Rice (The Theory and Practice of Interpolation) (1899)

THE
THEOKY AND PRACTICE

OF
INTEEPOL ATION:
INCLUDING
MECHANICAL QUADRATURE, AND OTHER IMPORTANT PROBLEMS

CONCERNED WITH THE TABULAR VALUES OF FUNCTIONS.
WITH THE REQUISITE TABLES.
BY
HERBERT
L.
RICE, M.
S.,
ASSISTANT IN THE OFFICE OF THE AMERICAN EPHEMERIS,

AND PROFESSOR OF ASTRONOMY IN THE COBCORAN SCIENTIFIC SCHOOL, WASHINGTON, D.C.
LYNN, MASS.
THK
NICIIOI, >
I'UESS
TIIOS.
1899.
P.
NICHOLS.
COPYRIGHT, 1899, BY
L. RICE,
HERBERT
WASHINGTON, B.C.
6?A
PREFACE.
IN preparing the following
the
treatise
originality, either of subject matter
author has attempted no marked
method.
or
been written of Interpolation, Quadratures,
Indeed, sufficient has hitherto

to
etc.,
firmly
Yet of the numerous contributions
such an endeavor.
one from
dissuade
to these allied subjects,
there has appeared thus far no distinct treatise covering the entire ground.
a consequence the author has
repeatedly felt the need of a work which would
exclusive of other matter
give
sion
of
that
all
useful
is
As
a simple, practical, yet comprehensive discus-
concerning Differences, Interpolation, Tabular Differ-
and Mechanical Quadrature;

a work, moreover, which would include
tables appertaining to the text which are required by a practical computer.
entiation
all
To supply
the want thus conceived, the author offers the present volume.
But while viewing the matter

work
no mere compilation.
in this practical sense, the writer
regards his
and developments are

while the same
original, so far as he is concerned, and possibly altogether new
remark applies to a few of the minor results. In fact, if adverse criticism be
forthcoming, it will probably result largely from the somewhat unusual or indias
Many
of
the
processes
vidual methods which in
many
On
the customary forms of analysis.

the extent
gestions
of
indebtedness
his
and he
have been employed in preference to

the other hand the author realizes fully
instances
to
previous writers for valuable
ideas
KNCKE, LOOMIS, NEWCOMB, and SAWITSCH as most valuable sources

tion, to which frequent reference has been made.
Concerning the
includes the
several
of
foregoing
the
earliest
STIRLING, MAYKU,
of
the
bibliographical
general
names),
writers
it
list
is
such
as
of
this
state
of informa-
volume (which
that
references
to
WALLIS, MOUTON, COTES,
have purposely been omitted because
their works.
that any
to
BHIGGS,
As
present century, however, the author believes that
have been included, and trusts
close
but proper
WALMESLEY, LALANDE
inaccessibility of
the
at
and sug-
works of BOOLE, CHAUVENET,
desires especially to mention the
regards
all
omissions
detected will be regarded merely as oversights.
the writings
of
the
contributions of importance
of
consequence hereafter
PREFACE.
IV
Special care has been given to
with the
preparation and printing
the
absolute accuracy.
of
At
of
the
tallies.
a considerable cost of labor,
hope
securing
and by wholly independent methods, the computations were all made in duplicate; and in every case the tabular values are true to the nearest unit of the
last
place.
published
Though
for
the
a few of these tables have appeared before, several are here
first
time,
and
it
is
hoped
they
will
prove
useful
to
the
computer.
In conclusion, the author desires to express
ciation
to
POKTER,
of the
Feelings
skill
thanks and appreOffice, and to Mr. M. E.
cordial
Naval Observatory, for much valuable service and many useful
suggestions received during the
word
his
Mr. E. C. RUEBSAM, of the Nautical Almanac
of gratitude further
various
inspire
phases
of
preparation of this treatise.
simple justice even demands
a special
commendation of the publishers, whose uniform courtesy, accuracy and

have done much to enhance the general value of the work.
in
H. L. R.
WASHINGTON,
D.C., December, 1899.
CONTENTS.
CHAPTER
I.
OF DIFFERENCES.
Section.
.........
..........
1.
General remarks concerning tabular functions and the construction of

mathematical tables,
2.
Fundamental
definitions
and differences,
Method
and notation.
4.
N functions
5.
Effect of inverting a given series.
6.
Differences of two combined series.
7.
Irregularities in the differences of functions
8.
Detection of accidental errors by differences,
9.
10.
Numerical examples
Numerical examples
11.
General properties of differences.
yield
Theorem
n nth differences.
Theorem
Theorem
III,
which only one function requires correction,

involving two or three erroneous functions,
in
Expression of
n)
//i
in
equation
(4),
Determination of the
13.
Remarkable formal relation between the expressions for
14.
The wth
15.
Converse of the foregoing proposition.
16. 17.
Convergency of
differences of
are constant.
coefficients
B, C, D,
2
4
5
5
(5
which are not mathemati-
12.
etc.,
7
9
11
13
terms of the
...
in equation (4),
and d'u
//J'
15
18
21
........
....
any rational integral function of the Mth degree
Theorem V,
differences.
Theorem VI,
Magnitude of tabular
and rhnrNumerical illus-
............
terms of tabular differences,
18.
Expression of
19.
Change of the argument interval from <a to m<a

nitude of the successive differences,
u>F'(t), <a"F"(t), etc., in
effect
.
Theorem VII,
Practical result of the foregoing investigation.
21.
Numerical example
22.
Expression of any difference in terms of tabular functions,
23.
Expression of any tabular function in terms of
..........
,
d'
28
.30
....
reduction of tabular interval,
25
upon the mag-
20.
23
24
interral
aeter of function the principal elements involved.
Examples,
I,
....
....
II,
Theorem IV,
th and higher derivatives of /*(<);
trations,
...
...........
.....
cally exact,
General schedule of functions
of checking the numerical accuracy of differences.
Page.
A'J,
J'J', etc.,
34
34
35
30
.38
CONTENTS.
vi
CHAPTER
II.
OF INTERPOLATION.
.........
Section.
24.
Statement of the problem,
25.
Rigorous proof of
differences
26.
some particular order are constant,

Second demonstration of NEWTON'S Formula, restricted as
27.
Formula for computing the interval
28.
Example
Backward
interpolation
end of a
80.
31.
n,
in
of
25,
41
43
43
..........
of interpolation by
being constant,
29.
40
......
NKWTON'S Formula, assuming that
series.
NEWTON'S Formula, the fourth
differences
44
......
by NEWTON'S Formula.
Numerical example,
Interpolation near the
44
General investigation proving that NEWTON'S Formula is sensibly accurate as applied to series whose differences practically
though not
...
...
vanish beyond the 4th or 5th order,

absolutely
Numerical example illustrating the foregoing discussion,
in the use of
NKWTON'S Formula,
32, 33.
Practical examples
34.
Transformations of NEWTON'S Formula.
46
57
....
notation of differences.
Modification of the foregoing

STIRLING'S Formula. Schedule of differ-
ences referring to same.
Example,
61
......
...
....
......
...(>"
62
35.
Backward
36.
Further example in the use of STIRLING'S Formula,
65
37.
The
66
38.
Derivation of BESSEL'S Formula.
39.
interpolation by STIRLING'S Formula.
algebraic mean.
Practical precepts,
Backward
41.
Property of BESSEL'S Coefficients,
42.
of the
Comparison
relative
BESSEL'S Formulae,
...
...
68
........
advantages of
69
NEWTON'S, STIRLING'S, and

.
69
71
..........
43.
Magnitude
Simple interpolation.
second differences,
44.
Interpolation by means of a
Backward interpolation by
45.
Numerical application,
Second example of interpolation by BESSEL'S Formula,

Example,
interpolation by BESSEL'S Formula.
40.
65
Example,
of
error
arising
corrected first difference.
means
of
corrected
from
neglect
Example,
of
72
73
difference.
first
74
Examples,
46. 47.
Correction
of
Example,
48.
tabular
Systematic interpolation of
terval,
49.
Interpolation
50.
Precepts for
61.
............
erroneous
to halves.
functions
by
interpolation.
Reduction of a given tabular
series.
.
direct
Practical rule,
in-
......
....
Schedule showing
systematic interpolation to halves.
arrangement of quantities. Numerical example,
Derivation of general formulae for reducing the tabular interval from
w to m<a, m being the reciprocal of a positive odd integer,
.
76
78
80
81
83
CONTENTS.
Vll
.....
.....
Section.
52.
thirds.
Systematic interpolation
to
53.
Systematic interpolation
to fifths.
54.
On
Example,
Example,
...........
............
the best order of performing
88
89
successive interpolations to halves,
thirds, etc.,
55.
Page.
91
Interpolation, with a constant interval n, of an entire series of functions.
Example,
Examples,
CHAPTEK
91
94
III.
DERIVATIVES OF TABULAR FUNCTIONS.
56.
............
Concerning the close relation between differences and differential

efficients,
.....
....
57.
Practical applications of formulae resulting from this relation.

tance of tabular derivatives in Astronomy,
58.
Derivation of the required formulae in general terms,
59.
60.
Formulae for computing derivatives at or near the beginning of a

Examples,
Formulae applicable at or near the end of a series. Examples,
61.
Derivatives from
62.
Examples,
Derivatives from BESSEL'S Formula.
STIRLING'S
64.
Application
constant.
65.
Examples,
Impor98
series.
101
.
105
F'(t+ $<a).
'
derivatives.
first
Practical rule for this case.
Example,
differences
115
121
are nearly
....
............
in
97
109
preceding method when second
Regarding a choice of formulae
97
compiiting F'(f).
Simple expression for
Interpolation by means of tabular

of
Rule for
Formula.
Applications and examples,

63.
co-
Examples,
any given
CHAPTER
case,
124
127
128
IV.
OF MECHANICAL QUADRATURE.
130
<i~.
Statement of the problem. Important applications of the method,

Derivation of formulae for single integration from NEWTON'S Formula.
The auxiliary series 'F. Schedule of functions and differences,
68.
Numerical applications illustrating two of the foregoing formulae,
137
69.
Precepts for computing a definite integral when either or both limits

are other than tabular values of the argument T.
Necessity of
66.
interpolation in this case,
131
.138
CONTENTS.
Vlll
Section.
70, 71.
Page.
Transformation and extension of the fundamental relations of

67,
such that integrals whose limits are non-tubular values of T are
........
expressed directly in terms of Interpolated values of
Formulae and examples,
etc.
72.
'F,
F,
1
,
Generalization of preceding formulae to include integrals of
.146
limits.
any
151
Example,
74.
75.
.........
Formulae for single integration from BESSEL'S Formula.

any limits. Examples,
Double integration. The conditions involved,
Extension to
Derivation of formulae for double integration from XKWTOX'S Formula.

Introduction of the series "F. Schedule of functions and differ-
77.
Value of the
General formulae and relations,
ences.
nition
formulae for double
all possible cases.
Examples,
Derivation of formulae for double integration from STIRLING'S and
BESSEL'S Formulae.
Schedule referring to same.
Precepts and
integration covering
examples,
79.
of
Collection
quantity H^.
.100
Introduction and defi-
first integral at the lower limit.
of the
Change
Examples,
............
............
//,
CHAPTER
166
173
Y, due to an arbitrary change
in Value of the double integral
in the constant
153
160
76.
78.
140
Formulae for single integration as derived from STIBLING'S Formula.

Schedule of functions and differences. Examples,
.
73.
J"
d",
188
189
V.
MISCELLANEOUS PROBLEMS AXD APPLICATIONS.

80.
Introductory statement,
81.
Problem
82. 83.
Problem
find the
sum
Application to
S=
To
I.
gers.
II.
Given the
signed value of
solutions.
84.
Problem
III.
unknown
85.
Problem IV.
maximum
86.
series
solve
quantity.
To
or
the &th powers
+3 +
4
F_ 2 F_ lt
,
..
Concluding remarks,
Examples,
and an
the corresponding interval
of the
n.
function.
argument
Example,
191
as-
Two
192
195
corresponding to a
Problem V.
Given a series of values, F^, F_lt F
some function F(T) analytically unknown; to
mate algebraic expression for F(T~). Examples,
Geometrical problem,
etc.,
191
first r inte-
any numerical equation containing but one

Example,
minimum
88.
the
find the value
87.
....
of
+ r*,
F Flt F
......
.......
.....
to find
Examples,
To
of
l 4 -H2 4
Flt F
find
.
196
etc., of
an approxi.
..........
198
200
202
203
CONTENTS.
IX
APPENDIX.
ON THE SYMBOLIC METHOD OF DEVELOPMENT.
..........
Section.
89.
Introductory remarks,
90.
Definition and operation of the symbols
91.
Definition and operation of D,
D2
A,
A A
2
8
,
etc.,
92. 93. Proof that
94.
.....
obey, in
Eemark
Demonstration of Theorem
97.
Fundamental relation between

Expression of
98.
A,
A A
etc.,
and
Expression of D,
D 2 D8
,
D,
terms of ascending powers of
etc., in
100.
Reduction of the tabular interval
101.
Effect of the operator
102.
Definition of the symbol
103.
Derivation of NEWTON'S Formula for backward interpolation,
Expression of
o>.
terms of ascending powers of A,
+ A. NEWTON'S
terms of ascending powers of
in
Demonstration of Theorem V,
99.
......
III,
"d,
d*,
Formula of
V.
of operation
205
general, the
concerning results established in the preceding sections,
205
206
the foregoing symbols

fundamental laws of algebraic combination,
and D,
Consideration of negative powers of
95.
D", etc.,
of operation
96.
Pace.
<?
8
,
A,
interpolation,
Its relation to
208
209
209
210
D.
.
210
211
etc., in
.
206
and
D,
211
211
212
213
104.
Expression of any difference in terms of the given tabular functions,
213
105.
Derivation
the fundamental
of
Single integration,
The fundamental formulae
106.
relations
of
.
mechanical
.
quadrature.
'
.
214
214
of double integration,
TABLES.
TABLE
I.
NEWTON'S
TABLE
II.
STIRLING'S coefficients of interpolation,
TABLE
III.
TABLE IV.
coefficients of interpolation,
BESSEL'S coefficients of interpolation,
NEWTON'S
coefficients for
.....
.....
......
....
TAI;LE V.
STIRLING'S coefficients for computing F'(2"),
TABLE VI.
BESSEL'S coefficients for computing F'(T),
For finding n when
TABLE VII.
Giving y
TABLE VIII.
Coefficients for interpolating
....
computing f(T),
Fn
is
given,
.....
by means of tabular
first derivatives,
218
220
222
224
226
228
230
232
IllItLIOGRAPHY,
233
CHAPTER
I.
OF DIFFERENCES.
1.
in
In
applications of the exact sciences, and of
many
particular,
it
is
often necessary to
tabulate
a series
Astronomy
of numerical
values of some quantity or function, corresponding to certain assumed

values of the element or argument upon which the functional values
depend.
In the more purely mathematical tables, the function is analytithe argument is then the independent variable of the
cally known
;
given expression.
The common
functions,
cubes,
squares,
and
of logarithms, trigonometrical
reciprocals, are examples of tabular
tables
functions of this class.
second and larger class includes those functions which are not
related analytically to the argument, but which are either determined
by experiment, or based wholly or partly upon observation.

The final results are usually obtained from the fundamental obserdirectly
vations by suitable mathematical
transformations or reductions, which
frequently include the process of adjustment
known
as the
method of
Empirical values are also occasionally introduced in the

of
functions of this class, to supply some theoretical
development
least-squares.
deficiency.
In the great majority of such cases, the time is the argument of the
tabulated function. This is particularly the case in astronomical tables.
Thus
the Nautical
of the sun and
Almanac
gives the right-ascensions and declinations
the planets
for every
Greenwich mean noon
in the
case of the moon, these coordinate's are given for every hour, because
of the rapid
is
motion of our
satellite.
tabulated for every twelve hours
The moon's
horizontal parallax
the sun's for every ten days.

In like manner, the readings of the barometer and thermometer
;
THE THEORY AXD PRACTICE OF INTERPOLATION.
are recorded for certain hours of the day,
The
as functions of the time.
and therefore may be regarded
velocity of the wind, the height of tide-
and rate of a clock, are further instances of a

large number of physical quantities which are tabulated as functions
water, the correction
of the time.
As
examples of tabular functions of the physical or observational

kind, whose arguments are elements other than the time, we may
mention
:
The force of gravity (determined by

(a)
as a function of the latitude ;
pendulum experiments),
The atmospheric pressure (determined by
(ft)
a function of the altitude
The angle of
(c)
the barometer), as
refraction in a particular substance, as a function
of the angle of incidence.
Although differing thus fundamentally
in
the character of their
mathematical tables are alike in giving the

numerical values of the functions for certain assumed values of the
respective functions,
all
argument, so chosen that intermediate values of the function may

For this purpose
readily be derived by the process of interpolation.
not essential, to have the assumed argument
it is convenient, though
values proceed according to some law ; and since as a rule the greatest
simplicity is attained where the argument varies uniformly, it is nearly
always so taken. The interval of the argument is decided

by the rapidity with which the given function varies.
We
shall
in general
assume throughout these pages that the given values of
the argument are equidistant.
The
present chapter will be devoted to the subject of (Jij}'ri-<-nces,

as defined below.
The student should become thoroughly and practically familiar with this
upon the chapters

2.
Definitions
fundamental portion of the work before entering
that follow.
and Notation.
If
we have given
series
of
proceeding according to any law, and take the difference

of every two consecutive terms, we obtain a series of values called
the first order of differences, or briefly, first differences.
quantities
THE THEORY AND PRACTICE OP INTERPOLATION.

If
form a
we
new
difference the
first
in
series called second differences.
tinued, if necessary, so long as
We
differences
shall apply this process
the
same manner, we
The process may be con-
any differences remain.

of differencing to
the tabular values
of functions given for equidistant values of the argument.
Let
designate the argument; w, its interval; F(T), or simply

the given values
F, the function
t, t -j- w, t -(- 2w, t -\- 3w,
,
of T;
the corresponding values of F(T);
F^ 2 t
,
,
;
/",
lent
F F
,
J'", A-",
is
the successive orders of differences.
then shown in the following; schedule
Argument
The arrange-
THE THEORY AND PRACTICE OF INTERPOLATION.

Then, from the definition of differences, we have
Hence, by addition, we find
which
the
is
be
obviously
statement of Theorem
algebraic
an independent
as
applied
of
the
accuracy
differencing.
THEOREM
4.
are
taken,
If
the
values
of
differences
">
are
may
check upon the numerical
of
derived
values
it
JF(T)
of
beiny
assumed
N>n.
that
_/V
For,
of
II.
This theorem
I.
,/",
N3
values of
functions evidently yield _ZV 1 values of /!', JW- 2 values

values of //"', etc. ; hence
values of F(T) yield
n
J<">.
Inversion of a Series of Functions.

It is sometimes necessary
or convenient to invert a given column of functions, thus bringing the
5.
last value into the position
of the
position
given in
ences.
2,
second,
of the
find
F(T)
the next to the last into the
For example,
etc.
and observe the
Thus we
first,
J'
effect of
let
us invert the series
this inversion
upon the
differ-
THE THEORY AM)
To prove
two
that
such
series below, the
F(T)
an
I'll
A(
TICK
effect
is
<>K
INTKKH H.ATH >N
true generally,
we consider
second series being an inversion of the
first
the
:
To prove
F F F
generally, let
denote the two series of functions

will
be
F -\-f
u
^4/1
>
series
values are hence as follows
We
3'
F -F
fv rp
I
1 o
Also,
by
and
//,, /
J',
8',
let
is
and D', respectively
first
their
/o
therefore have
in
but since
the same
that the latter are derived from the given functions, the theorem
also true for
2.
Similarly with the following differences, each
order being the first difference of the order just preceding.

the theorem is true generally.
As an example we
F
series
us designate the
1
These relations prove the theorem directly for n
the second differences are formed from the first differences
manner
two
J'
*
then the sums of the
fi+fs,
differences of these three
f\
F.
write
Hence
THK THKOKY AM) PRACTICE OK IXTKKPOI.ATION.
differences frequently
e\liil>it
the omission of decimals
in
a small
the
degree of irregularity, owing to
approximate values of the functions
As
an example, we take the following values of T*, true to

the nearest unit of the second decimal
employed.

T
10
To
investigate the principle that underlies the method, let
denote the correct values of any function

(T) (tabulated for equidistant values of T), and let the differences be as shown in the
schedule below
F<r)

orders of differences,
with
is
alternate
easily seen
11
be that of the binomial coef-
to
in
Hence,
we have
only to
carry the differencing to that order at which the differences of the
correct functions would vanish, or sensibly so
the location and magficients,
signs.
practice,
nitude of the error will then be clearly shown by a succession of

and
terms, following the binomial law.
if
Thus,
fifth
the values of
differences
-lOe, +5e,
e,
J v vanish
the correct table
in
of the incorrect series will be

0;
the initial value, -(-,
0,
-j-e,
above, the
+10e,
oe,
therefore the error sought,

required function is found by
is
magnitude and sign. The

tracing backwards and downwards along the line of heavy type from
e,+ e to -F6 -\-e, which is the incorrect function and since the corboth
as
to
rection
is
the negative of the error,
we have
(J^+e)
the true value of the function in question.

