MATHEMATICAL TRACTS

PAET I.

BY

F. W. NEWMAN, M.RA.S.
EMERITUS PROFESSOR OP UNIVERSITY COLLEGE, LONDON

HONORARY FELLOW OF WORCESTER COLLEGE, OXFORD.

Cambridge :

MACMILLAN AND BOWES.

1888

[All Rights reserved.]

 TKACT I.

ON THE BASES OF GEOMETRY WITH THE

GEOMETRICAL TREATMENT OF J^l.

CONTENTS.

(1) On the Treatment of Ratio between Quantities Incommensurable.

(2) Primary Ideas of the Sphere and Circle. Poles of a Sphere.
(3) Definition and Properties of the Straight Line.

(4) Definition and Properties of the Plane.

(5) Parallel Straight Lines based on the Infinite Area of a Plane Angle.
(6) On the Volume of the Pyramid and Cone.

I. THE RATIO OF INCOMMENSURABLES.

1. IN arithmetic the first ideas of ratio and proportion, and

the laws of
passage from one set of 4 proportionals to another, ought
to be learned, as preliminary to geometry; but in geometry the

doctrine of incommensurables requires a special treatment, unless

the learner be well grounded in the argument of infinite converging

series. Repeating decimals may perhaps suffice. Another, possibly
better is open by the introduction of VARIABLE quantitu
way, ".

which will here be proposed.

2. Nothing is simpler than to imagine some geometrical quantity

to vary
in shape or size according to some prescribed law. This

must imply at least two quantities varying together. Thus, if an

equilateral triangle change the length of its side, its area a Is..

changes. If the radius of a circle increase or diminish, so does the

length of the circumference. In general two magnitudes X and Y

together: they be either the same in kind, as the

may vary may "

radius and circumference of circle is each a length; or the two

a may

be different in kind, a length and an area. In general it is a

say,

N,
1
2 VARIABLES WHICH INCREASE UNIFORMLY.

convenient notation to suppose that when X changesto X', Y changes

to r.

3. Again, if X receive successive additions a^ayCg ...#", the responding

cor-

additions (ifadditions they be) to Y are well denoted by

'Vn-
y 'i2/*2/3-- An obvious and simple case, if it occur, will deserve

notice; namely, if the two variables are so that equality

regulated,
in the first set of additions (i.e.
#^=3^= x3 "

...
=#") induces equality
in the second set; (i.e.
y^=yz=y^= ... =yn). The variables X and Y

are then said to increase uniformly. As an obvious illustration,

pose
sup-
X to be the arc of a circle,
and Y the area

of the sector which it bounds, evidentlythen if

the arcs #,,#" are equal increments of the arc X,

the y^yz which
sectors are bounded by x^ will

be equal increments of Y. Then the arc X and

the sector Y increase togetheruniformly.

4. We may now establish a theorem highlyconvenient for ap-

plication
in geometry, alike whether quantitiesare commensurable
or incommensurable.

THEOREM. "
If X and Y are any two connected variables,which
beginfrom zero together,and uniformly; then X increase varies

proportionablyto Y. In other words, if Y become Y' when X comes

be-

X', then X is to X' as Y is to FV

Proof. First,suppose X and X' commensurable, and f a common

measure, or f (m times f) and X'

X= m . n". We may then sup-pose =

X and X' made up by repeated additions of f Every time .

that X has the increment ", Y will receive a uniform increment

which we may call v\ then Y is always the same multipleof v that

X is of f; thus the equationX =
mj; impliesY=mv, and X' =
n^
impliesY'=nv. Hence X : X'=m : n= Y : Y'.

Next, when X' is not commensurate X, yet f is

with some snh-

multipleof X, such that n% = X, and X' contains f more than "///

times, but less than (m + 1) times ; evidentlywe cannot have

X : X' = Y : F

(when the four magnitudes are presented to us) unless, as a first

condition,on assuming nv= Y, we find Y' to contain v more than

m times and l"-s than (m+ 1) times: and unless tin's condition were

X and
fulfilled, Y would not increase uniformly. We therefore
may
 DISTANCE OR SHORTEST PATH.

assume X9X8 on oppositesides of X', with values

likewise F2F3 on oppositesides of F',with values

Then by the first case have

F2 and X : X3= Y : F3.
we X : Xz = Y :

But X^-X^j;, and F3" F2=u. Let n perpetually increase,then f

and v perpetually lessen. Xz and Xz run togetherin X', F2 and \\
run togetherin Y'. Thus each of the ratios X : Xz and X : X3 falls
into X : X', and each of the ratios F : F2) F : F3 falls into F : )".

Inevitablythen, X : X' = Y : Y', even when these last are mensurate.

incom-

Q.E.D.

II. PRIMARY IDEAS OF THE SPHERE AND CIRCLE.

For the convenience of POSTULATES

beginners, may be ad van cvd

concerning the straightline and the plane,as well as concerning

straight
parallel lines. But in the second stage of study the whole

topicought to be treated anew from the beginning:a task which is

here assumed.

On Length and Distance.

THEOREM. "
All lengthsare numericallycomparable."To make
this clear,it is simplestto thin, flexible
imagine a thread indefinitely
and appliedupon any given line,will become an
inextensible. This,if
exact measure of its length;and if any two lines be then measured

by two threads,the threads are directlycomparable, shewing either

other and ln"w
that they are equal,or that one is longer than the
safelyassert the same fact concerning
much longer. Hereby we
any givenlengths.
two
that i
Obviously,lengthis continuous magnitude: which means,
throw jh ail
a pointP run along from A to B, the lengthAP passes
magnitudefrom zero to AB.

space may be joined

either
THEOREM. Any two given pointsin
by one path which is shorter than any other possible,
or by si'\
other is short. For of all possible
equalpaths than which none so

be long; yet
needlessly unless there
pathsjoiningthem some must

the distance would be nil; the |"

is some limit to the shortening,
would not be two, but would coincide and become one.

1"2
4 THE SPHERE.

DEF. A shortest path that joinstwo pointsin space gives a

ire of their DISTANCE. The same argument applies,if the two

given pointsand the line that joinsthem must lie on a given surface;
or again,if two surfaces that do not touch be given,and we speak of
the shortest distance of the two surfaces.

Assume a fixed point A and a second point S so movable as

always to be at the same distance from it. It will be able to play all
round A : therefore its locus will be a surface enclosingA. The
s""lid mass enclosed is called a SPHERE (Globe or Ball) and A its
( 'KNTRE.

THEOREM. Every pointoutside the sphere is further from the

"

centre and every pointwithin the sphereis nearer to the centre, than

are the pointson the surface." For if T be an exterior point,every

path joiningT to A must piercethe surface in some point$; there-

fore
the path TSA is longerthan SA by the interval TS. Again, if
R be within the sphere, we may imagine an interior sphere whose
surface is at the common distance AR from A. Then 8 being
exterior to the new sphere,SA is longerthan RA ; that is,R within
the sphereof S is nearer to A than is the locus of 8. Q. E. D.

DEF. Two such concentric spheresenclose within their surfaces

a solid called a spherical
shell.

THEOREM. "The two surfaces are equidistant,each from the

other." For if the shortest distance from a pointS to the inner face
sur-

is the
path SR, symmetry all round shews at once that if from a
second pointS' the shortest path will be S'R', the two distances SR,
S'R' will be equal. Indeed it is not amiss to remark, that if any

spherical surface be rigidly attached to its centre, the entire surface

may glideon its own ground without disturbingits centre, because

th"- distances SA, S'A nowhere change. Hence also we may justly
imagine the sphericalshell to glideon its own ground, while the
centre suffers no displacement, and any shortest path S'R joining
the oppositesides of the shell may assume the place which was

previously h eld by SR. Actual superposition thus attests equality of

nee.

TIIK"II:F:M. "If a sphericalsurface be given,its centre is deter-

mined."
For if an inner pointR be assumed ,-it a ^ivcn distance /)
from the surface, its focus is an interior continuous surface. Witliin
tins, at distance //. imagine a point I\' to ^vnerafe a second c.,n
 POLES AND PARALLEL CIRCLES. 5

tinuous surface,and it will be interior to the precedingand so con-

tinually.

The series of surfaces must then necessarilyconverge

towards a singlepoint,which will be the centre of the given surface,
because the sum of the distances is the same, from whichever point
we calculate. The same argument proves that all the surfaces are

concentric spheres.

Poles of a Sphere.

THEOREM. To
pointon a sphere one oppositepoint lies
"
every
at the longestdistance along the surface." For if the pointP be
given,and we take a point$ at any distance from P alongthe surface,
and suppose S to vary under the sole condition that its distance
from P
(alongthe surface)shall not change,the locus of $ is a self-
rejoiningline enclosingP. (We call this a circle.)Next, beyond
S, along the surface, take a new point T, which moves without

changing its distance from S and from P. This generates an outer

circle, cuttingoff a part of the surface which was beyond the circle
of S. Beyond this we may similarly form a third circle,and this
series of circles ever lesseningthe finite area beyond it,will neces-
sarily

converge towards a point Q on the sphere. P will then be

farther from Q (alongthe sphere)than any of these parallel circles.
We call P and Q opposite polesof the sphere. The distance between
them is evidently the halfgirthof the sphere.

Every pointon the sphere has not only its own oppositepole;
but also its system of equidistant (or parallel) circles. The middle

one of these (that is,the one equidistant from the two poles), is

called their equator.

If in an equator whose poles are P and Q, you fix any point
0, and then proceedinghalf round the equator a fix second point
D, C and D are evidentlyoppositepoles.
If you imagine a sphereto glideon its own ground,with centre

unmoved, P to pass over to the site held viously

pre-
you may suppose
P.
by Q. This carries Qplace previouslyheld by
to the
is
Thus the poles are exchangeable,while the sphere as a whole
unchanged and the same equator is attained.

THEOREM. "
If P and R be any two pointson a spherethat are
not oppositepoles,
one equator,and one only,passes through them
both."
 6 THE STRAIGHT LINE.

Proof. Through P and its oppositepoleQ (justas above through

the polesC and D) an equator may pass. If this half equator PQ

become rigidand be rigidly attached to the fixed centre A, it still

may sweep over the spherical surface (without change of P or Q)
until it passes through R ; but after passingonce through E, it does
not corne back to it,except in a second revolution. Q. E. D.

