·Dg.
~ #Itt# N
Furthermore,
4.4 The Bigeometric Average 27
If f is bigeometrically-uniforna, then ~f has a constant

value equal to the bigeometric slope of f. Indeed, only bi-
geometrically-uniforna functions have bigeometric derivatives
that are constant on R+.
It is also worth noting that if h(x) = exp(x) for x > 0,
then f>h = h.
We noted above that
£P£1 (a) = exp{a• [Df] (a) /f (a)}.
Since economists refer to the expression within the braces as
the elasticity of f at a, we call [~f] (a) the resiliency of f
at a. We believe that resiliency will prove to be more use-
ful than elasticity because the former is the derivative in a
complete system of calculus (the naturalness of which is ex-
hibited in this chapter), whereas it appears to be impossible
to construct a complete, natural system of calculus in which
the derivative is the elasticity.
Perhaps the psychophysicists will find some interest in
the bigeometric calculus, for one of their basic laws may be
stated thus: The resiliency of the stimulus-sensation func-
tion is constant. (That constant is determined by the nature
of the stimulus.)
The biqeometric calculus may also prove to be useful in
biology, for a fundamental law of growth is the following:
If f is the function relating the size of one organ to the
size of any other given organ in the same body at the same
instant, then, within certain time limits, the resiliency of
f is constant.
A physicist who preferred not to settle on specific units
of time and distance could, nevertheless, assert that the hi-
geometric speed of an object falling freely to the earth is
constant.
Heuristic guides for selecting appropriate derivatives
are discussed more fully in Section 9.2.
4.4 THE BIGEOMETRIC AVERAGE

The bigeometric average of a continuous positive func-
tion f on a positive interval [r,s] is denoted by B~f and de-
fined to be the positive limit of the convergent sequence
whose nth term is the geometric average of f(a 1 ) , ••• ,f(an),
where a 1 , ••• ,an is then-fold geometric partition of [r,s].
The geometric and bigeometric averages are not identi-
cal; for example, if f(x) = x 2 for x > 0, then
exp{2/(e- 1)},
whereas
#VMef
~1
= e.
M8 is multiplicative:
Nevertheless, the operator ,__r
M8
~r
(f•g) = M8 f·Msg.
~r ~r
Also,
The bigeometric average of a bigeometrically-uniform

function on [r,s] is equal to the geometric average of its
values at r and s, and is equal to its value at the geomet-
ric average of r and s.
The bigeometric average is the only operator that has
1. For any positive interval [r,s] and any positive
constant function h(x) =bon [r,s],
~ h
5
~r
= b.
functions f and g that are continuous on [r,s],if
f(x) < g(x) for every number x in [r,s], then
-
any positive function f continuous on [r,t],
[ln(s)-ln(r)] "'t [ln(t)-ln(s)]
[Ms£1
,...,r • [tJsf]
"'t [ln(t)-ln(r)]
- [tJrf 1 •
4.5 THE BASIC THEOREM OF BIGEOMETRIC CALCULUS

In view of the following theorem we say that the bigeo-
metric average fits naturally into the scheme of bigeornetric
calculus. (We forego the discrete analogue.)
4.6 The Bigeometric Integral 29
The Basic Theorem of Bigeometric Calculus

If Dh is continuous on a positive interval [r,s],
N
then its bigeometric average on [r,s] equals the

bigeometric gradient of hover [r,s], that is,
h(s) 1/[ln(s)-ln(r)]
•
h(r)
the bigeometric integral.
The Basic Problem of Bigeometric Calculus
Suppose that the value of a positive function h
is known at a positive argument r, and suppose
that f, the bigeometric derivative of h, is con-
tinuous and known at each number in [r,s]. Find
h ( s) •
Solution
h(s) = h(r). [Msf] [ln(s) -ln(r) 1.
~r
4.6 THE BIGEOMETRIC INTEGRAL

The bi9eometric integral of a continuous positive func-
tion f on a positive interval [r,s] is the positive number
[Msf][ln(s)-ln(r)]
'Wr
A's Air
and is denoted by / f. We set / f = 1.
~r .._,r
The bigeometric integral is a weighted bigeometric aver-
•
age, s1.nce
Ns
j
f = ~:{f[ln(s)-ln(r)1}.
~r
Als
Furthermore, J
~r
f equals the positive limit of the convergent
sequence whose nth term is the product

where a 1 , ••• ,an is then-fold geometric partition of [r,s],
and kn is the common value of a 2;a1 , a 3;a 2 , ••• ,an/an-l"
Ns
The operator J
is multiplicative:
~r
s ....,s
1 (f•g) .NrI g.
~'~s
- f
~r
Also,
41Vs ~s c
/ (fc) - '- f I c constant.
-vr 4Vr
The bigeometric integral is the only operator that has
constant function h(x) =bon [r,s],
1h
""'s
= b ln ( s) -ln ( r) •
~r
2. For any positive interval (r,s] and any positive

functions f and g that are continuous on [r,s], if
f(x) < g(x) for every number x in [r,s], then
-
Ns ~s
f ~ g. 1
4Wr
I4Wr
any positive function f continuous on [r,t],
""s t
1f
._, r
•
""s
f =
t
f•
4.7 THE FUNDAMENTAL THEOREMS OF BIGEOMETRIC CALCULUS

The bigeometric derivative and integral are 'inversely'
If f is positive and continuous on a positive inter-
val [r,s], and
,_.,X
g (x) =/ f, for every number x in [r, s] ,
"'r
then 1/W
Dg
_.., = f, on [r,s].
4.8 Relationships to the Classical Calculus 31
If ~his continuous on a positive interval [r,s], then
Ns
(Dh) = h(s)/h(r).
"'
The preceding theorem is useful for evaluating bigeo-
metric integrals.

G8 £ be the bigeometric gradient of a positive func-
Let ._,r
ti0n f OVer a eositiVe interval [r,S] 1 and Set f(X) = -
ln(f(ex)), r = ln(r), i = ln(s) I and a= ln(a). (Of course,
G~l is the classical gradient off over [f,J].) Then we have
the following uniform relationships between the corresponding
operators of the bigeometric and classical calculi.
(1) ~s
c..; f
s-
= exp{Gff}
---r
(2) [Df]
N
Ill#
(a) = -
exp{ [Df] (a)}
( 3)
(4)
N
Remark. For (2) we assume that [Df] (a) exists; for (3) and
IV
(4), that f is continuous on [r,s].

theorem in bigeometric calculus, and conversely.
Chapter 5
SYSTEMS OF ARITHMETIC
5.1 INTRODUCTION
In this chapter we discuss the general concept of an
arithmetic, without which it would be impractical to con-
struct the non-Newtonian calculi in Chapter 6. We also
present here one specific arithmetic, the geometric arith-
metic, which will be used in Section 6.10. In Chapter 7
we shall discuss and use the quadratic arithmetic, and in
Chapter 8, the harmonic arithmetic.
5.2 ARITHMETICS
The concept of a complete ordered field evolved from
the axiomatization of the real number system, whose basic .
ideas are assumed known to the reader. * Informally, a
complete ordered field is a system consisting of a set A,
• • • •
four (binary) operations, +, -, x, I for A, and an order-
•
inq relation < for A, all of which behave with respect to
A exactly as +, -, x, /, < behave with respect to the set
of all numbers. We call A the realm of the complete or-
dered field.
There are infinitely-many complete ordered fields,
all of which are structurally equivalent (isomorphic) •
The real number system is nowadays conceived as an arbi-
trarily selected complete ordered field whose realm is
denoted by R and whose operations and ordering relation
are denoted by+, -, x, /, and<.
By an arithmetic we mean a complete ordered field
whose realm is a subset of R. (Although the term "arith-
metic" is commonly used for referring to the system or
study of the positive integers, we have taken the liberty
of using the term in the sense just indicated.) There are
infinitely-many arithmetics, one of which is the real num-
* Complete ordered fields are discussed in many textbooks
on modern algebra and in some works on analysis. A detailed
new axiomatic treatment of the real number system is pre-
sented in the authors' Axiomatic Analysis, obtainable from
Lee Press.
32
5.3 a-Arithmetic 33
ber system, henceforth called the classical arithmetic.
• • • • •
The rules for handling any arithmetic (A,+,-,x,j,<)
are exactly the same as the rules for handling classical
arithmetic. For example, + and x are commutative and as-
sociative; x is distributive with respect to +; < is tran-
sitive; and there are two unique numbers 0 and i in A such
• • • •
that y + 0 = y and y x 1 = y, for every number y in A. The
notation "y <• z" is, of course, an abbreviation of "y <• z
-
or y = z."
Although all arithmetics are structurally equivalent,
only by distinguishing among them do we obtain suitable
tools for constructing all the non-Newtonian calculi. But
the usefulness of arithmetics is not limited to the con-
struction of calculi; we believe there is a more fundamen-
tal reason for considering alternative arithmetics: they
may also be helpful in developing and understanding new sys-
tems of measurement that could yield simpler physical laws.
In his Foundations of Science (formerly titled, Physic~:
The Elements) , Norntan Robert Campbell, a pioneer in the the-
ory of measurement, clearly recognized that alternative arith-
metics might be useful in science, for he wrote, "we must rec-
ognize the possibility that a system of measurement may be ar-
bitrary otherwise than in the choice of unit; there may be ar-
bitrariness in the choice of the process of addition."
5. 3 a-ARITHMETIC
A generator is a one-to-one function whose domain is R

and whose range is a subset of R. For example, the following
are generators: the identity function I, the function exp,
and the function x 3 •
Consider any generator a with range A. By a-arithmetic
we mean the arithmetic whose realm is A and whose operations
and ordering relation are defined as follows.
34 Systems of Arithmetic
a-addition ••••••••••• y +z = a{a- 1 (y) + a- 1 (z)}
a-subtraction •••••••• y ~ z = a{a- 1 (y) - a- 1 (z)}
a-multiplication ••••• y X z = a{a- 1 (y) x a- 1 (z)}
a-division (z ~ 0) ... y i z = a{a- 1 {y) I a- 1 {z)}

•
a-order •••••••••••••• y < z if and only if
a- 1 {y) < a- 1 {z)
We say that a 9enerates a-arithmetic; for example, the

identity function generates classical arithmetic, and the
function exp generates geometric arithmetic, which is dis-
cussed in Section 5.4. Each generator generates exactly one
arithmetic, and, conversely, each arithmetic is generated by
exactly one generator.
All concepts in classical arithmetic have natural coun-
terparts in a-arithmetic, some of which we now discuss.
The a-positive numbers are the numbers x in A such that
• • • •
0 < x; the a-negative numbers are those for which x < 0. The
• and the a-one, 1,
a-zero, 0, •
turn out to be a(O) and a(l) •
The a-integers consist of 0• and all the numbers that result
• • •
• .
by successive a-addition of 1 to 0 and by success1ve a-sub-
traction of 1 from 0. Thus the a-integers turn out to be the
following:
• • • I a(-2), a(-1), a(O), a(l), a(2), •••

For each integer n, we set n• = a(n). Of course, if n•
is an a-positive integer, then
• • • • •
n =1 + ••• + 1.
n terms
The a-absolute value of a number x in A is
• •
• •
if
X X > 0
- •
IX I 0 if X = 0
•• • 0•
O-x if X <
5.3 a-Arithmetic 35
• • •
The a-average of n numbers ul' ••• , un 1n A l.S the un1.que
number u in A such that •
• • - • •
u + ••• + u ul + • • • + u n •
n terms
It turns out that u equals
which equals
-1 -1
a{(a (u 1 ) + ••• +a (un)]/n}.
We shall occasionally refer to the a-average as the nat-

ural average in a-arithmetic. The natural average in clas-
sical arithmetic is the arithmetic average; the natural aver-
age in geometric arithmetic is the geometric average.
Although the a-average is widely known, we have seen no
reference to it as the natural counterpart in a-arithmetic of
the arithmetic average in classical arithmetic. Indeed, the
a-average has the same properties in the context of a-arith-
metic as the arithmetic average has in the context of classi-
cal arithmetic. (An important example will be given in Sec-
tion 10. 4.)
An ~-progression (or natural pro9ression in a-arithmetic)
is a finite sequence of numbers u 1 , ••• ,un in A such that
•
ui+l- ui is the same for every integer i from 1 to n-1. In
classical arithmetic the natural progressions are the arith-
metic progressions; in geometric arithmetic they are the geo-
metric progressions.
•
For any numbers r and s in A, if r < s, then the set of
• •
all numbers x in A such that r < x < s is called an a-inter-
• • - - •
val, is denoted by [r,s], has a-extent of s - r, and has an
•
a-interior consisting of all numbers x in A such that r <
•
X < s.
An a-partition of an a-interval lr,sf is any a-progres-
sion whose first and last terms are r and s. An a-partition
is n-fold if it has n terms.
Let {un} be an infinite sequence of numbers in A. Then
there is at most one number u in A such that every ~-interval
with u in its a-interior contains all but finitely-many terms
36 Systems of Arithmetic
If there is such a number u, then {u n } is said to
be a-convergent to u, which is called the a-limit of {un}·
For a = I, a-convergence reduces to classical conver-
gence.
5.4 GEOMETRIC ARITHMETIC
The arithmetic generated by the function exp will be

called geometric arithmetic rather than exp-arithmetic. Sim-
ilarly, the notions in geometric arithmetic will be indicat~
by the adjective "geometric" rather than by the prefix "exp."
For example, the natural average will be referred to as the
geometric average, a usage that is consistent with generally
accepted terminology.
In Section 6.10 we shall show that by apt use of geomet-
ric arithmetic one can readily obtain the geometric calculus,
the anageometric calculus, and the bigeometric calculus.
Geometric arithmetic has the following features. (The
letters y and z represent arbitrary positive numbers.)
Generator ••••••••••••••••• exp