9.
shall now consider several examples,
We
may be
EXAMPLE I.
process
in
e,
or jFG
for
order that the
fully understood.
Find the error
in the
following table of
F(T)
~ =
3
:
12
Finally, the
of the error
series
in
./
is
iv
,
number 206 is too small by 10 units, since the

shown by the leading or initial value of the binomial
10.
namely,
correction of -[-10
is
therefore to be
applied to the incorrect function, giving 210 for its true value.
In the column c, following z/' v in the above table, are given
the corrections to j |v due to the correction of -[-10 to the function.
,
The column J + e
lv
therefore
gives
the 4th differences
of the true or
always well to re-difference the series after

correction has been applied, to check the accuracy of the work.
corrected series.
EXAMPLE
II.
It is
Find the error
in the following table
of logarithms:

four values of /r
are
these are here sufficient.
available,
inspection shows that a correction of
13
slight
13". 0, as applied to the latitude
13".0 being a trifle

very nearly serve the purpose;
too great numerically, we soon find by trial that
12".8 produces the
best result.
Hence, the moon's latitude for May 11.0 should read,
for
+1
May
11.0, will
10' 10".6.
of Errors when More titan One Function is

Thus far we have considered examples of an error in one
Correction
10.
Affected.
are in error,
difficult.
It
When
two or more consecutive or neighboring values

the problem of correction becomes more complicated and
may even become indeterminate in some cases, since only
function only.
accidental errors
sive functions,
do not
can be detected by the differences.
and possibly
affect the regularity
all,
may
contain
Several succes-
systematic
errors
which
of the differences.
In general, the correction of a group of errors by differences may

be considered practicable only when the law of the function is not
obscured or altered by the presence of those errors. More definitely,

the method may be regarded as available in the case of two or per-
haps three neighboring functions, provided the errors be accidental in

character, and of sufficient magnitude to produce a distinct and definitive irregularity in the differences.
EXAMPLE
F(T) EE2T
I.
Correct the errors in the following tabulation of
25T
40:
TIIK TIIKOKY
14
column
C|,
series for
M
we
/'".
T=
-|-4,
we
EXAMPLE
correction
shown under
S, as
of third differences,
F(T}
the
place
correction of
value
AM) PRACTICE OF INTERPOLATION'.
for
with
T=
the
c.,
Correct,
-12
1S98
clearly indicates
this gives a final corrected
value
read
the errors
:
column
latter
instead of
ephemeris of the sun's declination

Date
This gives a corrected
-\-~).
constant
+2, should
should have
II.
The
./'"+<,.
applied in
of
-74
-
of
-|-12.
column
Hence,
instead of --7!
the
for
1.
which occur
in
the following

correction corresponding to
indicates a
the binomial coefficients of a
lower order than that of the difference
This means the
question.
in
of the order
correction
z/",
in
of an error in some earlier order of difference, rather than

the column of functions.
For example, if z/ v requires a
existence
an error
15
since A*
is
1,
3, 1, it
3,
the third difference of
follows
More
z/".
an error exists in
that
generally,
when
requires a correction according to the binomial coefficients of the

error exists in A*- m \
order, an
z/ (n)
th
These remarks imply the necessity of
some caution on the part of the beginner.

It will
is
be observed that when either the
in error, only the first or the last
term
in
first
or last function of a series
each order of difference will
be affected, and only by an amount numerically equal to the error.

in such cases, the method above explained is of little value.
In
general,
it
may
covered by differencing,
when
question,
be stated
it
is
that
when
errors
Hence,
have been dis-
advisable to re-compute the
values
in
the data for the calculation are available.
GENERAL PROPERTIES OF DIFFERENCES.

Let F(f), F(t+<a), F(t+2u),
of tabular functions, whose differences are taken
11.
below
Function,
F(T)
F(t)
4'
4,"
4"'
z/<">
F[t+(i
z/."'
/I"
"
/in/
represent any series

in the schedule
as
16
Till-.
We
shall
F (T)
assume that
and continuous function,
a finite
is
capable of expansion in a series of powers of

within the limits of the given table; then, denoting the successive
and that
s<a,
AM) PRACTICE OF INTERPOLATION.
TIIKOKY
/^(rf-j-.s-w)
of
derivatives
is
F(T)
by F'(T), F"(T},
Theorem, the following expressions

F(t)
etc.,
we
have, by TAYLOR'S
28
F(t)
F(f)+
F(t)
F(t)
+ 3uF> (t)+
~F"
<aF'(t)+
<
F(t+3u)
2,/* (0
F" (0 +
~ F"
(t)
1
+ 27 - F
12.
4aF>
(t)
(t)
16
(t)
81
"
(t)
(#)
li
+ ItZ-F" (t) +
(t)
^ F*
Lii
F(f)
F'"
64 ?L JP'"(<)
+ 256
iv
^-
Differencing these values of the functions in the usual manner,
we
obtain successively the expressions for
4'
J/
4'
J/
=-
F'(t)+
F"(t)
F"'(t)
J',
J",
//'"
F"(f)+
-^ F"'(f)+
15
^F
o)^'()+
F"(t)+
19
^ F'"(t)+
65
^J?
<oF'(t)+
7^
F"(<)+ 37
as follows
F(t)+ ....
F'(t)+ 3
<
F'"(t)+ 175
iv
1|T
^-
(t)+
....
(<)+
....
7^(0+ ....
+3<a*F"'(t)
i'"
It will
,*
F "'(<)+
be observed that
all
4
iv
f<o /'
()
+ ....
(3)
terms of the expansions (0) are of

where
denotes a numerical factor,
the general form, K<JF (t) ;

and r an integer which increases
by unity as we proceed from any
term to the next term following.
Hence, the differences
(r)
will
contain
17
We
thus see, a priori, that any difference

only terms of this form.
"'
of the
order must be of the form
J<">
A<a F<"(t)
Bar" F lr+" (t)
r
C<o + s
F"(f) +
D<a r+ *
+ ....
<r+s}
(t)
Let us now assume what appears from (1), (2), and (3) to be
the general law; that
is
A =
leaving the coefficients

We therefore assume
z/i"'
a,"/''""
(if)
B<a "+ 1
B, C, D,
F "+
Since the value of
making
l>
is
(t)
C<a ''+*F
undetermined for the present.
"+
+ Do>"+ F<"+ >(t) + ....

3
(t)
we may write
right-hand member of
arbitrary,
this substitution in the
get the expression for the n

-that is, the value of j;;;.
ttt
difference
for
t-\-a>
(4),
we
(4)
t;
by
evidently
immediately following
4*"',
Hence we have
Developing the functions of the right-hand member by TAYLOR'S
Theorem, we
find
(#)
()
+ uF
"+ ii
F'"
ai
,F"^-'(t)+ ^-V"'+ (<)+

1
"
Collecting the coefficients of
/'
(t) 4-
...
F' n> (t),
.1
(n
+(t),
get
n
we
.,
obtain
t
mW
(t)+(B+l)** F< (()+(c+B+\f**F***(t)
Subtracting (4) from (5), and observing that
....i
M "^F'"^'(f')
lix
li
/i+i
(5)
/I"'
,/i"
+
",
we
THK THKOKY AXD PKACTK'E OF INTERPOLATION.
18
If,
therefore,
we put
(6)
we have
...
(7)
Hence, if the general form of expression assumed in (4) is true

n -\- 1 ;
for the index n, it follows from (7) that it is also true for
we
but
see
by equations
hence
respectively;
1, 2, 3,
The expression
indefinitely.
and
(1), (2),
= 4;
therefore true for
is
(4)
holds for
it
the law obtains for
(3), that
and
on
so
positive inte-
all
gral values of n.
We
12.
haVe
now
to determine the coefficients
B, C, D,
of equation (4). These quantities are evidently functions of n and s,
and will be determined in the following manner
and determine the constants for ,/;/", which
0,
First, we take s
.
B
n
....
Cn
n
purpose by
These values are found by induction, thus: the relations (6) give
Cn D,,,
Cn+ i, .Z>n+i, .... in terms of
n
Making
n+1
we
shall denote for this
we
1,
take
{ ,
(7,.
Z>,,
derive
for
for
(7n ,
Hence,
B _^, B
n
tlie
first
gives
of the
succes-
Similarly,
we
Hence, the coefficients of (4) become known
0.
/;;"
Second, the coefficients of
is
clear from the schedule
it
F(t-}-sa>)
we
/>, ....
from
therefore
((>)
directly
a continued application of
equations (1)
s
B.,,
B,
sively the values of
in
if for
precisely the
brevity
we
easily follow
of
manner
that
11
that
/;;"
write
shall have, since the value of

/;"
is
from those of
arbitrary.
,/y
is
is
related
related
to
to
F(t).
19
in a series of powers of sa>,

we arrive
Then, expanding
^(^-f-sw)
at an expression of the form (4), in which the coefficients are fully
determined functions of n and
To perform
s.
the steps indicated,
we
from the
take
first
of the
equations (1) the following values:
To
A =
find B
n
Ci
= A
..-.
repeated application of the
By
first
of (6),
= B, + i
= B +i
B = B, + i
B,
B,
Hence, by the addition of these
To
find
Cn
Using the second of
we get
equations,
(6),
we
obtain
whence, by addition, we find
Since
(7,
î
this gives
r=-l
C.
= l(B + B,+
t
/?_,)
2
r=
To
find
B =
we have
But, from (9),
Again, from
2>,
(6),
hence we get
we
derive
A + i Ci + 1 -i + A
(8)
we have
THE THEORY AXD PRACTICE OF INTERPOLATION,
-!<>
whence
C> +
From
(10),
2B
we have
or
l>.
g(
+ l)
(11)
In like manner, the process might be extended to the values of

but the results already obtained are here sufficient.
n ,....;
n
Cn and />,,, given by
n
Substituting in equation (4) the values of
0), we
(9), (10), (11), (remembering that these values suppose s
E F
,
have
(12)
z/<->
= w"F<-> (0 +
n\
M^+> /-'(+
<w
We
now
proposed, we
/fj%
fl
(t)
+ WT:
J^(s+l)
obtain from (12) the
tt
,-+>/'i-+
!'
(o
!L
"4o
(+i)
expression for
,+/'<"+>(<)
As
/;-'.
already
write
Then, as shown above, we
shall
*(<) +S(0 *'()
have
+lJi
li
*"(^)
+l!i
E
L.
cS,.,8
w,
Upon arranging
we get
this expression
according to ascending powers of
Hence, substituting the foregoing values of .&,
and 1>
(1
we

of
values
the
that
find
S,
D,
C,
in
equation
21
(4)
are
as
follows:
"
*-H
These
are
results
(14)
easily verified
by substituting
special values of
and comparing with the coefficients in equations (1), (2), (3);

s
and taking n
1, 2, 3,
1,
thus, putting
successively, we obtain the numerical coefficients in the expansions of J,', ,7",, and J/",
n and
s,
respectively.
Itemur'kablf
13.
and J
The
'.
Formal Relation between
coefficients
S C D
n
the
Expressions for J
in the expression for
'">
z/p "",
be determined by the following method, which not only is

shorter than the above, but also possesses the advantage of showing
may
also
between the expressions for J "" and /7 ', respectively.

Retaining the above notation, we write (12) in the form
a direct relation
./'-'
We now
be
an
,"/''
>(<)
nU>'
+i F"' +l
<7,X'
+2 ^''' +2
'
(0
let
auxiliary expression, such
coefficient of
>(f)+
n+r
o>
(n+r}
and using the relations
(t)
(6),
the
that
in (15).
coefficient
Writing
+1
of
for
n+r
y
n
in
is
the
(15),
we have
Again, since the coefficients of

tain from (1),
g,
(y)
are those of
',
we
ob-
22
By
we
re-arranging the terms of (16),
c" y
E.
li
I?.
B.
find
li
li
'
'
'
'
li
Hence, by (15) and (17), we have

T.+I
Taking
1, 2, 3,
<J 4
Multiplying
cancelling the
.... n
these
common
1,
Ti
'/..
we
find
member
for
successively,
fits
equations
we
factors,
Therefore, by (17),
together
obtain
ti)"
member, and
(18)
we have
(19)
we
(15o) and (19),
coefficients in
Comparing
23
find
Substituting these values in (15), the latter becomes
^"'=
.....
<o+
^^"
+l
>(t)
+ ~ (3+
1)
***J**(0+.jg
(+i
which agrees with (12).
These
results
may
be
conveniently expressed symbolically: thus,

us represent the quantities J J ", J '",
J "> by Ja z/ 2 J 8
Jj
and for <af" (t), to*F"((), o?F'" (f), .... ofl (t) let us write the
let
symbols
J.
',
J92 , IP,
I),
....
---h U* + IS +
//-
D-\
Lr
D",
then
respectively;
ft.
have
shall
+ ....
ft
It
we
ft
(21)
D'1
14.
expression of the n
function
value of
The nih
THEOREM Y.
F(T)
is
.f
th
"'
of any rational integral

If the general form of the
differences
degree are constant.
is w"
aT n -\-pTn-
-2
-\--yT
the
constant
.
[
For, from the nature of the function,
and
F'^'d)
F<" + -<(t)
we
= ....
have, evidently,
=0
Hence, from (4), we have

z/j>
The theorem
interval
oj.
is
w "]<
.....
(f)
,"[
(22)
therefore true, whatever the value of the constant
Several examples have
already occurred:
in
we have
24
F (T) = T
the differences of
102"
20;
w==4, a ==
here
1, a>
l.
Hence, by (22), we get
/"
It
24
the value already found by differencing.
In
of
Example
F (T)
9,
r
-
'-
3
,
= 1;
we
obtained
there
for the value of the third difference

./'"
(>
which agrees with the theorem.

in
Again,
of
Example
F (T)
10,
E=
2T
25T
40,
w ==
1;
whence the theorem requires
J"'
which
is
2li
12
the result already obtained.-
THEOREM VI.
15.
[i
the n
If
th
differences
of a
series
of quantities
the given quanti(tabulated for equidistant values of T) are constant,

ties are the tabular values of a rational integral function of the form
"
F(T) EEa2 + 8r- +y7
7
7n
-2
This proposition
is
....
the converse of
THEOREM
V, and
is
proved as
follows:
denote the function whose true mathematical values,

tabulated for the given values of T, form the given series of quantities.
Let
From
F(T)
(4) and (5),
only in their
in
(a
first
we
see that the expressions for
having unlike
order that
J""
First,
<u
(<)
coefficients.
that
n
(a
(n)
(t)
agree
the remaining terms of like order
Hence, the conditions necessary in
does not vanish;

?
2
2
o "+ JF '"+ '(0
1>
U "+ 1*"+ (<)
and
j;-"
throughout are as follows:
shall be constant
Second, that
But, since
o"jP
term,
T(w)
= 0;
cannot vanish, these conditions reduce to the form

.....
<
A-<-H>(C) =
>
If
then,
now we put
(23)
/'<+!(*)
= .... =0
T =
+r
by TAYLOR'S Theorem, we have
(24)
F(T)
F(f + r)
F(t)
rF'(t)
F"(t)
+ -/'<" (Y)+
II
By
12
III
(23), this gives
F(T)
we
F(t)+rF'(t)
.+F<-"(t)+F<
observe, the coefficient of T" cannot vanish.
in
which,
in
(25) the value of T given by (24),
F(T) =
we
Substituting
obtain
+ (T-t)f'(t)+(T-t?-F" !+.
F<"
(t\
F(t)
lt\
+ (T-f)--
li
12
has a fixed value, the right-hand member of this equation

an expression of the t th degree in the variable T, and hence may
Since
is
25
be written in the form
which establishes the theorem.

16.
In the discussion
Convergence of the Differences in Practice.
of Theorems
and VI, we were concerned with the true mathematical
values of the quantities
involved.
In practice, however, the absolute
mathematical values of functions are seldom employed; fre-
or true
quently, as previously noted, a function
tabulated
only to a certain
enough decimals being retained to give the
observe that in such cases there is a tendency
is
degree of approximation,
desired accuracy.
We
of the differences to decrease numerically, and usually to vanish sensibly,

as the order of difference progresses.
The explanation of this tendency
readily from
follows
thus:
for
any given function, the

derivatives
have definite values;
F""(t),
(0,
(), .....
of
the
value
w
be
chosen sufficiently small to render all
hence,
may
the terms in the second member of (4) insensible, except the firxt.
equation
1
/'
When
this
and equal
cally
is
condition
to
brought
"'+ I>
(n+2)
obtains, the value of
n
/<"""(
(4),
t ).
The
differences
to a termination at
the
th
of
J<"'
is
F (T)
sensibly constant,
are thus practi-
order, whether the function
algebraic or transcendental.
In
rapidly
of the successive derivatives converge

the chosen value of w may then be quite large, and yet allow
many
;
cases the values
the differences to sensibly vanish at an early order.
This
is
equivalent
THK THEORY AM) PRACTICE OF IXTEHl'Ol.ATIOX.
26
to the obvious statement that,
when
a function
is
to
be tabulated so
as to difference readily, the interval of the argument must be decided
by the manner
To
in
which the given function
exemplify these
seven-figure logarithms:
principles,
we
varies.
take
the
following
table
of

due
to be
to the small value of the interval w,
u F'"
appreciable,
(t)
insensible;
w^
4
but renders
accordingly,
/'"
is
7lT
27
which makes the term
(0,
ai'F v (t),
the last difference which
....
quite
we need take
into account, the
may
remaining differences being practically zero.

add that if the values of T in the present table were
etc.,
instead of the given values, the interval w would become
We
100, 101,
1 instead of 0.01,
and hence
w,
'-',
w3
4
*
,
....
would not converge

100 instead of 1,
We
should then, however, have

t
which would cause the successive derivatives to converge rapidly, as
is obvious from the
general expression
as
above.
Furthermore, the differences of
F(T)
contain only terms of the
form
K denotes
where
a numerical factor; hence, since the values of
and
both increased one hundred-fold by the assumed change, it is

evident that the general term Kw"F H} (t)
is
not altered thereby.
The differences are therefore unaltered by the proposed change; this
t
are
conclusion
tion in
to 2,
confirmed by the consideration that the assumed alterawould merely change the logarithmic characteristic from
is
and thus would not
servations illustrate
the
affect
case
derivatives converge rapidly,
the
resulting differences.
These ob-
of a tabular function whose successive
whereby
a comparatively large
argument
may be used, and yet allow the resulting series of differences

to converge as rapidly as may be required.
17.
As a second example, we consider the following table of
interval
cubes:
T
28
We
have already seen (Theorem
matical values of
T*
that
when
the
true mathe-
are tabulated, the third differences are constant,
the fourth differences being the
order to vanish.
first
In the present
3
whereas
however, only two decimals have been retained in
To this degree of approximation, the
the true value involves six places.
table,
this follows
third differences are entirely insensible;
which gives for the constant value of
In this example
from Theorem V,
,.]'"
we have
a,
0.05
and hence
J"f=
which
insensible
is
(0.05)
X6 =
0.00,0",
when only two decimals
are concerned.
Thus,
in
frequently used in practice, the differences

generally terminate (either absolutely or approximately) at some order
earlier than would occur if the true mathematical values of the function
the
so
approximations
were employed.
be added that the above example affords an illustration of
Theorem VI. For, since the second differences are here absolutely
It
may
follows from this theorem that the tabular quantities are

the true mathematical values (corresponding to the given values of T)
constant,
of
it
some function of the form

F(T) =uT* + /3T+
Thus,
T = 5.16,
for
he
To Express
The problem
Let
n
consists
terms of
in
n,
5.21,
l?
us
numbers
M
in
....
and retains
5.46,
(n
\i)
in
Terms of
<o
n+l
w r F' r \t)
for
in
" +lt
(t),
by xr
(15),
decimals involved,
all
identical with the
reversing the series
<o"F w (t),
denote
-f-2,
....
will find his tabular
18.
student tabulates the function
in particular, if the
F(T) =
above
.,/<",
./<+",
series.
./;;+-',
(15), which
etc.
expresses
....
then,
we have
writing
successively,
THE THEORY AND PRACTICE OF INTERPOLATION
29
(28)
from which we obtain, by transposition,
~ -"!
Jt
x/("-t-ll
**
r
*
..2
v
*
- T)
-^n
(29)
The second
of the equations (29) gives a value of x n+l

substituted in the first equation, gives xn in terms of Jy", J;n+1) x^ t
,
we
xn
find
terms of
in
x a+2 given by
expression the value of
substituting in the latter
of (29),
J;"
./<"',
42
+1)
JJ,"
',
tinuing this process of elimination indefinitely,
The
',',"'
&,
cn
<o"F ^(t)
(
coefficients
(15) we
x n ^, x f+4
we
x,, +ls
the third
....
Con-
arrive at an expres-
form
sion of the
*
which,
+
,
''
dn
""
A'""
d,,/]^
obtain the following group of equations
//"
+B
+ ....
must now be determined.
....
M>
'',,''(""
(30)
From
y"+*+
(31)
Comparing (28) and
we observe
(31),
that the latter group
may
r
r>
be obtained from the former by writing
r and
y for //O and av,
respectively; the algebraic relations in both groups are otherwise identin
in terms of
cal.
Hence, if from (31) we seek to express y
,j
7,,,
f(
that
n+l
(in+t
which gives
terms of
in
./
y"
the process of reversion will be identical with
.,
<i,.
/'?+,
/;,">,
"
//;,"+",
<r-n:
the coefficients being those of (30).