III. POINTS LYING EVENLY.

In Simson's Euclid, the line whose points lie evenlyis called

"
; but the phrase lyingevenly is not explained.We can
"
STRAIGHT
now explainit.
When the two poles P and Q, and the centre A, all remain
unchanged,nevertheless each of the parallelcircles associated with
r and Q can glideon their own ground. Evidentlythen, if P and
A be fixed,this suffices to fix Q. In fact while each circle spins
round its own line,Q onlyspinround itself.Also, to fix P arid Q
can

fixes A. "
These parallel circles excellentlydefine to us the idea of

rotation,which is a constrained motion, still possible,even when P,

A, Q are all fixed. Now suppose that a line PMQ internal to the

sphererigidly
connects P with Q. Then if the system revolve round

P, A, Q, PMQ may generate a surface within the sphere.

self-rejoining
Again within this new surface line PNQ
a rigid may connect P with

Q, and the line PNQ by rotation round P, A, Q may generate a

third surface interior to the preceding; and so on Since

continually.
tliere is no limit to the constant thinningof the innermost solid,we
see that a mere line without thickness connects P with Q and passes

through A, which line is interior to all the solids and during rotation
remains immovable. It is called an axis,and can only turn about
itself. Hence every pointin this axis lies evenlybetween P and Q.
And since P and Q may representany two pointsin space, we

now discover that between any two there is a unique line lyingevenly.
This continuous line,while we talk of rotation round it,is entitled an

axis; but ordinarily

we call it simply STRAIGHT.

On the StraightLine and its "

Direction."

We now infer that

1. Any two pointsin space can be joinedby a line.

straight
2. Every part of a line
straight is straight.
 THE PLANE. 7

3. A line
unique straight is determined,when its two end-points
are given.
4. Any part of a line,if removed,
straight may take the placeof
any equal part of the same. Hence it easilyfollows that a straight
line,glidingalong itself, will prolongitselfindefinitelyfar,either way,
along a determinate course.
We are sharpen our
now able to idea of direction. Hitherto we

might say vaguely, Imagine a path "

to proceedin any direction" that
guidance. But
is,without particular now we see, that if ever so short

a straightline be drawn, it pointsto a definite prolongation

beyond
of indefinite
itself, extent. This we entitle its direction. If this

direction bechanged,a deviation there occurs, and a sharp corner is

recognizedat the point of deviation. The amount of deviation

suggests a new kind of magnitude, which will presentlyneed

attention. Now it suffices to remark on the case in which a new line

AZ equallyfrom a previousline PA and from AQ the pro-

deviates longation
of PA. The equality is tested by imaginingAZ to become

an axis of a sphere. Then if P and Q revolve in the same circle, ZA

is equallyinclined to AP and to AQ. It is called perpendicular to

PAQ. EvidentlyZ (on the sphere)is at the distance of a quarter

girthfrom every pointof the equator traced by P and Q.

IV. THE PLANE.

We return to the sphere. When any two polesP, Q joined

are

by a straightline,it has been seen that this passes through the

centre A. The line PAQ is called a diameter of the sphere,and
its half (AP or AQ) is called a radius.

Evidently all the radii of the same sphere are equal; and of

different spheresthe greater the radius, the greaterthe sphere.

is midway between the polesP, Q, and D
If an equator CDEG
diameter PQ, too the meter
dia-
is the pole oppositeto C, then as the so

CD, passes through centre A. This is true, whatever pointin

the equator is assumed for C. Therefore CAD is a varyingdiameter,
whose extremities trace out the equator,while the diameter traces

out surface in which the equator lies. This surface is called a

a

is the planeof the equatorialcircle.

PLANE, and particular
in
while the
It was seen that P and Q might exchange places,
centre A, and the sphere'ssurface as a whole,
remain unchanged.
It is
Necessarilyalso the plane of the equator remains unchanged.
 Till: PLANK HULED HY A STRAIGHT LINE.

then symmetricalon its

oppositesides,or in popularlanguage,the
plane turns the ^mne face towards P as towards Q.
The axis PAQ is called perpendicularto the plane of the equator,

g perpendicular to- every radius of the equatorial

circle.

THEOREM. "
No other line but AP can be perpendicular
to the
plane of the equator,}*
Proof. For if AR be some other radius of the
sphere,some one

of the
parallelcircles, whose pole is P, passes through R, and every
pointof this circle is nearer to the equatorial circle than is the poleP.
Theivt""iv the distance of R from the equator is less than a quarter
of the sphere'sgirth,a fact which shews RA not to be dicular.
perpen-

THEOREM. "
Through any two radii AP, AR of a sphere,that are
not in the same straight line,one planeand one only may pass."
It has Urn seen that through P and R only one equator can
pass. The plane of this equator is the plane that passes through
the two radii.

Cardinal Property of the Plane.

THEOREM. "
If M and N are any two pointsin a plane,no point
in the straightline which joinsM and N can lie offthe plane on
cither side."

Symmetry suffices to establish this truth. Our

hypothesissupplies
data to fix what line is meant by MN, but givesno reason why any
point of it should lie off the plane on one side rather than on the

other; for the whole line is determined by merely the extreme

pointsM, Ny of which neither can guide any pointtowards P rather

than towards Q. Thus there is no adequate reason for deviation
towards either side.

Symmetry of data is in other mathematical topicsacceptedas

an adequate argument for symmetry of results. Otherwise, "the
want of sufficient
reason for diversity"
passes as refutation of alleged
sity. Therefore the argument here presentedhas nothingreally
,

novel.
We have now a new method of generating a plane that shall
pass through two intersecting straightlines LM, MN. Along ML
let a point E run, and along MN similarly a point F. Join EF
while the motion of E and of F continues. Then EF (by the last
 THE PLANE EVERYWHERE LIKE ITSELF. 9

Theorem) always continues to rest on the plane LMN. This mode of

generatingthe plane supersedesthe idea of rotation. For simplicity

we might suppose ME : MF to retain a fixed ratio.

THEOREM. "
A plane has no unique point or centre."
For if we from
given sphericalradii AP, AR
start through
which passes an (equatorial) plane,in AP take M arbitrarily, and
in AR take N arbitrarily. Then we have seen that the locus of the

moving line MN is our given plane. But again,in this plane take
a fixed point 0, and join 0 to fixedpointsM and N. Then from the
lines OM, ON we can (as in the last)generate the very same plane,
which can glideon its own ground sphere did ; thus
as the the point
A can pass to 0 without changing the ground or surface as a whole.
The plane is the sphere is finite ; but
infinite, as with the sphere,
so with the plane,no pointof the surface is unique.
After this,no impediment from logicforbids our passing to the
received routine of Plane Geometry, until we are arrested by the

difficultyof parallelstraightlines,to which I proceed, after one

remark on the definition of an angle.

Above, a sharp corner or turn was identified with deviation,or

change of direction. In geometry it has the name of an angle,and
we measure magnitude by aid of the circular arc which it sub-
its tends

at the centre or by the sector of that arc. But no insuperable

logicforbids our estimatingthe magnitude of an angleby the portion

of the infinite area which it interceptsfrom a plane; which indeed
is suggestedby a perpetualelongationof the radius of the circle
whose sector was assumed as measure of the angle.

Monsieur Vincent in Paris (1837) adopted this definition as

adequate to demonstrate the equivalentof Euclid's Twelfth Axiom

without new axiom at all. Has this method received due

any
attention in England ?
Monsieur Vincent was not the suggest acceptingthe
first to

infiniteplane area cut off by two straightlines,as the

intersecting
measure of the angle which they enclose : but perhaps he was the

first to introduce the method into a treatise Elementary Geometry,

on

that obtained in high institution as the University

acceptance so an

of France.

Two lemmas alone are wanted, and these every beginnerwill find
natural.
10 PROOF OF Tin; TWELFTH AXIOM.

LEMMA "Every angle is a

I. finite fraction of a rightangle;"
that is,some finite multipleof it exceeds 90". For the circular arc

which subtends it,is always some finite fraction of the quarter of

the circumference.

DEF. straightlines AM, BN

When two in the same plane are
both perpendicular to a third straightline AB, we call that portionof
the planearea which is enclosed between MA, AB and BN a BAND.

LEMMA II. Then, I say, whatever the breadth (AB) of the band,
UK area of the band is less than any finite fraction of a rightangle.

Proof. Prolong AB indefinitely to X, and along it take any

number of equal lengthsAB BC CD DE, "c., and through
= = =

G, D, E... draw perpendicularto ABODE... straightlines CO, DP,

.c. Evidentlythen the successive bands are equal,by superposi-
tion.
Thus, whatever multipleof the first band be deducted from
the plane area marked off by the right angle MAX, the loss is
iiiM nsible ; for,as remainder,we find the area marked off still by a

rightangle (such as QEX, if only four bands were deducted). Any

two rightangles embrace areas which can be identified by super-
position,
and have no appreciabledifference. The matter may be

concisely summed up by remarking, that every band is infinite in

one direction only, say, horizontallybut the area

" embraced by any "

rightangle is infinite in both directions,horizontally and also verti-

cally.
Thus it is no paradox to say, that no finite multipleof the
band can, by its deduction from the area of the rightangle,lessen
that infinite area in our estimate. Q.E.D.

Euclid's Twelfth Axiom is now an immediate ; viz.

corollary If
MABN be any band ; and, within the right 3
N
angleNBA, any line BT
straight be drawn,
T
11 be prolongedso far as to meet the

prolongationof AM. For the angle NBT

'r.'H-tionof a right Jin-lf,while
i li"-ban"l J//;.LY is less than Unite fraction of the same; hence
any
the angle NBT is greater than the band MABN, but unK-ss /""/'
AM this would be ials" .
Thus of necessitythe two lines
do cross, as we assertril.

I cannot see new axiom involved in this

proof:therefore
any
I am forced to abandon several other speciousmethods and give it
 VOLUME OF PARALLELEPIPEDON. 11

preference. Surely we may bow to the authorityof the University

of France in such a matter.

On the Volume of Pyramids and Cones.

The treatment of this

topicin Euclid is very clumsy. It demands
and it admits much improvement.
1. For parallelepipedaprove first,
that if two such solids differ
solelyin the length of one edge,which we may call x in the one

and a in the other, then their volumes are in the proportion of

x : a.

2. Next, if they have a solid angle in common, but the edges

round it are in one xy y, zt and in the other a, b, c, then the two
volumes are in the proportionof xyz : abc.

3. After easilyshewn that parallelepipeda

this it is on the same

base and equalheighthave equalvolumes.

4. Therefore finally, that the volume of a parallelepipedon
is

measured by its base x its height. Con. The same is true of any

prism.
From this we proceedto approximateto the volume of a pyramid.
5. Divide the height(h)into (n) equal parts by (n" 1) planes
all parallelto the base (B). Establish, on these (n 1) bases, -

upright walls, and you will find you have constituted a double
system of prisms,one interior to the pyramid, one exterior; the
latter has the lowest prism in excess of the other system. Every

base is similar to every other, by the nature of a pyramid. The

volume here of every prism is - x its base, the number n and -

being

the same for all,but the base varying.