Realm. • • • • • • • • • • • • • • • • • • • • R+
Geometric zero •••••••••••• 1 [ = exp(O)]
Geometric one ••••••••••••• e [= exp ( 1)]

Geometric sum exp{ln(y) + ln(z)}
•••• • • •••••
of y and z = yz
Geometric difference
••••• •
exp{ln(y) - ln(z)}
between y and z - y/z
Geometric product exp{ln(y) •ln(z)}
•••••••• • ln (z) ln(y)
of y and z = y - z
Geometric quotient exp{ln(y)/ln(z)}
•• 1/ln ( z)
of y and z, (z .;. 1) - y
Geometric order ••••••••••• Identical with classical order
Geometric positive numbers Numbers greater than 1
Geometric negative numbers Positive numbers less than 1
Geometric intervals ••••••• Identical with
positive intervals
5.4 Geometric Arithmetic 37
Geometric extent of [y,z] ••••• z/y

Natural average ••••••••••••••• Geometric average
Natural progressions •••••••••• Geometric progressions
Geometric partition of [y,z] •• Any geometric progression
whose first and last
terms are y and z
Geometric convergence is equivalent to classical conver-

gence in the sen$e that a sequence {pn} of positive numbers
geometrically converges to a positive number p if and only
if {pn} classically converges to p.
Geometric arithmetic should be especially useful in sit-
uations where products and ratios provide the natural methods
of combining and comparing magnitudes. Of course, geometric
arithmetic applies only to positive numbers, since the range
of exp is R+. But it is a simple matter to construct a geo-
metric-type arithmetic that applies to negative numbers: one
simply uses the generator -exp, which assigns to each number
x the negative number -ex.
Chapter 6
THE *-CALCULUS
6.1 INTRODUCTION
For the remainder of this book, a and B are arbitrarily

selected generators and * ("star") is the ordered pair of
arithmetics (a-arithmetic, B-arithmetic). The following
notations will be used.
a-Arithmetic a-Arithmetic
Realm •••••••••••• A B
•
Addition ••••••••• +
Subtraction •••••• -
•
-
••
• ••
Multiplication ••• X X
• ••
Division ••••••••• I I
• ••
Order •••••••••••• < <
It should be understood that all the definitions in

Chapter 5 apply equally well to a-arithmetic. For example,
a-convergence is defined by means of a-intervals and their
a-interiors.
In the *-calculus, a-arithmetic is used on arguments
and a-arithmetic is used on valuesJ in particular, changes
in arguments and values are measured by a-differences and
a-differences, respectively. The operators of the *-calcu-
lus are applied only to functions with arguments in A and
values in B. Accordingly, unless indicated or implied oth-
erwise, all functions are assumed to be of that character.
derivatives, integrals, and averages are discussed in Chap-
ter 9.
The *-limit of a function f at a number a in A is, if
it exists, the unique number b in B such that for every in-
finite sequence {an} of arguments of f distinct from a, if
{an} a-converges to a, then {f(an)} a-converges to b. We
write
*-lim f(x) = b.
x-fa
38
6.2 The *-Gradient 39
A function f is *-continuous at a number a in A if and

only if a is an argument of f and
*-lim f(x) = f(a).
x•a
When a and S are the identity function I, the concepts
of *-limit and *-continuity are identical with those of clas-
sical limit and classical continuity, but that is possible
even when a and a do not equal I.
The isomorphism from a-arithmetic to B-arithmetic is
the unique function t (iota) that possesses the following
three properties.
1. 1 is one-to-one.
2. 1 is on A and onto B.
3. For any numbers u and v in A,
• ••
t(u + v) = 1 (u) + 1 (v) ,
1 (u .:. v) = t (u) :.: 1 (v) ,

• ••
1(U x v) = 1 (u) x 1 (v) ,
• •• •
t (u I v) = 1 (u) I 1 (v) , v ~ 0,
•
u < v if and only if
••
\ (u) < 1 (v) •
It turns out that 1(x) = 6{a- 1 (x)} for every number x

in A, and that t(n) =
.
n
for every integer n.
-1 .•
Since, for example, u + v = 1 {t(u) + 1(v)}, it should
be clear that any statement in a-arithmetic can readily be
transformed into a statement in a-arithmetic.
6.2 THE *-GRADIENT

• •
The B-change of a function f over an a-interval [r,s]
is the number f(s) ~ f(r) in B.
A *-uniform function is a function that is on A, is *-
continuous, and has the same a-change over any two a-inter-
vals of equal a-extent. Clearly every constant function on
A is *-uniform.
The *-uniform functions are those expressible in the
fornt 1 { (m x• x) +• c} , where m and c are constants in A and
40 The *-Calculus
• •
x is unrestricted in A. By choosing m = 1 and c = 0, we see
that 1 is *-uniform.
It is characteristic of a *-uniform function that for
each a-progression of arguments, the corresponding sequence
of values is a a-progression.
The *-slope of a *-uniform function is its a-change
•
over any a-interval of a-extent 1. For example, the *-slope
• •
of the function l{(m x x) + c} turns out to be t(m). In par-
••
ticular, the *-slope of 1 equals 1, and the *-slope of a con-
••
stant function on A equals 0.
• •
The *-g;adient of a function f over [r,s] is the *-slope
of the *-uniform function containing (r,f(r)) and (s,f(s)),
and turns out to be
••
[f(s) ~ f(r)]/[t(s) ~ t(r)].
The preceding expression yields the indeterminate form

••
o/o when r = s.
6.3 THE *-DERIVATIVE

defined at least on an a-interval containing the number a in
its a-interior.
* (a)
If the following *-limit exists, we denote it by [Df]
call it the *-derivative of f at a, and say that f is *-dif-
ferentiable at a:
*-lim{[f(x) ~ f(a)]/[t(x) ~ t(a)]}.
x+a
* (a) is necessarily in B.
If it exists, [Of]
*
The *-derivative of f, denoted by Df, is the function
*
that assigns to each number tin A the number [Df](t), if it
exists.
The operator D* is B-additive and B-homogeneous, that is
D(f * ..+ Dg,

* ..+ g) = Df *
* ..x f)
D(c = *
c ..x Of, c constant in B.
* has a constant value equal
If f is *-uniforin, then Of
to the *-slope of f. Indeed, only *-uniform functions have
*-derivatives that are constant on A.
6.4 The *-Average 41
Now we generalize the concept of classical tangent de-
fined in Section 1.3.
The *-tangent to a function fat the point (a,f(a)) is
the unique *-uniform function g, if it exists, which possess-
es the following two properties:
1. q contains (a,f(a)).
2. For each *-uniform function h containing (a,f(a))
and distinct from g, there is an a-positive number
• • •
p such that for every number x in [a ~ p, a + p]
but distinct from a,
•• •• •• ••
Ig (x) ~ f (x) f <t Ih (x) ~ f (x) I.
* (a) exists if and only if f
It can be proved that [Df]
* (a) does exist,
has a *-tangent at (a,f(a)), and that if [Of]
i t equals the *-slope of that *-tangent.
We say that two functions are *-tangent at a common
point if and only if they have the same *-tangent there.
* (a) do not necessarily
The derivatives [Of] (a) and [Of]
coexist and are seldom equal; however, if the following
exist,
* (a) do coexist, and the *-tangent to f

then [Of] (a) and [Df]
at (a,f(a)) is, if it exists, classically tangent there.
6.4 THE *-AVERAGE

• •
The *-average of a *-continuous function f on [r,s] is
denoted by ~sf and defined to be the a-limit of the a-conver-
r
gent sequence whose nth term is the B-average of f(a 1 ) , ••• ,
• •
f(a ), where a 1 , ••• ,a is then-fold a-partition of [r,s].
n n
The operator ~~ is a-additive and a-homogeneous (Section
6. 3).
• •
The *-average of a *-uniform function on [r,s] is equal
to the a-average of its values at r and s, and is equal to
its value at the a-average of r and s.
The *-average is the only operator that has the follow-
ing three properties.
42 The *-Calculus
• •
1. For any a-interval [r,s] and any constant function
• •
h(x) =bon [r,s],
*s
Mrh = b.
• •
2. For any a-interval [r,s] and any functions f and g
that are *-continuous on fr,sJ, if f(x) < g(x) for
• • -
3. • •
For any numbers r,s,t in A such that r < s < t,
• •
and any function f *-continuous on [r,t],
~
•• 1
[t(s) t(r)] + [1 (t) !.• 1 (s)
•• •• *t
- [ \ (t) - 1 (r)) X M f.
r
6.5 THE BASIC THEOREM OF *-CALCULUS

We begin with the discrete analogue.
The Discrete Analogue of the
Basic Theorem of *-Calculus
Let h be a discrete function whose arguments a 1 , ••• ,an
are • •
an a-partition of [r,s]. Then the B-average of
the following n-1 *-gradients is equal to the *-gradi-
ent of hover fr,sf:
••
[h(ai+l) :.• h(ai) 1/ [1 (ai+l) ~ 1 (ai) 1,
i = 1, ..• ,n-1.
In view of the following theorem we say that the *-aver-
age fits naturally into the scheme of *-calculus.
The Basic Theorem of *-Calculus
If
*Dh
is *-continuous on
.[r,s],
.
then its *-average on
• • • •
[r,s] equals the *-gradient of hover [r,s], that is,
*s*
Mr(Dh) = [h(s) :.:
..
h(r) 1 I [1 (s) :.: 1 (r) 1.

the *-integral.
6.6 The *-Integral 43
The Basic Problem of *-Calculus

Suppose that the value of a function h is known at
an argument r, and suppose that f, the *-derivative
of h, is *-continuous and known at each number in
• •
[r,s]. Find h(s).
Solution
h(s) = h(r) •• J
+ 1 [t(s) ~ t(r)] ••
x
*s J
Mrf •
6.6 THE *-INTEGRAL

• •
The *-integral of a *-continuous function f on [r,s],
*
denoted by j(sf, is the following number in B:
r
[t(s) ~ t(r)]
*
We set / r f = 0.
r
The *-integral is a weighted *-average, since
*
/
*s 1J [ t
sf -- Mr (s) :.: t(r)] X f J.
r
*sf
Furthermore, / equals the B-limit of the B-convergent
r
sequence whose nth terut is
• •
where a 1 , ••• ,an is then-fold a-partition of [r,s], and k 0
is the common value of a 2 ~ a 1 , a 3 ~ a 2 , ••• ,an ~ an-l·
(If a is classically continuous and B = I, then the
*-integral is a Stieltjes integral.)
The *-integral is B-additive and S-homogeneous (Section
6.3) and is the only operator that has the following three
properties.
• •
1. For any a-intervpl [r,s] and any constant function
• •
h(x) =bon [r,s],
*
sh = [l(S):.! t(r)] ••
X b.
/
r
44 The *-Calculus
• •
2. For any a-interval [r,s] and any functions £ and q
• • ••
that are *-continuous on [r,s], if f(x) < g(x) for
• •
every number x in [r,s1, then
-
* *
Jsr f ~ /sr g.
• •
3. For any numbers r,s,t in A such that r < s < t,
• •
and any function f *-continuous on [r 1 t 1 1
* * *
/sf
r
••
+ - ltf
s
/t f.
r
6.7 THE FUNDAMENTAL THEOREMS OF *-CALCULUS
The *-derivative and *-integral are 'inversely' related

in the sense indicated by the following two theorems.
• •
If f is *-continuous on [r,s1, and
* • •
g(x) = /xf, for every number x in [r, s 1 ,
r
then
*Dg = f, on
.[r,s].
.
* • • •
If Dh is *-continuous on [r, s 1 , then
*
Js
*
(Dh) - h (s) ••- h (r) •
r
In this section we indicate the uniform relationships

between the corresponding notions of the *-calculus and c1as-
sical calculus.
For each number a in A, let a = a- 1 (a). Let f be a
function with arguments in A and values in B, and set f(t) = -
a- 1 {£(a(t))}.
Then *-lim f(x) and lim f(t) coexist, and if they do
x•a t•~
exist,
*-lim f (x) = a~lim l (t) J •
x~a t~a
6.9 Relationships Between Any Two Calculi 45
Furthermore, f is *-continuous at a if and only if f is clas- -

sically continuous at a.
*sf • •
If G
r
is the *-gradient of f over [r,s], then
*s
Grf = { s -}
B Gff ,
where
-
G;f is the classical gradient off over [f,§].
The derivatives * (a) and [Df]
[Of] - (a) coexist, and if
they do exist,
* (a)
[Dfl = B{ [Of] (a)}.
• •
And if f is *-continuous on [r,s], then
~~f = a{M~f}, and
The following facts are also worth noting. Let a E A

and b £ B. If a and a are classically continuous at a- 1 (a)
and a- 1 (b), respectively, then
*-lim f(x) =b if and only if lim f(x) = b.
x-+a x+a
If a and a are classically continuous at a- 1 (a) and
a- 1 (f(a)), respectively, then f is *-continuous at a if and
only if f is classically continuous at a.
6.9 RELATIONSHIPS BETWEEN ANY TWO CALCULI