//"
Taking
i\
w=l,
.,.,
'
'
rf
I,,
<h
.,/"
'\t?
e,
hence we must find
....
'/-,++
Therefore, by (18),
42
1'
,/,/"
r +
(32)
we have
:;
'/'/
(30) and (33),
in
and
<rr
<', t
we
,/,./"
(33)
obtain
+ W\*+d^+
(34)
'
(35)
THK THKOKY ANH PR ACT CK OK INTERPOLATION.
30
From
(17), by adding unity to each

1
member, we have
or
//
.'/
fc
= -i
= +
<>
F'(t)
(37)
-**' + * ft* -i?i4 +
4,'-
we
(38)
find
*--!
we
Substituting these values in (34),
x,
+ g,)
(35) and (38),
coefficients in
Comparing
*i
log,(l
(3)
obtain
J" J
4+
J
^--^- + ....
'"
IT
(40)
Again, from (38), we derive
(42)
Eqiiating coefficients in (33) and (42),
A.
-|
U-.i
+y4 (3
+ 5)
d.
These values being substituted
in
J(n
we
find
+ 2)(n+3)
(30), the latter
....
becomes
+s '4-
Using the symbolic notation adopted

ing expressions
D -
4 - */v +
in (21),
(43)
we have
(44)
the follow-
J -
i.
v+
(45)
19.
Effect of a
Magnitude of
Change in
the Several
the
Argument
Orders of Differences.
Interval
upon
the
Let us now suppose
THF THEOKY AND PRACTICE OF INTERPOLATION.

that a second tabulation of
F(T)
31
has been made, differing from the

first only in the value of the interval,
Let
mj be the interval of the argument in the second table; denoting the differences
'
by
8',
8", 8'",
the
new
table will run as follows
THE THEOKY AM) I'KACTICE OF INTERPOLATION
Now
let
61
be an auxiliary expression, such that the coefficient of

in (47).
coefficient of x H+r
From
we
(33)
is
the
g-T
we must perform
tion
+r
obtain, in succession,
....
+1
to eliminate
Now,
y"
?/", //"
from (51), by means of (52),
same algebraic steps as in the derivawe shall therefore obtain

of equation (49) from (47) and (48)
precisely the
and, for
1,
we have
-i
Now
li
the equation (51)

H
(my)"
+B
Pi<f>*
....
7iVi*+
(54)
be written
may
n
Whence, by (15), we have

and hence,
also,
2,
or,
by (17),
lH.l2.li
Also, from (36),
.-.
Sl
e""
(56)
we have
1
(f,
e>
the combination of which with (56) gives

,
m(m-'l)
- 1
2
ni(m,
1)
(w
1)/
nnin
V
1)/
(in
5
+ 1)
'
'
or
m(m
r+1)
fr,

Comparing (54) and
we
(57),
33
find
m(Hi-l)
rn(m-l)(m-2)
~F~
~1T
we
Substituting these values in (50),
obtain the following funda-
mental relation:
1)
m(m
V = '^ '+- L g- - Jo"+ -...+-
Again, using the relation

*.
<j, (
we
.;;-,
<r.0y)
....
-
1)
///(;/i
(in
+ !)
M +r)
(59)
obtain from (55),
\<ti(y)\*
*r
Hence, from (57), we find
m(m
-
Expanding and
m(m
~
l)
factoring,
we
l)(m
^
2)
obtain
=
.
coefficients
Equating
0.
2"(i-l)
y.
of
like
powers of
^m"(-l)r(3
in
(53) and (01),
+ l)m-(3+5)l
(61)
we have
....
(62)
Substituting these values in (49), the latter becomes
m" Ji
Finally,
sions
m" (m - 1) 4V' +I) -f
we may symbolize
--
"
(in.
- 1)
(3n +
!)- (3 + 5)
J<"+2
'
these results by the following expres(64)
m(m-l}. (m-3)
.
12
m(m-l)
'4+
- 7)
-^.
(63)
34
20.
THEOREM
VII.
n th differences of a given
the
//'
serif*
<>j
as compared with all the following

differences, then, if the series be re-tabulated with the argument interval
times its original value, the n th differences of the new series will
functions are numerically
/n/-</<'
be
approximately
times the
corresponding
n th
differences
of the
original series.
The theorem
J^, ....
+1)
x/J"
is
are
proximate value of
COROLLARY.
th
stant, then the n
equal
to
For,
if
of equation (63).
For, if
small in comparison with J<,n) , then the ap-
a direct
If
all
<n)
the
is
interpretation
mn j
nth
differences
is
differences
of
times the original

//""
(n>
constant,
the
new
of the given
series
are con-
series are also constant,
and
th
differences.
J +,
(a
ji->,
....
are
we
take
all
zero,
and
hence (63) gives, rigorously,

8<">
21.
To
illustrate
table of cubes:
= wBJ
"'
the foregoing results,
the
following
in the latter
we
vanish,
V =
From
-4 4," + A 4,"'
first
'
Whence, from
'
and observing that the differences beyond
3,
//'"
find
44,'
the
35
= 44,"-A4/"
of the above tables,
+92727
(65),
30909-618
4,"
we
we
V" =
A 4,'"
(65)
take
J >" = +162
+5562
derive
+ 10 =
8 "
30301
-12
618
606
'
which agree exactly with the values found in the second table above.
"
and 8 come within s'o part of equaling
It will be observed that 8
'"
$4/ and ^4,", respectively; while 8 = ^-4'", exactly. These rela'
Theorem VII.
tions are in accord with
To Express
22.
the Differences
of
Terms of
in
F(T)
the
Given
series be F
F F F
then
F
F
F
F
F
F
the
second
the
the third differ^ 2^, + ^,, F 2F -\-F^
differences,
ences,
^33^+3^^0, Ft 3F +3F F .'...; and so on.
Let the given
Functions only.
first
differences are
l}
coefficients
The
evidently follow the
t ,
binomial
l}
law.
Thus we have
generally
^> =
F,- F_,+
-1
(66)
F _2 n
+ -1'.C,F
which, according to the usual notation, we put n Cr for the coefficient of of in the expansion of (1 -)-#)".
To prove (66), let us assume it true for the index n; then the
in
th
expression for the w difference immediately following 4,"
will be obtained by increasing the subscripts of
M H _ t)
We therefore have
(66) by unity.
F F
(
Subtracting (66) from (67),

/
= Jr-4," =
F.
we
find
{i.e., //"')
....
- 1)" F,
in
(67)
36
TIIK
THEOKV AM) PRACTICE OF INTERPOLATION".
But, as proved in Algebra,
we have
Cr+l
a^r
h+l^V+
and hence the preceding equation becomes

(68)
It follows
n,
n -f-
also holds for
it
sion
so
from (68) that
true for
is
on
indefinitely;
1,
1.
if
But
2 and
the law expressed in (66) holds for

we have seen above that the expres-
Hence
3.
the equation (66)
is
it
is
true for
therefore true for
and
-1,
all
positive
integral values of n.
23.
To Express Any Function of a Given
Particular function
(f ), and
As
which follow that Function.

denote the given
below
the Differences
before, let
series, the differences
(oo,
Terms of Some
&, c
.)
f)
F F F F
,
} ,
Let
J"
A"'
be required to express
the nature of the differences,
it
in
....
being taken as in the schedule
F(T)
From
of
Series in
Jvl
Jiv
terms of
we have
3a
&,
c?
The
and so on.
/'
To prove
index
(69),
since
the
.
find
fU,
^Jn(n
1)
this value
= '-+
in
a,,
relation
n(n !)(
2)
0+ - I-2L _J
we assume
==
we
same
Co
that
(69)
true
is
it
for the
ô+ (+l)o+
from
as the relation of
to
obtain
n(n
1)
....
+^-^c +
that of F given
+ni
of a n to
terms of
here the
is
thus
a,.
Adding
Moreover, we evidently have
n.
na,
(69) by induction,
We may now
,
which sug-
coefficients again follow the binomial law,
form of the general term
gests for the

F.
37
^
E,
by (69), we find*
*o
i^
)^-
-)
Co+
(70)
Thus, having assumed the relation (69) to be true for the index
we find by (70) that it is also true when n-\-\ is written for

but we have shown directly that (69) holds for
n
1, 2 and
The formula (69) is therefore true for all positive integral values of
,
We
here omit the proof for the general term, since the process
is
the
same
as in
22.
n;
3.
n.
THE THEORY AM) PRACTICE OF INTERPOLATION.
38
EXAMPLES.
Tabulate the five-place logarithms of 25, 30, 35, .... 65, 70,
and take the differences to the fifth order inclusive. Retain a c-opv
1.
of the table for further use.
Tabulate
2.
F(T)
74, 77;
53, 56,
log cosT, to five decimals, for

difference to the fifth order, as in
E
T= 50,
Example
1.
Retain a copy of the table.
Verify the accuracy of both the functions and their differences

in Examples 1 and 2, by noting the degree of regularity in ,r, accord3.
ing to the method of
8.
Also, rigorously check the differencing in the above examples,

by taking the algebraic sum of each separate order, as explained
4.
in
3.
the two series of functions tabulated in Examples 1 and

difference the new series as before, and see that the resulting
5.
Add
values of /f are the
sums of the
fifth
differences of the other series,
according to Theorem IV.

6.
Correct the errors in the following tables by the method of
differences:
()
T
(*)
(')
.1

(d)
CHAPTER
II.
OF INTERPOLATION'.
Given a series of numerical values
Statement of the Problem.
of a function, for equidistant values of the argument, it is required to
find the value of the function for any intermediate value of the argu24.
ment, independently of the analytical form of the function, which may

or may not be given.
Interpolation
is
method by which the required
process or
the
values are found.
Without
certain restrictions or assumptions as to the character of
the function and "the interval of

is
polation
from a
station,
of temperatures
would be impossible
tabulation, the problem of inter-
Thus
an indeterminate one.
series
it
its
it
is
evident, a priori, that
recorded for every noon at a given

to obtain
by interpolation the tempera-
per contra, the thermometric

readings were recorded for 7.00, 7.10, 7.20, 7.30, .... P.M., it is highly
probable that the temperature at 7.14 P.M. could be interpolated with
ture at 8.00 P.M., for
given day.
If,
accuracy.
The Nautical Almanac gives
the heliocentric longitude of Jupiter
every 4th day; but, because of the slow, continuous, and systematic character of Jupiter's orbital motion, it is found sufficient to
compute the longitudes from the tables direct for every 40th day only.
for
The intermediate
which equals,
if
places are then readily interpolated with an accuracy

indeed it does not exceed, that of direct computa-
tion.
Nautical Almanac for every

for the moon's orbital motion is so rapid and compli-
The moon's longitude
is
given
in the
twelve hours;
cated that it would prove inexpedient to attempt the interpolation of
accurate values of the longitude from an ephemeris given for whole
day
intervals.
41
appears that, to render the problem of interpolation

determinate, the tabular interval (w) must be sufficiently small that
It therefore
nature or law of the function will be definitively shown by the

tabular values in question. The condition thus imposed will be satisfied
the
given table, the differences become either rigorously or

This follows from the
sensibly constant at some particular order.*
fact, soon to be proved, that for all such cases a formula of interpoin
when,
lation
can be established, either rigorously or sensibly true, according
to the foregoing distinction.
Extension of Formula (69) to Fractional and Negative Values

Provided the Differences of Some Particular Order are Constant.
2.1.
of
n,
We
have shown (Theorem
gral function vanish
V)
that the differences
beyond a certain
We
order.
that, for any such function, the formula (69)
is
of a rational inte-
proceed to prove
rigorously true for all
values of n.
denote any function whose differences become constant at the order i, and let j^
1
J?(T) and its differences are
Let
F(T)
then shown in the schedule on following page.
Excepting, of course, any periodic function whose tabular interval (w) differs but
its period, P.
An example of such a series is the following
some multiple of
Date, 1898
little
from
42
THKORY AND PRACTICE OF INTERPOLATION.
T11K
J"
F(T)
\
*;
From
we
(30)
<jF^(t)
With
./;,"
1,,. 7,';+'
-"
'+<-,., A,"-
the condition assumed, these equations give
in
Hence,
this
It follows
term.
that,
TAYLOR'S Theorem
Hence, for
table.
Let us
v
= F(t +
now
~
expansions
for
at
the
(i-(-l)th
capable of expansion by
values of n within the limits of the given
F(t -\-nui)
words,
all
'
is
we have
such values,
all
end
(0)
under the present assumption, the expansions
in other
(0) are valid;
the
case,
(0
+.
F"(t)
.+'
(71)
consider the expression
'
'
|
(721/
\
i^
Substituting, successively,
0, 1, 2, 3,
i-|-3,
in
(72),
we
get, according to (69),
= F Flt F
(,>
Substituting these
same
results,
FI,
same values of n
F^,
in
respectively.
(71),
we
evidently obtain -the
namely
F,
= F Flt Flt F
,
i+s
in succession.
THE THEORY AND PRACTICE Or INTERPOLATION.
F and Q are equal

But F and Q are both
Hence,
//.
more than
to each other for
expressions of the degree
43
in
values of
Now,
n.
when two
expressions of the degree i in n are equal to each other for

more than i values of n, they are equal for all values of n. There-
of
fore, for all values
n, fractional
and negative, we have

.
"
= =
provided that J
of interpolation, and
1
constant.
1H
is
known
This
is
the fundamental formula
NEWTON'S Formula.
as
Second Proof of NEWTON'S Formula, for Constant Values of

Formula (73) is readily proved by means of equation (59), in
26.
.
/'".
m may
which
The only
have any value.
condition necessary for the

But
validity of (59) is that the expansions (0) are themselves valid.
since we assume that the differences beyond j<" vanish, it follows (as
that the expansions (0)
in the last section)
proved
are, valid.
Hence
(59) gives, rigorously,
From
the definition of 8
V =
.:
'
(see schedule, p. 31),
we have
= Fm -F,
F(t + m*)-F(f)
F. =
,
which
is
the
same
as formula (73), except that
is
written for n.
To Find n, the Interval of Interpolation.

The binomial coefficients of NEWTON'S Formula are given in Table I, for every hundredth part of a unit in the argument n. The quantity n is called
27.
and
the interval of interpolation,
To
in practice is
obtain an expression for n, suppose that
we
always
less
than unity.
are to interpolate the
value of the function corresponding to the argument T, whose value

lies
between
and
t-\-u;
then
f\ = F(t + n<o)
we
F(T)
and therefore
"
which determines the interval
n.
shall
,
or
have
t
+ n<a = T
(74)
44
28.
of (2.8)
EXAMPLE. From the following

by NEWTON'S Formula:
T
table of
4
,
find the value
THE THEOEY AKD PRACTICE OF INTERPOLATION.

therefore,
is
to find
in the schedule
before
F_ n
below
for this purpose, let
F (T}
45
be differenced as
the values of J<" being supposed
constant as
46
given in this case, the formula (7a) must be used; n being the interval of the required function from
toward F_
From
EXAMPLE.
of (13.26)
= 14,
we
find
14-13.26
3-
0.3,
F =. 38416.
the interval counted backwards from
is
from Table
I,
we
Hence,
obtain
=
B =
A
a_,
6_2
I)
+0.06333
-0.04164
by (75), we have
= +17680
= + 6944
we
Therefore, by (75),
=
=
+0.37
-0.11655
for the differences required
c_3
d_<
= +2112
= + 384
derive
-Aa_,
+ Bb_i
-Cc_*
+Dd_j
F =
.-.
By
given on page 44, find the value
And
Taking
which
the table of
(13.26)
we
direct calculation,
= +38416.00
= -- 6541.60
= - 809.32
=
133.75
-
=
=
15.99
+30915.34
find
(13.26)
30915.34492 +
Application of NEWTON'S Formula, when the Differences Beand 26)

have proved
come only Approximately Constant.
that (73) is true for all values of n, provided the differences of some
30.
We
particular order
that, if
(25
We
rigorously constant.
are
the value of n
lies
and
between
-(-1,
more frequent case
approximately true for the
now propose
in
to
the formula
is
show
very
which the differences
of some order become approximately, but not absolutely constant. The

example given on page 8 is typical of this case; the numbers involved
are not the true mathematical values of the quantities represented, and
hence the
Let
irregularities, as already explained.
F F F F
,
v ,
mate tabular values of any function
denote a series of approxi-
F(T),
given for equidistant

values
FU
T, and true
of
FZ,
....
the
to
matical values of the series, which
F = F -\-f
let
also,
approximate values, due to
unit of their
last figure;
let
denote the corresponding true mathe-
nearest
47
we
shall designate generally as
F;
being the difference between the true and

the omission of decimals in the tabular
quantities.
The
are
now
differences of F, and those of the series
/,
f.2
defined by the two schedules below:

A<
F(T)
J'"
t+
(0
(A)
ft,
ft
ft<
J"
t
t
0)
ft
Vo
ft
(B)
ft
ft
73
ft
Then, since
differences of
F=F-\-f,
it
follows from
are as given in the
appended table
that the
J"'
F(T)
Theorem IV
"
O)
<
+
(C)
= F +f
t
a,
ft,
ft
"3
m,
-A
Let us now suppose that the differences J< i+1 in Table

(C) are
either alternately -(- and
or that -j- and - - signs follow each other
>
48
II
THEORY AND PRACTICE OF INTERPOLATION.
Moreover, the foregoing definition of
irregularly.
terms
IK
in // ((+ "
are sufficiently small
no errors exceed-
indicate that
to
that the
requires
ing half a unit in the last place exist in the functions F(T). The
\ alues of J<"
are then approximately constant, and therefore Table (C)
proceed to investigate tinrepresents the typical case in practice.
We
accuracy of XEWTON'S Formula as applied in this case; assuming that

n is always taken within the limits
and -\-\, and that terms beyond
dw
are neglected.
Applying (73)
beyond
F.
in
.//>,
//.tli
order.
cover
its
from Table (C), and omitting the terms
we have
(F +f )+A(a
which
to find
+ nJ + B(!, + /3 +
)
C(r
l>
+y +
l}
.+Z(/ + A )
(76)
A, S, C, .... L denote the binomial coefficients of the

Let us now examine the approximate formula (70), to dis-
maximum
The formula
error
when
all conditions conspire to that end.
may be
(76)
written
(77)
For
brevity, let us put
Q =
RE
.:Fn
F + Aa + M + .... +U
+ ^c> + BP + .... + M,
/o
=Q+R
It will
be observed that
(77a)
is
the
value obtained for
when
We
(73) is applied to Table (A), terms beyond J being neglected.

leave the discussion of Q for the present, to consider the quantity R,
<()
which evidently expresses the error of interpolation due to the unavoidable errors, f, contained in the tabular functions F.
22 to the differences of Table (B),
Applying the formulae of
we have
fi
-fo
/ -3/ + 3/,-/
/4-4/ + 6/ - V1+/.
- 5/4 + IQ/i - 10/ + 6/, - /
/c
8
(78)
THE THEORY ASTD PRACTICE OF INTERPOLATION.
49
Hence, from (77a), we obtain
R = /+
.-.
+ &+ C7o +
Aa,
R = f (l-A + B-C+D-E+
+f,(B-3C+6D-10E+
+MD-5E+
Now the binomial
.
L)+f (A-2B+3C+f (C-iD+10E-
DB
.)
(79)
+ME- ....)+....
coefficients
A, B,
G,
are connected
by
the following relations:
A =
Hence, since we have assumed that n

follows that
A, B, C,
are
alternately positive
ABODE
thus:
We
therefore
coefficient of
since the values of
is
/,
-\-Q.5, in
sum
+0.5
/,
We
number.
of
"
-;
-0.5
if
we
/,
f may
take
+0.5
f = -0.5
....
all
1
(l-x)- (1-*)" E (1-z)"-
into the
right-hand
-(-
all
form
(80)
member of
sign.
in (79)
the
us
consider
For this pupose, let
identity
which, for
the terms after the first in the

(79) will be numerically a maximum, with the
shall now show that the coefficient of
the
and negative,
have any value between

terms of the same unit; hence, it follows from the
foregoing conclusions that

/,
it
are supposed true to the nearest unit
of the last decimal figure, the quantities
and
-j-1,
....
it
"
0.5
and
draw the following conclusions respecting (79)

The
Now,
between
lies
is
a positive
values of x numerically less than unity,
may be expanded
50
Upon
equating
we
identity,
the
the
L =
the
two members of
(-1)'
member of this equation is the

the final member contains only
coefficient of
we
in
(79)
is
positive
a positive quantity.
in
it
Ac-
conjunction with the values of

designated in (80), the value of
given by (79)
take
-f-0.5,
in
now append
.a
table
of the quantities
as above determined, with their differences
factors,
/i>/2>/3>
will then be the greatest possible under the assigned conditions.
We
this
(-0
(-l)(-2)(-S)
first
(79) ; and since

follows that the coefficient of
cordingly, if
of x* in
find
1-A+B- C+
Now,
coefficients
f f f f
,
t ,
THE THEOEY AND PRACTICE OF INTERPOLATION.
To evaluate (81) for different values of n between

we make use of the following abridged table:
n
=A
51
and
-}-l,
52
Whence
is
it
seen that the greatest possible values of R, under
the assumed conditions, are

RZ
.R,
0.6
0.8
1.1
1.6
RI
2.3
3.4
....
While it is obvious that the combination of accidental

shown in Table (B'), is very improbable, yet approximations
errors /,
such
to
In such cases the
combination- will occur occasionally in practice.
by NEWTON'S Formula may be a

These values show
considerable part of the values given by (83).
errors (7?) in functions interpolated
that
when the
be greater than
this
cannot
beyond /f are neglected, the error
In all probability
units of the last place in F.
differences
1.6, in
error will not
exceed one unit
and when
it
is
considered that
the results of an average logarithmic computation are uncertain within this amount, we are justified in neglecting the error R, provided
that fifth differences are practically constant.
Beyond
fore
the limiting values of
increase rapidly.
We
conclude that, aside from the inconvenience involved,
by NEWTON'S Formula when the
practicable to interpolate
be neglected.*
consider the expression Q of (77a), that
beyond /r are too large
We
now
is
im-
differences
to
Q=F + Aa
t
Now, because
it
there-
+ b.+
the differences of
in
+L1
is
(84)
Table (C) become approxi-
mately constant at J \ notwithstanding the irregularities they contain;

in Table (A) become sensibly
so, a fortiori, must the differences of
(<
constant at
Hence the
z/ (i) ,
the quantities of this table being mathematically e.w7.
differences
//<+
in
Table (A), namely,
form a series of continuous, but very small terms, whose values

Per contra, we have assumed that the
are nearly equal to each other.
will
differences
Excepting the case where F(T)

mathematically exact.
is
a rational integral function of T, whose tabular values are
53
of Table (C) either are alternately -(- and

or that -(- and
terms
succeed each other irregularly. It follows that the quantities m must
be numerically less than the maximum value of p. in the series
,
quantities
p.,
mask
the former would
bined series
the quantities
if
For, otherwise,
exceeded the greatest of the
the effect of the latter in the com-
hence there would be no general alternation of
m-\-p.;
signs in the series
But
our
assumption that the diiferencing in

Table (C) has been carried to an order J +1) which does exhibit a
therefore conclude that m is numerigeneral alternation of signs.
this
is
to
contrary
('
We
cally less than the
maximum
value of
Table (B'), we
Xow, from
p..
observe
under
that
assumed,
The maximum value
of
J")
y (=//'")
0(=
"
"
"
is
(=z/ )
ft(=J
( '
+1)
)
=
=
=
conditions
the
(2);
1
(2)
(2)
(2)'.
Hence, m^ is numerically less than 2'.

We have observed above that, as a consequence of the conditions
herein assumed, the differences of
in Table (A) are converging, being
beyond J>; hence the fundamental expansions (0),

deduced from these, are valid in this case. The formula
practically insensible
and
(59)
all
is
relations
therefore applicable to the series
in (59),
hence, writing n for
we have
ao
in
F(T);
'
Aa
Bb
Cc
+L1 +
Mm + Nn +
which as many terms should be retained as accuracy requires.