6. The base whose distance from the vertex is -.h, is to the

r2
is which givesfor the
original(B) as r2 : n2; hence its area
jji"ft
volume of the prism standing on it (^BJ . .
-

.
Hence the sum of

l2 + 22 + 32+...+ rca h "

the volumes of the external prisms is -

-

.
-

.
#, f
a
^

the numerator of the largerfraction we obtain

by omitting ?i2from
\'2 YoU'Mi: OF PYRAMID.

thi- Mm i
""t' volumes for the internal prisms. Now since -
.,'

when It, /" arc finite and n infinite, the difference of the

ot' prisms vanishes when is infinite. But the volume

systems n

of the pyramid is less than the exterior system and greater than the

interior: hence each system has the volume of the pyramid for its

limit, when n
increases indefinitely.

7. Let be the unknown numerical limit to which the fraction

fj,

1s + 22 + 32 + ?i2
" "
approximates when n
thus increases. Then the
,

volume of the pyramid h B. Since is the limit of a

numerical
=
/j, . .
//,

tract i""n, which remains the same,

whatever the form of the pyramid's

base, we
shall know the value of for all pyramids, if we can
find
/z,

it in one.
Meanwhile the result V = /j,.h.B at once shews that

pyramids with equal base and equal height have equal volume, since

is tin- same for all.

IJL

8. When this theorem has been attained, we have only to divide

a triangular prism into three pyramids, and instantly infer that the

3 equal themselves; therefore that each has volume

are among a

just " of the prism, i.e. equal $h.B.

This, being proved of a pyramid whose base is a triangle, shews

that the unknown is there exactly J.

//,

Hence universally ji = J, and volume of

every pyramid = J/t B, .
or

is equal to J of the prism which has the same base and height.

COR. Every cone also is one third of the cylinder which has the

same
base and height.
 TEACT II.

GEOMETRICAL TREATMENT OF V~l-

1. In algebra, concerned with number

pure only, the symbols +
and "
denoting addition and subtraction, in an early stage needed
,

elucidation when the mark of minus was doubled. It is found natural

that (+ a) and (" a) should both

"
+ mean "
a, but that "("a) should

mean + a,
and (- a) (- 6) should be + (db) surprises a beginner, and
.

is illustrated by urging that to subtract a debt increases the debtor's

property, and to subtract cold is to add heat. But as soon as we

apply algebra to geometry, the symbols + and " are still better

interpreted of reverse direction; also time past and time coming

afford equally good illustration. Distinguishing positive and negative
direction along a line, we find no mystery in the fact that to reverse

negative direction is to make it positive,so that " (" a) gives + a,

as reasonably as " (+ a) gives "

a. If we know beforehand whether

a given distance is to be counted positively or negatively along a

given axis, no ambiguity is incurred, and the sign + or " generally

gives the needful information. For this reason some are apt to think

of + a and "
a as different numbers, instead of the same number

differently directed. Out of this rises the learner's natural complaint

when he meets V" 1 or V~ 5. "There is no such number: you confess
it is imaginary: a proposition involving it has no sense." So murmurs

every scrupulous and wary beginner: and the teacher's reply, "Some-
how

work out useful results by \/- 1," sounds like saying: "Out
we

of this nonsense useful truth is elicited."

The be made is: No ought to desire

2. first reply to one any

number for iJ"\ except the unit itself; the V~ which precedes,
double has the force of symbol only. The next
though a symbol, a

is decisive, the double symbol V- points to direction

reply "
a new

the direction perpendicular to + 1 and

in geometry; namely, "

But to explain this fully,it is better to make a new beginning.

14 WARREN'S RADII.

Suppose that radii issue in many directions from a fixed pointin

a plane,and that distances along them. So are counted long as we
know along which radius we are to count, nothing new is involved,
and of course no difficulty. Suppose one of these radii to be our
ordinary positive axis, another to be called the m radius, and for
distinct ion write the index m under every number to be counted

along it,so that lm is its unit,in length 1 = as estimated absolutely.

Then ambmcm...along the m line,and combine am " bm,
we deal with
and interpret5aw, obm " 6cm, without difficulty or fear,since all are
lengthsto be counted along the same radius. But such a product

as am bn would need careful interpretation.

.
In abn no obscurity is

found, whether the a be linear or numerical. If linear,we proceed

as in interpreting ab3,though space has only three dimensions. If

we put A for the value of ab3, we attain it by the proportion

b* : A, so that A
= is the same in kind as b3. Similarlyif

A'=ab", we are able to count A' along the n radius,whether a be

simply numerical, or when it is linear,by aid of the ratios 1 : a bn: A'. =

But if we proceedthus with am. bn,usingthe proportion

we confound ambn with abn\for 1OT: am is the same ratio as 1 : a.

3. Mr Warren in 1826 laid a basis

logical for this matter by his
ise on ^"l, which I here follow,and wonder
substantially that
it is not found in all elementary works. He
virtuallydistinguishes
between proportionatelengthsand lines. In the former,
proportionate
DIRECTION is not regarded;with the latter, it is essential. Thus if

4 1 ,
/",C, D are proportionate but are
lengths, drawn along our radii,
"
viz. A along the positiveaxis,B along the m radius,C on the n
radius and D on the p radius,we do not pronounce these lines pro-
portional,
unless also their directions justifyit; that is,the p radius
b"- disposedtowards the n radius,as is the m radius to the
axis. This amounts to sayingthat the p line must He on

tin- -aine side of the n line,and at the same

angular distance from it,as the m line pared

com-

with the axis.

positive Then, if OL,
""M, ON, OP I*- tin 4 radii,and LMNP a cular
cir-

arc, we need that the arc PN shall

= arc ML before we admit that the units
v k.

OL, OM, ON, OP are lines.

proportionate
Alt"T this condition of the directions is fulfilled,
we concede that
 WHEN PROPORTIONAL. 15

the proportionate counted

lengths along them are also proportionate
lines.
If the arc NP = arc LM, add MN to both, then arc PM = arc NL.
This enables us to exchange the second and third terms in the portion,
pro-
agreeablyto the process called Alternando. Also the arc

PL = PN+NL=ML+NL. Hence if we count from L, and call

arc LM" m, arc LN =

n, arc LP =
p, the test of necessary direction
is p = m + n.

The simplestcase is,when the four proportionalsbecome three

by the second and coalescing,

as third
if M and N run togetherin Q.
Then if arc ZQ arc QP, we have OL : OQ= OQ : OP.
= If further

arc PR = arc PQ, then OQ : OP OP : OR] and so on. =

4. Apply now this to the case in which the arcs PQ and QL are

both quadrants.Then OQ is the mean portional

pro-
between OL(= 1) and OP=(" 1).
The received symbol for a mean proportion
is V, as in OQ = *JOP .
OL. Here then

OQ =
V(- 1 1) = V- I- This is only a lowing
fol-
.

analogy with the symbol;

out of

though, previously,V expressedthe mean

without
proportionbetween numbers, or perhaps lengths, cognizance
of direction.

Now, our first care must be, to inquirewhether V~ as a symbol

of direction,has the same properties
as when it operates on a pure
number.
the order is indifferent,
First,in combining factors, as ab ba,and =

a (1) a =1 a. We ask, does V~ fulfilthis condition?

=
.
Evidently,
.

a. ^-1 =
^-1. a, each measuring the lengtha, directed along the
OQ.
perpendicular Similarly

We had OQ OL*/-I "/-!. OL. Also

Next, repeat */-" = or

OP=J-I.OQ, because QOP = 90", .'. OP = V~ 1 .(V -

1 "
OL).
But OP = - OL or -
1 .
OL. Evidentlythen V -
1 .
V - 1 is valent
equi-

1. This further the change

justifies of to -*J-\.
to -

-j"^
i.e. in
5. But a new arises in adding unlike quantities,
difficulty
If along radii m and n we have two lengths
connectingthem by +.
and b what meaning can we attach to am + bj. urgently
This
a ,

that the symbol + (plus)receive

needs explanation.It may seem
1C SIGN + NOW FOR TOTAL RESULT.

a now sense. " Now in fact when (a + 6) =

zero, the + does not strictly
mean addition; it reallyexpresses a difference,
not a sum; but not

to embarrass
generalization, we call it a sum, and say that either a or
b is negative. They may mean the very same line OL estimated in

oppositedirections,as OL and LO. If OL mean the line as travelled

from 0 to L, and LO the same as travelled from Z to 0, the ment
state-

OZ + LO =
zero, clearlymeans that the total result of such
travel is nothing;since the travel neutralizes
itself.Thus if,instead
of sayingthat the sum is zero, which givesonly a numerical idea, I
call total 7"esultzero, you will gain a geometrical idea. At this we
must aim, when we deal with lines differing in direction. Evidently,
if,starting from any pointin the outline of a limited surface,a point
travel round the circuit, until it regain its original place,we may
justlysay, the total result of such change is zero; and no one will

suppose it to mean that the lengthof the circuit is zero. So if there

be a triangleABC, we may say, the total result of the travel
AB + BC + CA = 0, if it be understood that each line is to be esti-
mated
in a different direction. Indeed, lengthsof the
suppose the
three sides are c, a, b, then in the equation cm 4- an + bp 0, the =

symbol -f cannot mislead us, though its sense is evidentlyenlarged

from sum to total result.

Again,since AB + BC+CA = 0 and GA = -

AC, when direction
is considered, we have AB + BC " AC =
0, which further justifies
AB + BG = AC. The last is interpretable, Motion "
"
along two
successive sides of a triangleyieldsthe same total result as motion

along the third side."

The word resultant characterizes mechanics, but there seems no

ctinn
ol.j( adopting it in geometry
to also for the total result,as
from the sum.
distinguished

In fact we have unawares made a great step forward; for the

symbol *J" now en;il"lcs us distance in every direction. If
to express

our parallelogrambecome a and AB

rectangle, is the ordinarypositive

before), of AB, BC, CA

the lengtfis are c, a, b, we have

BC=a"J " 1 when direction is estimated, and A C = total result of c

and a V" 1, or AC (an obliqueline) = c + a \/" 1. Since c and a are

AC
independentlengths, may have any direction whatever.

lint again,we must inquirewhether the symbol + ,

thus extended,

can be worked in the received method. First, does it fulfil the

fundamental condition expressed

l"yA + B = B + A ?
Assuming (as
 CONFORMITY OF V~ TO ALGEBRAIC LAW. 17

we must) the doctrine of parallelstraight

lines,
and considering
any
whose
rectangle sides are a and c, we find
that c+a"J" 1 = th e diagonal = a \/ "
1+ c

from the
opposite sides. Next, does it-
fulfilthe condition h(a + B) hA + hBI =

The doctrine of similar triangles

at once

affirms it. Let OM =

x, MP =
y, pendicular
per-
to it ; joinOP, then if 0 M is
the positiveaxis,and y expresses mere

length,we write
OP or OM + MP =
x + y"J-\.