In this section we indicate the uniform relationships
between the corresponding notions of any two given calculi,
-
the *-calculus and the *-calculus.
For the *-calculus there are two generators, a and S,
which generate arithmetics with realms A and B. For the -*-
calculus there are two generators, a and a,
which generate
arithmetics with realms A and B.
Since all arithmetics are isomorphic, there is a unique
isomorphism T1 from a-arithmetic to a-arithmetic and a unique
isomorphism T 2 from a-arithmetic to a-arithmetic. For each
- and for each
. A,
number x in A let -x be the number T 1 (x) 1n
number y in B let Y be the number T2 (y) in B. For each func-
tion f with arguments in A and values in B, let f be the -
46 The *-Calculus
function consisting of all ordered pairs (x,y) , where (x,y)
is an arbitrary ordered pair in f. Clearly f has arguments
- and values in B.
in A -
-
B
(i,Y> = (T 1 (x) ,T 2 (y))
-
y •
-1\
I I
I I
I I
B
I I
I I
T21 I -
I I X -
~,~------.---~A
I
I I /71
I ,
I
I I /
y I I /~
(x,y) // 1
~
/
------~~--------~A
X
The following notations will also be used.

*-Calculus -*-Calculus
Limits ••••••••• *-lim -*-lim
* -*
Gradients •••••• G G
* -*
Derivatives •••• D D
Averages ••••••• *
M
-M*
* -*
Integrals •••••• I I
•
Let f be a function with arguments l.n A and values l.n I
B
and let the number a be in A. Then the following coex·ist:
*-lim f -*-lim -f (t);
x~a
(x) ,
t-ta -
and if they do exist, then
*-lim f(x) = *-lim
- f(t).
t-ta
x-ta
Furthermore, f is *-continuous at a if and only if f is *-
continuous at a.
6.10 Applications of Geometric Arithmeti.c 47
If r and s are numbers in A such that r <• s, then

*s -*1-
Grf = Grf·
* -*- -
The derivatives [Df] (a) and [Df] (a) coexist, and if they
do exist, -
* (a) = [Df]
[Df] *- (a).
-
Finally, i f f is *-continuous on [r,s], then
and
The preceding observations clearly indicate that each

theorem in any given calculus has an analogue, or correspon-
dent, in every other calculus. In particular, each theorem
in classical calculus has an analogue in *-calculus, and con-
versely. As an illustration we state the followi~q theorem.
First Mean Value Theorem of *-Calculus
• •
Let f be *-continuous on (r,s] and *-differentiable
at each number x in A for which r <• x <• s. Then
• •
there exists a number c in A such that r < c < s
*
and such that [Df] (c) equals the *-gradient of f
• •
over [r,s].
6.10 APPLICATIONS OF GEOMETRIC ARITHMETIC
By appropriately specifying a and B one can obtain from
the *-calculus the calculi developed in Chapters 1-4:
Calculus a s
classical ••••• I I
geometric ••••• I exp

anageometric •• exp I
bigeometric ••• exp exp
Thus, one uses geometric arithmetic on function values

in the geometric calculus, on function arguments in the ana-
geometric calculus, and on both arguments and values in the
biqeometric calculus.
48 The *-Calculus
In defining anageometric slope (Section 3.2), we used
geometric intervals of geometric extent e, one reason for
which appears in Note 3. But the fundamental reason stems
from our desire that the anageometric calculus be the *-cal-
culus for which a = exp and B = I, since there the *-slope
•
is defined by means of a-intervals of a-extent 1, which
equals e.
Similarly, geometric intervals of geometric extent e
were used in defining bigeometric slope (Section 4.2) to
assure that the bigeometric calculus be the *-calculus for
which a = B = exp.
Our definition of the geometric derivative (Section 2.3)
required that it be positive. The fundamental reason for the
restriction rests upon our desire that the geometric calculus
be the *-calculus for which a = I and B = exp, since there
the *-derivative,if it exists, must necessarily be in the
realm of a-arithmetic, that is, in R+. A similar reason un-
derlies the stipulation that the bigeometric derivative be
positive (Section 4.3).
The operators of the geometric calculus are applied only
to functions with positive values. However, a geometric-type
calculus for negative (valued) functions can easily be ob-
tained by choosing a = I and 8 = -exp.
By choosing a = -exp and B = I, one obtains an anageo-
metric-type calculus for functions with negative arguments.
By choosing a = -exp and S = -exp, one gets a bigeometric-
type calculus for functions with negative arguments and val-
ues. Or one could, for example, choose a = -exp and S = exp,
thereby obtaining a bigeometric-type calculus for functions
with negative arguments and positive values.
Our emphasis, thus far, on the geometric family of cal-
culi may be explained by the fact that these calculi possess
gradients and derivatives which are independent of the units
used in measuring certain magnitudes. The invariance of gra-
dients and derivatives is discussed more generally in Section
9.2.
Of course an endless variety of calculi can be obtained
by using geometric arithmetic in tandem with other arithmetics
For instance, one could choose a = exp and B = tanh. Other
6.11 Graphical Interpretations 49
specific calculi are treated in Chapters 7 and 8. In Note 4

we discuss briefly the sigmoidal arithmetic and related cal-
culi, which may prove useful in statistics and biology.
6.11 GRAPHICAL INTERPRETATIONS
Except for the concept of •-tangency introduced in Sec-

tion 6.3, we have thus far treated the •-calculus not geomet-
rically but analytically, which was the way we conceived it.
In this section we give two graphical interpretations of the
•-calculus.
By •-paper we mean paper that is ruled off in squares
and marked thus:
B
I ~
.,..
•
-1 .d 6 i '
/ A
••
-1
For example, if a = I and 6 = exp, then •-paper is semi-

log paper that is logarithmically scaled on the vertical axis.
A •-point is an ordered pair of numbers in A and B, in
that order.
* 1 ,P 2 ), between two *-
Inforuaally, the •-distance, d(P
points P 1 and P2 is deteru,ined by plotting them on •-paper
and measuring their separation with the 'S-ruler' provided
by the vertical axis. The result is a a-nonnegative number.
(The concept of •-distance will be discussed formally in
Section 10.2.)
The •-sraph of a set of •-points is the result of plot-
ting them on •-paper. As expected, the •-graph of every •-
uniform function is a straight line.
Consider the •-graph of a •-uniform function g whose •-
slope is 6-positive. Let P1 and P 2 be two distinct •-points
on g, as shown below, and let P 3 be the vertex of the indi-
cated right triangle.
so The •-Calculus
----~--~----------------------~A
•
1
Then the •-slope of q equals

*d(P .. *
I d(P
2 ,P 3 ) 1 ,P 3 ). With obvi-
ous adjustments an interpretation may be provided if the •-
••
slope of q is B-negative. If the •-slope of g is 0, then the
•-graph of g is horizontal.
In Section 6.3 we observed that the •-derivative of a
function f at an argument a is equal to the •-slope of the •-
unifornt function g that is •-tangent to fat (a,f(a)). The
•-graph would look like this:
••
1
--~--~~------------------~A
•
1
The purely ari~hmetic nature of the •-average surely ob-

viates a graphic interpretation thereof.
Now we interpret the •-integral. All references here to
geometric figures are intended to apply to figures as they
appear on •-paper. The •-area of a rectangle is defined to
be the B-product of its '•-length' and '•-width.' The •-
area of a unit square (i.e., a square whose sides all have
•• ••
length 1) turns out to be 1. If a rectangle can be decem-
••
posed into n unit squares, then its •-area equals n, as ex-
pected.
• •
Let f be a B-positive, •-continuous function on [r,s],
6.11 Graphical Interpretations 51
and let S be the region, on •-paper, bounded by the •-graph
of f, the horizontal axis, and the vertical lines at r and s.
By usinq a-partitions, rectangles, and the S-limit process,
one can readily define the •-area of S, which turns out to
*
be /sf.
r
Another graphic interpretation of the •-calculus may be

obtained by using Cartesian paper, by which we mean paper
that is ruled off in squares and marked thus:
I ~
-1 0 1
,'
_,
The C-graph of a •-point (a,b) is the graph of the point

(a- 1 (a),B- 1 (b)) on Cartesian paper. The C-qraph of a set of
•-points consists of the C-graphs of all the individual •-
points. The C-graph of each •-uniform function is a straight
l.ine.
If P1 and P 2 are •-points and t is the Euclidean dis-
tance between their c-graphs, then the •-distance between P1
and P 2 equals B(t).
If f is a •-uniforiu function and m is the classical

slope of its C-graph, then the •-slope off equals S(m).
• •
If f is a B-positive, •-continuous function on [r,s],
and w is the (usual) area of the region bounded by the c-
graph of f, the horizontal axis, and the vertical lines at
a- 1 (r) and a- 1 (s), then *
/sf = B(w).
r
Chapter 7
THE QUADRATIC FAMILY OF CALCULI
7.1 THE QUADRATIC ARITHMETIC
In this chapter we obtain three specific calculi from

the *-calculus by using the classical and quadratic arith-
metics.
Quadratic arithmetic is the arithmetic generated by
the function that assigns to each number x the number
rx
if X > 0
1/2 -
X 0 if X = 0
-1-x if x < 0
That function is on R, onto R, and one-to-one; its in-

verse assigns to each number x the number
- x2 if x > 0
2 -
X 0 if X = 0
2
-x if X < 0
For any numbers y and z, we have

1/2 •z 1/2 ,
(yz)l/2 -- y
(y2)I72 - y -
Quadratic arithmetic has the following features. (The
numbers y and z are arbitrary.)
Realm ••••••••••••••••••••• R
Quadratic zero •••••••••••• 0
Quadratic one ••••••••••••• 1
-- -
2
Quadratic addition •••••••• y(f) z - (y + z2)1/2
- -
Quadratic subtraction ••••• Y e z - (y 2
- z2)1/2
Quadratic multiplication ••
Quadratic division •••••••• Identical with
Quadratic order ••••••••••• corresponding
Quadratic intervals ••••••• classical notions
Quadratic convergence •••••
Quadratic extent of [y,z]. z e y
52
7.2 The Quadratic Calculus 53
The quadratic average of n numbers u 1 , •.. ,un turns out
to be
which reduces to the well-known root mean square when the u.

1
are non-negative. The definitionsof quadratic progressions
and quadratic partitions specialize readily from those of
a-progressions and a-partitions in Section 5.3.
Quadratic arithmetic is a member of the infinite family
of power arithmetics discussed in Note 5.
7.2 THE QUADRATIC CALCULUS
The guadratic calculus is the *-calculus determined by

the ordered pair * whose members are classical arithmetic
and quadratic arithmetic, in that order.
In the quadratic calculus, the operators are applied to
functions with arguments and values in R, *-limits and *-
continuity are identical with classical limits and classical
continuity, the isomorphism 1 is the generator of quadratic
arithmetic, and the a-change of a function f over [r,s] is
equal to
==
{[f(s)1 2 - [f(r)] 2 ,
==}172
which is negative if f(s) < f(r), zero if f(s) = f(r), and
positive if f(s) > f(r).
For the remainder of this section we shall use the ad-
jective "quadratic" instead of the prefix "*."
The quadratically-uniform functions are expressible in
the form (rnx + c) 1 1 2 , where rn and care constants and xis
unrestricted in R. The quadratic slope of the preceding func-
.
t ~on equals m172 •
The quadratic gradient of f over [r,s] equals
-2 -2 I72
[f(s)1 - [f(r)]
•
s - r
0
The quadratic derivative of f at a, denoted by [Df] (a),
=
coexists with [D(£ 2 )1 (a), and
54 The Quadratic Family of Calculi
£1)£] (a)
- rn
= { [D(f2 ) 1 (a)} •
0 0
If [Df] (a) exists, then [Df] (a) exists; and if [Of] (a)
0
exists and f(a) ~ 0, then [Df] (a) exists. Moreover, [Df] (a)
exists if and only if f has a quadratic tangent T at (a,f(a)L
0
If [Df] (a) does exist, it equals the quadratic slope of T,
and if, furthermore, f(a) ~ 0, then T is classically tangent
to fat (a,f(a)).
0
Of course, Df is constant on R if and only if f is quad-
ratically-uniform.
The quadratic average of a continuous function f on
[r,s] equals
1/2
,
which reduces to the root mean square when f is non-negative.