But we also have*
S
'
F(t + n)
F(t)
= Fn -F,
and therefore
Bb
* See
26,
Cc
+L1
where the same relations were similarly employed.
54
Now, by
may be
(84), this equation
= Q+ Mm
F,
written
+ Nn
tt
or
Fn The
Mm
= Mm + Nn +
Jn
(85)
expresses the difference between the true mathematical value of the interpolated function
and
its
ences
in
series
series
But
are nearly constant,
Mm
Mm
above observed, the differ-
since, as
follows that the differences
it
Nn
Hence,
therefore
-)-
approximate value Q.
comparison.
brief,
-f-
small
is
Mm
compared with
as
are small
-(-
Nn -)-....
may
in the right-hand
member of
(85)
(82)
we
Sw - S =
the last of these,
we
we
have shown above that
(87)
is
(88)
numerically less than
2';
this
expressed in the form
may have any
2'
sin 6
and 2n.
value between
From
this
relation
obtain
Mm =
or,
-l) (-If)
(2-l) (+E)
(2
= Sw -S
m =
where
-l
obtain
2'M
may be
(2
(2'-l) (-l)'Jf
condition
(86)
derive
,-H = + 3(7 =
- R = - ID =
.-R<= +15JS =
We
The
therefore be written, without sensible error,
FH - Q = Mm
From
in
represents, very nearly, the value of the rapidly converging
latter equation
From
by (88),
Mm =
Substituting this value of
(R
Mm
Fn - Q =
2'MsinO
W -R,M) s'uiO
in
(B
(86),
M -R
(89)
we get
M)
sin^
(90)
From
55
we have*
(77a),
Fn -
(1+
sin 0)
which, subtracted from (90), gives
- Fn = R
F*
sine
(Jfl
From Table (D) above we

cannot exceed 0.04, which
we
discussion;
is
see that
beyond
sin<?
the
z/'"
coefficient
an inappreciable quantity in the present
therefore write the last equation
Fn -F = R^sine-Q + aiufyR,
K
The quantity
(92)
numerically greater than R^ and both are

alike in sign; this condition may be
expressed by the relation
i+l
is
R = Rt+l sin
t
in
which
|>
2
!/'
has a definite value depending upon the value of
stituting this expression for
F - F, = E
n
becomes
in (92), the latter
i+1
[sin 6
sin
Sub-
i.
V (1+ sin 0)]
or
Since
cos2 t/>
is
necessarily positive,
lows that the coefficient of
when sin#
i+1
in (93) will
sin
\ *
in
(93),
negative,
|>
be numerically a
attains its greatest negative value; that
Taking
and
is,
fol-
it
maximum
when
=\
F,-Fn = ^(-cosV-sinV) = -RM

which
is
the
maximum
IT.
we have
(94)
numerical value possible to
all
n,
con-
ditions favoring.
the true mathematical value of the required function.

n is
the approximate value of this quantity which is obtained by applying
n is
NEWTON'S Formula
to
Table (C), neglecting differences beyond jw.
being assumed, (1) that the given functions

are true to the nearest unit of their last digit;
it
F F F F
lt
(2) that
dent that
is
....
positive
K defined in (77o) is not distinguished by a subscript in the earlier part of this

1! 3
Considered as a particular term of the series II*
J{ 4
however, it is evishould be designated as lit .
The quantity
discussion.
56
and
than unity;
less
(3) that the
differences
constant; and (4) that the differences
and
signs following irregularly.

from (94) that the computed value
J< +
Under these
can never
are
z/">
are quite
approximately
small, .with
conditions,
differ
it
-(-
follows
from the true
by more than the quantity lii+l

One point further, however, must be considered.
value
In computing
n by (76), we should, in practice, obtain the values of the several
terms to one or two decimals further than are given in F, to avoid
accumulation of errors in the
n,
the extra decimals
final
dropped, the
are.
Thus we
nearest unit, as in F.
tained rigorously by (76), but
which we may denote by
expresses the
maximum
this expression
(Fm )
we
quantity
Jii+l
being taken to the
ob-
Accordingly, the relation
0.5
discrepancy between
with (94),
sum,
approximation to that value,
close
and
(Fn ).
Combining
finally obtain
F.-(FJ = The
result
in writing the
r
actually use, not the quantit}
(Fn ).
p.
But
addition.
0.5
(95)
therefore represents the final limit of error
0.5
of an interpolated function, in units of the last decimal

From the value of
of F.
6
given in (83), we find that when
2.8 units.
Since it is
zf is nearly constant, the limiting error is
in the value
highly improbable that
produce
this
practically
maximum
terminate
With
the
at
occasionally be in error
and never by
necessary conditions will conspire to

error, we may add that when the differences
the
all
fifth
by one
order,
interpolated
unit, only rarely in error
functions
by two
will
units,
three.
sixth, seventh,
employed, the results

most cases would be intolerable,
or higher differences
become subject to errors which in

and which would probably be obviated by a
direct calculation of the
function.
From
the foregoing
tabular
therefore
appears that, for

functions should always be given
investigation
it
purposes of interpolation,
with an interval sufficiently small that differences beyond
<//
may be
57
generally fulfilled in practice. As already

stated in 24, the longitude and latitude of the moon are given in the
Nautical Almanac for every twelve hours; from the values thus given,
neglected.
This condition
is
intermediate positions can always be safely interpolated by using differences no higher than the fourth or fifth order. On the other hand, a
of the moon's longitude for every 24 hours would yield differences of the eighth or even ninth order; the use of which in NEWTON'S
table
Formula might produce an error of several units
in
an interpolated
position.
In
all
that follows,
we
assume that differences beyond the

This assumption made, it follows from
shall
may be
fifth
order
taken less than unity.
neglected.
the preceding investigation that the fundamental formulae, (73) and
(75), may be applied in all cases without sensible error, provided that
is
31.
We
shall
now
solve an
example which
points of the foregoing discussion.
for
T = 0,
values
in
....
the
tabulate the function
- 199841.772 T + 50804.968 T
2169.395 T +
116.817 T +
5645.715 T -
1, 2, 3,
terminate
we
9,
fifth
we
find
decimal.
that
the true
1.507
(96)
mathematical
These values of
given in the table below, with their differences:
the main
606607.920
If
illustrates
F(T)
are
58
of each
only slightly from the quantity

0.00500, or half a unit in the second decimal place; and, moreover,
last
decimals
three
that the actual difference

excess
and
defect.
is,
differ
excepting the
This condition
first
function,
will rarely obtain,
and
is
nlli nnilih/
in
here selected
only to illustrate the limiting case.

If now we drop the last three decimals of F, we obtain a series
of approximate values, denoted by F. The following table gives F,
true to the nearest unit of the second decimal, together with its
differences
THE THEORY AND PEACTICE OP INTERPOLATION".

T
59
60
THE THEORY AND PEACTICE OF INTERPOLATION.

where the numerical term
With
is
now expressed
the above determined value of
J2-a
61
the same unit as
in
(= -(-1-526),
the last equation
becomes
F,
it
units,
instance
F.
1.56
7
we were
obliged to write (J ,,) gi'eater than

follows that the actual error of interpolation in
Finally, since
0.23
by
this
approximately 1.8 units in the second decimal place; which agrees with the result previously obtained.
32.
is
1.56
As
-f-
0.23, or
more
practical
application of
NEWTON'S Formula, we
take the following
EXAMPLE.
for April
20
0".
Date
1898
From
the.
appended
table, find the sun's right-ascension
62
April 21, by means of (75).

the tabular differences,
A =
+0.20
we
Thus, from Table
find
(for
= 0.20),
and
To
we
the above schedule,
of
J", j">,
T
,
56,
+ Dd +
Cc 2
S, C, D, E, ....
where, as before,
J*,
find for the value of
F = F + na +
n
Applying NEWTON'S Formula
Formula:
derive STIRLING'S
i(a'+oO
Ee3
(100)
Let us now put
= H' + 'i)
to
represent the binomial coefficients
respectively.
63
*('+,)
(101)
from which, with the relations

a'
a,
we
l>
c'
c,
e1
^rf
e1
obtain
d
Using the equations (102), together with the
we
(102)
relations given in
23,
find
= a +P
= *0 +C = +
= + 2rf + = c + J <+ e +
c
= d + 2e +
^ = ^ + 2^4== e +
= e 4e
!
*!
c'
e,
.
substituting these
Upon
latter
values of
,,
(103)
Jj, c2
....
(100), the
becomes
Substituting in the last equation the values of B, C, D,
__
n(n-l)
__
E,
namely,
(-!).. (n-3)
e.
ii
n(t.-l)(-2)
E =
|3
we
in
n(n-l)
(n-4)
(6
finally obtain
(104)
known
which
is
ployed
in this
as
STIRLING'S Formula.
The even
differences
formula are those falling on the horizontal
line
em-
through
64
the odd differences are the means of those which
above and below
this line, as defined
fall
immediately
by (101).
Table II gives the values of STIRLING'S coefficients for the argu-
glance at this table shows how much more rapidly these

coefficients converge than those of NEWTON'S Formula.
ment
n.
EXAMPLE.
April 20"
0".
Date
1898
From
the
table
below, find the R.A. of the sun for
65
When the
Bm-kiwird Interpolation by STIRLING'S Formula.
forward interval approaches unity, it will be more convenient to proceed backii'di'dx from the following function by the formula
35.
''_= ô
na
-=-
ba
-e+
120
24
the coefficients of which are taken from Table II with the

//,
as before.
It
will
(105)
argument
derived from (104) by

or, by supposing the given
be observed that (105)
is
n for n in the latter;

merely writing
series to be inverted, and hence (Theorem III) changing the signs of
and
a, c,
e.
EXAMPLE.
Solve the preceding example by (105);
the sun's R.A. for April 20
Taking
= April 21,
The
as
Table
II,
?z?
find
0.20
above for April 20; thence, taking the

with n
0.20, we find
same manner
coefficients
F =
A = +0.20
is,
interpolation.
differences are formed for the date April 21 in the
found
that
by backward
we have
rt
m
56 55.84
from
66
and we therefore obtain
We
67
now
pass from STIRLING'S Formula

to another, somewhat similar, wherein we employ the odd differences
BESSEL'S Formula.
38.
which
c n e,,
i,
on the horizontal
fall
between
line
and
l ,
and
the means of the even differences falling immediately above and below
Using the schedule on page
this line.
Then, since
J,
i(*
+*0
= 6-ic,
6.
and
Ci,
62, let us put
(106)
these equations give
e 1}
= d-ie,
*(*,+<$
(107)
Let us write the formula (104), for brevity,
F = F +na+fb +Ce+Dd,+ JEe + ....
(108)
where
(n-l)
__
__
~6~
1>
da
e
Upon
_.
we
of (102) and (107),
=
=
=
=
=
a,
- ^d =
c,
b
c,
d
e,
(4
derive
ct )
04
e,)
cl
ca
ê
-J-
(d
J.
a, b
c,
Finally, substituting in the last equation
we
n (n
--
l)(n
~
40
^4~
which
in (108),
the values of
we have
C,
D, E,
obtain
n (n 1)
s -2
'
(110)
substituting these values of
from (100),
""
n('-l)(n'-4)
~2T~
Now, by means
a
n*(n-l)
--120
"
'
BESSEL'S Formula of interpolation, commonly regarded as

the most convenient and accurate of the several forms in use.- The
is
odd differences here employed are those which fall on the horizontal
line between
and F^ as shown in the schedule on page 62; the
even differences are the means of those falling immediately above and
below" this
line,
as
denned by (106).
68
THE THEOKY AND PRACTICE OF INTERPOLATION.

Table III gives BESSEL'S coefficients for the argument
EXAMPLE.
Use BESSEL'S Formula
the table below:
to
compute log
sin
n.
22'
from

Taking
7,
we have
n
Therefore
we
find
=2
0.60
69
70
be observed that the coefficients are here numerically the same

for the arguments n and 1
n; having like signs for the even orders,
It will
and opposite signs
More
for the
odd orders of
the values of BESSEL'S coefficients
generally, let us denote

T
differences.
.... taken with the argument n, by B,C,D,E,

respectively; and the corresponding values taken with the argument
An inspection of Table III then
1
n by
d, DI, E!,
for
J",
//"',
J'
/f,
t ,
shows that we have
S = +B
G! = -C
= +1)
Z>,
E = -E
1
(112)
To
F = F + Ma, +
n
Now,
we
establish these relations generally,
+ Dd +
<7c,
of
may
(113)
also
be exactly the same as in (113).

formula (Ilia), we have
will
Fn =
But we have,
Ee
be obtained by interpolating backwith the interval 1

n; the differences thus involved
the value
ivards from
Bb
write (111) in the form
-F,
(1
-M) a, +
BJ>
C.c,
Hence,
+ Ddt
after
E&+
the
manner of
(114)
also,
F,
(1-11)0,
OF, -a,)
a,
= F +
na,
Whence, (114) becomes
Fn = F + Tia, +
Q
BJ>
(7,0,
E^ +
Djl
(115)
which, subtracted from (113), gives
= (B-E )b+ (C+Qc, + (D-D )d+ ....

1
The equation (116)

interpolation
are
is
true in
applicable;
cases to which the formulae of
all
is
it
therefore
function of the
true
when J^(T)
second degree.
But, in the
case, the second differences being constant, we have
rational
integral
c,
e,
The equation (116) then becomes
(116)
(B-Bfib
is
latter
71
Hence, since b cannot vanish, we have
S = +B
l
This result reduces (116) to the form
= (C+C
l -)c l
+ (D-D )d + (E+E )e +
Again, we may suppose

d
that
J"> constant;
el
is,
(117)
we may put
The equation (117) then becomes
= (C+Qc,
or
C,
= -C
repeated application of this reasoning,

tions (112) are true generally.
By
It follows
backward from
When
numerical process involved in finding

n by
identical whether we interpolate forward from
is
or no check
that the rela-
that the
BESSEL'S Formula
or
we prove
lf
except for the terms in
and
A'.
Hence
little
by performing the interpolation by both methods.

check is deemed necessary, BESSEL'S and STIRLING'S
afforded
is
such a
Formulae should both be used.

42.
Relative Advantages of
NEWTON'S, STIRLING'S, and BESSEL'S
In practice, the only important application of NEWTON'S

Formula consists in interpolating functional values near the beginning
Formulae.
or end of a
given series.
The
selection
of this formula
is
then a
matter of necessity rather than of preference.

In all other cases, either of the more rapidly converging formulae
of STIRLING or BESSEL should be employed.
Regarding a choice
between these two, when Tables II and III are available there would
appear to be very little advantage one way or the other. The form
given by BESSEL is more commonly used, and is perhaps a trifle more
accurate in practice than STIRLING'S form, particularly for values of
neighborhood of one-half. When n is quite small, however,
STIRLING'S Formula will probably be found more convenient.
in the
72
TIIE
Suppose we have given
F(D
a limited table of functions, as follows

are insensible,
we observe from BESSEL'S Formula
the approximate value of
The maximum value
that the
73
error
of
will be
of
whence we have the following
which obtains
>
result
for
= ^,
is
maximum
WTien second differences are sensibly constant, the
error
of functions obtained by simple interpolation is i//".

Thus, in Tables I, II, and III, the values of the coefficients for
J" (designated above as .Z?) can never be in error by more than 1 of
10 units, or 1.2 units in the fifth decimal, when found by simple
interpolation.
44.
Corrected First
or
stant,
Involving
Interpolation
When
Difference.
nearly
but
so,
Second
too
neglect,
included (and hence an accurate value of
ing simple method
Means of a
second differences are con-
the
to
large
by
Differences,
their
effect
may be
obtained) by the follow-
Since third differences are 'supposed insensible, BESSEL'S Formula
becomes
F =
m
which may be written
in the
F,
Now, because
(121)
F,+ na +
n (n
1)}
form
F.
!-
-*
third differences are negligible,
(121)
we may
write &
for b in
then, putting
we have
(122)
na
The value of
interpolation.
is
is
thus obtained almost as readily as in simple
In forming the quantity
the complement of
value of n
/*
with
respect
ordinarily required.
to
25
(which
unity),
The value
of
is
simply one-half
only an approximate
c^
the corrected first
74
difference, is thus
found by an easy mental process amounting almost
to mere inspection.
EXAMPLE.
Find
(8.2)
from the following values of T*

Taking
=
T =
=
t\
a,
= 30720,
find
_ VQf)
00
-
0.03919 (backward from
*/Ov/
30682.38
a'
960
Using 0.04
we
we have
11.8315
75
= -3683
= + 68
as a sufficiently accurate value of
in
determining
a',
by (124),
'
.-.
F_ a
=
=
+ (0.48 X 68) = -3650

11.8315 - [0.03919 X (-3650)] =
-3683
11.8458
In the present example the algebraic signs of the several quantiNow it is important to
ties of (124) have each been considered.
remark that
in the majority of
cases no
need be given to
attention
for in this fact lies the chief practical
advantage of the
method. Thus, in the present example, we are interpolating from the
third function toward the second; the value of j' to be corrected is
these signs;
two functions, or 3683; the sign we disregard.

be applied to this number is 0.48 X 68, or 33. Again
the difference of these
The
correction to
neglecting signs, we simply apply this quantity to 3683 in such a

manner as to obtain a result falling somewhere between the numbers
3683 and 3615 of the column

thus
na
obtaining 3650 for our
= 143,
j'.
Hence, we decrease 3683 by 33,
corrected
first
difference,
a'.
Finally,
by which amount we increase the function 11.8315 (giving
11.8458), since we observe that the functions are increasing in the

direction of the interpolation.
mechanical method of procedure is to

that is, when J'
have opposite signs
partial exception to this
be observed when
changes
sign in
must be noted;
a,
and
a'
In this case the sign of a

passing the function
we then have, as in (122) and (124),
.
76

For example, given the values below
T

only the values for each midnight.
abridged series
Date
1898
Thus we obtain
the
77
following
78
method
at
once serves for the determination of the several values in
Thus,
question.
in the series
F F
us suppose that
and difference the series
let
t ,
and Fj are
F F
s
the required values are easily found
in error.
we may proceed by
given series; thus we obtain
from which the values of
= i and
|,
by
tabulate
interpolation.
and
require cor-
tabulating every third function

the abridged series
and
are
if
Otherwise,
respectively.
we
Again, when two adjacent functions, say

rection,
if
Then,
series are too large for accurate
of the
found by interpolating with

the differences of the latter
interpolation,
original table every alternate function only, as
we may omit from the

in
46.
The resulting
series,
FO>
will
-^4>
-ê
Fg,
....
but one incorrect value, namely

The cort
then be found by the method of differences, whereas
may
method might be impracticable
neously.
or,
therefore contain
rection to
this
Similarly,
we may
correct
if
by interpolation from the corrected
SYSTEMATIC INTERPOLATION
Thus
we have
t and
applied to
by the differences of
simulta-
series
SUBDIVISION OF TABLES.
considered interpolation as a process for

computing the values of functions for occasional or special values of
the argument, simply.
shall now consider the subject in a broader
48.
far
We

sense,
in a
and
find
that
interpolation
79
of great importance as applied
is
more extended and systematic manner.
When
a complicated function is to be computed and tabulated for

a large number of equidistant values of the argument, or when the
tabular quantities result from a long and laborious calculation, it will
be
much
make
shorter and easier to
the direct computation for a less
frequent interval than is finally required, and thence to obtain the

intermediate values by systematic interpolation. For example, suppose
the function
F(T)
is
700".43
to be tabulated for every 10'
computing
the values
F(T) for every 4th

of F(T) for T =
sin2r-l".19 siniT
from 30
to 60
degree of T.
22, 26, 30, 34,
70
we should begin by
Thus we should obtain
;
somewhat beyond the assigned limits

the interpolation which follows.
These quantities
the calculation being extended

in order to facilitate
having been differenced, and corrected for accidental errors if necesWe

sary, the middle terms are then found by interpolation to halves.
thus obtain the series
F(T)
corresponding to
26, 28, 30, 32,
....
T=
64
Interpolating again to halves, we have a table of F(T) for every

third interpolation to halves gives the function for
degree of T.
7
every SO
Finally, interpolating the latter series to thirds, we obtain
giving F(T) for every 10' of the argument T.

It is obvious that the labor of computation decreases rapidly with
each successive interpolation.
the required
table,
All of the extended tables
in
common
use, such as tables of loga-
rithms, sines, tangents, etc., have been subdivided in this manner, at a
saving of labor almost beyond estimation. In fact, interpolation has

undoubtedly done more for mathematical science than any other discovery, excepting that of logarithms.
The following
and precepts which
sections will be devoted to the derivation of formulae

will simplify the process of systematic interpolation
80
Instead of performing a separate and distinct calcu-
just described.
each interpolated function, we shall develop a method by

which the required values are obtained by successive additions of the
for
lation
computed differences of those values.