Next, along OH take Om = hx',that is 1 : h = x : Om (whether h is

linear or numerical).Erect mp perpendicular
to Om and meetingOP
in p. Then by similar triangles,
mp = h . y (inlength)and Op=-h. OP.
In this h .
OP "
Op we have supposedh to be numerical.

Also OP is equivalent
to x + "J" 1 .y,

and Op to Om + fnp"J" 1 or hx 4- \f "

1 . hy,
that is, h(x-\-"J" ly) "
hx + V - 1 " hy,

just as if V" 1 were numerical.

If further we change h from a mere number to a positive

length,
it affects every term of the last in the same ratio,and leaves equi-
valence
as before.
If we proved generallythat with any factor h (provided
have
it be counted along the positive axis)the producth (x + V" 1 y) is "

equivalentto hx + ij"l. hy, the same proved,if h be

is virtually

changed into hm, that is,if the numerical h be computed along an

m-axis. For we may transform our hypothesis,by choosingthe m-axis
as positive.If herebyx, y change to x'ty',we obtain a result the
same inform as the previousresult,and x' y'remain quiteas general ,

as were the x, y. Thus we may write

^ "
0 + V- ly) = hmx + J-l. hiny.

After this,we change the obliquehm into a + */-l.b, where a. b

can

are axis. Now

along the positive if the m-axis be perpendicularto

the positive,we may write simply hm kj- 1, where k is along the =

axis. Then V
positive 1 hmy \/ 1 V 1 kyt and since each V 1-

"
= -
.
-
.
-

2
N.
 18 CONFORMITY OF ^ - TO ALGEBRAIC LAW.

denotes revolution of the ky through 90", the "J" 1 ^-1 shews

.

revolution through 180", or is equivalent to the symbol "

Thus
.

= /y7" 1 kx " ky

"lyas?yV" 1 were numerical. Evidently then the same holds

good in multiplying out (h + "J" Ik) into (x + \/" I?/).

THEOREM. If ^1 + " V- 1 =
"? + ^ V- 1, this implies Zwo ties,
equali-
viz. A =
(7 and B =
D. The geometrical proof appeals at once to

the
eye.
If OA be the positive axis, and

the binomials are denoted by OP and OQ,

viz. A + V-15 =
OP and (7 + V- ID = OQ,
we do not account OP = OQ until they
have the same direction, as well as the

same length. This requires Q to coincide \ v

with P. Of course then, if PM, QN are

dropt perpendicular to OA we have Oif = B"/"1, ON= C,

QN=D \/" 1, and as soon as OP= OQ in our hypothesis, P coincides

with Q, therefore also M with jJV. That is A =

(7 and B = D.

The reader will now see the geometrical meaning of the "imaginary
roots" (so called) of a quadratic equation. As a very simple case,

take first

a? -
IGx + 63 = 0, which yields x =
8 " 1.

Here both roots lie along the positive axis. But change G3 to 65,

then a? "
I6x -1-65 = 0, whence x =
8 " V" 1.

In the latter the two roots are equal radii drawn from the origin
at equal angles on opposite sides, radii which terminate where the

coordinate along the axis is 8, and the transverse coordinate is " 1.

 TRACT III.

ON FACTORIALS.

SUPPLEMENT II.

Extension of the Binomial Theorem.

1. THE following appeared in Cauchy's treatise,

elementary as

early, I think, as 1825, but without the new Factorial Notation, which

adds much to its simplicity. Boole writes #(2) for (x 1), #(3) for
x -

a? (a;-1) (2?-2), and xw for x(x- 1) (a?-2) (x- n+ 1), whence

...

a)(n+l] =
x(n) (x "
n). Better still it is, to place the exponent in a
.

half-oval, since a parenthesis ought to be ad libitum. I

propose
which are quite distinctive. Then the Binomial Theorem is

In this notation 1 2 3 4 n or n (n - 1) 2 1 is n^. Guder-

... .
. . . ...

for this has but (n 1)' is less striking to the than

mann n ; "
eye

\n, In "
1 introduced by the late Professor Jarre tt. This exhibits in

the Binomial Theorem its general term, by

1 + xn = + n + + n + .
. ... .

of course x^L" is equivalent to simple x.

The Exponent (of a power) is already distinguished from an

Index. In a Factorial x^ for x (x -

1 ) (x - 2) . . .
(x - r - 1) one may

call (which be the Numero, stating the number of

r must integer) as

factors.

2"2
20 EXPANSION OF (x -f J))^.

2. If 771, n, p be all and

positiveintegers, p = m -f n, then

or, in condensed expansion,

This equationbeing of the (m + n)thor pilidegree of x, and being

true for values of x indefinite in number (thereforein more than

p values),
must be equal term by term for every power of x. Now
when we multiplyany two such series,

(1 + Mvx + Mj? + M3x* + etc.)by (1 + Np + Nj? + etc.),

we have a product of the same form

by the routine of in
multiplication, which

P^M^NJ P^MI+M^ + NJ P^M^

and the law of the indices is so visible,that we get generally

This being true for all series of this form may be

appliedto the
three series (1 +x)p, and
(1 + x)m,(1 + a?)", at once it yieldsto us the
result

n mC n mC n
+ '"" + ''""" "

an equation true for more values of m and n than are counted by the

integerr, therefore it is true also when m and n are arbitraryand

fractional. Write x for m, h for n, and x + h for p, and you have an

extension of the ordinary(x + h)r. For, this latter may be written

of of'1 h x*-* ^ aT3 ^ "

rr-j 'I* r-"2' 2+ r-;r':l4 r'

with exponents replacingthe numeros of the preceding.

NOTE. The reader must carefullyobserve in that which follows, that the upper
i tub's of /', Q, A, /.',('. is not an exponent.
 POWERS IN SERIES OF FACTORIALS. 21

Powers in Series of Factorials.

3. Since x^L" =
x(x"l)=xz "
x,

of which the generallaw is

x^ (x-r) = a*iJ,
conversely #2 = x^L + x.

Again oxL = x (x -
1) (x -
2) = x3 -

But

Evidentlywe can thus in succession obtain x4,xs,...

and generally

xn in series of x^, x1^, x^,... x"^, x.

Since only x^ contains xn, its coefficient must be 1. In general,

with unknown coefficients P, dependent on n, but not involvingx,
we may write

xn =
P"Q x^.
+ Pr1 tfO1 + Pr VxJ + ...
+ P^_2^ i, + Pi^a?. .(a).
.

Here the lower index denotes the placeof the term in the series ;

the sum of upper and lower index =

n, the exponent of xn ; and the

upper index is the same as the numero of its term. We have also

seen that P"= 1, whatever n may be. It remains to calculate P"~r.

Multiplythe left member of (a)by x, and on the rightmultiplythe

successive terms by the of
equivalents x, viz.

(x-ri)+ n, (x -
n + 1) + (n -
1), (x -
n + 2) + (n -
2),etc.,

and apply to each term the formula x^ .(x "

r) = xr^2. Then

But if in (a)we write (n -f 1) for n, we have

and we cannot identifying

be wrong (6)and (c); that is,in equat-
in ing
the coefficients belongingto every particular
numero (r). At the
right hand end PJ,= lPln-l}coefficients of x\ i.e. since when n = 2,
ic2 = oxL + x, so that P\ 1, =

.
pi i Pl I "

-1
"

^a"-1?
_

-t "

. "
2 "
 INVESTIGATION OF TABLE

and P,l,=1,justas Pj
universally = 1. Also in generalwe find

P"- _
^pr i pr-1
"*" ~~
' "*" *
n+l-r n-r " ti+l-r"

otherwise, P^rP^ + PJ'1,

if p = n + 1 "
r.

This enables us to fillin the vacancies of a table,beginningfrom

one horizontal row and one vertical,

each
consistingof units. Each
P is computed from the P above it,multipliedby its upper index

(which is the number of its column) + its companion to the leftin

the same row. Thus to form the second row from

Evidentlythis second row is

1; 1 + 2; 1+2 + 3; 1 + 2 + 3 + 4;...

of which the generalterm P^=\n. (n + 1). Similarlyfrom

is the
second row we form the and the
third,workingfrom leftto right, law
is manifest.

Heuce a table to any extent requiredcan be made, such as is here

presci
 TO CONVERT POWERS INTO FACTORIALS BY P. 23

When solelyto evaluate the coefficients of (a),

this table is used
the indices of P^Pp1, P"~2... warn us that the numbers will be taken

out diagonally.Thus for a? we take out (beginningfrom PJ at the

top on righthand) 1, 15, 65, 90, 31, 1.

It will be observed that the second column is 21 "
1, 22 "
1, 23 " 1

and in generalP* = 2r+1- 1.

Such is the Table of Direct Factorials.

The letter P being almost appropriatedfor the Legendrian

functions,I see an advantage in supersedingP"rby I /^ .

sarily
Neces-

n and r are both integers,

n index of the column, r + 1 index of

the row.

Collect the results for P

and in general P=

which yieldidentically

#C2 + etc.

Factorials in Series of Powers.

4. This is a mere problem of common multiplication,

0^ = "(*-!) (a?-2)...(ff-.n+ l),
and their combinations often recurring,
yet the factors being special
24 INVESTIGATION OF TABLE

the work of one computer may avail for many after him. We may
assume

x^ = Qn0xn Qrlxn~l+ QrV-"

-
- etc. ...
" Ql,x ......
(");

where, as before, obviouslyQj=l. Also dividingby x, and then

making x = 0, you have

Since (x "
n) x^ = xn^y
multiply(a)by x "

n,

Also in (a) write n + 1 for n ;

But (6)and (c)ought to be identical. We have anticipatedthe

remarks that Q"+1 = Qo =
1, and Qln= nQ\_ltinasmuch as

But generally, Q'^ = nQnr~ + Qnr^', r

or, if n -
r =
m, Q?+l =
(m + r) "" + $".