The quadratic integral of a continuous function f on
0
[r,s], denoted by /sf,

r
equals
s
1r
(f2)
al: 1/2
,
which, for f non-negative, reduces to
1/2
•
That expression appears often in the mathematical literature,

though we have never seen it identified as an integral.
All the operators of the quadratic calculus are homoge-
nous and quadratically additive; for example,
0 0
D(c•f) = c•Df, c constant,
0 0 0
D ( f ~ g) = Df ~ Dg.
The Basic Theorem of Quadratic Calculus

0
If Of is continuous on [r, s] , then·· its quadratic
average on [r,s] equals the quadratic gradient
off over [r,s].
7.3 The Anaquadratic Calculus 55
The First Fundamental Theorem of Quadratic Calculus

I f f is continuous on [r,s], and
0
9 (x) = /xf, for every number x in [r, s],

r
then
0
Dg = f, on [r,s].
The Second Fundamental Theorem of Quadratic Calculus
0
0
fr s cDh > = h cs > e h cr > •
7. 3 THE ANAQUADRATIC CALCULUS
The anaquadratic calculus is the *-calculus determined

by the ordered pair * whose members are quadratic arithmetic
and classical arithmetic, in that order.
In the anaquadratic calculus the operators are applied
to functions with arguments and values in R, *-limits and *-
continuity are identical with classical limits and classical
continuity, the isomorphism 1 is the inverse of the genera-
tor of quadratic arithmetic, and the a-change of a function
f over [r,s] equals f(s) - f(r), since here S = I.
jective "anaquadratic" instead of the prefix "*·"
The anaquadratically-unifonn functions are expressible
-=
in the form mx 2 + c, where m and c are constants and x is
unrestricted in R. The anaquadratic slope of the preceding
function equals m.
The anaquadratic gradient of f over [r,s] equals
f(s) - f(r)
•
s
2 - r
2
The anaquadratic derivative of f at a, denoted by
[Df] (a), coexists with, and is equal to, the classical de-
r~vative of the function f(x) = f(xi72) at a 21
:-=
[Df] (a) = [Df](a 2 ) •

0
If [Df] (a) exists, then [Df] (a) exists; and if [Of] (a)
0
exists and a ~ 0, then [Df] 0
(a) exists. Moreover, [Of]0
(a)
exists if and only if f has an anaquadratic tangent T at
(a,f(a)). If [Of] o
(a) does exist, it equals the anaquadrat-
ic slope of T, and if, furthermore, a ~ 0, then T is clas-
sically tangent to fat (a,f(a)).
Also worth noting is the fact that [Df] 0
(a), if it exists,
equals the classical ~erivative, at a, of f with respect to
the function h(x) = x 2 •
Of course, Dfo
is constant on R if and only if f is ana-
quadratically-uniform.
Let f be continuous on [r,s], and set f(x) = f(x 11 2 ),
- -
-r = r2, and s -- s 2 • Then the anaquadratic average of f on
-
[r, s 1 equals
-
-s -
1
J_sr,
-r r
s
and the anaquadratic integral off on [r,s], denoted by~ f,
- 0
equals J_sf, which is the Stieltjes integral of f with respect

r ==
2
to the function h(x) = x on [r,s].
All the operators of the anaquadratic calculus are addi-

tive and homogeneous.
The Basic Theorem of Anaquadratic Calculus
If Of is continuous on [r,s], then its anaquadratic
0
average on [r,s] equals the anaquadratic gradient
of£ over [r,s].
The First Fundamental Theorem of Anaquadratic Calculus
If f is continuous on [r,s], and
X
g (x) = [ f, for every number x in [r ,s],
r
0
then
Dg
0
= £, on [r,s].
7.4 The Biquadratic Calculus 57
The Second Fundamental Theorem of Anaquadratic Calculus
0
Js <ph)
r
= h(s) - h(r).
0
7.4 THE BIQUADRATIC CALCULUS

The biquadratic calculus is the •-calculus determined by
the ordered pair * both of whose members are quadratic arith-
metic.
In the biquadratic calculus the operators are applied to
functions with arguments and values in R, •-limits and •-con-
tinuity are identical with classical limits and classical con-
tinuity, the isomorphism 1 is the identity function I, and
the a-change of a function f over [r,s] equals
{ [f(s)]
==
2 -
==}172
[f(r)] 2 •

jective "biquadratic" instead of the prefix "*·"
The biquadratically-uniform functions are expressible in
the form (mx~ + c)I7!, where m and c are constants and x is
unrestricted in R. The biquadratic slope of the preceding
function equals m!7!.
The biquadratic gradient of f over [r,s] equals
1/2
[f(s)]'2- [f(r)]2
•
52- /2
0
The biquadratic derivative of f at a, denoted by [Df]
0
(a) ,
-=
2 ==
2
coexists with [Df] (a) where f(x) = [f(x 11 2 >1 and a= a ; and
0
If a ~ 0 and f(a) ~ 0, then [Df] (a) and [Df] (a) coexist.
0
0
Moreover, [Df·] (a) exists if and only if f has a biquadratic
0 0
tangent T at (a, f(a)). If [Df] (a) does exists, it equals the
0
biquadratic slope of T, and if, furthermore, a ~ 0 and f(a) ~
0, then Tis classically tangent to fat (a,f(a)).
0
Of course, Df
o
is constant on R if and only if f is bi-
quadratically-uniform.
Let f be continuous on [r,s], and set
r -
= r 2,
and i 2
=s • Then the biquadratic average of f on
[r,s] equals
1
-s ,
.. -
r
and the biquadratic integral off on [r,s], denoted
equals
All the operators of the biquadratic calculus are homo

qeneous and quadratically additive.
The Basic Theorem of Biquadratic Calculus
0
If Df
0
is continuous on [r,s], then its biquadratic
average on [r,s] equals the biquadratic qradient of
f over [r,s].
The First Fundamental Theorem of Biquadratic Calculus

Iff is continuous on [r,s], and
0
g(x) = j[xf, for every number x in [r,s],

r
0
then
0
~ = £, on [r,s].
The Second Fundamental Theor~m of Biquadratic Calculus
0
If Db is continuous on [r,s], then
0
0
s o
l
0
r
(Db) - h(s)
o
e h(r).
Chapter 8
THE HARMONIC FAMILY OF CALCULI
8.1 THE HARMONIC ARITHMETIC

In this chapter we obtain three calculi from the •-cal-
culus by usinq the classical and harnonic arithmetics.
Harmonic arithmetic is the arithmetic generated by the
function that assiqns to each number x the number
1 1/x if x ~ 0
= •
X 0 if X =0
We shall also use the notation l#x.
The preceding function is on R, onto R, one-to-one, and
identical with its inverse, since
for every number x.

Harmonic arithmetic has the following features. (The
numbers y and z are arbitrary.)
Realm. • • • • • • • • • • • • • • • • • • • • R
Harmonic zero ••••••••••••• 0
Harmonic one •••••••••••••• 1
Harmonic addition ••••••••• y [±] z = l;(l#y + l;z)
Har1nonic subtraction •••••• ytj z = l#(~y- l#z)
Harmonic multiplication ••• Jldentical with corresponding
Harmonic division ••••••••• lclassical notions.
Harn~nic order •••••••••••• y ~ z if and only if l#y < l#z.
0 0
Ha~1uonic interval 1 [y, z] •• All numbers x such that y ~ x ~ z.
0 0
Harmonic extent of [y 1 z] •• z E1 y
The followinq facts should also be noted: the symbol;
does not stand for harmonic division, which is identical with
classical division; harmonic order is not identical with clas-
sical order (for example, -1 BJ -2); the generator of harmonic
arithmetic is not classically continuous at 0; and harmonic
convergence is not identical with classical convergence (for
instance, the infinite sequence {n} is hantonically conver-
gent to 0).
59
60 The Harmonic Family of Calculi
Harmonic sums occur frequently in science. For example,
if two resistors of resistance y and z are connected in par-
allel, the combined resistance equals the harmonic sum of y
and z.
Since the natural average and natural progressions in
harmonic arithmetic are direct extensions of the well-known
harmonic average and harmonic progressions defined for pos-
itive numbers, we shall use the same names for the exten-
•
s1.ons.
The harmonic average of n numbers u 1 , ••• ,un equals
1
•
••• + (l#un )
n
Harmonic arithmetic is a member of the infinite family

of power arithmetics discussed in Note 5.
8.2 THE HARMONIC CALCULUS
The harmonic calculus is the •-calculus determined by

the ordered pair * whose members are classical arithmetic
and harmonic arithmetic, in that order.
In the harmonic calculus the operators are applied to
functions with arguments and values in R, •-limits and •-
continuity are NOT always identical with classical limits
and classical continuity, the isomorphism 1 is the generator
of harmonic arithmetic, and the a-change of a function f over
[r,s] equals
1
,
l#f (s) - l#f (r)
which is 0 when f(s) = f(r).
jective "harmonic" instead of the prefix "*·"
The harmonically-uniform functions are expressible in
the form l#(mx +c), where m and care constants and xis un-
restricted in R. The harmonic slope of the preceding func-
tion equals l#m.
The harmonic gradient of f over [r,s] equals
8.2 The Harn~nic Calculus 61
1
•
l#f(s) - l#f(r)
s - r
0
The harmonic derivative off at a, denoted by [Df] (a),
coexists with [D(l#f)] (a), and
0 1
[Of] (a) - •
[D(l#f)](a)
0
If f(a) ; 0, then [Df] (a) and [Of] (a) coexist. More-
a
over, [Df] (a) exists if and only if f has a harmonic tangent
0
Tat (a,f(a)). If [Df] (a) does exist, it equals the harmonic
slope of T, and if, furthermore, f(a) ~ 0, then T is clas-
sically tangent to fat (a,f(a)).
0
Of course, Of is constant on R if and only if f is har-
monically-uniform.
The harmonic average of a harmonically continuous func-
tion f on [r,s] equals
1
,
1
s - r
Js r
(l#f)
which reduces to the traditional harmonic average when f is

positive.
The harmonic integral of a harntonically continuous func-
os
tion f on [r,s], denoted by ~ f, equals
r
1
•
/s
r
(l#f)
All the operators of the harmonic calculus are homoge-

neous and harmonically additive; for example,
D(c•f) = c•Df, c constant,

D(f (±]g) = Df [±) Dg.
62 The Harmonic Family of Calculi
The Basic Theorem of Harmonic Calculus
If ~f is harmonically continuous on [r,s], then
l ts har1uonic average on [r, s] equals the harmonic
qradient off over [r,s].
The First Fundamental Theorem of Harmonic Calculus
I f f is harmonically continuous on [r,s], and
ox
g(x) =~ f, for every number x in [r,s],
r
then
0
Dg = f, on [r,s].
The Second Fundamental Theorem of Harmonic Calculus
0
If Dh is harmonically continuous on [r,s], then
as
~ cHh> = h(s> EJ h(r>.
r
8. 3 THE ANAHARMONIC CALCULUS
The anaharutonic calculus is the •-calculus determined

by the ordered pair * whose members are harmonic arithmetic
and classical arithmetic, in that order.
In the anaharmonic calculus the operators are applied
to functions with arguments and values in R, •-limits and
*-continuity are NOT always identical with classical limits
and classical continuity, the isomorphism t is the (self-
inverse) generator of harutonic arithmetic, and the a-change
0 0
of a function f over [r,s] equals f(s) - f(r), since here
B = I.
jective "anaharmonic" instead of the prefix "*·"
The anaharmonically-uniform functions are expressible in
the form m(l#x) + c, where m and c are constants and x is un-
restricted in R. The anaharmonic slope of the preceding func-
tion equals m.
0 0
The anaharmonic gradient of f over [r,s] equals
f(s) - f(r)
•
l#s - 1//r
8. 3 The Anahartnonic Calculus 63
The anaharmonic derivative of f at a, denoted by lSf1 (a),

coexists with, and is equal to, the classical derivative of
the function l(x) = f(~x) at l§a:
[Df] (a) = [Df] (l#a) •
0
If a ~ 0, then [Df] (a) and [Df] (a) coexist. Moreover,
0
[Bf 1 (a) exists if and only if f has an anahar1uonic tangent
Tat (a,f(a)). If lBf1 (a) does exist, it equals the ana-
hariiionic slope of T, and if, furthe:r1uore, a ~ 0, then T is
classically tangent to fat (a,f(a)).
Of course, Bf is constant on R if and only if f is ana-
harmonically-uniforlu.
0 0
Let f be anaharmonically continuous on [r,s], and set
f(x) = f(l#x), r s
= l#r, and = l§s. Then the anaharmonic
averaqe of f on Yr,sY equals
-
- 1 -
s - r
f_·t,
r
and the anahaxmonic inteqral off on (r,sY, denoted by )(sf,
- or
equals J_ sf..
r
All the operators of the anaharmonic calculus are addi-