The most convenient interpolation to perform, either in an isolated
case, or as applied to the subdivision of an extended series, is interpo-
which gives the function corresponding to the mean

This case will
of two consecutive tabular values of the argument.
lation to halves,
now be
considered.
49.
"put
Interpolation
i,
Halves.
l)
BESSEL'S Formula (111), we

and JT vanish, and we get
If,
the coefficients of j'"
F F =a
Since
to
in
we have
F +F
Also, by (106),
we have
Hence, (125) may be written in the form
which
the formula for interpolation
is
inclusive.
on page
The
differences are to
to halves,
true to fifth differences
be taken according to the schedule
62.
Supposing that fourth differences are so small as to produce no

sensible effect, we obtain from (126) the very simple formula
(127)
true
to
third differences
midway between two
inclusive.
Hence,
to
consecutive tabular values,
interpolate
we have
function
the following
RULE
From
mean of
81
two given functions, subtract oneThe

eiglth the mean of the second differences which stand opposite.
:
the
the
To
result is true to third differences inclusive.

to fifth differences
of
add
inclusive,
the corresponding
of the mean
above result T f
the
to
obtain the value true
fourth differences.
The forePrecepts for Systematic Interpolation to Halves.

going rule applies either to the interpolation of a single function into
For the latter
the middle, or to that of an entire series of values.
50.
purpose, however, the
work may be arranged
in
more expeditious
manner, as follows:
For convenience, we assume
may be
S
'
= F,- F
if
accordingly,
neglected;
V = F, -
F,
S2
for
present that 4th differences
we put
'
the
F,
F,
V=
F,
- F,
...
(128)
we obtain from (125),
v(129)
The
defined by (128) are evidently the first differare

of the interpolated series ; the alternate terms, 8 ', 8 2', 8 4 ',
8'
qiiantities
computed by (129) from the

series of functions
the values of
The method and arrangement

below
first
8'
8,',
of the
8 3 ', 8 3 ', ....
work
are
8"
F
So'
F,
,'
8,"
8,'
82
"
8,"
F,
and second differences of the given
F(T)
>
are not computed.
shown
in the schedule
THE THEOBY AND PRACTICE OF INTERPOLATION.
82
The
of the given series are placed in the last three

The column a is then filled in by
J", and J'".
differences
columns, under
j>,
The
writing opposite each of the quantities J' one-half its value.
column ft is also computed, each term being minus one-eighth the mean
of the two values of
z/"
The
which stand opposite.
alternate quantities
are then found, as in (129), by taking the sums of the

corresponding terms in a and /8 ; the results are written immediately
of column
8'
above the line of the latter terms, so as to
and F,
fall
between jP and
f F
t
etc., respectively.
by (128) we have
Finally, since
F =
t
+ S,'
F,
F^Ft + SJ
F = F +V
t
....
(130)
only necessary to add each computed value of 8' to the function

immediately preceding, to obtain the required middle functions. Having thus completed the interpolation, the remaining or alternate values
it
of
is
are filled
8'
by direct differencing. The second differences are

the column 8", their regularity proving the accuracy
in
then written in
of the work.
The given
functions, also the computed first differences, etc., are

distinguished in the above schedule by heavy type.
When it is necessary to take account of 4th and 5th differences,
we have only to form an extra column