As before,this enables us to continue the table,when the firstrow

and firstcolumn are known. To compare our formula with that of

the first table,we may write it

In fact,the first row is unity as before. The firstcolumn is

1, 1, 1.2, 1.2.3, 1.2.3.4,

when r =
0, Q? = mQ? + Qf1,
also in the former table,when
n =
r, + Prl-1.
lJr1=rPf0
Hence the second row is the same in the new table as in the old.

To compute the third row from the second :

1 6 10 15 21

1.2 11 = 35 = 85 = 175
 TO TURN BACK FACTORIALS INTO POWERS BY Q. 25

Themultiplierr + (p"l) combining upper and lower index of

its Q distinguishes
the Q table from the P table: thus

1.2.3

1.2.3.4

or indeed

Inverse Factorials.

These too are used for

diagonally Thus

Again it seems better to supersede

5n"2+ etc.
by

in Arithmetic driven upon

"
decimals,"
recurring
5. Even we are

and learn that an infinite series may tend to a unique finite limit.
from
Nor can Elementary Algebra fail to recognize,
l-xn
l-x
26 WITH A NEGATIVE NUMERO

that when x is numericallyless than 1, with n indefinitely

increasing,
the series 1 + x + a? + . . .
-f xn tends to the limit
n
l "
x

After this it quickly follows (by Cauchy's process now perhaps

universal),from Binomial Theorem with n positiveinteger, that

(1\"
-J1 + with n has for limit
infinite,

+^=2-7182818...
17

which call e, and that f1 + has for limit

we
-J

which also = ea. On this we need not here dwell : but it will
be
presently assumed. Now let us propound

Factorials with NegativeNumero.

6. Analogy suggeststo define x^ as meaning"; --

: of course
-^
x \x ~H 1}
x^ will be identical with x~l. Then x^ will stand for

and x^" for [x (x + 1) (x+ 2) . . .

(x + n -
I)]"1.

Hence x^J = (x + n)~l #O. .

Now x^ = (x H-l);*;^,

and when x is " 1, we have, in descendingpowers

x^ = x~* " x~3 + x~4 - x~b 4- etc.

Also x^ = (x+3y.x^"

of the two factors here on the right,each can take the form of
a descending in powers of x.
series So then their product,the
"l\iivalent of x^l". By like reasoningwe claim a rightto assume
with coefficients independentof x,

x^ = A;x-n-l-A"1x-n-z+An,x-n-3-etc .......... (a),

and our task is to discover the coefficients when n is given.
 THE TABLE OF P SUFFICES. 27

First make n =
1,

But from the series alreadyobtained for x^ that every

we see

coefficient of the last is 1 ; in generalAlr=I. This

or
givesthe first
column of our table. Also universally
An, obviously 1 ; which
= fixes
the firstrow of the table.

Next, multiplyequation (a)by (x + n),and you get

s~M 2" etc. ...

+ A"x~n~r+ ..

But in (a) we write for 1, which

may n n + gives
x^ =
x-n-A^lx~n~l+... "A?~1x~n~r + ............ (c).
Now (6)and (c)must be identical,
hence

Al=n + Arl=nAl+Ar\
and generally A^nA"^ + Anr~l.
But this is exactlythe law of P in Art. 3, only there we had r, p for
what are here n, r. Now as the first row and first column are here,
as there, unity,and the law of continuation the same, the whole table
will be the same, and we may write P of Art. 3 in place of this A.
Thus we get
x" = x~n -
Pr V"-1 + PJ- V""1 - Vn'3
P3n- + etc. . . .

with the same values of P as before. But in the last equation we no

longer take the P's diagonally,but vertically,

down the column, as

the same upper index (n 1) above every P denotes "

; thus

T7 - etc. . . .

To multiplyby
verify, x + 2. But for convergence x ought to

exceed 2. So
4
f - 90af5 + etc. . . .

and for convergence, x ought to exceed 3. Evidentlyin the series for

,
x ought to exceed (n "

1).

7. Assume now the Inverse Problem, to developx'n in series of

Factorials.
With unknown coefficients B independentof x, we start from

x-n-l = x^+Btx^J + ]%x^J+ ...............

(a).
28 FOR THE INVERSE PROBLEM, Q SUFFICES.

Multiplythe lefthaud by x, and the successive terms on the right

by the equivalentsof x, viz.

(x + n)-n, (a-+ n + l)-(w + l)f (aj+ n + 2) -

(n + 2),etc.
Observe that (x +p) a,'-*"" x(~p) =
"

-...'

But in (a) we may write n for n + 1, which gives

^J+...+Bnr-l.x^r+ ...... (c).
Identify(6)with (c),
.-. J3T=n + "rl;
and generally Bnr (n + = r -
1)B*^ + Bn/1;

the same formula as for Q in Art. 4, only n and r here standingfor

what there was r and p. Also since .Bj=l, the top row is unity,
here as there.

We may further prove that the first column of our J5's is the
same as the firstcolumn of the Q's. For

Multiply by x on the left, also by (x + 1) "

1, (x + 2) -
2,
(x + 3) "
3, for the successive terms on the right; then

Obviouslyx~l x^ =
; and the other terms must annihilate selves,
them-
making ^-1=0; B\ -

2B\ = 0 ; B\ -
SB] = 0 ; etc.,

or B\ = I, ^ =
2^ = 1.2;

JBJ=3B;=1.2.3, and Bln=l .

2 .
3 ... (n- l)n.

Thus B\=Q\, BI =
Q13)etc. exact. Therefore the B table is the
same as the Q table throughout.
Here also we take the to obtain
Q's vertically, x~n in series of
factorials as x~4 = x^ + Ga;^ + 35^ + 225#^ -f etc.

generalx~" x^ + Q^x^
In + Qrlx^~? + + Q;*1.^C^r + etc.;
=
...

but specialinquiryis needed concerningconvergence. Apparently

it converges more rapidlythan
x*+(js -hi)'2+ (a+2)-f+ etc.
 NEW PROBLEM. 29

*"
8. To develops (e'-l)" or {? + + ^ jj+etcj"
+ in powers

of xn. We assume for this the form

may

Mnxn + Mn+lxn+l+...+Mrx' + etc.,

where Mn manifestly = 1, and if r is less than n, Mr = 0. Now by

the Binomial Theorem where n is a positive integer

(e* - 1)- = e- -
? ""-""
+ ^ .
""-"* _
etc. "
? e* +1,

out of which we have to pick up the partial coefficients of xr which

make up Mp observing that

1 '
1 2 13
in which we have to make p successively n, n- 1, n- 2, ...
When

make the only term that here concerns us is When

we p n,
- -

=
.

A^ f/yr "I )** fyF

--
n 1, we get rf- When p = n 2, we have
p = - -

j .
- " -

.
-

and so on. All have r in the denominator. Hence

of which the last is + -

.
I2.

N 1) what if is to then, with the for both,

Let be to (n + n ; r same

terms.
l,....
If I-
pair that with the
Add this to the preceding, coupling every

value of p has (n "

p)r as factor ;
same
30 COMPARE J/r+ .Vr WITH N^.

n + 1 Ti + 1 7i - 1 71 + 1 n - 2
,

Observe that " -- 1 =

n, - ---- 1 = "
-- 1, =

-^ 9 ,
^ ^
"

and so on ;

.-. |r (Jl/r
+ J"TJ
.
= (n + 1)" -

j
(n)'+ *-"^"
(" -
1)' etc., -

or, in shorter notation,

= (TI
+ I)'- ^ (")r
.
+ ^ .
(n -
l)r -

^ .
(n -
2)'+ etc. to (n + 1)terms.

But in Nr change r to r + 1 and you have

r + 1 .
A;., (n + IP
= -
^ (")"'+ **\iC" n
-
V"

l
- "^^o '-
Q" "
(" -
2)r"+ etc- " " "
to (n + 1) terms

n--"~--1(n
(" + 1) {(n1)"- 1(ny
+ + -
1)'

w. n-1 .
w " 2 ,

("-2
1.3.8

-f-1)
(T? . |r(Mr + Nr),from above. Divide by (n + 1) |r-f 1 . ;

Our r always exceeds n. We simplifythe notation somewhat by

assuming as coefficients after dividingby xn,

z!V=i +
^H .
+ _
^ _

4 etc.
.
7i + 2 . . .
n -f p

Multiplyby xn. The generalterm becomes

'

(n + 1) (71+ 2)...(n+p- 1) (71+ j")

 THE TABLE OF P AGAIN SUFFICES. 31

Put n + p r, to identify
this term with
-
our previousMr .
xr ;

.'. M =
*5="
w + l.w-f 2-. 7i + 3 ... (r-1) .r'

Cn+l
that
so
^rr=n+ 2tn-^3~1
r__l r
with one less factor in denomina-
tor:

but when both and in

n r Mr are increased by 1, r -
n undergoes
no change; and N ^
= r~n

" + 2.n + 3... r" l.r.r-hl'

Introduce these values into (a)and multiplyby

Change r -
n to p and n + 1 to n, .'. Cnp Cnp~l
+ nC;^=
; the same

law as for the P's in Art. 3.

Also when n =
lt

But we assumed as equivalent

01* OX cx_ ex
2 "^S.S^S.S.^"1" "t"2.3...(r
+ l)

which identifies (7Jwith 1, for all values of r. Just so Plr = 1 in its

first column. Also evidently"7j l; just as PO=!" = in tne whole
first row. Thus, with firstcolumn and first row identical with those
of P and the same law of continuation,the whole table is the same.

Finallythen we obtain

-iy pj.a?
= i+

in which the increasing

numerators are pulleddown by increasing
denominators.
The generalterm may be written n^ . Pnp xp, or equally
.

(n

[The course of analysishere pursued forestalls that of A" .

Or to

the learner.]
Thus

\\"?
+ l n + 1 .
n + 2 n + 1 .
w + 2 .
n + 3
o2 PROBLEM OF LOG (1 -f ?/).

9. To
investigatelog (1 -f x) in series,with no aid from (1 + x)n
except in the elementarycase of n being a positiveinteger,
and no

aid from the Higher Calculus.

Let y = e* "
1, then x =
log (1 + y). Also

x of xs

by elementaryalgebra.
From the last we see that for minute values of x, as a first
approximation,
y x, and = a? =
y*. As a second, y = x + \x*" = x + J y2,
/. x=y- Jy2. Whatever the series in powers of x, x can be thus
reverted into powers of y, at least within certain limits. Hence we

may assume with unknown coefficients called A,

x or log (1 + y) =
y -
A^f+A3y*-A^4 + etc ....... (a),
which (with the appropriatenumerical values of A^A3A4...)will be
an identical equation.