tive and homogeneous.
The Basic Theorem of Anaharmonic Calculus
[] []
If Df is anaharmonically continuous on [r,s], then
0 0 C
its anahar1uonic average on [r ,s] equals the anabar-
monic qradient off over ?r,sY.
The First Fundamental Theorem of Anaharmonic Calculus
Iff is anaharmonically continuous on ?r,s9, and
o a
for every number x in [r,s],
then
0 0
Dq
0
= f, on [r,s].
The Second Fundamental Theorem of Anaharmonic Calculus
If Dh is anaharmonically continuous on ?r,sY, then
0
J s (Db)
or o
= h(s) - h(r).
64 The Harnonic Family of Calculi
8.4 THE BIHARMONIC CALCULUS
The biharmonic calculus is the •-calculus determined by
the ordered pair * both of whose members are harmonic arith-
metic.
In the biharmonic calculus the operators are applied to
functions with arguments and values in R, •-limits and •-con-
tinuity are NOT always identical with classical limits and
classical continuity, the isomorphism 1 is the identity func-
o 0
tion I, and the 6-chanqe of a function f over [r,s] equals
1
•
l#f (s) - l#f (r)
jective "biharmonic" instead of the prefix "*·"
The biharmonically-uniform functions are expressible in
the form l#[m(l#x) +c), where m and care constants and x
is unrestricted in R. The biharmonic slope of the preceding
function equals ~m.
0 0
The biharn1onic gradient of f over [r, s 1 equals
1
•
l#f (s) - l#f (r)
l#s - Vr
0
The biharmonic derivative of f at a, denoted by [Df] (a),
0
coexists with [Df] (a), where f(x) = l#f(l#x) - = l#a;
and a and
0
[Df1 (a) = l#[Df1 (a).
0
0
If a ~ 0 and f(a) ~
0, then [Of] (a) and [Df] (a) coexist.
a
0
Moreover, [Df] (a) exists if and only if f has a biharmonic
0
tangent Tat (a,f(a)). If rB£1 (a) does exist, it equals the
0
biharmonic slope of T, and if, furthermore, a ~ 0 and f(a) ~
0, then Tis classically tangent to fat (a,f(a)).
0
Of course, Df is constant on R if and only if f is bi-
o
harmonically-uniform.
0 0
Let f be biharmonically continuous on [r,s], and set
f(x) = l#f(l/x) I r s
= l#r, and = 1/s. Then the biharmonic
8. 4 The Biharntonic Calculus 65
0 0
average of f on [r 1 s] equals
1
- I
-s
1
-
/_sf
- r r
0
and the biharmonic integral off on ?r,s), denoted by }(sf,

equals or
1
•
All the operators of the biharmonic calculus are homo-

geneous and harmonically additive.
The Basic Theorem of Biharmonic Calculus
0 • . . 0 0
If Of 1s b1harmon1cally continuous on [r, s] 1 then
0 0 0
its biharnonic average on [r,s] equals the bihar-
0 0
monic gradient of f over [r,s].
The First Fundamental Theorem of Biharmonic Calculus
0 0
If f is biharn~nically continuous on [r,s], and
0
0 0
q(x) = /x f, for every number x in [r,s],
or
then
0 0 0
Dg
0
= f, on [r,s].
The Second Fundamental Theorem of Biharmonic Calculus
0 0
is biharmonically continuous on [r,s], then
OS
1
or
c8h> = h(s) a h (r).
o
Chapter 9
H E U R I S T I C S
9.1 INTRODUCTION
Since it is not unreasonable to suppose that there are
situations where a non-Newtonian calculus might be useful,
we shall propose some guides that may be helpful in selec-
ting appropriate gradients, derivatives, integrals, and av-
erages. Of course, one is always free to use any operator
that is meaningful in a given context.
We continue to let • be an ordered pair of arbitrarily
• • • • ••
chosen arithmetics, (A,+,~,x,;,<) and (B,+,~,x,/,<), with
generators a and a, respectively.
9.2 CHOOSING GRADIENTS AND DERIVATIVES
Since the choice of a derivative tacitly involves a

choice of the related gradient, we shall restrict our atten-
tion to gradients, for which three heuristic principles will
be offered.
Principle I
If the natural methods of measuring changes in
arguments and values are provided by a-differ-
ences and a-differences, respectively, then the
•-gradient may be appropriate.
For example, the classical gradient is appropriate for
analyzing motion because changes in time and position are
naturally measured by differences. The geometric gradient
is appropriate for analyzing stock-price movements because
changes in time and price are usually measured by differ-
ences and geometric differences (i.e., ratios). The bigeo-
metric gradient may be useful in psychophysics because
changes in stimulus and sensation are often measured by geo-
metric differences. ("Equal stimulus ratios produce equal
sensation ratios," according to s.s. Stevens.)
Principle II
If the functional relationship between two magnitudes
would be, or is assumed to be, •-uniform under normal
66
9.2 Choosing Gradients and Derivatives 67
or ideal conditions, then the •-gradient may be
appropriate.
For example, the geometric gradient may be appropriate
for analyzing radioactive decay because the relationship of
mass to time is assumed to be geometrically-uniform (i.e.,
exponential) under ideal conditions. Similarly, the sigmoi-
dal gradient may be useful in the study of growth that is
normally sigmoidally-uniform (Note 4). The harmonic gradient
may be appropriate for analyzing the relationship between two
magnitudes that would normally be inversely proportional.
Finally we observe that another reason for using the classi-
cal gradient in the study of motion is the assumption that
under ideal conditions, i.e., the absence of forces, there-
lationship of position to time would be linear (Newton's
First Law).
It is worth noting that each one-to-one function f on R
is •-uniform if one chooses a = I and S = f. This implies
that each such function has a derivative that is constant
on R.
The statement of the third principle require~ some pre-
liminary definitions.
•
An a-translation is a function of the form x + k, where
k is a constant in A and x is unrestricted in A. For example,
the classical translations have the form x + k, and the qua-
dratic translations have the form
{x2 + k2}1/2.
• •
Each a-translation x + k transforms 0 to k and preserves
a-differences, that is, for any numbers u and v in A, the a-
difference between u and v equals the a-difference between
the transforms of u arid v. Therefore one may say that each
a-translation effects a shift in origin.
•
An a-similitude is a function of the form p x x, where
p is an a-positive constant and x is an arbitrary a-positive
• •
number. Since each a-similitude p x x transforms 1 to p and
preserves a-quotients, one may say that a-similitudes effect
a change in unit.
Of course, a-translations and 8-similitudes are defined
68 Heuristics
similarly.
The classical similitudes are exp-translations, that is,
geometric translations. Indeed, every similitude is a trans-
lation (each a-similitude is a y-translation, where y(x) =
a (ex)).
It is an important fact that the •-gradient (and •-de-
rivative) is invariant under all a-translations of the argu-
ments and all a-translations of the values. For example, the
classical gradient is invariant under all classical transla-
tions of arguments and of values, and the geometric gradient
is invariant under all classical translations of arguments
and all geometric translations of values.
Principle III
If one desires a gradient that is invariant under
all a-translations of arguments and all a-transla-
tions of values, then one should consider using
the •-gradient.
It is an interesting fact that the mixed second deriv-
~
ative D(D) is invariant under all classical translations and

,.,.,
classical similitudes of values, and D(D) is invariant under
""'
all classical translations and classical similitudes of argu-
ments and of values.
9.3 CHOOSING INTEGRALS

Suppose that a scientific proposition has been expressed
by a •-differential equation, that is, a differential equa-
tion involving •-derivatives and perhaps related arithmetic
operations. To solve the equation one would certainly try
to use •-integration. If obtained, the solutions would at
first be expressed by means of •-integrals with variable up-
per limits, which in turn would be evaluated in terms of
known functions or new functions defined for the purpose.
(Many functions were originally defined as classical inte-
grals.) It is even conceivable that solutions in closed form
could be obtained for certain intractable classical differen-
tial equations by re-expressinq them with appropriate non-
Newtonian derivatives and related arithmetic operations.
9.3 Choosing Integrals 69
A differential equation containing several types of de-

rivatives can readily be transformed into an equation involv-
ing only classical derivatives or only non-Newtonian deriva-
tives of the same type. (See Sections 6.8 and 6.9.)
Integrals are also used for defining scientific concepts.
Consider, for example, the physicists' concept of work. If
the force f is constant, the work on the position interval
[r,s] is defined to be (s- r)•f, which we denote here by
W5 f. When f is constant the following conditions are clearly
r
satisfied.
1. W~f = (s- r)•f.
2. Work is monotonic increasing with respect to force.
3. Work is additive with respect to displacement, that
is, for any positions r,s,t such that r < s < t,
Desiring to extend the concept of work to the case where

•
the force f is continuously variable, the physicist stipulates
that the extended work concept should satisfy Condition 1 when
f is constant and should satisfy Conditions 2 and 3 in general
The solution is now uniquely determined; for according to Sec-
tion 1.6, the operator W satisfies those three conditions if
and only if W is the classical integral. Thus, the physicist
must adopt the following definition:
The foregoing well-known example illustrates the fact

that the characterization of the classical integral in Sec-
tion 1.6 is a heuristic guide for its appropriate use.
Now suppose that a scientist is concerned with a posi-
tive magnitude g, called 'gorce,' which may be continuously
variable with respect to time. If g is constant, he defines
the 'toil' on the time interval [r,s] to be qs-r, denoted by
T8 g. (Somewhere we have seen such a concept, perhaps in eco-
r
nomics.) When g is constant, the following conditions are
clearly satisfied.
70 Heuristics
s
1. Tr g = q
s-r •
2. Toil is monotonic increasing with respect to gorce.
3. Toil is multiplicative with respect to time; that

is, for any instants r,s,t such that r < s < t,
s t t
Tg.Tg=Tg.
r s r
To extend the toil concept to the case where the gorce
g is continuously variable, we stipulate that the extended
toil concept should satisfy Condition 1 when g is constant
and should satisfy Conditions 2 and 3 in general. The solu-
tion is now uniquely determined; for, according to Section
2.6, the operator T satisfies those three conditions if and
only if T is the geometric integral. Thus, we must adopt
the following definition:
""s
T~g = I
r
g.
It should now be clear that the characterization of the

•-integral in Section 6.6 is a heuristic guide for the use
of integrals.
9.4 CHOOSING AVERAGES

Most of this section is devoted to a-averages rather
than •-averages, since the choice of the latter depends pri-
marily on how one would average finite sequences of values,
although the method of partitioning the argument intervals
is certainly a factor. (In contrast to the •-integral, the
•-average is not independent of the type of partitions used
for the domain. )
First let it be noted that our definition of the B-
average corresponds to a definition of the arithmetic aver-
age of u 1 , ••• ,un as the number u such that
u + ••• + u = u 1 + ••• + un.
n terms
Accordingly, the S-average has the same properties relative
to a-arithmetic as the arithmetic average has relative to
classical arithmetic. (One important illustration of that
9.4 Choosing Averages 71
fact is given near the end of Section 10.4.)
The predominant use of the arithmetic average may be
attributed, perhaps, to the custom of measuring deviations
by differences rather than by ratios or some other B-differ-
ence. However, any statistical justification for the use
of the arithmetic average relative to classical arithmetic
can be matched by a justification for the use of the S-aver-
age relative to a-arithmetic.
Scientists often invoke a kind of "normalcy" principle
in choosing averages. For instance, it is commonly stated
that the geometric average is appropriate in the study of
those populations which normally increase in geometric pro-
gression. This idea can be extended as follows: If under
normal or ideal conditions the measurements would be in a-
progression, then the a-average may be appropriate.
Some scientists, notably the psychophysicist S.S.Ste-
vens, favor the use of invariance principles for choosing av-
erages. An interesting discussion may be found in the book
Basic Concepts of Measurement by Brian Ellis.
Now consider n measurements u 1 , ••• ,un of magnitudes that
naturally combine by S-sums; that is, the measure of a com-
bination of two or more magnitudes equals the S-sum of their
individual measures. Surely, then, the B-average of the ui
would be physically meaningful, though the arithmetic aver-
age of the ui might well be physically meaningless. For in-
stance, since the total resistance of n resistors connected
in parallel equals the harmonic sum of the individual resist-
ances, one would expect their harmonic average to be physi-
cally meaningful.
The intended use of an average is, of course, a basic
factor in its choice. Suppose, for example, that u 1 , ••• ,un
are the measurements of one side of a square whose area is
to be estimated. The standard procedure is to square the
arithmetic average of the u l..• But a different estimate of
the area is usually obtained if one computes the arithmetic
average of the squares of the ui. Although some consider
that to be no serious objection, it should be noted that the
geometric average always yields identical answers both ways.
72 Heuristics
The point here is that since the area is the square of a
side, one should seriously consider using a multiplicative
average rather than an additive average.
As is the case for integrals, a fundamental heuristic
guide for choosing averages is provided by the three charac-
terizing properties stated at the end of Section 6.4.
In many situations, averages are intuitively more sat-
isfying than integrals. Consider, for instance, a particle
moving rectilinearly with positive velocity v at time t. The
distance s traveled in the time interval (a,b] is given by
Although the derivation of this result may be clear to a stu-