above. Under y we write the terms
3
(Ad\
~~~
128V~2~~;
of
the values
terms
in a,
/8,
8'
and
EXAMPLE.
.... 42
are
find
3
128
y, to follow /8 in
the schedule
(d
\~2~
'
then formed by adding the three corresponding
y.
Given the values of log
sin
for
T= 30, 32, 34,
the value for every degree of
from 32
to
40,
we arrange
the
inclusive.
In accordance with the method above outlined,
given functions, with their differences, as follows
83
84
in
which the argument interval
nun, or
is
the two adjacent
Now,
T-.
functions of this interpolated series, which, as a pair,

tween
and
are
,
that
ittitfu-itt/
be-
is
aild
'ro
'a?)
Hence,
if
we put
V = F
- F
{.TV
it
fall
follows that
lated
series
which
we now
If
series.
on the
falls
midway between
line
and
JP,
we
quantity a middle first difference of the required
this
shall designate
of the first difference of the interpo-
the value
is
8,'
(131)
(rr)
let
l
+m =
we have
T
1
and (131) becomes
OT
(132)
-M
V = *-*;_
(133)
Hence, to express 8,' in terms of the differences of the given series,

we have only to express the values of
_ n by BESSEL'S
n and
Formula;
we
thus, abbreviating coefficients,
F,
= F + na^ + Bl +
Cc t
have, as in (113),
+ Dd + Ee^ + ....
Also, by virtue of the property of these coefficients established in
(134)
41,
we have
F^ = F,+ (l-n)a
<
The
+ Bb-Cc +
l
Now, by
= F -F _n =
(132),
(2n-l)a -^2Cc
1
hence, from (111),
E =
Jn(n-l)(n-i)
T i T (n
+ 2Ee + ....
l
we have
H
ei
(135)
difference of these equations gives

8,'
Dd-E + ....
we
+
-T-
771
find
^ (i*-l)
+ l)n(-l)(n-2)(n-i) =
& (n
(136)

Substituting these values of
n,
C,
and
in
(136),
we
85
obtain the
formula
by which the middle

is
provided
first
a positive
differences
odd
may be computed
integer.
Let us now consider the schedule below
in
any case,
80
F(T)
interpolating the given series
thus virtually reduced to that
is
of interpolating the computed values of 8' in precisely the t*<nn<- nnnnn-r.

"
Now, let 8 denote the second difference of the interpolated series
series
and
the
8',
8,';
the second difference opposite F^,
8,",
"
the middle first difference of the interpolated

which falls between S'_, and S'
8,", that falling between
It follows that 8
etc.
8,'
which stands opposite
is
Hence, we may
and so on.
computed
series
8'^,
8,',
8,',
find
.
that the latter quantities are derived from
", 8,",
in
F_ l}
82",
precisely
from
the
manner
F F ,....;
,
that
by application of the general formula (137), mutatis mutandis. For

this purpose, we must form the differences of the computed series
is,
S /
"J
SI
-J
Accordingly,
let
S
"J
>
us put, for brevity,
'-lXw'-O)
and (137) becomes

S4
'
mih +
Cl
M'ei
provided differences beyond /f are disregarded.

of the quantities
8'_ s
Function,
8'
k ',
8
S
',
.,
and
(138)
(139)
We
now form
a table
their differences, as follows:
87
'"
of the interpolated series F(T). Finally, we shall denote by 8

the
third difference of the interpolated series F, which stands opposite
The quantity 8"' is therefore the middle
the given schedule.
8,' in
4
first
difference of the completed series 8", which falls between 8
bears the same relation to
8,";
it
Hence, to find
The
differences of
Function,
= 8"
8,", that 8
8 "
S"_,
and
bears to
"
and
and
us put
8"', let
and (140) becomes
80"
'
(141)
"
,
8,"
are therefore as follows
88
II
which
serve
IK

reduce the tabular interval
to'
to
times
its
original
being the reciprocal of a positive odd integer. It will be

observed that the differences required in computing each of the quantities
value,
same
8 are always found on the

52.
Interpolation to
in the formulae
line
For
Thirds.
(143), and
with that quantity.

this purpose,
we
take
.-5I3 e
(144)
These formulae are more conveniently computed
in the
form
= m-A<*o)
v = K.-S;")
of
from 27
between 33
According
work
as follows
Given the value of

to
inclusive
to
the precepts
48,
and 42.
find
v ==
V
*i-
EXAMPLE.
(145)
log tan
find
of the
for
every third degree
the function
last
section,
for every degree
we arrange
the
89
necessary to compute five values of 8'", four values of 8", and only
These quantities are computed to one more than the
three of 8'.
number of decimals given
in
F(T),
to
the final additions.
appreciable error in
avoid accumulation of any

Having obtained for 8'" the
series
+ 3.1
the
terms
intermediate
necessary, however, to
are
see
2.6
2.3+2.2
+
+
2.0
If these
relations
lated values of 8'"
The column
1.9
+1.9
1.9
readily inserted,
that
with the computed values of

2.8
2.2
2.6
the completed
=
=
=
2.0
1.9
series 8'"
is
it
is
consistent
Thus we must have
8".
2.5
shown above;
as
-(18.0-25.9)
-(11.5-18.0)
-(5.7-11.5)
=
=
=
+7.9
+6.5
+5.8
not satisfied exactly on first trial, the interpomust be adjusted to fulfill the necessary conditions.
are
8"
now completed by
is
successive
additions
of the
necessary to see that the completed series

For we must have
8" agrees with the computed values of 8'.
8'".
quantities
Again,
it
is
-(20.5 + 18.0 + 15.7)
Since these
are
relations
1569.6-1623.8
seldom exact
in
-54.2,
the
etc.
beginning,
the
pro-
visional values of 8" will usually require slight alterations.
by successive additions. As
before, an agreement must subsist between the values of 8' and the
Thus we should
that is, between 8' and //.
given set of functions
From
the final series 8",
we
obtain
8'
have
2TS'
1646.9+1623.8
1603.3
+4874.0
J', etc.
In the latter case, however, a discrepancy not exceeding four or five

units in the added decimal may be tolerated.
Our final series 8' is
therefore satisfactory
required values of
53.
(143),
whence we obtain by successive additions the
log tan T.
Interpolation to Fifths.
we
Taking
-\
in
the formulae
obtain
= &(V-A*)
V =
(146)
In practice it will suffice to put \e for both

formulae (146) then become, very approximately,
l
v'êj
and
THK THEOKY AND PRACTICE OP INTEBPOLATIOK.
90
= AP.-&4,)
V =
V'
EXAMPLE.
every
ten
Sept. 23
The
(147)
The following ephemeris gives
hours.
Obtain the
value
for
the
every
second
20" to Sept. 25' 12", inclusive.

details of the
Date, 1898
computation are as follows
moon's R.A. for
hour,
from

Here we extend the computation of 8'" and
mals
one of which is dropped in computing
8" two places of deci-
same as
in the last
Successive Interpolation
to
to
Halves,
is
unnecessary.
when a Series Requires
Folloiv,
Thirds,
in
and method being the
principle
example, further explanation
Order of Interpolation
54.
The
and the other
8',
forming the required functions.
91
etc.
When
table
of
be interpolated, successively, one or more times to

halves, and also to thirds and fifths, the easiest method is to proceed
in the order named.
Thus, if the interval of the original series is w,
functions
is
to
and that of the
final table is &/,
we may suppose
the relation of these
quantities to be
first,
and
then be found most expedient,

then to thirds, I times
to interpolate to halves, k times
and
where
X',
/,
are integers.
It will
finally to fifths, in times.
For example,
being given for every degree, and required for

every minute of arc, we should first interpolate to 30', then to 15',
then to 5', and finally to every minute of arc.
To
55.
Interpolate ivith a Constant Interval n,
of Functions.
follows
Let
the
given
series,
with
its
an Entire
differences,
Series
be
as

T
<p
93
94
EXAMPLES.
Tabulate the five-place log cosines of 15, 18, 21, 24, 27, 30 ;
17 43', 23 8', and
for
from these values interpolate log cos
1.
28
15',
2.
respectively.
Given the following table
T=
95
the
series twice to halves; the first result to include the

values from July l d .5 to 4 d .O, and the final three-hour ephemeris to
extend from July 2 d O h to July 3 d 12", inclusive.
Interpolate
7.
third
The ephemeris below
day
1898
gives the sun's true longitude for every
96
logarithms of 47, 54, 61, .... 96, by computing the logarithms of

the consecutive numbers between 61 and 75.
Show
formulae (143) were extended to include

the middle differences of order i, we should have (using the symbolic
11.
that
if
the
form of notation employed
in
in the
analogous formulae (64))
which
may be
either
odd or even
symbolize the tabular differences which

line with S'.
and where
fall
z/',
//'+*,
J l+t
....
upon the same horizontal
CHAPTER
III.
DERIVATIVES OF TABULAR FUNCTIONS.
56.
It
differential
is
often required to find certain numerical values of the
of functions either
coefficients
analytically
unknown,
or
In the majority of such cases the function

complicated
has been previously tabulated for particular (equidistant) values of
the argument.
The required derivatives are then readily computed
in expression.
from the differences of the tabular functions.

with certain limitations
have already seen that
We
particular
values of a function, with their differences, practically determine the

character and law of that function, thus enabling us to determine
intermediate values
by
The trend
interpolation.
of the function being thus defined
by
its
or law of variation
differences,
it
is
but natural
to suppose that the successive derivatives are quantities closely related

to
these
differences
since the
derivatives
are themselves
direct
in-
dices of the character of variation of the function.
Practical Applications.
The most useful application is in
finding the change or variation in F(T) corresponding to an increase
of one unit in T, supposing the rate of change in
to remain con57.
stant from
for
to T-j-1,
this
is
and equal
simply the
quantity
with respect to T, which
we
to the actual rate at the instant
first
shall
differential
coefficient of
T;
F(T)
denote by F'(T).
For example, having observed that a

sixteen feet during the first second of
its
freely falling
body describes
descent, forty-eight feet the
second second, and eighty feet the third, its velocity at the end of
two seconds is easily found to be sixty-four feet per second. This
velocity of sixty-four feet is nothing more than the first differential
with respect to the time, computed for the inthe space which would be described during the third
coefficient of the space

9
stant 2 .0
it
is
98
second, supposing the action of gravity to have ceased at the end of

the second second.
The most frequent and important
applications occur in
Astronomy.
An
astronomical ephemeris contains a great variety of tables giving the

positions and motions of various heavenly bodies, and of certain points
of reference.
From
the given positions, tabulated for every hour or from
day to day, are derived the motions per minute, per hour, or per day,
according to circumstances. For instance, the Nautical Almanac gives
the
declination
sun's
The hourly motion

computed from the dif-
every Greenwich noon.
for
in declination (also given for every
noon)
is
its value is the differential coefferences of the tabular declinations

ficient of the tabular function at the date in question.
:
In the following sections the various formulae employed in computing the derivatives of tabular functions will be derived.
Development of the Required Formulae in General Terms.

The variables T^ and n are connected by the fundamental relation
58.
T =
which
and
<j
(150)
Accordingly,
we
In like manner,
we
constants for a given series.

have hitherto written, under varying circumstances.
in
are
<o
F(T)
as
the
lowing equivalent forms
*<*)
When
successive derivatives of
F(T)
by the
fol-
= F<
= F'
(T)
F(T)
5^1
F(t + n)
of the same quantity.
equivalent expressions
shall hereafter denote
f(T)
=
I
F "( T
F"
= t
(151)
convenient to proceed baclowarda from the argument

with the interval n, we shall use the expressions
FL =
a
it
F'(t-nu)
is
F'l>
F"(t-n)
F'l'n
F'"(t-n^)
(152)
THE THEORY AKD PRACTICE OF INTERPOLATION.
Now, by means of any one

polation, we may express Fn in
F =
n
-F.
where, in any given case,
and where
J3,
D,
C,
successive derivatives of
of the fundamental formulae of interthe form
+ Bb +
na
+ Dd + Ee +
Cc
a, J, c,
S,
C, I),
(153)
known
are
differences;
functions of n.
are definite
99
Let the
taken with respect to
/?,
be denoted by
J)l
D
ri
\s
D'
E'
"Till
Kill
fit
\s
**
>
fin
\s
D"
E'i
D'"
E'"
Then, observing that the coefficient of J w

in n,
is
we have
*<1
dn
d* B
'
//
'
C"
dn 1
dn>
'
'
^'
dn*
dn*
'
dn
C'"
D'
D"
^-
dn
E"
E">
(154)
dn*
we have
F'
we
d*E
5
(150)
E'
th,
dn 6
Reverting to (151),
^=
Z)'"
'
dn*
dn*
From
always of the degree
dn
dF.
dT
'
'
dn
dT
(155)
derive
^
(156)
dT
whence
F'
dF
<a
.
'
dn
(157)
100

In like manner
VII
we
obtain
101
Introducing these changes, and operating as before, we obtain the

required formulae, namely,
F'_ n
= .!(..
F"
-*-
C'y
- D'S + E't -
(161)
now remains
It
of interpolation, of which (153)
several formulae
the
It is
type.
C", C",
apply (159) and (161) specifically to each of
to
obvious that a particular set of coefficients,

etc., will result in each case.
is
the
general
B', 23",
To Compute Derivatives of F(T) at or near the Beginning

The formulae adapted to this purpose are derived from
of a Series.
NEWTON'S Formula of interpolation (73), which is
59.
F =
(162)
where
n(-1
n(n
re
2
l)(n
2)
n8
=
~Ji~
re(re-l)(re-2)(M-3)
w2
_=
~^~
?i
~^ +
rit_
_i
~
(1G3)
"
24
24
120
12
+ 24
Differentiating these expressions successively with respect to n, as
indicated in (154), and substituting the resulting values of B', B",

in the general formulae (159), we obtain
etc.,
C", C",
.
THE T1IEOHY AND PRACTICE OF INTERPOLATION.
102
F>
*-+ *).+
(n-l) Co + (j'(164)
These formulae determine the derivatives of F(T~) for any or all

values of T between t and t-\-a>, according as we assign different
As
values to n.
in
preceding applications, n
is
always a positive
proper fraction.
frequently the case, derivatives are required for some
tabular value of the argument, say t, we have only to make n
as
When,
in (164)
we
is
thus derive the following simple expressions
F'(t)
F(t)
(165)
The
employed in (164) and (165) must be taken

according to the schedule on page 3, as in direct applications of
NEWTON'S Formula.
differences
The formulae (165) have
18;
for
it
be observed that (45) and (165) are identical, since in the former
Z> 2 Z> 3
are used symbolically to denote
<*!"(), <JF" (t),
will
Z>,
already been established in
Owing
to the special practical importance
the coefficients of
B<
of the first derivative,
namely,
F'(t-\-n<a),
= n-
103
D>
_
(166)
have been tabulated in Table
IV
for
argument n. By means of these

from the formula
F'(t-\-n<a)
the
every hundredth of a unit in

quantities,
we
readily compute
(167)
The formulae
(164), (165), and (167) are especially adapted to

the computation of derivatives at or near the beginning of a tabular
series.
We
shall
EXAMPLE
compute
I.
F"(T)
T
now
solve a few examples to illustrate their use.
From
for
the following table of
T = 2.S.
F(T) =0.3T
2T +4,
2

This result
the function
easily verified
is
thus, since
F(T)
we
from the known analytical form of
0.3T*
22"
derive
F'(T)
T= 2.8
Substituting
-4T
1.2T 8
F"(T)
in the last equation,
F(T) =
we
3.6T 2
-4
obtain
+24.224
as found above.
EXAMPLE
F'(T)
From
II.
table
of
the
last
example, compute
T=Q.
for
Here we employ the

a
the
-3.2
first
of (165).
+51.2
Making
+172.8
0,
rf
we have
+115.2
AVe therefore obtain

F'(f)
The
result
is
(-3.2-^+HM-p) =
obviously correct
F'(T)
which vanishes for
EXAMPLE
compute
III.
F'(T)
T
for
we have
= 1.2T'-4T
T = 0.
Given the following
for
T=83&.
table
of
F (T}
EE sirfT:
we
Taking the coefficients 13',

and the differences
0.15,
C",
,
D and E'
b
....
from
=
B'
1
=
=
-0.35
+0.19458
[>
= +8891
= - 638
L
C
IV
from the given
accordance with (167),
find, in
Table
105
+0.023858
with
table,
10G
Comparing (168) with the general formula (160), we have
Therefore,
If, B",
we
the
substituting
C', C",
previously determined values of

in the general formulae (161),
etc.,
obtain
;-&'+ *'-* + i) e -6-
_,--_4
Making
w.
(0
==
in
(169) ,
-2n+
169 )
we have
-,
<0
(170)
As
ative
in
above,
we emphasize
practice
the relative importance of the first derivthus, for brevity, we write the first of equations
(169) in the form
F'(t- nu,)
the coefficients
B',
-(a_ -B'/>_3 + C'c^-D'd^ + E'e^l
C", IX,
E'
being taken from Table
argument n.
Formulae (169), (170), and (171) are
(171)
.)
IY
with the
particularly useful
in
the
computation of derivatives at or near the end of a series of functions.
THE THKORY AND PRACTICE OF INTERPOLATION.
107
Moreover, when the interval n approaches unity, formulae (169) and

(171) are convenient for computing derivatives corresponding to the
argument
argument
-\-nia, since
examples to
-\-
<a
with the interval 1
I.
From
ascension (a), compute

20h 24m
Date
1898
n.
We
shall
now
solve several
illustrate these applications.
EXAMPLE
.
they enable us to proceed backwards from the
the following ephemeris of the moon's rightthe hourly change in a at the instant Feb. 3 d
108
EXAMPLE
From
II.
the hourly variation in
the preceding table of moon's H. A., compute

for Feb.
J^
3'
12 h ; where, as above, J,
the change per minute in R.A.

Regarding one hour as the unit of time,
of F"(t) given by (170)
is
sixty times
it
is
denotes
clear that the value
the ex-
the quantity sought:
pression for the required variation is therefore ^\ F"(f), where

h
3' 12
Accordingly, using the second of (170), we find
<
Hr. Var. in J^, Feb. 3 d 12 h ,
EXAMPLE
find
(~ 12
62
- 5 37 +
'
H X 0.62 + | X 0.54) =
Given the following values of
III.
for
F'(T)
I =75.
7
-0-.00196
F(T)
i~\oge T:
The
true value of this quantity
is
109
-0.00022.27
(67)'
61.
Derivatives
from STIRLING'S Formula.
When
differences
both preceding and following the function J?(t) are available, formulae
more convenient and accurate than the foregoing may be employed.
The most useful and important of these are derived from STIRLING'S
Formula of interpolation (104), which
Fn = F +
na
Cc
is
Ee
(172)
where the differences are taken according to the schedule on page 62,
and where
a, c, and e being the mean differences defined by (101)
have the values
J3, (7, ....
;
THE THEORY ANB PRACTICE OF INTERPOLATION.
110
Making
in
(174), the latter become
(175)
= -*(-
(o
Again, writing
Fi (tn<a)
F'"(t-n<a)
= Vfe-1
O)
in (174),
for
we
obtain
(ai
\
(176)
(-)
The
= -
coefficients for the
B'
computation of F'(tna),
namely
n
(177)
with the argument n. The quantity F'(T} is

thus readily computed (for any value of T) by either one or both of
the formulae
are given in Table
F'(t+no>)
F'(t-no>)
+ E'e)
(178)
(a-n^+C'c-D'd^+E'e)
(179)
'rf
which the odd differences are algebraic means of the tabular

ences, taken as indicated below
in
differ-
Ill
112
in the
form
F'(t+n) =
Substituting
(.0
F"(f),
(t+noi)
in
these
as given
equations
the
expressions
for
a>F' (f),
by (181a), we get
i r
(a-
(182)
']
(t+nv))
^[K-
)+(-
)+
These expressions, upon being arranged according to the successive orders of differences, will be found identical with the formulae
the present form is more con(174). For some purposes, however,
venient.
It is quite
common,
tabulate the values of
F(T}.
Such a
table
an astronomical ephemeris, to
corresponding to the tabular values of
particularly in
F' (T)
would run
T
as follows :*
The
113
of the formulae (175) is almost invariably used for this

purpose, because of its simplicity and rapid convergence; this formula
first
most important and useful of those which pertain to

the computation of derivatives.
For this reason we formulate the
is,
in fact, the
following
RULE
computing the
for
first
derivative
of a tabular function
From the mean

corresponding to one of the given functional values
of the two first differences which immediately precede and follow the
:
function in question, subtract one-sixth
Q)
the
mean of
the correspond-
ing third differences, and divide the result by the tabular interval.
This rule neglects only 5th and higher differences. To include 5th
and 6th differences, add to the above terms (before dividing by w) onethirtieth (g 1^) the
by
mean of
the corresponding fifth differences,
and divide
as before.
a)
It will evidently suffice, in
most
cases, to apply only the first part
of the above rule.

Several examples will
now be
solved as an exercise in the use of
the preceding formulae.
EXAMPLE
nation (S)
10, 13,
Date
ISO*
and
I.
Given the following ephemeris of the sun's
compute the hourly

16.
difference
in
decli-
for the dates Jan. 7,
114
of
sixth
<a
the
= 72 hours,
corresponding mean
we
by summing the
We
differences.
Finally,
since
write in the last column J% of the quantities formed

corresponding terms of the two preceding columns.
thus obtain the hourly differences required.
EXAMPLE
the
third
(I
II.
a II i/ motion in
Compute, from the ephemeris of the last example,

h
m
declination for the date Jan. 6' 13 30
1
We
proceed backwards from Jan. 7, using the formula (179), and

with the argument
taking the coefficients from Table
--
7 d O h O ra -6 d 13"30 m
Thus we
find
10.5
'
14583
EXAMPLE IT.
for
T=
From
the table of
and
= 0.62,
Example
III,
compute
115
F" (T)
0.462.
Taking
= 0.4
)>
we
obtain,
by means of the second
of (174),
J
+0.014930
116
F'
/""(+) =
ft
Putting
(t)
in (185),
(185)
we
get
=
(186)
Again, putting
formulae
in
(185),
we
obtain the following simple
(I)
(187)
which determine the derivatives of

the tabular values
The
of the function.
coefficient of e l vanishes.
F(T)
It
is
at
points
midway between
important to observe that.

unless
F'
third
(t-\-\
are
differences
*"(<
which
differs
close
approximation
to
given by the simple expression
is
w)
considerable,
117
from the
e^
^ ^
+ *) = ^ =
i,\
^
(187o)
f
i.i
formula only by the omission of the small
quantity
;<-*>+ ---The formulae
for
the derivatives of
(t
are deduced from
n<a)
Let us put, for brevity,
(111).
i(i e +*0
= *K+<O
<*
( 188)
and (Ilia) becomes

F_n
Comparing
that
this
a, ft, y, 8,
I\
na'
expression
,....,
+ Bb-
+ Dd -
Cc>
Ee>
(189)
with the general formula (160),
in the latter, are replaced
by
a', b,
c, d,
we
find
e',
....
hence, observing these changes, and substituting the above

(189)
determined values of J?, B",
C', C",
etc., in the
formulae (161), we obtain
in
_L("
T Vfl
n"
15>"34
(190)
The values
sions
of B', C', I?, and E', as
-"-f -l + A
computed from the expres-
^-I'-l'-ft + A
^'
= fi-f4 + ft-Ti T
118
THE THEORY AND PRACTICE OF INTKUPOLATION.
VI
are given in Table

efficients,
with the argument
F'(T)
values of
are
readily
n.
By means
computed from
of these co-
either
one of
the formulae
'
(t
o>)
F' (t-nw)
in
+ IVb + C + D'd + E<e.^
(192)
^(a'-B'Z+C'c'-D'd + E'e'}
(193)
(a,
'c,
which the even differences are means, taken as indicated below

T
EXAMPLE
F"(T}
We
for
we
the preceding table, compute
the value of
T=4A48'.
take
of (185),
From
II.
119
= 44
hence n
= 0.40.
Accordingly, from the second
obtain
= -0.0008614
= +
30
C"'c,
l>"cZ = 2
= -0.0008586
F" = -0.70465
b
C"
Z>"
= w-A = -0.10
= f - - ^ = -0.203
c,
= -300
= + 11
.-.
The
actual value
is
F"(T)
EXAMPLE
III.
= -sinT =
-The
table
-sin 44
48'
Date
1898
-0.70463
below gives the Washington mean time
of moon's upper transit at the meridian of

U'ASHINGTON
Washington
MOON CULMINATIONS.
120
retardation to remain constant between the
what
moment
two
transits,
and equal to
Thus, if Z) is the value of

for the instant of transit at Washington on Mar. 24, the local time of
it
is
at the
of the
first.
moon's transit at a station 20 minutes west of Washington

with sufficient precision by the formula
r
Now, by
the
first
D =
t
Mar. 24" 1" 47 m .29
of equations (186),
F'(f)
j t (47.59
we
-+
is
given
Z>
find for the value of Z>
-TT) =
l m -954
Hence the preceding equation gives

T
Mar. 24"
47"'.94
simply and accurately

determined for any number of stations within half an hour of the
In this manner the local time of transit
is
Washington meridian.
To
of moon's transit over a meridian 3 hours
find the local time
west of Washington, on the 24th day of March,
we have
only to interpolate the Washington time of transit between the tabular values
for Mar. 24 and Mar. 25, as given above, the interval from the former
being
3h
24
were required to compute the local time of transit

for several stations whose longitudes range from 2| to 3^ hours west
of Washington, we should find the time for the 3 hour meridian by
Finally, if
it
as
direct interpolation,
D = F'(T)
the local time
for the
explained
same meridian;
is
3'"'
-f-
that
"" 1
,
is
TJ
is
is,
should
for
also
compute
= 0.125.
Then
given by the simple formula

A
where
We
any adjacent meridian, whose longitude
of transit at
from Washington
above.
the time of transit at the 3 hour meridian.
EXAMPLE IV
From
the preceding ephemeris, compute the difference in time of transit for 1 hour of longitude (D) at the instant of

moon's transit over the meridian of San Francisco, Mar. 25, 1898;
h
m
s
3 h .025.
longitude from Washington being taken as 3 l 30
121
the
Here we use the formula (192)

thus, taking the coefficients from
Table VI (with the argument n
3.02a -^ 24
0.12604), and the differ:
ences from the given ephemeris,
we
obtain
B'
+48.95
THE THEORY AND PKACTICE OP INTERPOLATION.
122
We
shall
assume that the differences of
F(T)
lT
beyond J may be
disregarded; hence the differences of F'(T) beyond y may be neglected in the above schedule.
Now, by TAYLOR'S Theorem, we have
F =
m
F,
nF <+
F"+
F +
F +
>
(194)
Again, since
dF'
~dt
we
obtain,
cPF'
7f HI
II
~W
'
Jf IT
'
d*F*
~df
by means of the formulae (175),

v
L
-,
in
which we have put,
for brevity,
Substituting these expressions for

latter
F F
",
'",
and jP iv
in (19-4),
the
becomes

(197)
By means
of
F'(T),
of this formula
F in terms of the differences

of F (T) direct, as in the usual
we compute
instead of the differences
formulae of interpolation.
Substituting
F_ n
The
for
in
(197),
= F - n*
9
(/-.'-?
we have
+ f ft- A (f-1)
y)
(198)
values of
BEE"
2
,
r=
T\(J
-l)
(199)

are
given
coefficients
Table VIII with the argument
in
we
readily
By means
n.
123
of these
compute
(200)
W-l
The
coefficients in
since by this
method
the nearest function
EXAMPLE.
it
From
(201)
Table VIII are not extended beyond n

0.60,
is invariably more convenient to proceed from
.
the
American Ephemeris
for
1898 we take the
longitude of Mercury, together with the daily motion in

The differences of
longitude, for a portion of the month of October.
the daily motion are then taken, as shown below
heliocentric
Date
1893
124
Differencing the given series of longitudes and applying BESSEL'S

Formula of interpolation, we find
F.
64.
194
15' 18".2
Application of the Preceding Method of Interpolation when
Second Differences of the Series F(T) are Nearly Constant.

When the 3d and 4th differences of _F(T) are small enough to be
neglected, we may omit the terms containing /3 and y in the formulae
we therefore obtain