Consequently,writingy + z for y,

log(l + y + z) =
(y + z) -
Az(y + z)*+ A3 (y + z)3+ etc. ...
(6).

Subtract (a)from (b)developingthe powers

then \og(l+y + z)-\og(l + y)=z-z

+ A, (tyz + Zyz*+ *3) A, (Itfz+ Qy*z*+ 4^2 -
+ z4)+ etc. . . . (c).

But the left hand = log .

" -
or log( 1 -f :="" jwhich again,

by (a),if we write - -- for y, has for equivalent

f^-AJ ^ Y+ ^fVi^- V-etc..

\i yt + \i + y" 3
+ yi

of which the first term alone contains the simplepower of z, and there

of the
its whole coefficient is
hj" J "
This then must be the sum

coefficients of
partial z found in (c).

That is -
= 1 -

2A^y + 3A jf -
4A4y*+ etc.

But = 1 "

y + etc. when
?/2 ?/3-f "
y i" less than
numerically 1.
 PROBLEM OF [log1 + a?]".

Hence 2AZ = 1, 3AZ = 1, = 1, or in generalAn = -

Finallythen, log(1 + y) =
y -
\y^+ Jy8 \y*+ - etc. . . .
while y2" 1.

N.B. This furnishes the means of computing logarithms in

Elementary Algebra,before knowing the Binomial Theorem with

negativeor fractional exponent. If Differentiation or "Derivation,"'

as far as (x) xn, "fi
"f" "
(x) = nxn~l,be admitted, this equation at once

proves that when ""(z) means log (z),""'

(z) = -

: a process which

eases the next Article.

10. PROBLEM. To develop[log1 +x]nin .

a series of powers of x.

That this is when

possible, x2 is " 1, the precedingArticle shows.

We may then assume, with unknown coefficients X? A"

,
A* . . . ing
depend-
,

on n alone, where the upper index is not an exponent,

+ etc (a).
2 ...
w + r

If = uny dz-nun~ldu, and [It is

hardly worth while to disguiseDifferentials by a more elaborate and

tedious Algebra.]
Differentiate both sides of (a),then drop the common factor dx :

n (log1 + XT'* fe
'tc----
hereby, -
_

etc.

Multiplyby 1 + x, and divide by n,

.-. (logf+^r1 = *"* -

*? "
;r
+ ^i -etc.

etc. (6).

N.
34 THE LAW OF Q SUFFICES.

But writing (n "

1) for n
in equation (a) we get

AT1.

Identifying (6) with (c), we

obtain X^w + XJ'1; and generally

XJ = (n + r -
1) x;_j + X?-1 or X?+1 = (n + r) X? + X?;}, the same law as
of

Q in Art. 4. Also evidently X" 1 whatever n be, as Q'j I.

=
may =

Thus the top row

is 1. Again by (a) making n = 1,

XX \lx*

But this is known to be x "

\s? + J#3 "
\x* + etc., whence

Thus the first column also of X

agrees
with that of Q. In short

then, the two tables are

the same. Finally :

(logl+aO* =
aft-
^^ +n
+
i'tt+2""tt 1 + 2 3
+ CiC'}
n n +

with coefficients already known.

The analogy to the series of Art. 8 deserves notice.

iv r\nu I\M u*
(!^--i-l^+Jj^l5 "
I"NJ
..+et.
 TRACT IV.

ON SUPERLINEARS.

SOME Apology seem needful from since Dr Todhunter

may me,

in his volume Higher

on Algebraic Equations has treated the same

subject under the name of "Determinants." Of course

I do not

pretend to add to him, nor indeed he to Mr Spottiswoode, who

carries off all merit on this subject. I read the details with much

admiration as treated by the latter, but found his notation by accents

very dazzling to the eye, in so much as to make it hard to know

by sweeping over half a page,

what was the meaning of the formula

presented to one. Also I found the chief strain in argument and

chief liability to to turn the question, whether this

error, upon or

that resulting term would require a plus or a minus. By reasoning

from linear functions in own though less direct, I was able
my way,

to avoid this danger and lessen fatigue to brain. When I

my changed
ex-

words with then colleague in University College,

my

London, the late Professor De Morgan, on

the question, why this

topic was not admitted into common Algebra, he replied, that it

was too difficult for beginners. I have since thought that he might
not have judged, if of the otherwise treated;
so some arguments were

and in fact I have found, that some

whom I supposed to speak with

authority thought slight change of method easier to learners.

my

At the time I must add, that the rare occasions in

same on very

which I have tried to teach an elementary class of mathematics,

mode which pupil easier, to another

a of reasoning to one was

seemed less teacher ought to have "

two
satisfactory. Perhaps every

strings to his bow."

3"2
 M N
BILINEAR TABLET
P Q
1. Equations of the first degree are
simple,but when called
three or more letters (x,y, z...)are involved,complexityarises with
much danger of error, even when there is no difficulty
of principle.

To solve two equationsof the form alx + bly clt a"jB+ b$ c9 is = =

always the same process. If we could alwayscertainly remember the

solutions

x =
y

and never confound the indices,nor mistake between + and " this
,

alone might have much value. The modern method, due eminently
to the genius of William Spottiswoode, is quiteadapted to Elemen-
tary

Algebra; but its vast range of utilitycannot there be guessed.

First study the denominator D alb2 a,ib1.It arises from the = "

left-hand of the equation (alx+ bly c1, in which the four letters =

stand in square, as Here D is the difference of the two

products formed diagonally,

and the diagonalwhich slopesdownward

from leftto rightis acceptedas the positivediagonal. This is a

cardinal point.Remember it,and you will not go wrong on + and "

Understand then, that in square with vertical sides

M N means MQ-PN.
P Q
Of course then, so does M P which exchangesrows into columns.
,

N Q
But if you change the order of the rows, or the order of the columns,
into P Q or N M you change the result to PN"NQ, i.e.you
M N Q P
change D to " D.
After this is fixed in the mind, it is easy to remember the
common denominator above,viz. at"2 a^, in the form "
Call

* "

it C. Then the numerator of x is obtained from C by changingthe

column ata2 into the column c^, making A =

So B for the numerator of y is found by changing the column

bpt into c^, yieldingB =
a. c.

/\ t\

Finally,
x=
"" , y= ,
without mistake.
V
C/
 LINEAR FUNCTIONS. 37

Observe,if the equationsbe presented in the form

aix + ty + c1 = 0 ; a2x + bjj+ c2 = 0 ;

this is equivalentto changing the signsof c1 and C2, which does not
affect C, but exactly reverses the signs of A and B. Previously
we had
(x : A) =
(1 : C) =
(y : B),

or (x :
y : 1) =
(A : B : C) =

But from the two equationsa:x b$

new + + c^z =
0| you get
V + ty + c/ = OJ

6. c. ci ai
a? : y : z =

C2 a2
in circular order.

We may also present the solution as follows :

M,

Simple equationsare
2. often called Linear, by a geometrical
metaphor. If a quantity w is dependent on x,
so y, z... that how-
ever
the values of these may vary, yet always u = ax + by + cz -K . .

(where a, b, c... are numerical),then u is called a linear functionof

#, y, #... its constituents. [One might have expected u to be called
a dependentor a resultant : but for mysterious reasons of their own,
the French have adopted the strange word function ; and it cannot

now be altered.] It is convenient now to set forth a few properties

of linear functions. We here suppose the function to have no lute
abso-

(constant)
term.

I. To multiplyevery constituent by any number (m), multiplies

the function by that number.

[For,if u = ax + by + cz+..., then

mu = amx + bmy + cmz +....]

II. If two linear functions have the same number of tuents

consti-

and these have the same coefficients [as U=ax

 T.EMMAS OX LINEAR FUNCTIONS.

Ut a.rt-f byi 4- cz{]

=
; you will add the functions if you join the
constituents in pairs[forhere

U+l\ = u + a-J+ b (y + y t)+

(./.- c (s + *,);
whatever the number of constituents].

3. Observing now that a M or aN " bM is a linear function

b N
of X and J7",we see that to multiplya column MX by m multiplies
the function by m. Or ja, mM =
m a M .

Ib, wX b N

The same is true if we multiplya row ; for we may regard a and

M as the variables and b and N as constant ; then

ma, mM = m a M

b, X b N

This leads to the remark, that our function ought to be called

superlinearrather than linear ; for it is open to us to suppose stituents

con-

alternately
constant or variable.

4. Next, by making a column or row binomial, we can times

some-

blend two tablets

superlinear into one. Thus

A x G x\
B y "" y
,

having the second column the same, yield,

(A y -

Ex) + (Cij Dx )
-
=
(A + C )y -

(B + L) x = A +"7,
y

Here the column which was in both, remains as before, but the
other columns are added and make a binomial. Evidentlythe same

process holds, if a row, instead of a column, is the same in both.

when
Conversely, a givencolumn is binomial, we can resolve the
tablet into two tablets ; and when each column is binomial,we can

resolve the tablets into four.

Thus A+x, C .1, C! + z x, C + z

B, D + v y,

by " firstprocess. By a second, each of these tablets becomes two,

as result,
A C .1 x C
11 1" /; H '/ l" H ''
 VANISHING BILINEAR. 39

5. THEOREM. If one column (or row) is identical with the other,

the tablet = zero. For obviously A A by definition,
is AB-BA
,

or zero. B B

COR. Equally the tablet =

zero, if one column (or row) be portional
pro-
to the other. Thus if A : B = C : D, this proportion
yieldsAD = BO; hence

A C or A B meaning AD-BC, vanishes.

,

B D C D

6. This sometimes usefullysimplifies

a tablet. Thus

A"mC, C
B " mD, D

is resolvable by Art. 4 into A C mC, C But the second

B D mD, D
tablet is zero, because its two columns are proportional.Hence the

given tablet simply == A C

B D

COR. Hence an important inference. The value of a tablet is

not changed if to one column (or row) you make addition tional
propor-
to the other column (or row) ; nor again if you subtract instead

of adding.
Further, if in the original equations of Art. 1, the absolute

the solution for and y is simply x 0 and

terms c15 C2 become x =
zero,

0. This is indeed the only general solution, unless also the

y =

denominator vanish ; which

makes x "

^ , y =

^ ;
deciding nothing

to the value of or In that case we can equate the two

as x y.

i a
values of ^; viz. -
= " -
-*
Conversely,this shows afa = aj)1
X

equivalent to = 0. The last is the condition which provides

that the two equations shall be mutually consistent,though x and y

do not vanish. It results from eliminatingx and y, leaving their
value arbitrary.
"tO TRILINEAR TABLETS.