dent, he may nevertheless find that the following formula
conveys a more immediate meaning:
s = (b- a).~v;
a
that is, the distance traveled equals the product of the time
elapsed and the arithmetic average of the velocity. This
version is a direct extension of the case where v is constant.
Other examples will readily occur to the reader. (For the
last example of the preceding section, we may write
a formula that is a direct extension of the case where g is

constant.)
9.5 CONSTANTS AND SCIENTIFIC CONCEPTS
In this section we illustrate the thesis that the inven-

tion of a scientific concept may depend on the isolation or
discovery of a suitable constant. We also suggest that new
scientific concepts may arise from the constants provided by
the slopes of uniform functions.
Consider the concept of average speed. The definition
"distance traveled per unit time" is incomplete because it
fails to provide a method of determining the average speed
9.5 Constants and Scientific Concepts 73
of an accelerated particle. The definition "distance divided

by time," though not incorrect, is a gross oversimplification
that fails to reveal the underlying issues. Fortunately
there is a completely satisfactory definition, which was un-
doubtedly known to Galilee.
Let us begin with Galilee's definition of unifornt motion
as "one in which the distances traversed ••• during ANY equal
intervals of time are themselves equal." (Continuity of the
motion is tacitly intended here.) Then we isolate a constant
in each given uniform motion by defining speed to be the dis-
tance traveled in any unit time interval. Finally, for a
particle that moves non-unifornaly a distance d in time t, we
define the average speed to be the speed that a particle in
unifornt motion must have in order to travel a distance d in
•
time t. In our opinion, neither the simplicity nor the obvi-
ousness of the answer, d/t, justifies its use as the defini-
tion of average speed.
The critical first step in defining average speed is the

isolation of the constant (speed) in the phenomenon of unifornt
motion. Similarly, the critical step in defining the •-gra-
dient is the identification of the constant (•-slope) in a
•-uniform function.
Wherever a scientific phenomenon (or idealized version
thereof) is describable by a •-uniform function, one automat-
ically has a fundamental const.ant, the •-slope, which may
prove to be useful.
For example, in the idealized model of radioactive decay,
the relationship of mass to time is describable by a geomet-
rically-uniform function whose geometric slope provides a
constant that is independent of the unit of mass and equals
1 more than the percent change of the mass over any unit time
interval.
In the idealized model of the behavior of a gas, the re-
lationship of pressure to volume, assuming temperature con-
•
stant, is describable by a harn~nically-uniform function
whose har1nonic slope is a constant that equals the harmonic
change in pressure corresponding to a unit change in volume.
74 Heuristics
As a final example, consider a fixed light source radi-
ating light uniformly in all directions. Since the intensity
of illumination is inversely proportional to the square of
the distance from the source, the relationship of intensity
to distance is describable by a bigeometrically-uniform func-
tion, whose biqeometric slope is independent of the units of
distance and intensity.
Chapter 10
COLLATERAL ISSUES
10.1 INTRODUCTION
This last chapter contains, among other things, brief
discussions of some ideas that we have entertained but not
investigated in depth.
As before, * is an ordered pair of arbitrarily selected
• • • •• •• • •
arithmetics, (A,+,~,~,/,<) and (B,+,~,x,/,<), having genera-
tors a and B. The isomorphism from a-arithmetic to B-arith-
metic is denoted by 1, which was discussed in Section 6.1.
The a-absolute value of a number a in A is denoted by
. . -1
fal and equals a( Ia <a> I>; similarly,
.. ..
the a-absolute value
1
of a number bin B is denoted by lbl and equals B<IB- (b) J>,
(Section 5. 3).
The 8-square of a number b in B is bxb, which will be
••
2
denoted by b at a slight risk that it might be incorrectly
interpreted as b to the power 8(2). For each a-nonnegative
number t, the symbol ·vt will be used to denote
which is the unique a-nonnegative number whose a-square equals

t. For each number b in B,
• • ••
For each integer n, the symbolsn and n denote the a-

integer a(n) and the 8-integer S(n), respectively.
Henceforth a point is a •-point, that is an ordered pair
of numbers in A and B, in that order. The symbol Pi repre-
sents the point (a.l. ,b.).
l.
10.2 •-SPACE
By •-space we mean the system consisting of all points
and the following method of computing the •-distance,
*
d(P 1 ,P 2 ), between points P 1 and P 2 :
••
.:. a2)]2 + [bl ~
75
76 Collateral Issues
= a
This •-distance fonnula gives the same results infor-

mally explained in Section 6.11 and was initially designed
to facilitate the graphic interpretation presented there.
(When a = B = I, •-distance is obviously the usual Euclid-
*
ean distance.) Observe that d(P 1 ,P 2 ) is in B, as are the
values of all the operators of the •-calculus. Of course,
one could easily redefine •-distance so that it is in A,
but that would be less useful for our purposes.
* .. ..
If a 1 = a 2 , then d(P 1 ,P 2 ) = lb 1 ~ b 2 l1 and if b 1 = b 2 ,
* .. .. . .
then d(P 1 ,P 2 ) =I 1(a1 ~ a 2) I = 1(la1 ~ a2 1>·
Although d* is not a metric in the usual sense, it is a
•-metric in the sense that the following proposition is true
For any points P 1 , P2, and P 3 ,
•• ••
* 1 ,P 2 ) -> 0,
d(P
••
* 1 ,P 2 )
d(P -- 0 if and only if pl - P2,
* 2 ,P 1 ),
* 1 ,P 2 ) -- d(P
d(P
* 1 ,P 2 )
d(P
••
+ * 2 ,P 3 )
d(P
••
* 1 ,P 3 ).
> d(P
-
The points P 1 , P 2 , and P 3 are •-collinear provided that
at least one of the following holds:
* 2 ,P 1 )
d(P
••
* 1 ,P 3 )
+ d(P * 2 ,P 3 ),
- d(P
* 1 ,P 2 )
d(P
••
* 2 ,P 3 )
+ d(P - *
d(Pl,PJ) I
* ••
* 3 ,P 2 )
d(P 1 ,P 3 ) + d(P * 1 ,P 2 ).
- d(P
If a 1 = a2 = a 3 or b 1 = b2 = b3, then P 1 , P 2 , and P 3 are
•-collinear, but those are not the only circumstances that
engender •-collinearity, as we shall see shortly.
A •-line (or •-geodesic) is a set L of at least two dis-
tinct points such that for any distinct points P1 and P 2 in
10.2 •-Space 77
L, a point P 3 is in L if and only if P1 , P 2 , and P 3 are •-

collinear. A •-line is vertical provided all its points
have the same first member.
When a = B = I, the •-lines are the straight lines of
Euclidean analytic geometry in two dimensions. If a = I
and B(x) = x 2 , for example, then each nonvertical •-line is
the union of two parabolic parts. (The parabolic family of
calculi is defined in Note 6.)
Theorem
The class of nonvertical •-lines is identical
with the class of •-uniform functions.
Two •-lines are •-parallel provided they are identical
or have no common point. It turns out that two nonvertical
•-lines are •-pasallel if and only if they have the same •-
slope.
A •-line L 1 is •-perpendicular to a •-line L 2 provided
they have a common point P and for any points P1 on L1 and
* .. *
P 2 on L 2 , distinct from P, d(P 1 ,P) < d(P 1 ,P 2 ). If L1 and L 2
are •-perpendicular, then for any points P 1 on L1 and P 2 on
L2'
•• ••
* 2 •• *
[d(P,P 1 ) 1 + [d(P,P 2 ) 1
2
Furtherntore, two nonvertical •-lines are •-perpendicular if

••
and only if the S-product of their •-slopes is -1, that is,
B(-1).
The •-arc length of a suitably behaved arc can be de-
fined in an obvious way by partitioning the arc and using the
•-distance between the successive points. If f has a •-con-
• • • •
tinuous •-derivative on [r,s1, the •-arc length off on [r,s]
turns out to be
*s
As expected, the •-distance between two points P 1 and P 2 is

equal to the •-arc length of the •-segment connecting P 1 and
P2.
With the •-metric one can define all the •-conics, the
study of which may be facilitated by using the •-vectors de-
fined in the next section. (It would be interesting to know
whether one can select a and 8 so that every •-circle is an
ellipse.)
Since •-space is, in fact, a model of plane Euclidean
geometry, all the theorems of the latter have counterparts
in the former. Nevertheless, we believe that it is just as
profitable to distinguish these spaces as it is to distin-
guish the calculi.
Clearly the foregoing developments can be extended to
n dimensions.
In Note 7 we show how to convert the set of all points
into a system of •-complex numbers.
10.3 •-VECTORS
For any numbers a in A and b in B, the transformation
• ••
that maps each point (x,y) into (x+a, y+b) is a •-vector and
* *
is here denoted by v[a;b]. The sum of v[a;b] *
and v[c;d] is
*v[a+c,
. b+d].
..
Theorem
For any point (x,y) and •-vector * v[a;b], the
• ••
point (x,y) , its image (x+a, y+b), and the
• • •• • •
latter's image (x+a+a, y+b+b), are •-collinear.
*
We say that the vector v[a;b] is rectilinear provided
• ••
that for EVERY point (x,y), the points (xly) 1 (x+al y+b) 1
• • •• ••
and (x+a+a, y+b+b) are classically collinear; otherwise the
vector is curvilinear.
If a = B= I, then all •-vectors are rectilinear.
But,
for instance, if a = I and B = exp1 then most •-vectors are
*
curvilinear; e.g., v[7,2] is curvilinear since it maps (1,3)
into (8,6), which is mapped into (15,12), three •-collinear
points that are not classically collinear.
Various types of scalar multiplications and norms can
be defined for •-vectors.
10.4 THE •-METHOD OF LEAST SQUARES

The •-method of least squares is the same kind of exten-
sion of the classical method of least squares as the •-calcu-
lus is of the classical calculus.
10.4 The •-Method of Least Squares 79
Let f be a discrete function with arguments a 1 , ... ,an

in A and values in B, and let S be a set of functions whose
values are in B and whose arguments are in A and include
al, • .. 'an.
If there exists a member g of S such that for every mem-
ber h of S distinct from g,
•• ••
2 •• ••
••
- - f(a )] 2
••
[g (al) f(a 1 )] + • • • + [g (an)
n
•• ••
•• 2 •• 2
< 1 -
[h ( a ) •• f(a )]
1
••
+ • • • + [h (an) ••
f (an) ] , -
then g, which is obviously unique, is said to be the member
of S that is best fitted to f by the •-method and is denoted
by L* (S; f) •
When a = B = I, the •-method is identical with the clas-
sical method and the notation L(S;f) will be used instead of
*
L(S;f). Please note that L(S;f) is not necessarily a linear
function.
For each function e with arguments in A and values in B,
- be the function consisting of all ordered pairs
let e
(a- 1 (x), B- 1 (y)) for which (x,y) is in a. LetS be the set
of all functions -e for which e is in s. *
Then L(S;f) and
--
L(S;f) coexist, and if they do exist,
*
L(S;f) = - -
S{L(S;f)}.
The pattern of that relationship is already quite familiar
in view of the results given in Section 6.8.
*
Thus, if L(S;f) exists, it can be found by using the
classical method to select the member of S that best fits -f -
and then applying B to the result.
Any statistical justification for the classical method
can probably be converted into a justification for the *-
method.
If S is the set of all •-uniform functions, then we have
the following results.
*
1. L(S;f) exists (and is unique).
2. 1£L(s;f) 1 (a 1 ) ~ f(a 1 >J + ...
+ 1[L(s;f) 1 (an) ~ f(an)} =
••
0.
*
3. L(S;f) contains the •-centroid of £, which we
define to be the point whose members are the
a-average of the ai and the a-average of the
f(a.),
l.
in that order.
*
In this situation, the •-slope of L(S;f) may be useful for
indicating the "overall •-direction" of the discrete func-
tion f.
Consider now the problem of fitting a function of the
form exp(mx +c) to a positive discrete function f:{(a 1 ,b 1 ),
••. ,(an,bn)}. Since the classical method is extremely diffi-
cult to apply here, scientists use the following "lineari-
zation" technique.
First f is transformed into {(ay1n(b1 )) , ••• ,(an,1n(bn))}
and the functions exp(mx + c) are transformed into ln[exp(mx+
c)], that is, into the linear functions mx +c. Then the
classical method is used to fit a linear function to the
transformation of f, and finally exp is applied to that lin-
ear function. The result, which we have not seen explained
heretofore, is identical with that provided by the •-method
for which a = I and S = exp.
The idea for the •-method arose from the following ana-
logue of a well-known theorem concerning the arithmetic aver-
age.
If b is the S-average of n numbers b 1 , ••• ,bn in B,
then
(b ~ b 1 ) + ... + (b ~ bn) = 0,
and for any number x in B distinct from b,
••
••
••• + (b :.: b )2
n
••
< (X ..- b )2
••
+ •••
+ (x - n •
••
••
•. b ) 2
1
Perhaps other statistical concepts can be developed in
the context of •-calculus. For example, since the •-graph
(Section 6.11) of the function a{exp[-(a- 1 (x)) 2 )} is a nor-
mal distribution curve, we may call that function •-norutal.
If a function is not classically norntal, it may be •-normal
for suitable choices of a and B, in which case the tools of
the •-calculus would be available for analytic purposes.
10.5 Trends 81
10.5 TRENDS
Although least squares methods provide interesting ways