(197) and (198)
the
FH = F +no>(F +
'
F_,
It will
= F
n<a
(FJ
(202)
% a)
(203)
be interesting to determine the error of these approximate
formulae as applied when the 3d differences of

For this purpose we write (197) in the form
F = F+
n
(/'
1 a)
+f
if
we
=
Now, from
(175),
(f ,
- T"i)
are appreciable.
*>7
F(T), and
thus neglect
(204)
0)y8
we have
= - =
F'"(f)
also,
j8
disregard 4th differences of

follows that the error in question is
Hence,
y, it
wo.
F(T)
from (195),
Whence
w /3
J"
(205)
and (204) becomes

e
Since in practice the

the
maximum
maximum
J'"
value of n
(206)
is
0.50,
it
follows that
error resulting from an application of the formulae (202)
and (203), when 3d differences of
F(T)
are sensible,
is
tV^'"-
Hence,
even when third differences are considerable, these formulae are sufficiently accurate for
many
purposes.
125
That the formulae (202) and (203) are rigorously true when the
3d differences of F(T) are zero may be clearly shown from geometrical considerations, as follows
The 2d differences of F(T) being supposed constant, it

from Theorem VI that the function is necessarily of the form
follows
F(T) = a
Now,
the accompanying figure we draw the rectangular coand 01", and plot the curve denned analytically by
if in
OT
ordinate axes
(207)
parallel to
evident that
is
it
= F(T)
(regarding
abscissa T),
(207)
as the
we
ordinate
corresponding to the
a parabola whose
obtain
axis
is
OY.
Let us now take
OH =
OS = +
ON = + n<a
t
<D
Whence
MN
n<a
= F
NQ = F(t+ nta =
MP =
F(f)
Draw
the tangents
PA,
dF
Then, denoting
F,
-^ by
QL
n ',
draw
PD QL and PB MN.
\\
=
=
FJ
te
tanZPZJ
find
NA = MP + PD tan APE = F +
= MP + PB tan DPB = F +
evident that to find
It is therefore
NA
\\
we have
FJ
Hence we
also,
no>F '
n<aFn
NQ = F
>
which
lies
between
ND, we must employ a value of F' somewhere between the

F and Fn Now, let KE be the ordinate erected at the mid-
and
values
'
dle point of
'.
MN,
and
EII
the tangent at
E.
Then, by an elementary

theorem of the parabola, the chord
PQ
is
parallel to
EH,
and we
have, therefore,
F = NQ = MP + PB tan QPB = F +
n
na>F^
(208)
which agrees with the formula (202).

We have shown above that the maximum error produced by applying this formula when the second differences of F(T) are not constant,
is ^V//"'-
Hence, unless the 2d differences of
we may compute
RULE
derivative
by the following
Find by simple interpolation

which belongs midway between the
:
nearest tabular function
(-F'o)/
nation (8)
I.
1898
value
(T
t),
of the tabular
required function
and apply
(F'n)
the
and
for the date
July
9'
product
tJie
by the units
to
Given the following ephemeris of the moon's
compute the value

Date
the
multiply this quantity
contained in the entire interval
EXAMPLE
are considerable,
F'(T)
18 m .O.
decli-
127
Whence
3
24' 37".4
102
13".626
which substantially agrees with the above
EXAMPLE
1'
result.
From
the following table of the moon's horizontal

m
interpolate the value for July 10' 16" 24 .O.
II.
parallax
(TT),
27".55
Date
189S
128
EXAMPLES.
1.
Given the following
table of "Latitude
Reduction"

Date
1898
129
CHAPTER
IV.
OF MECHANICAL, QUADRATURE.
We
66.
have shown
in the
preceding chapter that when a series
of equidistant values of any function are known,
it
is
possible to
com-
pute special values of the first and higher derivatives of that function,
without regard to its analytical form.
shall now consider the in-
We
verse problem, namely
From a
series
of tabular values of
F(T),
to
Jfind
X=
where the limits T' and
CT
\F(T}dT
T"
are numerically assigned.

The solution of this important problem is effected by integrating
the expression for
as given by any one of the several
F(t-\-n<a),
formulae of interpolation, and then
giving to n the limiting values
which correspond to T' and 'T." The method is wholly independent

of the analytical form of the function F(T).
It is therefore of
especial advantage
(a)
When
and importance
in the following cases
the function is analytically
unknown.
This
is
the case
with graphical records of continuous observations, so frequently made

in physical experiments
and
tests.
As
common example we mention
the indicator diagrams of a steam engine. It is usually required to

find the area comprised between the "pressure" curve, a fixed base
line,
and two extreme ordinates.
generality of cases,
(&)
When
of the
Under
method
where
e is
grdble.
This
area
may be
found, in the
by the method proposed.
analytically known, but is non-intethis head are included the most important applications
the function
in question.
is
For example,
numerically given.
We
let it
be required to find
cannot express the indefinite inte-

gral in finite form.
a
If e
is
e sin T)-* in a
expand (1
and integrate each term of
sufficiently small
(say e
= 0.1),
131
we may
of ascending powers of ea sin 2 T,

a very few
this expansion separately
series
terms will then suffice to compute
X as
accurately as
may
be required.
= 0.9),
this
however, the quantity e is nearly equal to unity (say e
series does not converge with sufficient rapidity for practical use, and
hence the method of expansion fails.
If,
On
the other hand, given any value of e not exceeding unity, we

2
a
can readily tabulate F(T)
for a series of values
e sin T)-*
(1
such as
20, 24, 28, .... 52. Having differenced these values
T=
of F, it is then a simple matter to compute

from the numerical
data thus furnished. In the nature of the case, however, the process
must, in general, be an approximative one; depending, as does the
method of
interpolation,
upon a limited number of (usually approxi-
mate) values of the function in question.

The process by which the definite integral of a function
is
com-
puted from a series of numerical values of that function, is called

mechanical quadrature, or numerical integration.
proceed to develop the formulae which are commonly employed for this purpose.
We
67.
lation.
Quadrature as Based upon NEWTON'S Formula of InterpoSuppose that -}-l values of F(T) have been tabulated and
differenced as
shown
T
in the schedule
below
132
Since
T =
we have
,_ _
=
dT
n<a
-I-
(210)
<adn
and therefore
/(
F(t + nu>)dn
F(T)dT
Xl-HU
(211)
o>J
Now, by NEWTON'S Formula, we have

nJ + BJ
'
"
+ CJ '"
t
where
S,
order.
Multiplying by dn, and integrating,
(7,
D,
dn
nth
denote the binomial coefficients of the
we
obtain
C(F + nJ + BJ "+CJ '"+

>
l>
.)dn
or
where
M
/3
= CBdn
Cbdn
Whence we
x/
'
/?z/
....
(213)
+ ....
(214)
derive
" 4-
y^
'"
8//
lT
= CFy + i^
/0
Summing
the integrals expressed in (214) and (215),
The numerical values of
we
find
r=0
r=0
coefficients
we
/i
Ddn
^ o
F(t + n<a)dn
f^(< + no))dn
we put
r.F
*
/*
*/(
= ^+
-i-
for brevity,
If,
/o
then, from the preceding equation,
|^( + wo))rfw
-i-
*)
the constant of integration.
is
*ô
(sometimes called the

of quadrature} must now be determined. These may be
y8,
7, S,
133
found directly by integrating the expressions for S, C, D,

as expanded in (163), and then taking the limits of n according to
.
(213). But the following indirect method seems

adds a significance to the result. Let us put
=(l + y)"dn =(l+n
Q
where y
is
supposed constant.
+ B,f+Cy + D,/ +
Then,
if
we
&f
preferable, since
dn
it
(217)
also put
Q>
we
i/
have
shall
Q>
= H-ly + py' + yy +
l
&y*
& +
6
(218)
the coefficients being those defined in (213).
Again, put
(1+y)that
(219)
is
log(l + y)
and we
logs
find
log (1
+ y)
dn
or
*dn
We
Q =J
(220)
log(l + y)
,
therefore obtain
(1
Whence
+ y)
dn
>
zdn
|
Jlog(l+^)
__
y _j' +
Expanding the
direct division,
we
!_!+...
last
"
+ const. = ^
log(l + y)^
\og(l
+ i/y+ const.
_._._
2345-
expression by the Binomial Theorem, or by
obtain
6
Tgffy
-^li
Whence, comparing (218) and (221), we
iJ
(221)
find
T&ff
-m*
(222)
134
coefficients of formula (210).
which are the numerical values of the

appears that the fundamental
those in the expansion of [log (1+y)]"
It therefore
coefficients
of quadrature are
Let us now regard the functions

differences of an auxiliary functional
'F.
F F F
,
series
1 ,
which we
Ft
as first
shall designate
A schedule containing the new series may be conveniently arranged
as follows
J,
135
This formula possesses the disadvantage of involving differences

which are not furnished by the foregoing schedule.
4,", di'", ....
To
we proceed
obviate this difficulty,
as follows
Put
'F
+ kF +
$A>
84
yJ
(225)
and (224) may then be written
= q-'
dn
giving to
Upon
3,
4,
IF
/I
'
(226)
formula (75), the values
n, in
successively,
-\-~L,
0,
1,
2,
obtain
IF
=A
we
+ jd"
//'" 4-
/J
lv
/f1
+
(227)
expressions be substituted in (225), we shall have q in

terms of the known tabular differences, and hence obtain the required
If these
To avoid the labor of numerical reduction inciintegral from (226).

dent to this substitution, we derive the result in the following indirect
manner
e
Put
x- l
log(l+x)
lx
l>
px
+ rx
+8x* + &* +
&*+
(228)
Also, take
(229)
1-u
and we have
(1
M)-
+ II? + U*+
U4
M6
(230)
130
If
now we
these expressions
substitute
or
for
#,
ic
x,
second member of (228), we obtain in terms of u~ , u, u, ?/ ....

Bxit it will be observed that this operation is identical in algebraic
a
in the
form with the substitution above proposed with respect

(225)
to (227)
and
for the 6 operation involves the quantities
i*!
X
tr**
v
X
T3
X
1-*
X
it~1
U
tfl
U
U a , 4/
U8 ,
U, ?/
tt
,
while the q operation involves, in precisely the same algebraic relations,

the quantities
Fa a
A
'F
fit
n
?>
Hence the
ff
'
t i
/I" a
A"
a
t
t.
>
>
PA'
a
IF
-^f+ii f u
A"
<-i> ^t-ts
immediately follow when the result for

But we may obtain 6 as a function of u, in the
result for q will
has been derived.
form required, more simply than by direct substitution of the expressions (230) in (228).
For, by (229), we have
log(l+:r)
Therefore, by (228),
we
(231)
find
Su'-^+^M 6 -
Accordingly, writing q for

as justified
-log (1- M )
T=-Ilog n
(1-M)
r
log(l+a;)
6,
'Fi+l
for
u~
l
,
F for w,
_1 for
/j' i
(232)
u, etc.,
by the preceding reasoning, we obtain

>Fl+l
- F + (U'^ - yJlL* +
t
Substituting this
8,J^
- J^ + {J^ -
(233)
value of q in (226), and grouping like terms,
we get
)+
Whence,
restoring the values
(222), and applying (211),
F(T) dT
Xi-Hu)
of
/8,
y, 8,
(234)
as
given in
we have
ft
\F(t+n*) dn
ô)
iV^'.-!-^')
- A OC.+4,")
-J T)5
(235)
When
t ,
it
137
of the function extends beyond the value

sometimes more convenient to employ the following formula,
is
the
easily obtained
F(T)dT
Jl+iU
tabulation
from (224)
=
,..'-.
We
here emphasize the fact that the value of 'F
is
(236)
wholly arbi-
trary.
68.
As an example
in the use of formula (235), let
it
be required
to find*
.V
/"*
44
cos
TdT
t/20
using six places of decimals.
The
ployed.
first
It
is
em-
consideration concerns the tabular interval to be

desirable to tabulate
as few values of the function
are consistent with a convenient schedule of differences.
In
all
as
cases
the differences should sensibly vanish beyond the third or fourth order.
4 as a suitable interval in the present instance, we
Adopting a>
obtain the following .table of
cos
F(T)
138
= 6)
F + F = +1.659033
=
20556
4'
4,'
" = 8180
4," +J
v
+4j'
+F
(Ft
^,'
'F
'F,
(i
yV (J6
'
'
5 ? (4,"
4/
4,"
= +5.878635
= -0.829516.5
= +
1713.0
= +
340.8
F (T
values of
sider the
more general problem of finding
for
which
We
has been tabulated.
when
139
shall
now con-
the limits have
the form
I"
n'o,
where n and n" are non-integers

mixed numbers.
To
it
T"
that
n"<a
either proper fractions or
is,
of the problem in question, suppose

were required to find by mechanical quadrature the value of
illustrate the significance
46'
54"
cosTdT
X42
.!
37"
13'
Obviously, it would be impracticable to tabulate the function

21 13' 37"
T'
for a series of equidistant values of T, of which
42 46' 54" are two particular terms.
and T"
may, however,
We
employ the same table as was used

structed for
result
by
T = 20, 24,
28, ...
in
Thus,
interpolation.
in
.
preceding examples, con-
the
44,
and obtain the required
we
the
examples just mentioned,

20
from the lower limit T'
to
have computed the values of
44 and 28, respectively. In like manner,
the upper limits
T"
20, we may find the integral corkeeping the lower limit always
responding to each of the following values of the upper limit, viz.
T"
that
the
is,
20, 24, 28, .... 44, respectively
for each of the tabular values of T.
resulting
values of
the integral,
polation the values which
Then, having differenced
we may
readily find
correspond to the upper limits
by inter21 13' 37"
Denoting these interpolated values by X' and

respectively, we have
and
42
46' 54".
13'
87"
X"
cosTdT
X21
/~42
46'
X"
64"
cosTdT
/20
>'
and therefore
X=
/42'
I
/B"
We
W 54"
TdT = X"- X'
cos
13' 87''
of this example to the student as

a valuable exercise, exhibiting the spirit of the method employed in
problems of this type. The process actually used differs somewhat in
leave the detailed solution
140
form from the method here explained

same.
We
proceed to develop the general formulae.
Let us put
70.
/,
and
* (i)
where
but the principle remains the
>F
F,
=
+
$A!
denotes an integer.
(237)
JV(H-fMi)*i
-I-
y J/'
8J.'"
f^r
(238)
Then (224) becomes

(239)
2,
Let us now suppose that (239) has been computed for i - in succession.
Then, from the series of values
3, 4,
.
0, 1,
thus determined,^
it
= *(0)-*(0)
72
*(2)-*(0)
evident that any intermediate value, say
is
be found by interpolation.
we must express
(238), we have
*(0) =
pose,
To
24
/, can
derive a general formula for this pur-
of the series (240).
the differences
Now, by
'
*f1\
L
\ }
'
* i
-~
'
r-"î
f~i
'
{ C>A\\
whence, observing the general relation
we
derive the following schedule of differences

Function
=
=
+ n (lstDiff.) +
*(o)
+ B (//
7o
2d Differences
1st Differences
Therefore, applying
NEWTON'S Formula of
J5(2dDiff.)
(7(3dDiff.)+
3d Differences
interpolation,
we have

transposing the term
tuting for ^(0) in the second
By
we
to the first
^(0)
member
member, and
the expression given
1,
/8,
the
y, 8,
last
by (241),
becomes
it
expression according to the coefficients
Now, it will be observed that the first polynomial

member of this equation is simply the expression for 'Fn
derived from the series
We
-'F
'
,
'F^
by
in
the second
the quantity
interpolation.
the remaining parentheses contain the expressions for
larly,
.
substi-
find
Upon arranging
141
likewise derived by interpolation from their
Simi-
j'n ,
JB",
respective series.
therefore have
*(0)
'Fn
+F + fM'
a
+ yJn + &V"+
= *()
(242)
Whence
Cf(t+nm)da
/0
71.
In like manner,
9
then,
W =
'f> + i
*(w)-*(0)
(243)
we put
if
-\F< +
IU',-1
y^lL,
+ M<\ - ....
(244)
by (234), we have
'
*(0)
Therefore,
obtain
by interpolation (reasoning precisely

dn
g>()-*(0)
as
above),
we
(2*45)
142
in
(245),
the upper limit n in (243), and n" for
ri for
Again, writing
we get
CF(t+n m)dn
*(')-*(0)
jT(<
+ n>)rfn =
(")
- * (0)
the difference of which gives
f# (* 4- na>) dn =
VH'
y (n")
- * (')
(246)
equations (243) and (245) the expressions

as given by (238) and (244), and restoring the numeriand
for
from (222), we obtain
cal values of ft, y, 8,
substituting in
Upon
<J>
F(T)dT =
JI+n(0
uiFy + tH^dn
- ô) - A (-4'.-^'o) +
B
F(T)dT
A w-4,")
!il 5
(/IT- JJ)
(247)
XM-nd)
VtSW
= -K'^+.-'-K^+^-TV^'n-i-ô)-^
-T A(^ -ô"')-T?TT(-<'U + ^ T )-^!l 5 (^-,-^)-.
1
In like manner,
F(T)dT
we
.|
(248)
(249)
derive from (246),
X<-Hi"(0
-n'OJ
-^ll,,^^-^)-.
In these formulae the quantities n, ri and n" are either proper

fractions or mixed numbers; while the value of 'F is wholly arbitrary.
It frequently
happens that we have

A'
for several different values of
equal to
quantity
t.
'F
In such cases
,
in
(247)
it
is
to
compute
= C
remaining fixed and
convenient to determine the arbitrary
T; the lower
limit
and (248), such that the sum of the terms
having the subscript zero

these formulae as follows
will
:
vanish.
Accordingly,
we may arrange
143
Take
Then(a)
When
near the beginning or
upper limit falls
the
middle of the tabular
series, find
fit
F(T)dT = vj
Jt+tKl)
(b)
When
F(t + n<a)dn
upper limit falls near the end of the
the
(250)
series,
find
F(T)dT
Xf-HKU
o,\
EXAMPLE
Let
I.
it
be required to find
X=
Here we adopt the
T = 0.42,
interval
....
at
= 0.02, and proceed
to
form a table
Instead of tabulating the given

function, it is more expedient to tabulate at times this quantity. All
differences are thus multiplied by the same factor, and hence the final
for
multiplication
0.44, 0.46,
by
at
is
We
avoided.
F(T) =
0.02
0.54.
therefore
10
compute
0.2
T)
for the values of
given above.
The
result
is
as follows
144
The computation is now

0.42, we make 'F6
For the values of n and
readily effected
by formula (249).
and complete the auxiliary

we have
0,
n",
Taking
series
'F.
=
=
5.5270
Whence,
interpolating by
'F,,
F*
J'.,
Jn
',
'/
0.4730
NEWTON'S Formula, we
= +0.149636.4
= +0.404288
=
2054
= +
673
=
19
'Fn
,,
+l
n ,,
J'a ,,_,
=
=
=
+2.621373.8
+0.400748
659
642
Accordingly, by (249),
we
find
= +0.805036
= + 2713
= + 1315
= +
19
A'
To
obtain
verify this result,
we observe
= +2.471737.4
= -0.402518.0
= 226.1
= 54.8
= 0.5
= +2.068938
that
and therefore
X=
=
EXAMPLE
II.
20 (sin-1
V 0.53054 - sin"
V0.42737)
20(168303".25-146965".80)sinl"
Let
it
be
Z,
required
to
2.068933
by mechanical
evaluate,
quadratures, the integrals
/"*4.8
j GOT'dT
and
A\
Here we tabulate w times the given function

10, 12;
thus
we
obtain the following table of
/!!.
60
j
for
2, 4, 6, 8,
F (T) = 120 T
3
:

T
145
140
(2)
To
find
We
use the same formula as before, the value
of n in this case being
or
an interval of 0.40 counted forward from the quantities 'Flt
Ji, j", and
Ji".
Accordingly we
find
1}

the schedule of functions (including
below
F) and
147
differences be taken as
F(T)
t
2ft
w
'F,
4!
t+
t +
u
(i
4%
(i
Reverting now to (104), an inspection of this equation shows

clearly the law of formation of the successive coefficients in the second
member
hence, adding the term in J*,
we have
(251,
Multiplying by dn and integrating,
we
obtain
(252)
3f being the constant

n
between the limits
J', /I"', /f,
of integration.
and
If
evidently vanish, and
this
= -\-\,
we
integral
the
is
now taken
coefficients
of
find, therefore,
(253)
148
In like manner,
we
derive
(254)
)dn
= F + A 4" t
we
"Whence, by summation,
Upon
obtain
substituting the relations
"+'... +4" =
r
(256)
r=i
2^
=
"
in
formula (255), the latter becomes
^+.
Finally, therefore,
we
*-)-
(267)
obtain
|tl>
= K'^Vn-'^)+ A (^'Hi-^)-iirW-^')+
When
rfW> W (^.-4I,)- ..-.j (258)
of an integral are to be computed from a

|w, it will be more
given series, each having the lower limit t
convenient and expeditious to determine the arbitrary quantity 'F_
several values
such that the sum of the terms with subscript

| is equal to
The formula (258) may therefore be written as below
zero.
F(
X+fio+
t<o
dT =
(-H
(259)
As
an application of (258),
let it
sec 2
149
be required to find
TdT
X46"
.j'
Taking
F(T)
o>
= sec T
2
= 3,
= 31
as follows
30',
and
'F+
0,
we
tabulate
150
denotes the value of any coordinate at the instant T, and

In particular, let us put
value at the epoch
where
its
-j=,
jT
Z
=
=
=
=
the heliocentric longitude of

the daily motion in longitude
Mars
for
any date T;
1898 June 13, Greenwich mean noon;

1
47' 14".3
the heliocentric longitude for the date
be required to compute the longitude

mean noon of the dates
and
let
it
(1898) June 21, June 29, July
7,
July
15,
(I)
for
and July 23
the values of the daily motion being taken from the
Greenwich
American Ephemeris
for 1898.
The complete
follows
Date
1898
solution
is
conveniently arranged in tabular form as
Whence
the factor w having been previously applied.

for
151
the expression
becomes
Thus, the value of

constant
for aw?/ date
the integral
to
taken from
T being found
T to T,
is
it
by adding the
that
clear
we
have merely to increase the above value of 'F_^ by the quantity

in order to avoid the subsequent addition of this
Z
1 47' 14:". 3
constant to each computed value of the integral. Accordingly, under
the heading
quantity 1
-\-'F,
47' 19".5
on the
line
|eo
13), we write the

of this column are then
(= June
the remaining numbers
formed in the usual manner by successive additions of the functions

F. Each term of the series thus formed is evidently greater by Z
than if the latter constant had been excluded from the initial term.
Under
-fad
ponding terms
in
from the corres-
are written the values derived
-j-
//'.
The sum
-\-'F-\--fad
is
then tabulated in
which therefore gives the heliocentric longitude of

Mars for the dates indicated in column T.
the final column,
Z,
Applications in which the Limits Fall Otherwise than

If
Between Tabular Values of the Argument and Function.
73.
Midway
we put
*-....
(260)
the formula (257) becomes

)dn
Whence,
we
derive,
if
as
0(t
+ |) _*(_$)
(261)
before n denotes a fractional or mixed number,
by the general method of interpolation employed
in
70,
(262)
Upon
substituting n
and n" successively for
taking the difference of the resulting expressions,
tht
0(n")-6(n')
we
in
(262),
and
get
(263)
152
(262) and (203), according

obtain the following formulae
Finally, replacing the functions
we
to the expression
(260),
/*l-(-<0
F(T)dT
\F(
/( lu)
0,
in
(264)
/ n"
(+"<!>
XF(T)dT =
-
I
Ai'
<
.-n'<i>
(*+) db
{
where the quantity 'F.+
(265)
wholly arbitrary; and where 'Fn /]' , //;,", /;;,

the
similar
terms with subscripts ri and n") are to be
(and
....
is
found by interpolation.
When
of an integral are to be computed from a

by (264), the latter may be modified to the more ex-
several values
series
given
pedient form given below
Cf(T)dT = o>p
*/
mj
(266)
J(l)
^^
EXAMPLE.
O) (
fn
-\-
TJ^X
"
^n
ffT?"??
*^~ ^T
Find the value of
^=
Ce"dT
%/0.16
being the base of the natural system of logarithms.
Taking
lowing table
r
a>
0.1,
0.2,
and
F(T)
E e
r
,
we prepare
the fol-
THE THEOEY AND PEACTICE OF INTERPOLATION.

whence the column 'F
153
completed as shown above. Denoting the

assigned lower and upper limits by T' and T", respectively, we have
is
0.15
T" =
0.48
T'
Hence, at the upper
=
=
Accordingly,
we
find
the quantities
'Fu
above,
we
J'25
'Fn
and
2.8
J' n
0.20
0.28
the value of n
limit,
= t-û
= + 2.8
0.20-0.05
j't %
2.5
and
*>
is
0.30
by interpolating forward from
//;,"
From
with the interval 0.30.
the table
take
'K.
05
J'.
J 25
+4.058247
^*W%f*fmt
'
+0.156896
J'",
+0.001572
Hence, making the required interpolations by means

Formula, and proceeding according to (266), we find
of
BESSEL'S
= +4.535670.3
= +
6736.4
+ ^
=
4.8
-sHg^n"
2 = +4.542402
X = +0.4542402
>Fn
= +0.161674
1619
^" = +
J' n
//'
.-.
The
Ce T dT
we have
74.
1'ition.
true mathematical value of
X=
48
e- 15
is
easily
found
thus, since
0.454240159
Quadrature as Based upon BESSEL'S Formula of InterpoAnother set of formulae for mechanical quadrature, similar
to those already developed,
BESSEL'S expression for
may
be derived in the same manner from
F(t-\-n<a).
However, since these formulae
be obtained more conveniently by a direct transformation of those

developed in the preceding section, we choose the latter course.
may
Putting
n"
i,
and
0,
in
we have
formula (263),
(267)
We
also have,
by (260),
-.
(268)
154
Referring now to the general schedule on page 147,

observed that the quantities
J'
t ,
be
will
it
4", 4,
are not explicitly given, but must be found by interpolating to halves
between
'J^-j
purpose,
let
quantities
and 'F^,
us
by
denote
d'i-i
the
('-^), (Xi)>
and
//',-+|,
algebraic
'?
C^")>
Applying formula (126), we have,
etc.,
respectively.
means of the
that
is,
let
latter
For
this
of
pairs
us put
therefore,
(270)
Upon
substituting these values of
'F
t ,
j' ( , j'{ ",
in
(268),
and reducing, we get
(f)
Putting
C^,)
0,
this
(J',)
4-
(271)
becomes
.
(272)
Whence, from (267), we derive

C'r
(273)
It is evident
from
(111) that the coefficient for the
(n+2)(t+l) n (B
l)(n
E
which, for
i,
yields the value given in the text.
ni.rtli
2)(u
ilifi'i-i n>
3)
<
in
BESSEL'S Formula
is

n
Again, putting
Cp(t+**)4*
=
=
in
155
we have
(262),
e(i)-6(-$)
('^)_ Tv(^) +
y ,(^'')- 5 MU(^) +
In like manner, making
n"
= *-(-|,
and
ri
(274)
in (263),
0,
we
obtain
+ nu) dn =
_
F(t
X'+i
+ b)-6 (0)
*H r 2t Âti+1
6(i
IV
A< _L
H
17
5TB5
3B7
//
SSTffSTJ ^(+1
)ri'
Finally, substituting
=n
and
in
0,
(275)
(263),
the
latter
becomes
CFtf+n^dn =
-0(0)
(re)
'^.
A^'.-iH^:"+* ftv4:-J
'J
- AV K") + vitlv
W - ....
(276)
equations (273), (274), (275) and (276) give, respectively,

the following formulae of quadrature
The
dn
A [W- W] +
....
-'f-t
+ l(J
- A ^'- + *H*
^-
Vs[(^")-')]
(277)
^4- ....
(278)
5 H 5 4+1 + FrfV^TJ 4'+1 - ....

- to K") + aib C4) -
(279)
AV
/<-t-i
F(T)dT = aJ/^+ww)^
= '^ + + A ^Hi -
J(+iU)
Cô)
+A
(^'o)
/'l-HiW
\F(T)dT=<a \F(t + n< >)dn

= ,^Vn +^J'B 1
^"+ 5B 7V5Tj jv_

- TVa K') + tm*
5 fJ 5
- ....
(280)
156
is
denotes an integer and n a non-integer ; where 'Fwholly arbitrary ; and where ('F ), (j'<), .... and ('^ ), (X ), .....
are means of corresponding tabular quantities, as defined by (269).
in
which
If,
in the formulae (277),

tV
then the
sum of
W-
(279), and (280),
TV*
K") + *ili*
we
(40
But, since
the terms with subscript zero will vanish.
, the preceding condition
is
evidently satisfied
if
we
(4?)
The formulae
computed
take
(278),
(277),
as follows
(279)
and (280)
take
- ....
may
(281)
therefore
be
/<-H'
ij
(282)
/*
(283)
)
-\
= - i ^ + Tv (^' - 7VV K") +

)
>
XM-
dn
(284)
= - iô + A (^'o) - ^V K") +

dT =
'f-k
Several examples will

EXAMPLE
I.
Let
it
now
(285)
be solved as an exercise in the use of
be required to find
Tsin TdT
Tsi
X->
THE THEORY A5JD PRACTICE OF INTERPOLATION.

Here we take
F(T)~a>T sin
r
T,
o>
= 20 =
as follows
10
157
it
,
and tabulate
158
Here we take w
as
below
0.1,
0.9,
and tabulate
F(T)=(1+0.1T
)-*,
We
now
ing values of
take
a>
T}
=2 =^
-_
a>
sec"
= 45,
159
and tabulate the follow-
160
DOUBLE INTEGRATION BY QUADRATURES.

Having derived various formulae
75.
for the
mechanical quadra-
ture of single integrals, the corresponding formulae for double integraThese will serve to compute integrals
tion are now readily deduced.
of the form
Y = C CF(T)dT*
(286)
independently of the analytical nature of the function F(T), provided

more
T' and T" are numerically assigned. To define the quantity
explicitly, let us put
CP(T) dT
f(T)
(286a)
where
is
We
the constant of integration.
then have
(287)
T')
It
is
therefore evident that
unless
the constant
M has
a definite
In
will be indeterminate.
value in any given case, the value of
is generally known
practical applications, however, the quantity
from the fact that the first integral has an assigned value (usually
zero) corresponding to the lower limit of integration.
If
we now put
T =