7. Begin from the problem of three simpleequations,

[aBx
+ b5y4- csz =
OJ

jese present three equationsto be fulfilled by only two disposable

/v" /iy
viz.
quantities, - and -

,
if they are all divided by z. The three

equationsare not certainly self-consistent. If #, y, z be eliminated,

an equation of condition will remain. Our first business is to vestigat

in-

it.

From the two firstequationstreated as at the end of Art. 1, but

abandoningcircular order, we have

in which "

In the third givenequation,substitute for xyz the three quantities

now provedproportional
to them, and you get

(1) -fc a, c. 0.

Call it F3 = 0. Then F3 is linear in aj)9cs.This is the condition

that the three equationsshall be compatible. It is seen to result
from eliminating x, y, z. Professor De Morgan wished to call these
tablets Eliminants. Why Gauss entitled them Determinants, no one

explains.Spottiswoodeapparentlyintroduced the excellent notation

at bl cl Then Fn is linear in a8, 63,c3, that is,

(2) F3 =
a2 62 C2 in its third row. Hence also in any row.

as fy"c" ^3 ^s superlinear
of the third order.

for the value written above. The coefficient of aa is obtained in (1)by

the
obliterating constituents as last written that are in the same

column or row as aa, thus reducingF8 to

then we see the t.-ihk-tl"v \\liirli,/.. rnust !)""multiplied.

 EXPANSION OF A TRILINEAR. 41

The is used with 63and

same process c3, producing

...
c.

and

Finallythe signsof the terms of V3 are alternate

justas in the bilinear aj)2 "

Evidently F3 is formed of six terms, three positive

and three
negative; each term having three factors,but in no term is any
factor combined with another of its own row or its own column.

In the term aj)2c3

(the diagonal slopingdown
c0 from

leftto right)is positiveas before. Thus when F3 is given in tracted

con-

form, we can expand it into three apparent binomials.

8. In the three given equations you may exchange the position

of x and y ; then by eliminatingy, x, z you obtain

= 0;
an ca

but you cannot infer that U3 =

F3. In fact,to exchangethe a column
with the b column reverses the signof Thus it changesF3 into.

62 ,

6,0,

That is,to exchange the first and just reverses

second columns the

sign of F3. The same effect follows from exchangingany two tiguous
con-

columns or rows. Thus generally

Again,
"1 EXCHANGE OF COLUMNS OR ROWS.

Observe, that if three binomials of V3 are expressedby

maa + 4- pal
H-cra
the multipliers
mt n, p contain nothing of the column a1? a2, a3,
therefore V3 is linear in these three constituents. Evidentlyit is
linear in regardto any column ; as we before saw, as to any row.

9. THEOREM. Further, 7 say, To exchange rows and columns

does not alter the value of
V3. In proof,multiplythe three equations
givenin Art. 7, by disposable numbers m, n, p, and so assume m, n, p

that when the three productsare added togetherthe coefficients of y

and z may vanish. There will remain (ma^ + ?ia2-f pa3)# 0, and as =

we do not admit #=0, we have three equationsconnecting m, n,p\ viz.

a,m -I-a2?i+ a 3p = 01

Ji7H+ "g?l_l_ =o'

])^p
cjn + c.2n-f C3p =
OJ
and that these may be compatible,we need

k o,
Cl C2 "

by eliminatingm, n, p.

Here nothing but V3 with rows

S3 is changed to columns and
columns to rows, retainingthe same positivediagonalaj)2c3.Every
learner will easily find by developingV3 and S3 that theyare identical:
but there is an advantage here in general argument applicableto
higher orders. $3 0 and V3 0, being each a condition of compati-
= =
bility
of the previousequations, must contain the same relation of the
constituents. S3 is a linear function of its column aj)^ ; so is V3 a
linear function of its row ajb^. But $3 0, Fa 0 being derivable = =

one from the other, there is no possiblerelation but $3 /z V3 in =

which n must be free from a^Cj. But the same arguments will prove
that p is free from a262c2,also from a3b3c3.Therefore /-t is wholly
numerical. Make 6, 0, ct 0 and this will not affect ft. But this
= =

makes
and S =
M.

But that the two minor tablets are identical

implied in their was

definition. Hence, on this assumption for bland c,, we find F3 A\ =

or /Lt = l. This then is the umuersal value of /^ ur F;! #, in all =

"". i-;.I).
 ON A SUPPLEMENT VANISHING. 43

10. In the
developed value of F3 (Art. 7) if 63 0 and
=
c3 = 0,
you get simply F3 as \ 64 c, which does not contain
=
av or az. These

two constituents are made wholly inefficient

by the vanishing of b3
and
c3, and may be changed to zero. Thus

So *"**

0 0 a 0 0 ""

When
major square is divided into two
a minor squares and two

complemental rectangles,the vanishing of one rectangle obliterates

the other, and the greatest tablet has only the two for its
squares
factors.

11. Since F3 is a linear function of every row and of every

column, we can
argue as in the Second Order or Bilinears,that to

multiply any column, or row, by multiplies F3 by m ; and if

any m,

a column or row consist of Binomials, we resolve the tablet into two.

Converselytwo F3s which differ only in a singlerow or singlecolumn

can be joined into one universally,

A+m, A' A0 A A' A, mA' AQ

B + n, B' B0 B Bf B0 n B' B0
C G C' Cn P C' CQ

12. Further, if two contiguous columns (or rows) be identical,

For exchange them changes F3 to F3. Yet the
the F3 = zero. to -

exchange leaves F3 exactly what it was before. Hence F8 = "

F8.
This can only be when F3 0. =

It follows that if a column (or row) be proportional

to a contiguous
column (or row), the tablet is zero. For instance, if av bv ct are portional
pro-

to a,2,62,
c2, we may assume ax =
wa2, 6t=
mb.2,cl =
??ic2,then

rac

This gives a2 62 c,
44 TO SOLVE FOR X, Z IN SIMPLE EQUATIONS.
y,

and the last tablet is zero, because its first and second row are

identical.

What has been said in this Article of two contiguous rows or

columns is evidentlytrue of any two rows or any two columns, since

exchange of contiguous rows (or columns) does but multiply the

tablet by - 1.

13. The same argument as before in the Second Order now

proves that a tablet V3 is not changed in value, if any row (or

column) receives increase or decrease proportionalto some other

row (or column). Thus we have shown in V3 the same propertiesas

those enunciated in F2.

14. NEW PROBLEM : to solve for x, y, z in the three equations

ai" + b + cz + d = 0,

2
= 0,

Assume d, =
Ajc, d2 = A^x, d3 = A3x. Then

Eliminate the x, y, z here visible;then

= 0.

The last tablet may be resolved into two, namely:

a, I,c,
0.

Multiplythe former of these by x, and, as an equivalence,

multiply
each constituent of the first column of the latter by xt

a
 THE PROBLEM SOLVED. 45

In the first column of the last restore for Ajc, Ajc, Aax their
values dv d2,ds. Then if

you have Dx + = 0,

which solves for x. similar steps

By perfectly

a, dl Cl at b,dl
0; = 0.

These are easy to remember: each suggests the other, by entire

symmetry. The method succeeds in Higher Orders.

COR. If we make x =
y = -
z = -

-, , ,

Ifi = 0.

Then from

= a -Dz =

you obtain the proportion

" : * : f

"3 C3 ds

15. PROBLEM. To eliminate x from the two equations

(px2 + qx +r* =0)

where P} Q, R, p, q, r may involve y or other quantities.First,put
x2 = X, then we have

pX + qx + r = 0

Eliminate X\ solve for x;

i.-e.

= 0 .(1).
 TWO PROBLEMS OF KL1MINATION.

Again,writingin the original

'(Pa;
+ Q) x + R =
0]
(px 4- q) x 4- r = Oj
eliminate .r, as if P,r + Q and px + q were ordinarycoefficients. Then

Px + Q, R

px + ,
r

which expanded,gives
PxR + QR = 0, or PR x + QR
r r r r
px q p "

Eliminate x between (1) and (2),which gives

2_
that is, = 0,

the result required.

16. PROBLEM. To eliminate x from the two equations,

+ Sbx*
few;3 + Bex + d = 0,
\ax*+2bx
+c = 0.

First,multiplythe last by x and subtract ;

Compare the two last with the two equationsof Art. 15,

/. P =
a, Q = 2", R =
c, p = b, q = 2c, r = d.

Hence the elimination of x yields

a 2b 2bc a c 2=0, or ab be = \ a c

b 2c 2c d b d b c c d b d

This is the condition of two equalroots in the given cubic.

17. By conductingelimination from three givensimpleequations

by second attain relations which were not easy to
a process, we
Assume
-"niticipate. the three of Art.
"".jiiations 7. Kliminatc //
 AN IDENTICAL EQ. BETWEEN 9 CONSTITUENTS, ALL ARBITRARY. 47

twice,(1)from the two first;(2) from the second and third. Hence
we get

x =

from which, by eliminating obtain U=

#, we 0, if we define U by the
equation

aa ba

But, otherwise eliminated, we find,as the condition of compatibility

F3 = 0.

Therefore "7=0 contains the same relation of the constituents as

F3 = 0. By inspection,
we see that U, equallywith Fs, is linear in
a,, 6j,cr Since then ?7 and F vanish together,we have necessarily
U=fj,.V3,in which yu,does not involve alt b1nor cx. To determine /i,
suppose at 1, 6X 0, Cj 0
= = =
,

c2 = bn ba

and 100 100 ",',

O^a
so that in this particular
case U=
62F3,or /-t 52. This then is the =

value of //, for all values of altbiycltor universally

U=bz. F3, while
all the nine constituents are arbitrary. Q.E.D.

18. To remember this important equation, write the square

trilinear largerand mark out its minor square. The factor b2is in

the centre.

'
'2 "
3-

By interchangingrows or columns without alteringthe value of

F3, fresh relations are obtained. Indeed no one constituent can

claim the central placefor itself.

 QUADRILINEAR TABLET.

Fourth Order.

19. To eliminate x, y, z, u from four given equationseach of the

form a^x 4- b$ + ctz 4 dtu 0, we have = now much facilityfrom the
Cor. to Art. 14. First, eliminate from the three first equations
and get the proportionsof x, y, z, u. Next, insert these tionate
propor-
values in the fourth equation whereby you eliminate
entirely
all the four,and obtain an equationF4 = 0 ; if F4 stand for

6, c, dt

developedform of F4 can always be recovered (by attention

This
to the simple rule given for developingF3) from the conciser or
form
**ttdeveloped

Evidentlyin the definition F4 is a linear function of a4 64c4 d^.