of deterutining various "overall directions" of a discrete
function, there are other methods that are easier to calcu-
late and may prove to be more useful. For simplicity we be-
gin with functions defined on (closed) intervals, rather than
with discrete functions.
By a trend of a function on an interval we mean an aver-
age of a derivative of the function on the interval. (It may
be helpful to conceive of a trend as a kind of global deriva-
tive.)
For example, one could compute the quadratic average of
the classical derivative of g on [r,s]:
1 rs
}_
-=
(Dg) 2 •
s - r r
Or one could compute the biquadratic average of the anageo-
metric derivative of q on a positive interval.
Trends that do not depend on all the points of the func-
tion are trivial. For example, the arithmetic average of Dq
on [r,s] is a trivial trend, since it equals
[g(s) - g(r) ]/(s - r),
which depends solely on the points (r,g(r)) and (s,g(s)).
* . .
Similarly, the •-average of Dg on [r,s] is a trivial trend.
Nontrivial trends are obtained only by applying an average
from one calculus to a derivative from another calculus.
One can compute trends of a discrete function f by aver-
aging the slopes of the uniform functions joining successive
points of f, assuming, of course, that the arguments of f are
equispaced relative to the arithmetic used for the arguments.
For instance, one could compute the quadratic average of the
classical slopes of the linear functions connecting the suc-
cessive points. (The resulting trend is related to the root-
mean-square-successive-difference used by some statisticians.
We would like to know whether there is a trend operator
that assigns to each discrete function with equispaced argu-
ments the classical slope of the linear function that can be
fitted to f by the classical method of least squares.
10.6 CALCULUS IN BANACH SPACES
Each arithmetic can be converted into a Banach space.

For example, a-arithmetic is so converted by defining the
norm of each number u in A to be la- 1 (u) I and by defining
•
the scalar product of a number r and u to be a(r) x u.
Shortly after completing our development of the non-
Newtonian calculi in August of 1970, we had occasion to con-
sult the first edition of Jean Dieudonne's Foundations of
Modern Analysis. There we found a definition of the deriva-
tive of a function with arguments in a Banach space and val-
ues in a Banach space. If those Banach spaces are taken to
be converted a-arithmetic and converted a-arithmetic, re-
spectively, then that derivative is the •-derivative. For
the integral, Dieudonne apparently restricted his attention
to functions with arguments in R and values in a Banach space
If the Banach space is taken to be converted 8-arithmetic,
then that integral is the special •-integral for which a = I.
However, since we have nowhere seen a discussion of even
one specific non-Newtonian calculus, and since we have not
found a notion that encompasses the •-average, we are in-
clined to the view that the non-Newtonian calculi have not
been known and recognized heretofore. But only the mathemat-
ical community can decide that.
10.7 CONCLUSION
We trust that the basic ideas of the non-Newtonian cal-

culi have been presented in sufficient detail to enable in-
terested persons to develop the theory in various directions.
For example, one could study •-differential equations, inves-
tigate Taylor series in the context of •-calculus, or con-
struct non-Newtonian calculi of functions of two or more real
variables by choosing an arithmetic for each axis. It might
even be profitable to seek deeper connections among the cor-
responding operators of the calculi. One well-known theorem
about averages comes to mind:
10.7 Conclusion 83
Let a = I. For each positive integer n, let
xn if x > 0
- I
X if X < 0
and let *
n
be (a-arithmetic, Bn -arithmetic). Then
for any positive continuous function f on [r,s],
If the corresponding propositions for *n -derivatives and

integrals were true, then one could say that the geometric
calculus is the "limit" of an infinite sequence of calculi.
But, alas, those propositions are false. Nevertheless, it
is not unreasonable to speculate that there are nontrivial
sequences of calculi which do converge in some well-defined
sense.
Some years ago we encountered a remark by N.J. Lennes
that has become for us a constant reminder of the hazards of
premature judgment:
"Christiaan Huygens, one of the world's great

mathematicians and physicists, ••• never became en-
thusiastic about the calculus, and urged to the end
of his life that all problems solved by means of it
could be solved equally well by the older methods."
* * *
N 0 T E S
Note 1 (to page 12)

Those who prefer percents to ratios may find the follow-
ing concepts more appealing than the corresponding concepts
of the geometric calculus. However, the relative derivative
is not a non-Newtonian derivative according to our use of the
term and there is no integral that is the 'inverse' of the
relative derivative. It should also be mentioned that the
relative derivative is not the so-called logarithmic deriva-
tive, which also fails to be a non-Newtonian derivative.
The relative change of a positive function f over [r,s]
is [f(s) - f(r)]/f(r), which equals 1 less than the geometric
change of f over [r,s]. The relative slope of a geometrical-
ly-uniform function is its relative change over any interval
of classical extent 1. The relative gradient of a positive
function f over [r,s] is the relative slope of the geometri-
cally-uniform function containing (r,f(r)) and (s,f(s)), and
turns out to be
1
f(s) s-r
- 1,
f(r)
or 1 less than the geometric gradient off over [r,s].
In securities analysis and other applications the rela-
tive gradient is often called the comP9und growth rate. For
example, suppose that one paid $64 for a share of stock.
Three years later the price of the share is $216. The com-
pound (annual) growth rate of the price over the time inter-
val [0,3] is given by
[216/64]l/(J-O) - 1,
which equals 0.5 or SO%. The significance of that figure

stems from the fact that if an original investment of $64
increased 50% in each of three successive years, the final
value would be $216.
The relative derivative of a positive function f at an
argument a is the following limit, if it exists and is great-
er than -1:
84
N o t e s 85
1
f(x) x-a
x-e f (a)
The relative derivative of f at a coexists with, and equals

1 less than, the geometric derivative of f at a.
Note 2 (to pa9e 13)

We shall explain here how a simple algebraic identity
led us to construct the geometric calculus. First we re-
express the discrete analogue of the Basic Theorem of Classi-
cal Calculus(p.6) in a slightly different manner.
Consider any n points (a 1 ,v1 ), •.• , (an,vn), where a 1 <
a 2 < ••• <an and a 2 - a 1 = a 3 - a 2 = ••• =an- an-l• Let k
be the common value of the a.1 - a.1-1 , so that k(n-1) = a n -a 1 •
Connect the n points by line segments, of which there are
n - 1. Then the arithmetic average of their classical slopes
equals the classical slope of the line segment containing the
endpoints (a 1 ,v1 ) and (an,vn); that is
+ + ••• +
k k k
- •
n - 1
Now assume that the vi are positive. Then the following

identity is obvious:
1 1
1 1
-
v2 k
1
v3 k
- vn
-
k
n-1 v a- a
n n 1
• • •• v -
vl v2 n-1 vl
The left side of that equation is the geometric average of
the n-1 numbers (v./v. >1 /k, which, we imagined, must be
1 1.- 1
slopes of a new kind. We conjectured that the preceding
equation must be the discrete analogue of the Basic Theorem
of a new system of calculus, the geometric calculus.
Note 3 (to page 19)

Here we explain why, in the definition of anageometric
slope, we use positive intervals of geometric extent equal to
e, rather than to some other number greater than 1. (Geomet-
86 N o t e s
ric extent is necessarily greater than 1.)
Let w > 1, and assume we had defined the anageometric
slope of an anageometrically-uniform function to be its clas-
sical change over any positive interval of geometric extent
w. Then the anageometric slope of ln(cxm), the typical ana-
geometrically-uniform function, would equal rn•ln(w). Accord-
ingly, it is convenient to stipulate that ln(w) = 1, or w =
e. The choice of w = e is no more arbitrary here than the
choice of intervals of classical extent 1 in the definition
of classical slope (Section 1.2). In Section 6.10 we make
another comment on the choice w = e.
Note 4 (to page 49)

Many 'growth' calculi can be constructed by taking a =
I and choosing S from the growth curves, which include or
are related to the logistic curves, cumulative normal curves,
and the sigmoidal curves that occur in the study of popula-
tion and biological growth.
For example, let o(x) = (ex - 1)/(ex + 1). {Note that
a(2x) = tanh(x) .} The function a is of the sigmoid type (S-
shaped) and generates sigmoi~al arithmetic, whose realm con-
sists of all numbers strictly between -1 and 1. The sigmoid-
al sum and difference of numbers u and v in the realm turn
out to be (u + v)/(1 + uv) and (u - v)/(1 - uv), respectively.
The si9moidal calculus is the •-calculus for which a = I and
B = a. By choosing a = a and B = I we get the anasignaoidal
calculus; for a = B = a we obtain the bisigmoidal calculus.
It is entertaining to extend sigmoidal arithmetic by
appending -1 and 1 to the realm and defining the sum of num-
bers u and v to be •
(u + v)/(1 + uv) if uv ~ -1
u ~ v = •
0 if uv = -1
Then 1 ~ (-1) = 0; if u ~ -1, then u ~ 1
1; and if u ~ 1,
=
then u ~ (-1) = -1. Thus, -1 and 1 behave as negative and
positive infinity in the extended sigmoidal system.
If units are chosen so that the speed of light is 1,
then the extended sigmoidal sum of two speeds equals the
relativistic composition of the speeds, even if one or both
No t e s 87
of the speeds is 1.
Note 5 (to pages 53 and 60)
In this Note we show how to generate the infinitely-
many power arithmetics, of which the quadratic and harmonic
arithmetics are special instances. Let p be an arbitrary
-
nonzero number. The pth-power function is the function that
assigns to each number x the number
and negative
when x is negative.
For any numbers y and z,
- = -
(yz)P = yP.zP, and
<YP>1/p = Y = <Y1/p>P.
If p = 1/2, the Pth-power function is the function x 11 2 ,
which generates the quadratic arithmetic (Section 7.1); if
p = -1, the pth-power function is the function l#x, which
generates the harmonic arithmetic (Section 8.1); if p = 2,
- 2
the pth-power function is the function X ' which generates
the parabolic arithmetic (Note 6).
The Pth-power function is one-to-one, is on R and onto
R, has the 1/pth-power function as its inverse, and gener-
ates the Pth-power arithmetic. In that arithmetic the nat-
ural average of n numbers u 1 , ••• ,un turns out to be
{ (u11/p + • • • + un1/p) /n} p'

which reduces to the well-known pth-power average when the
u. are positive.
l.
Note 6 (to page 77)

The function p(x) = x 2
generates the 2arabolic arithme-
tic, which is one of the power arithmetics discussed in Note
5 above. The realm of parabolic arithmetic is R. The para-
bolic calculus is the *-calculus for which a = I and S = p;
88 No t e s
the parabolic!lly-uniforua functions are expressible in the
form {mx + c) 2 • By choosing a = p and B = I we get the
anaparabolic calculus; for a = B = p we obtain the bipara-
bolic calculus.
Note 7 (to page 78)

The system of •-complex numbers consists of all •-
points and two operations defined thus: the sum of (a 1 ,b 1 )
• ••
and (a 2 ,b 2 ) is (a 1 + a 2 , b 1 + b 2 ) and their product is
(a{a 1 a2 - b1 b2 l, a{a 1b2 a

+ b1 2 l),
where a- 1 = a -1 (a 1 ), a- 2 = a -1 (a 2 ), b- 1 = 8-1 (b 1 ), and b- 2 =

-1
B (b 2 ).
For this system, which is a field, one can define sub-
traction and division that are 'inverses' of addition and
multiplication. Furthermore, a-arithmetic (and hence every
arithmetic) is embedded in the system.
• •• • ••
Since the product of (0,1) with itself equals (-1,0),
* . ..
we may define ito be (0,1). Of course, the product of
• •• • ••
(0,-1) with itself also.equals (-1,0).
Various types of moduli can be defined for the •-complex
numbers.
LIST OF SYMBOLS
• • • •
Single-dotted notations, such as +, <, and [r,s],
pertain to a-arithmetic; double-dotted notations
pertain to a-arithmetic.
A, 32,38; A, 45 -
B, 38; B, 45 -
D, 4; ""'
D, 11-12; o, 19-20; ,..,
D, """' * 40;
26; D,
-D,
* 46;
,__.
0 a 0 0
D, 53; D, 55; D, 57; D, 61; 0, 63; D, 64
0 0 0 0
~
G, 16; G, 16; ,...,

G, 24; * *
G, 31; G, 45; G, 46
~ -
,...,
I, 1
*
L(S;f), 79; L(S;f), 79
~
M, 5; M, 13; M, 20; """'
M, 27; *M, 41;
-M,* 46
pl' p2' p3' 75