So then it must be of any other row, the order of the given
equationsbeing arbitrary. Also the firstterm of F4 as defined is
free from a/y^; each of the others is linear in a^a^ Therefore

F4 in square is linear as to its first column ; so then must it be

as to every column.

Being thus linear,if one column (or row) of such quadrilinear

tablet is binomial, it may be splitinto two tablets by the same

process as in the third order. Likewise to multiply any column

(orrow) by m the
multiplies whole tablet by m.

To exchange first and second column, exchanges the first and

second term of F4, but reverses signs. It reverses their
the sign
of the third term, also of the fourth. Thus to exchange the first and
second column reverses the sign of F4. Evidently then the same
must happen by exchanging any two contiguous columns. The

same argument appliesconcerningany two contiguousrows.

The reasoning of Art. 12 concerning F3 now appliesto F4 ,

showing that the tablet vanishes, if one column (or row) be pro-
 ITS PROPERTIES THE SAME. 49

portionalto another column (or row). From this it further follows

(as concerningF3 in Art. 13, and concerningF2 in Art. 6), that.

F4 is not changed in value, if any row (or column) receive iner
or decrease in proportionto some other row (orcolumn).

20. That F4 is not altered by exchangingrows and columns,

is generallyproved by elaborate inspectionof the separate terms
when F4 is resolved into 24 elements each of the form a1b^c^d4,
no

two factors of the same row or column, and showing that the + or -

of the term is never altered. It is,no doubt, a perfectdemonstration,

and more elementary than mine ; but I find the less obvious gument
ar-

of Art. 9 the easier for all the Multiply

higher orders.
the given equationsby 'inl^lplqi and assign to these multipliers
the condition that from the sum of the equationsthus multiplied

y, z and u shall disappear.

There will remain (mal+ na^-}-pa9 + qaj x Q. But = our thesis

hypo-
forbids # = 0, hence we have four equationsto determine

tnnpq, viz.

+ aji + a3p + a^q = 0,

bjn + b2n+ b3p 4- b4q = 0,

Cjiw.+ czn + c3p + c4"/ = 0,

djm,+ dzn+ d3p + d4q = 0.

When eliminate the result,which we may call S4 = 0,

we mnpq
exchange of rows with in the
shows $4 differingfrom F4 only
columns. Each of them is linear in ap^^. Each involves the

same relations between the constituents. The equation$4 0 must =

be deducible from F4 0. The only possiblerelation,making S4

=

and F4 vanish together, has the form $4 yu,F4, in which /z

is inde-
pendent =

of "VWV But symmetry proves p equallyindependent

therefore is numerical. To find it,we
of every other column; p

make the constituents on positivediagonalall

the 1, mM =

may
all the other constituents vanish. Then both $4 and F4 at6sc3rf41 = -
.

Universally then, /a = l, or S4=F4. Therefore F4 is not all

with columns.
by exchanging rows

holds, however high the order of the

Evidently this argument
law.
Tablet if the successive definitions follow the same

*
50 SOLUTION OF FOUR SIMPLE EQUATIONS.

21. We can now solve when four given simple equationscon-

necting

xyzu have on the left side absolute e1e2e3e4,with

terms zero

(asbefore)on the right. We proceed as in Art. 14. Let

el A jflj, A4cc.
=
e2 =
9x, e4 =

Then our equationbecomes

u = 0

(a4+ A4) x + b4y+ C4z + d+u = 0

Eliminate x, y, z, u, then

blCl

.
"4 C4

The first column being binomial, we can resolve this tablet into
two. Then multiplythe left tablet by x, and the first column of
the second also by x ; whence

A 1

Cj
0.

c*

In the second tablet we now replaceits column by its value

0te,e,04.

Thus we have solved for x. By perfectlysimilar steps we

solve for yt for zy and for u.

if
Finally,

M= """
N= P =
, ,

i"4
c4 dt e4 a4 c4 c?404

^i c.
Q= R =
,

a,

we have x : y : z : u : 1 = M : " N \ P : "

Q : R.
 COMPLEX DELATION LT =
(6.(C3bnc) V
-

51

22. Take four

our equationsas in Art. 19. From the three first
and also from the three last eliminate both y and z ; whence

b c

; and

To eliminate x from these two, have

we

a, 6. "

U=0

which we may remember by

Thus f/"40 = and F4 = 0 express the same condition of the stituents

con-

for the
reconciling four equations.

Inspectionshows that like F4, is

"7"4, linear in ajb^d^ Put
and //. will
C/4 yLtF4, be free from these four. Assume then l,
=
al =

*)
= c =
d = 0. and it will not affect u,. But it makes
111^

and U4 = 1

that is,

whence generally
f/"4 .
F.. This could not have been seen.
fore-

By varying the order of the elements, we have other results.

4"2
 GENERALIZATION OF THESE PROPERTIES

Observe that is the square in the centre of V4. The tri-

linears in ?7are squares cut from the four corners of F4; those of the

first term in U4 being from the positive diagonal.

23. By aid of Art. 21 we readilyproceed to the Fifth Order,

with the same law of continuous formation; whence in every Order

we have these same properties.

(1) To exchange rows and columns does not affect the value
of T..

(2) Fn is a linear function of any one row, or any one column.

(3) If a row or column be binomial, the Fn may be splitinto

(4) To multiplya row or column by m, multiplies

Fn by m.

(5) To exchange any row (or column) with a contiguousrow (or

column) changes Fn to "

Fre.

(6) If one row (or column) is identical with or proportional

to

another row (orcolumn), the Fn = zero.

(7) Fn is not altered in value when a row (orcolumn) receives

increase or decrease proportionedto another row or column.

(8) If Fn be divided along the diagonal,so as to fall into four

parts,two squares and two rectangularcomplements, the vanishing

of one complement makes the other wholly ineffective.

24. To prove the last it suffices

universally, to prove it for the
fifth order.

Call the two squares P, S and the complements Q, R. Then if

one of the
complements,as Q, hav"- :"11 its constituents zero, I say,
R is ineffective, and F=P S, justas if R also had all its constituents
.

zero. For every term of FB when fullyexpanded, has the form

ajbncpdqer,where mu/n/r an-, taken from I, '2, :",I, 5 and no two are

the same. Hence a

ajb4c4 nd (the constituents
"66ftcfi of R) are neces-
snrilymultipliedby one or nih"T ,,f il 'l.cjlf.^in Q, mid
 FOR ALL ORDERS.

all the products vanish. Consequently, if P is of the third order and

S of the second, the F in question is equivalent to

0 0

"2 00
a2 c2

a.
6. c.
00

000

000

and might arise from two equations separately, yielding P = 0, and

If Spottiswoode did not plant the first of this valuable

germ very

he first investigated the laws and exhibited its vast

theory, power.
 TRACT V.

INTRODUCTION TO TABLES I. AND II.

To these four Elementary Tracts I have added two Numerical

Tables, solely because their compilation and verification is elemen-

tary.

Table I. gives values of A~" to 20 decimal places. Here A means

the series 2, 3, 4, to 60, and the odd numbers from 61 to 77;

... up

and n means 1, 2, 3, continued until A~n is about to vanish. To

...

verify, use the formula

The reader may convince himself how searching is this test, by

applying it, for instance, to A"11 when A =
37 or when A =
71. Only
in the case of 2~n and 3~n, where the Tablets give only odd values of

n, we must apply the formula

~2m~1
A "
A
(A~l + A~3 + A~5 + +A-*m~l) = ~

... .
,
A. " 1

Table II. has values of xn with 12 decimal places, where x means

"02, '03, '04 up to '50 and n is continued from 1, 2, 3,... until xn is

insignificant. The formula of verification is (with ra any integer less

than r)
(at* + xm+l + aT" + + O (1 -
x) = x" -
af* '.
. . . .

I compiled this table while working at Spence's integral

but it has much wider use.

One who is sagely incredulous of printed tables can verity

fur himself any tablet which he is disposed to use with much greater
ease than he could
compose the tablet. Thus, too, he would dct" "" -t

jury error fn"m muCOpying or misprinting, against which I can leasl

give a L,ri"
TABLE I. TWENTY DECIMALS. 55
TAl'.I.K I. TWENTY DECIMALS.
TABLE I. TWENTY DECIMALS.
TABLE I. TWENTY DECIMALS.

*
For i6~" look to 4
TABLE I. TWENTY DECIMALS. 59
60 TABLE I. TWENTY DECIMALS.

For 25
"
look to 5 2".
TABLE T. TWENTY DECIMALS. Gl
62 TAl'.T.K I. TWENTY DECIMALS.

35-" 37"

i '02857 14285 71428 57142 i "02702 70270 27027 02702

2 81 63265 30612 24489 2 73 04601 89919 64938
3 2 33236 i5l6o 34985 3 i 97421 67295 12566
4 6663 89004 58142 4 5335 72089 05745
190 39685 84518 144 20867 27182
6 5 43991 02415 6 3 89753 16951
7 J5542 60069 7 10533 86945
8 444 0743" 8 284 69917
9 12 68783 9 7 69457
10 36251 10 20796
ii 1036 ii 562
12 29 12

13

For 36"" look to 6'

TABLE I. TWENTY DECIMALS.
(54 TAl'.LE I. TWENTY DECIMALS.
TABLE I. TWENTY DECIMALS.

"
For 49 look to 7'
66 TABLE I. TWENTY DECIMALS.
 TABLE I. TWENTY DECIMALS. 67

63-

i "01587 30158 73015 87301 i "01538 46153 84615 38461

2 25 19526 32905 01385 2 23 66863 9"532 54438
3 39992 48141 34942 3 364J3 29085 11607
4 634 80129 22777 4 56" 20447 46333
10 07621 09885 8 61853 03789
6 15993 98569 6 27751
253 87279 7 203 98888
8 4 02973 8 3 J3829
9 6396 9 4828
10 101 10 75

5"2
68 T.vr.LE I. TWEXTY DECIMALS.
 TABLE II. TWELVE DECIMALS. 69

TABLE II. Powers of -02,-03,-04, ...

up to -50,useful to compute
A3a? + "c., when x does not exceed J.

(Twelve Decimals.)
TABLE II. TWELVE DECIMALS.
TABLE II. TWELVE DECIMALS.
1-1 TABLE II. TWKLVE DECIMALS.
TABLE II. TWELVE DECIMALS. 73
74 TABLE 11. TWELVE DECIMALS.
 .

TABLE II. TWELVE DECIMALS. 75

76 TABLE II. TWELVE DECIMALS.
TABLE IT. TWELVE DECIMALS. 77