R, 1; R+ I 1
a, 33-34, 38
a- 1 is the inverse of a. (Page 1)
B, 38
6- 1 is the inverse of 6. (Page 1)

\ 1 39
*
v, 78
• • 0 0
[r,s], 1; [r,s], 35; [r,s], 59
e, 1
exp, 1
ln, 1-2
• ••
+, 32-J4; +, 38; <!> , 52; [±), 59
-,
• 32-34; -,
••
38; e , 52; EJ, 59
• ••
x, 32-34; x, 38
• ••
;, 32-34; ;, 38
89
90 List of Symbols
• ••
<, 32-34; <, 38; ~, 59
• ••
0, 33, 34; 0, 39, 75
• ••
1, 33, 34; 1, 39, 75
• ••
n, 34; n, 39, 75
• • •• ••
IX I I 34, 75; I f,
X 75
*, 38; -
*I 45
•-lim, 38; •-lim, 46 -

* 1 ,P 2 ), 49, 75-76
d(P
=
2
X I 52
1/2
X I 52
1
- 1 1#xl 59
X
••
b 2 1 75
·r, 75
""
f,?; [.14-15; f22; f29;
~
f. 56; fsa:
0 0
I N D E X
Most items are listed according to their qualifying

prefixes (a,S,•) or qualifying adjectives (anageo-
metric, anaharmonic, anaquadratic, bigeometric, bi-
harmonic, biquadratic, classical, geometric, harmon-
ic, quadratic).
a-Absolute value, 34,75 Anasigmoidal calculus,86

a-Arithmetic,33-34 Arguments,!
Banach space,82 Arithmetic,32-33
a-Average,35,70-71,80 Arithmetic average,5,70-72
a-Convergence,36 characterization,S-6
a-Extent,35 Arithmetic partition,S
a-Integers,34
a-Interior,35 (For 6-items not listed here,
a-Intervals,35 see like entries under a.
a-Limit,36 Also see comment on p.38.)
a-Negative numbers,34 a-Absolute value,75
a-One,34 S-Additive,40
a-Partitions,35
a-Positive numbers,34 B-Arithmetic,38
a-Progressions,35 a-change,39
a-Similitudes,67-68 6-Homogeneous,40
a-Translations,67-68 8-Similitudes,67-68
a-Zero,34 a-square,75
a-square root,75
Additive,4 S-Translations,67-68
Anageometric average,20 Basic Problem:
characterization,21 Anageometric Calculus,22
Anageometric calculus,l8-24, Bigeometric Calculus,29
47-48 Classical Calculus,7
Anageometric derivative,l9 Geometric Calculus,l4
Anageometric gradient,l9
Anageometric integral,22 •-Calculus,43
characterization,23 Basic Theorem:
Anageometric slope,lS-19, Anageometric Calculus,21
48, 85-86 Anaharmonic Calculus,63
Anageometrically-uniform Anaquadratic Calculus,S6
functions,l8 Bigeometric Calculus,29
Anaharmonic average,63 Biharmonic Calculus,65
Anaharmonic calculus,62-63 Biquadratic Calculus,58
AnahaLittonic derivative, 63 Classical Calculus,6
Anaharmonic gradient,62 Geometric Calculus,lJ-14
Anaharmonic integral,63 Harmonic Calculus,62
Anaharmonic slope,62 Quadratic Calculus,54
Anaharmonically-uniform •-Calculus,42
functions,62
Anaparabolic calculus,SS Bigeometric average,27-28
Anaquadratic average,56 characterization,28
Anaquadratic calculus,SS-57 Bigeometric calculus,25-31,
An·aquadratic derivative,35 47-48
Anaquadratic gradient,SS Bigeometric derivative,26,48
Anaquadratic integral,56 Bigeometric gradient,25-26,66
Anaquadratic slope,SS Bigeometric integral,29-30
Anaquadratically-uniform characterization,30
functions,SS Bigeometric slope,25,48,74
91
92 I n d e x
Bigeometrically-uniform Biharmonic Calculus,65
functions,25,74 Biquadratic Calculus,SS
Biharmonic average,64-65 Classical Calculus,S
BihaJ:tuonic calculus ,64-65 Geometric Calculus,l6
Biharn~nic derivative 64
• I
Harmonic Calculus,62
B1harmonic gradient,64 Quadratic Calculus,SS
Biharmonic integral,65 •-Calculus,44
Biharmonic slope,64 Function,!
Biharmonically-uniform
functions,64 Galilee, G., 9,18,73
Biparabolic calculus,88 'Generates,' 34
Biquadratic average,S8 Generator,33
Biquadratic calculus,57-58 Geometric arithmetic,36-37
Biquadratic derivative,57 applications,47-49
Biquadratic gradient,57 Geometric average,l2-13,71,83
Biquadratic integral,SS characterization,l3
Biquadratic slope,57 Geometric calculus,9-17 47-48
Biquadratically-uniform 85 I I
functions,S7 Geometric change,lO

Bisigmoidal calculus,86 Geometric derivative,ll,48
Calculus in Banach spaces,82 Geometric extent,l8
Campbell, N.R.,33 Geometric
• '
qradient,ll,66 I
67
J.nvarl.ance,68
Cartesian paper,Sl Geometric integral,l4-15,70
C-graphs,Sl characterization,lS
Classical arithmetic,33 Geometric partition,20
Classical calculus,J-8,47 Geometric slope,ll,73
Classical change,3 Geometrically-uniform
Classical derivative,4 functions, 10,73
Classical extent,! 'Gradient,' 3
Classical gradient,3,66,67 Graphical interpretations
• •
1.nvar1ance,68 of •-calculus,49-51
Classical integral,7,69 Graphs
characterization,& c-, s1
Classical method of least
squares,79 ·- 1 49
Classical slope,3 Harmonic arithmetic,59-60,87
Classical tangents,4-5 Harmonic average,60,61,71
Classically tangent func- Harmonic calculus,60-62
tions,S Harmonic derivative,61
Classically-uniform func- Harmonic gradient,60-61,67
tions,3 Harmonic integral,61
Complete ordered field,32 Harmonic progressions,60
Compound growth rate,84 Harmonic slope,60,73
Harmonically-uniform
Dieudonne, J., 82 functions,60,73
Discrete analogue,6 Heuristics,66-74
Discrete function,! averages,70-72
Domain,l constants and scientific
concepts,72-74
Ellis, B. , 71 qradients and derivatives
Exp, 1 66-68 ,
integrals,68-70
First Fundamental Theorem: Homogeneous,4
Anageometric Calculus,23 Huygens, C. , 83
Anaharmonic Calculus,63
Anaquadratic Calculus,56 1 (iota), 39
Bigeometric Calculus,30
I n d e x 93
Identity function,! Relationships between calculi:
Invariance,68,71 anageometric to classical,24
Inverse of a function,! any two calculi,45-47
Interior,! bigeometric to classical,31
Interval,! geometric to classical,l6-17
Isomorphism (\), 39 • to classical,44-45
Relative change,84
Relative derivative,84-85
Relative gradient,84
Least squares methods,78-80 Relative slope,84
Lennes, N.J., 83 Resiliency,27
Linear function, 1,3 Rockland, c., 5
'Linear operator,' 4 Root mean square,S3,54
Ln, 1
Logarithmic derivative,l2 * 1 ·3 8
*-Arc lenqth,77
Mean value theorem *-Area,SO
of geometric calculus,l7 *-Average,41,70,82
of •-calculus,47 characterization,41-42
Measurement,33,66,71 •-calculus,JS-51
graphical interpretations,
Natural avera9e,JS 49-51
Natural progression,35 *-Centroid,SO
Newton, I., 18,67 *-Collinear,76
n-Fold,S,20,35 *-Complex numbers,88
Non-Newtonian calculi, *-Continuity,39
see •-calculus *-Derivative,40,50,82
Number,! invariance,68
*-Differential equations,
'On, ' 1 68-69
One-to-one,l *-Distance,49,51
• Onto, • 1 forntulas, 75-76
*-Gradient,40,67-68,73
Parabolic arithmetic,87 invariance,68
Parabolic calculus,87 *-Graphs,49
Point,l *-Integral,43,50-51,70,82
Positive function,! characterization,43-44
Positive interval,l8 *-Limit,38
Power arithmetics,87 *-Lines,76-77
2th-Power average,87 *-Method of least squares,
pth-Power arithmetic,87 78-80
*-Metric,76
Quadratic arithmetic,52-53,87 *-Normal function,80
Quadratic average,S3,54 *-Paper,49
Quadratic calculus,SJ-55 *-Parallel,77
Quadratic derivative,SJ-54 *-Perpendicular,77
Quadratic gradient,SJ *-Points,49
Quadratic integral,S4 *-Slope,40,50,51,73,80
Quadratic partitions,SJ *-Space,75-78
Quadratic progressions,S3 *-Tangent,41
Quadratic slope,S3 *-Tangent functions,41
Quadratically-uniform *-Uniform functions,39,73,77
functions,53 *-Vectors,78
Range,! Second Fundamental Theorem:
Realm,32 Anageometric Calculus,23
Real number system,32 Anaharmonic Calculus,63
94 I n d e x
Anaquadratic Calculus,S7
Bigeometric Calculus,Jl
Biharmonic Calculus,65
Biquadratic Calculus,58
Classical Calculus,8
Geometric Calculus,l6
Harmonic Calculus,62
Quadratic Calculus,SS
•-Calculus,44
Sigmoidal arithmetic,86
Sigmoidal calculus,86
Sigmoidal gradient,67
Similitudes,67-68
Speed,72-73
Stevens, s.s., 66,71
Stieltjes integral,22,43,56
Translations,67-68
Trends,81
Values,!
Vertical •-line,77

Michael Grossman, Robert Katz - Non-Newtonian Calculus (1972 - 2006, Kepler Press) PDF

Uploaded by

Copyright:

Available Formats

Michael Grossman, Robert Katz - Non-Newtonian Calculus (1972 - 2006, Kepler Press) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Michael Grossman, Robert Katz - Non-Newtonian Calculus (1972 - 2006, Kepler Press) PDF

Uploaded by

Copyright:

Available Formats

Non-Newtonian Calculus

We are pleased to present here the

101 Granite Street

Quoted, without citation,

A popular method of creating a new mathematical system

shall not pursue that course. Instead, in Chapter 1 we for-

6.5 The Basic Theorem of •-Calculus, 42

Chapter 10: COLLATERAL ISSUES

Since complex numbers are not used in this book, the

positive; a discrete function is a function that has only a

1.2 THE CLASSICAL GRADIENT

1.3 THE CLASSICAL DERIVATIVE

Throughout this section, f is assumed to be a function

The classical derivative of f, denoted by Of, is the

lg(x) - f(x) I < lh(x) - f(x) I .

1.4 THE ARITHMETIC AVERAGE

The arithmetic average of a linear function on [r,s] is

1.5 THE BASIC THEOREM OF CLASSICAL CALCULUS

The Basic Theorem of Classical Calculus

metic average fits naturally into the scheme of classical

The number (s- r)•M;f in the foregoing solution arises

1.6 THE CLASSICAL INTEGRAL

The classical integral is a weighted arithmetic average,

Furthermore, ls f equals the limit of the convergent sequence

a 1 , ••• ,an is then-fold arithmetic partition of [r,s], and kn

[sr (c• f) =c ·1r f,

The classical integral is the only operator that has

2. For any interval [r,s] and any functions f and g

3. For any numbers r, s, t such that r < s < t, and

The classical derivative and integral are 'inversely'

q(x) = [xf, for every number x in [r,s],

During the Renaissance many scholars, including Galileo,

We shall see that the expression within the brackets is of

2.2 THE GEOMETRIC G~DIENT

The geometric change of a positive function f over [r,s]

The geometric slope of a geometrically-uniform function

If one plots a geometrically-uniform function h on semi-

Clearly the geometric gradient is independent of the

2.3 THE GEOMETRIC DERIVATIVE

additive; however D is multiplicative:

Recalling that the function exp equals its classical de-

2.4 THE GEOMETRIC AVERAGE

3. For any numbers r, s, t such that r < s < t, and any

2.5 THE BASIC THEOREM OF GEOMETRIC CALCULUS

Our investigation of non-Newtonian calculi began with

The Basic Theorem of Geometric Calculus

If Dh is continuous on [r,s], then its geometric

In view of the preceding theorem we say that the geo-

2.6 THE GEOMETRIC INTEGRAL

The geometric integral of a continuous positive func-

gent sequence whose nth term is the product

The operator l s is multiplicative:

The geometric integral is the only operator that has

1. For any interval [r,s] and any positive constant

The geometric derivative and integral are 'inversely'

Just as the Second Fundamental Theorem of Classical Cal-

2.8 RELATIONSHIPS TO THE CLASSICAL CALCULUS

The relationship of the nth-geometric derivative to the

3.2 THE ANAGEOMETRIC GRADIENT