n<a
T'
n'u
T"
n">
we have
dT*
= o>W
(288)
and hence (286) becomes
Y
upon which
76.
~F(T)dT* =
2
>
Fty + n^dn*
(289)
relation the subsequent formulae are based.
Double Integration as Based upon NEWTON'S Formula of
Interpolation.
If
we
and n"
substitute, successively, n'
and take the difference of the two

CF(t+n*)dn
/'
results,
we
for
in (243),
obtain
= *(")-*(')
(290)
From
the
of (290)
form
follows that
it
the
161
expression
for
the
indefinite integral is
or,
by (238),
the constant
of
integration
being contained
upon the arbitrary quantity 'F

integrating,
we
\S
Multiplying this
tS
up
-*
"F
which depends
equation by dn, and
,
'
up"1
up2
*s
up
* 31
'
*s
namely
series,
>
*S
*s
Let us now consider a new
the term
'Fn
get
CC
*s
in
'
'
'
up
-*
rf-f-2
being arbitrary, and the subsequent terms so determined
that the quantities

>Ft
'F,,
'F,,.
>F,+i
of the proposed series. The manner of arranging the series "F, 'F, and F, together with the differences
of F, is shown in the schedule below
are the successive first differences
Jlv
F(T)
up
*
'F,
>F2
"Ft
f\
'F,
4?
(t-l).
A?
162
Now,
functions
the differences
since
whose
1st,
....
2d,
may be regarded
(r>
are
differences
a series of
as
J<-+",
<4<'+
it is clear that formula (291) may be applied successively to each of

the integrals in the second member of (292). Accordingly, we have
f'F.d
.
1"
dn
dn
('*
f CF (t + no,) dn* =
"/-;
+ 'Fn +
(i
we
find, in
+ 2/3) .F. +
accordance with (292),
+ 2 y) J'
(/3
Upon
(293)
tfiese expressions,
Summing
Sz/i"
'
z/:"+
substituting the numerical values
of
y8,
y, 8,
(294)
from
(222), formula (294) becomes

(294a)
the
expansion
to
We
of J' n reducing to zero.
coefficient
which the
in the
brevity, let us write (294)

dn* ==
Now, from
"Fn +
(228),
of
coefficients
proceed to determine the

For
this formula belong.
form
>Fn
bJ'
(295)
we have
(296)
Also,
let
us put
w =
x-*
x~ l
ax"
bx
ex 2
dx>
(297)

in
which the
in
(293),
follows
is
it
taken as in (295)., Whence, since the

the combined sum of the second members
are
coefficients
second member of (295)
163
is
evident that
may be
(297)
resolved,
conversely,
as
w = x^ +
x~
3x+x + Sx
....
4-
w =
x-1 (x~ l
+ Sx
(a;-
yx
+ px +
yx
+ k z + /3 +
x"
Therefore, by (296),
8x*+
S.r
we have
Comparing (297) and (298), it follows

former, and hence, also, those of (295),
2
expansion of
[log (1 -f-#)]~
troducing these values of a,
>Fn +
as
was found directly

Let us now put
as
b,
7<;- 2:
T',
c,
U î'+
in part
ji
in
ff
(298)
that the coefficients of the
the coefficients
are
developed
d,
in
in
(298).
(295),
4V'- B &!U
v
^!.
the
in
Whence,
we
in-
obtain
+ MT, *l
(299)
(294a).
= "F + 'F, + aF + 1>J', +

+
+ eJ* +..'..
= "F, + 'F + T g F. + OJ' - 5 U J: + ,1, j;" - ,H JLT +
'
X(n)
c,-1'
'
t
(Ld','
(300)
and (299) becomes

(301)

Whence,
and
limits, ri
if
the
n",
we
taken between the
be
integral
two
fractional
have
shall
C CF(t + n<a)dn' = X(n")-\(n')

1
And
if
we make
the upper limit an integer, say
f p^-f wo/) dn =
3
A (i)
*/ fc/n'
The
ences J
F(T)
n"
i,
we have
- A (n*)
(303)
formula involves the disadvantage of employing differj", J-", .... which are not given when the tabulation of
last
{ ',
ends with the quantity
ceed as follows
v
(302)
and substitute
To remedy
this defect,
we
pro-
Put
\(i)
"F,
'F,
aFi
"F^ 'F
for
t ,
bJ'
Ft,
eJ'
'
/t",
z//,
dj' "
t
-f eJJ T
....
+ ....
(304)
the expressions
'Pi
F,
F,
(305)
"'
"Whence the integral (303) may

available differences,
j' i_ l
/l'-_ z
-\-
-t-
once be expressed in terms of the

However, to avoid direct
J'^
at
substitution, let us put, as in (229),
(306)
and we
have
shall
--1
,,-1
W(l
It)-
= U+H*+U +U +
S
(307)
X*
MV1-W)-4
M 4+
165
Again, from (297), we have
w =
,
X'2
aj-
bx
ex*
dx*
....
ex*+
(308)
the expressions (307) be substituted

of (308), the algebraic process will be identi-
evident that
is
it
+ x- +
if
in the
second member
cal in
form with that of substituting the expressions (305)
The
in
(304).
operation involves the quantities
x"2 x~
,
x 1 x8
x", x,
....
u~2 u~
u3
while the v operation involves, in exactly the same manner, the quantities
,.
.
1
lip
f tl ip
fit-fit
Hence,
if
known.
(308)
From
is
/tin
aAlit aA"tt**{t'
up
f i+V>
.
>
ptt
f
Al
-^ i11
aAlli-
"l-i
{log(l+*)j-
the expansion of which
u~ l
u-*
)log(l-u)\-*
au
Thus we
bu
+ cti*
find
dit*
....
eu 4
Therefore, according to the preceding reasoning, the

v is
= "F^ -
Denoting
>FIJrl
c,
at\
//',._!
cA'^
d/T^
expression by TT(), and
this
b,
4-
()= "F, -
eJ
(309)
expression for
^ - ....
1
restoring the numerical
from (300), we have
w + aF
'F
u for x
immediately obtained by writing
is
second member of (297).
w =
find
w = |_log(l-w)J- =
a,
we have
Whence, by (306), we
values of
obtained with greater expedition by the following process
(298),
Ain
2
we perform the w operation, the result for v is at once

But the expression which results from substituting (307) in
w =
in the
If
-Fi+l!
bA\_,
cA'^
d/J'^
+ eA^ -
166
Whence, by (304) and (310),

A.
(l)
7T
(i)
and the formula (303) becomes, therefore,

'
f
In the formula
C'F
dn*
7r(i)-X(')
proved the quantity
just
(311)
denotes an integer.
general method of interpolation employed in

easily shown that (311) is true for non-integral values of
this formula becomes
writing n" for
the
Now, by
70,
i.
it
is
Thus,
*',
dn*
We
now
and (289),
*(n") -X(n')
(312)
bring together equations (300), (310), (302), (312)

the. order named
observing that in the first two of
in
we may write "Fn+l for "FH -\-'Fn and for "Fn+z

Thus we obtain the following group
tively.
these
'Fn+l
respec-
^dn*
Cfft+n*)***
\(n")-\(n')
(313)
'C
'
TT
(.")
*Jn>
From
this
X (')
group are immediately derived
all
of the formulae given
in the following section.

77.
We
have already remarked that
in
the process of single in-
tegration the value of the definite integral is wholly independent of

which may therefore be assigned arbitrarily.
the absolute value of 'F
Similarly, in double integration, the quantity "F may be taken at
,
the integral being independent of its absolute value. Per

contra, the double integral will evidently vary with the value assigned
pleasure,
to
'F
Hence, unless 'F
value of the
is
double integral
derived from (287).
fixed
is
by some
special
indeterminate
consideration, the
a conclusion
already
Now,
as
was previously remarked, the value of the
corresponding to the
cations.
We
lower
limit
is
therefore denote
shall
which results when
usually
by
upon restoring the numerical values of

and transposing,
first
integral
in practical appli-
the value of
^F(T)dT
Then, by (291), we have
substituted for T.
is
or,
'f'\
known
167
/3,
^ + i ô + rV <*',- A 4' + r'& 4"- lis ^
y,
8,
i&Jtff
from (222),
4f-
(314)
which determines 'F1} and hence, also, the double integral Y, provided
// is known. In practice the value of // is frequently zero.
conjunction with the relations (313),
the several groups of quadrature formulae given below
I
<ing
in
(314)
we
obtain
IL
/
j
/(+iOJ
iF^dT
cu
/(
(315)
j j
o~
-A
(316)
/
(
/'l+i(U
\F(T)dT* =
w"i
fi
\
F(t +
(317)
JET.
= ^\ /"n"
F
J Jn/-
(318)
THE THEORY AND 1'KAOTICK OF INTKIM'OLATIOX.
168
The foregoing formulae
are
applicable
when
the upper limit
falls
near the beginning of the tabular series. When the upper limits falls
likeat or near the end of the given series, the following formulae
wise derived from (313)
may be employed:
'F,
2
a.
pp (*+>) d
(319)
^5
<l>
(320)
/ /*l+l)
1
F(T)dT* =
/<
/
<u
J (*+.) <*
1
/ /
*/(-4-nO)
(321)
CF'(T) dT* =
f ^/+n'u
2
a)
*^
fF(if
In applications of
(or of
"F when
+ no))
<Z
(322)
*/7i'
the
all
employed)
is
preceding formulae, the value of
wholly arbitrary, and therefore
assigned at pleasure in every case.
and (320) are applicable,
it
is
may be
But when (315), (316), (319)
frequently convenient to determine
such that
The formulae
"J*,
in question then take the
form as follows
"F
rib
CD
(323)
/ /f+nU)
(
-'
\F(T) dT*
^r
(324)
(325)
/* /* (+<!>
F(T)dT =
The
(326)
/*
o,
differences which appear in
the foregoing formulae, together

and "/P, are to be taken according to
with the auxiliary functions 'F

The symbol i denotes a positive integer,
the schedule on page 161.
while n designates a fractional or mixed number
so that all functions
:
and differences
whose
their
series
respective
notes
as
subscripts
by
interpolation.
previously defined
substituted for
involve
so that
must be derived from
Finally, the
the value of
quantity 7/ de-
\F(T)dT
when
is
we have
(327)
It
while
known
may happen
the
for
value
a
of
occasionally that the value
\F(T}dT
particular
of //
corresponding to
value of n.
Denoting
this
is
T=
unknown,
t-\-n<o
quantity by
is
170
we may, by any one
of the foregoing methods, compute the
definite
integral
Jl+nUl
dT =
- 1L
//
(327*)
and hence find
= H.-X
we proceed as before.
examples will now be solved
with which value
Several
as
an
exercise to illustrate
the formulae given above.
EXAMPLE
I.
Let
it
be required to find
Y
on the supposition that
We tabulate
T
fcosTdT=2
when
--
- 0.
and difference the following values of
F{ T) = cos T
The column
also,
the
'
is
now completed by
column "F, having
first
assumed
(319), the remainder of the computation
"Fw
+136.04398
successive additions
is
"
F=
l
as follows
0.
:
171
hence,,
Whence, by
172
We
have, therefore,
2.0
<o
//
0.1
whence, proceeding by (326), the computation of

follows
'JF
and
"J<\
is
as
= +0.12500
^' = _
193.7
+
_ & J- = _
12.9
1.3
+ 7 g ^" = =
+0.12292
'f\
.,i
7<;
T>J
- 5 j ff
.-.
...
we now
the completed table
=
=
F =
n
(2.468- 2.0)
4.68
-f-
ij'
From
F = 0.02083.3
1.3
<= +
<'= +
"f\
0.1
"Fn+1 =
= 5-0.32
+
+0.16418
'
T5
easily verified, for
T+ CT\
[H2.468
log,
+2.36025.6
is
0.02082
find
.-.
This result
0.2
Y =
1368.2
+2.37393
+0.0237393
we have
= -
log. 1.234
+ 0.468 C
_fe
also
Hence
F =
-log.1.234
0.234
-0.2102609+0.234
+0.0237391
with which the above result substantially agrees.
EXAMPLE
III.
From
the table
the value of
r
Y =
I
^2.
i
of the preceding example, find

Here we employ formula (324),
n
2.15
- 2.0
173
which we take
in
1.50
0.1
We
therefore obtain
(n
j;'
J'"
= +0.24992.0
= + 1802.8
=
1.0
= 2+
= +0.21633
= +
235
= 38
)
0.2
2'
...
The
true mathematical value of
Y =
0.075
- log, 1.075 =
Y =
+0.26794
+0.0026794
is
+0.0026793
Double Integration as Based upon STIRLING'S and BESSEL'S

Formulae of Interpolation.
Let the schedule of functions (including
78.
'F and "F) and
differences to be used in the subsequent formulae of
quadrature be as follows
T
THE THEORY AND PRACTICE OF
174
Now, by
KPOLATION.
II
we have
(260),
= '^+^J;- s Hs<'+0AYS D^-
6(n)
and hence the preceding equation becomes

no,)
For
dn
'Fn + jf J'n
,tt v
W+
v jfj
/*;
(328a)
brevity, let us put
= +ls
-**
+Ww
329 >
and (328) may be written
= Crn dn =
CF(t + nu>)dn
'Fn
+ aJ' +
n
bJ'^
+ c/l'n +
the constant of integration being contained in
we
equation by dn, and integrating,
C'Fn dn
we
+ a.Cj n dn + bC^'dn + cCj'n dn+

l
restoring the values of
as in (327),
obtains for
each
to
of the
(331)
integrals
a, b, c,
from (329), and
obtain
JJ> (+<") <^ "

If,
Multiplying this
second member of (331), we obtain
expressed in the
reducing,
(330)
get
Applying formula (330) successively
Whence,
'Fn
....
T'=
"F*+-hF*-ih4'+*MnJ:-
we denote by //
t,
then,
the value of
fF(T) dT
( 332 )
which
by (328), we have
T=I
=,0.0(0)
and hence, by (272),

(333)
Upon
substituting
in the first of equations (269),
175
we
get
which, together with (333), gives
(334)
where the differences enclosed within parentheses are means of the

corresponding tabular quantities, as defined by (269).
employing simultaneously the relations (332) and (334), and
By
assigning various limits to the integral,

of formulae
we
obtain the following group
/(+('!!
7
/XT
dT 2 =
F(T) dT*
/*
/*f-t-"i>
F(T) dT'
/i-Hi>
f*
a
o>
J F(T)dT*
(*
(335)
<>)
fn
/
(336)
jJ
= ^
J
= - + i /; +
/ /I+n"d)
j
'^
/*!
S*t+nlo
/
o,
/"
j
fm
(337)
>
T5
/
(J'J
'i")
+ *Jt*w(40-
/"
Ml *
(338)
We
In the preceding group the value of ''F is wholly arbitrary.

may, however, determine the quantity "F such that the sum of the
terms in (335) and (336) having the subscript zero will vanish
formulae may therefore be written
these
THE TIIEOHY AND PHACTK'K OF INTERPOLATION.
176
/I+l>
\F(T)dT*
= ji/
(339)
xi
(
&a^'--
(340)
/l+n<D
J (F(
Let us now denote the second member of (332) by y

is,
let
(ri)
that
us put
y (n)
Making
= i-^4,
y
It will
d"v+\i
respective
"F,
A ^- -
this
sin
^' +
, A'STS
^' -
becomes
(342)
(i
"
be observed from the foregoing schedule that

ai'e not explicitly given, but must be derived from their
series
by interpolation
to halves.
For
this
purpose,
let
us
put, in analogy with (269),
(343)
then, after the
manner of (270), we
shall
have
(344)
Upon
substituting these
(342), and reducing,
we
expressions
in
the
second
member of
find
(345)
177
Again, by means of (332) and (341), we derive

n1
by
denoting
Finally,
we
$to,
//_*
shall have,
- ('**-.
the
value
of
(346)
F(T~)dT
when
by (328a),
A J'-i - ** J*
y(n"} -y(w')
^+
-H
)
which gives
(347)
assigning various values to the limits ri and n" in (346), and

employing either (341) or (345) as required in each particular case ;
and finally, by using either (334) or (347) to determine the series
By
we
the assigned
as
'F, according
lower limit
is not or is
derive the group of formulae given below
equal to
(348)
F =
11
awy convenient value
arbitrarily assigned.
(349)
//
'
i 'F_,
4-
/* ffrlnt
I
|.F(T)rf2"
'
+/
+/t\'j>
A (/_,) - T ib
=
(^,)
/ /(
!
+'
/(*+) d
/ \
(350)
J
178
f> /*t+*<0
I
I F(T)dT^
*/ */(
2
tu
*^
(to
(351)
*'
TT
IV
"^ =
f
I
./
~* _ J
*-*"
/I'
^
-T4_
~|^~
2
(^(T)dT =
a>
/
|
tio
367
/*'"
awy convenient value;
xv+ii+DU)
i'
5T6TJ -^-J
SffTifgflr
//
J.
^-J+
arbitrarily assigned.
/i
|^
(352)
=
'[
The
last
formula
may
also be written in the following
form
TT
(353)
It
well to again point out the fact that the functions and
enclosed within parentheses denote the means of corre-
may be
differences
sponding tabular quantities, as defined by (269) and (343). Further,

that Jff and H^ denote the values of the first integral of F( T) when
we
substitute
that if in
any case
for
to
and
p is
compute
X=
|-co,
respectively.
Finally,
given and IIq required,

/V+pW
I
F(T) dT
it
is
we may add
only necessary
= Hp - H,
^l+?(D
and thence
find
If,
is
It,
-X
In the process of double integration by mechanical quadrature

sometimes convenient to tabulate, not the given function, but
it
s
<u
times that quantity. By this means all differences are multiplied by

a
&> , and thus the
final multiplication by that factor is avoided. However, in order that the quantities 'F and "F shall be multiplied by the
TT
same
factor,
it is
evident that the independent term
10
(which has the

same fixed value whether we tabulate
wise be multiplied
by w
a
:
becomes necessary to take
so
that,
(oil in
F(T)
179
must
or
this
proceeding by TT
like-
method,
it
which occurs
place of the term
The computer is cautioned against

neglecting this precept in case he tabulates a?F(T) instead of the
in all
given function
We
F(T}.
close the chapter with
several examples which
illustrate
formulae given above.
EXAMPLE
I.
Find the value of

2.0
Y= on the supposition that the
We
T
2TdT'2
first integral vanishes for
tabulate the given function as below
T = 2.2.
the
180
"Ft
F<
J't
'
= 0.000000
F = -0.129011
^ = - 1746
"Ft
-0.907502
= -0.086353
= 994
.-.
Verification
we have
Integrating directly,
f
= -0.907502
= +
3554.8
= 3.1
= -0.903950
Y = -0.00903950
(F -"FJ
+^(F -F
- g g (z/;'-x/;')
2
2TdT
1
T2
'
12.6
Y =
]"
J2.2
\_
whence
//.
+ T )2
(1
.-.
+c
Jr=2.2
[_
-0.17123288
Finally, using the relation

tan" 1
tan" 1 a
the preceding expression for
tan"1,fa-b\
r
1
^.
Y becomes
which gives
Y =
EXAMPLE
From
II.
-0.00903949
the table of the preceding example, compute
2TdT*
12 2
(1+7
Here we employ (349), taking

t
Thus we
2.2
ô
(2.23-2.2) -^ 0.1
find
^)
)
= -0.090739
= 1068
('%)
(F8i )
= -0.717599.5
= +
3780.8
+
-0.713828.2
0.30

Also
(
181
182
We
proceed by formula (353), taking as our data

t
=
=
32
4
that
Whence, observing
= 4 =
= 2M =
<o
//_,
we must now
J'_t
And
for
"Fn we
_
)
)
tr)
= 8.782752
= + 2088
= + 468
0.868589
take
H-^-ô) in (353), the computation of 'Flog o>#_,
ir-M5
instead of the term
wJBT_$
as follows
is
= +0.060639.0
87.0
-3^-, =
I-*
+*^-' = +
= +0.060553.4
u>//_j
find
= +0.030276.7
309.8
+ ? (F_,) =
7.9
-rJ^W) = +
\'F^_
= -0.007436
= 887
= _
367
3>
.-.
Upon completing
the
+0.029974
shown above, and continuing the
as
table
"F
computation by (353), we obtain

(i
(F
4i )
=
=
= +0.240524.0
97 2
-AWi) = +
('%)
4)
-0.002332
Y =
.-.
We
+0.240620
easily verify this result analytically as follows
-McosTdT
McosTdT
"
sinT
smT'"
'+
'
CT+
C'
'
gl
Jr=o.=
But
M
C
F=
log w tan
/50\
f--
/30
log lo tan
THE THEORY AND PRACTICE Or INTERPOLATION.
Now we
find
log tan 25
.-.
9.428052.5
0.240620
which agrees exactly with the former
EXAMPLE IV.
From
- 10
- 10
9.668672.5
=
Y =
log tan 15
183
result.
the table and data of
Example
III,
compute
the integral
Y=
f/McosTd
//
sinT
^f ^sn.
Here we employ (351), taking

n
for the value of
at the
We
(45
upper
-32)
= 32
we then have
as before;
limit,
-=-4
3.25
3-+ 0.25
therefore obtain
........
"Fn
= -0.002993
163
4;= -
+^F
F,
= +0.189420.3
=
249.4
= +
0.7
Y = +0.189172
-
-4'
.-.
Verifying this result as in the last example,
Y =
log,,
EXAMPLE V.
tan 22 30'
As
final
log w tan 15
exercise,
we
find
+0.189172
combining both single
and
double integration, and illustrating, moreover, the use of formula (339)

assigned in succession to the integer i, we
shall conclude these examples with a complete and detailed solution
when
several values
are
of the following problem

of unit mass
particle
:
is
impelled along a straight line
AB
is the
where
by a varying force whose expression is SOOOOT"
time in seconds after a definite epoch, and the implied unit of length
is one foot.
It is required to find by quadratures the velocity, v, and
3
the distance,
AP = x,
T =
assuming that v
seconds.
for the times
102, 104, 106, 108
= 0.6
and 110 seconds, respectively;
feet per second
and xa
=8
feet
when
T = 100
184
Since the mass of
whence by a
is
unity,
we
have, simply,
</-.!
20000
dT 1
~~T~
single integration
dx
dr
/yOOQQdT
"
r5
T.
and by double integration
We
compute the required values of x as given by equaThe

tion (/3), effecting the double integration by means of (339).
details of the computation are shown in the following table
shall first
TABLE
T
(A).
THE THEORY AKD PRACTICE OF INTERPOLATION.
85
jT
by the same factor
so
that, writing
(= 7/ )
we have
equation of (339), and omitting insensible terms,
>Ft
>
we
'F
ô + iVW
(y)
=
F = 0.04000,
'F = +0.61980, and thus
vn
Hence, substituting
0.00240,
"o
find
in the first
for
0.6,
(z/ )
(.r_ ,
+ 4) =
complete the series
as given above.
The second equation of (339) gives simply, "F^-=.

^ F the
term in j" being insensible. But since, by equation (/3), we should
,
afterwards have to add the constant
double integral taken from

place of
"F
F +
to
x-
T,
computed value of the
to each
it
expedient to tabulate in
is
the quantity
?J>
= -A/^+4.0 =
"F,+ $r,
and thence complete the
as given
series
reason for this procedure
is
tion of (339) gives (since
easily
-0.00333
under
"F-\-
made apparent
+3.99667
a.
The
for the final equa-
w must now be replaced by w)
20000dT
ffand substituting
this expression in equation
///ET
<
FI
+
I
T'J
ET\
-f*
4-
^
^
==
Therefore, upon forming the column
we have from
A+
/// 1?
(/3),
we
obtain
/\
()
T?\
+ iVA)
-4- y ^
J^ = E
J,
as given above,
(8)
whence the required values of x are derived and tabulated

column of Table (A).
in the final
186
For the computation of the

the
or,
first
we employ formula
(282),
equation of which, gives
by adding
But we
velocity v
to both
members,
avoid subsequent additions of the constant r c required

0.6
by equation (a), if we increase this value of 'F by the term v
that is, if we take
shall
which
is
series
'F
the same as the expression (y), used for determining the

in Table (A).
The latter series is therefore to be employed
computation of which
in finding v, the
TABLE
T
(B).
is
as follows

This completes the solution of the problem.
is
derived, however,
by observing that equation

/
(a) gives
a-
(t)
series
TABLE
T
interesting check
= vdT+
whence x may be obtained from the

For this purpose we make f(T)
table below
An
187
(C).
<av
v by single integration.
and "thus form the
= 2v,
188
79.
It is
worth while to inquire what change takes place
in the
value of the double integral
T = CCF(T) dT*
when,
in
particular problem, the quantity
new
assigned value //' to a

if
we change
//' to
H",
the value of the
to any particular value of

//" //'; or, what amounts
first
integration,
fore,
by (287),
value //".
This
first
II
is
is
changed from an
easily answered.
integral
For,
corresponding
thereby increased by the quantity

to the same thing, the constant of the
in
(28G),
it
is
evident
is
thus increased by II" //'. Thereis increased

that
by the quantity
is
189
EXAMPLES.
Given the semi-major axis of an
1.
semi-minor
axis,
ellipse,
1,
and the
to find the length of the elliptic quadrant.
0.8,
Ans. 1.41808.
[NOTE
Take the
eccentric angle
E as
independent variable, and hence find
IT
= f Vl A>
where
e is
2.
the eccentricity, and
e*
cos 2
^ dE
the required length.]
Given the equation of a
cardioid,
= I-\- cos
to find,
by
mechanical quadrature, the length of that part of the curve comprised
between the
initial line
and a
the initial line.
3.
The equation of
line
through the pole at right-angles to

Ans. 2.82843.
curve being
= a? V2
sin x
find
the
area included between the curve, the axis of x, and the two ordinates,
4.
and
Compute
Ans. 0.180518.
n.
the value of
==
0.82
assuming that the
sinT
first integral vanishes at the
lower
limit.
Ans. 0.139727.
5.
relation
Given
curve
in
a vertical plane whose points
satisfy the
THE TIIEOBY AND PRACTICE OP INTERPOLATION.
190
the axis of y being vertical. Find the difference of level between

two points whose abscissae are 1.000 and 1.473, respectively ; assum-
ing the direction of the curve to be horizontal at the
first
point.
Ans. 0.044228.
By what amount would
the preceding result be changed by

supposing the tangent to the curve at the first point to be inclined
45 to the horizontal ?
G.
[NOTE:
This question should be answered mentally.]
CHAPTER
V.
MISCELLANEOUS PROBLEMS AND APPLICATIONS.
short chapter will be devoted to the solution of

of problems and examples involving certain principles and
The present
80.
number
precepts hitherto established.

81.
PKOBLEM L
To find
EE l*-f
2*
+3* -f .... +
*,
where
k and r are integers.
The method
lar value to k.
of solution
Thus,
is
let it
S =
I4
best illustrated
by assigning a particu-
be required to find
+
24
+3 +
4
r4
TVe tabulate below and difference the values of T* which correspond
to
T = 1, 2, 3, 4, 5
T
and
6.
Thus we
find
192
consider the auxiliary series 'F

defined as in Chapter
have, by the fundamental formula (73),
-l)
we
IV
shall
(r-4)
(24)
11
Theorem
Therefore, by
S
which
the
first r
>F,-'F9
we have
^ (r+l)(2r+l)(3r+3r-l)
the required expression for the
is
sum of
(355)
the fourth powers of
integers.
PROBLEM
82.
FI,
I,
Given a
II.
series
and an assigned
offunctions,
F_ 3 F_
,
intermediate value,
F_^
To
n :
find
the corresponding* interval n.
First Solution
The
simplest
tion an approximate value of n,
method
and then
is
find
to determine
by
direct
by inspec-
interpolation
the values of the function corresponding to three or four closely equiThe latter
distant values of n that shall embrace the required interval.
is then readily found by a simple interpolation.
the following ephemeris find the time when the
that is,
9.7968280
logarithm of Mercury's distance from the Earth
The tabular quantities are here
n
9.7968280, to find n.
given
EXAMPLE.
From
F =
given for eveiy second Greenwich mean noon.

Date
1898
193
(from the former date) is somewhat greater than 0.4. Hence we take
the functions
9.7905482, and interpolate
by BESSEL'S Formula
n
F=
corresponding to
= 0.38, 0.41, and 0.44.
ferencing these values,
we
n
find
Thus, computing and
dif-
194
To
find z
from
this
equation,
we
neglect the
first
and thus obtain an approximate value which we

this manner we find
s?,
small term in
= F.-F-
(358)
now suffice
we obtain
This approximate value of z will

last term of (357).
Accordingly,
In
shall call x.
for substitution
the
in
(359)
whence, putting
(360)
we have
x
and equation (356) becomes
= m+x
Fm
Finally, to express
(*F'm ,
and
(361)
in terms of the differences
AZC
be expedient to employ STIRLING'S

Formula of interpolation, together with the expressions for F'm and
61.
The above solution may then be expressed
F'n as developed in
of the
given series F,
as follows
will
it
Determine
m =
Thence
Fm = F + ma +
an approximate value of
n, true to
the nearest tenth
of a unit.
find
D =
K
t
'
and
Here the
uPF^.
&
Bb
me
Cc 4- Dd^
(362)
= $x>K
= m x
-\-
differences are to be taken according to the schedule
on
the coefficients B, C, D,
being taken from Table II,
page 62
and C", I/, .... from Table Y. Finally, Table VII gives the value
and side argument *
of y for top argument
observing that y has
the same sign as K.
.
EXAMPLE.
Here we
Same
find
82.
as in
'==
195
0.40
and hence take from the given
table,
and from Tables II and Y, the quantities
m =
B =
C =
D =
E =
The computation
lows
of
=
ma =
-
= +144461.5
= + 15725
=
4137.5
rf
==
=-
0.40
+0.080
-0.056
-0.0056
+0.01075
9.7905482
Fm D
,
+
+
t
457
105
=
=
E =
C'
D<
1
and Z>2 by (362)
is
-0.08667
-0.02267
+0.01440
therefore
as fol-
196
EXAMPLE.
Solve the transcendental equation

I
T - 20
where
sin
T =
45
expressed in degrees of arc.

This equation may be written
is
F(T)
= T-
20 sin
T-
45
which by trial we find to be satisfied by a value of T not far from

63
hence we tabulate
62, 63, and 64, as follows
(T) for
T=

condition
Therefore, since the

that
F\T)
we
0,
of
maximum
or
minimum
197
requires
have, by neglecting 5th differences,

|
(a
^c)
^d^n + \cn^ + ^d n* =
(6
which determines the value of

the point of
maximum
or
n,
and hence,
also, the value
minimum of F(T).
This equation
(363)
of T, at
may be
solved by successive approximations, by first neglecting the

terms containing n? and n3 and afterwards substituting therein the
readily
approximate value of n thus found, and so on ; or, we may consider

the solution of (363) from the standpoint of Problem III,
which
may be regarded as the more direct of the two methods.
EXAMPLE. -- The following ephemeris gives the log radius vector

of Mars with respect to the Sun (log r)
Find the time of perihelion
.
passage of the planet.

Date
1898
198
Neglecting the
approximate value of
n,
and since
two terms of
last
135-:- 2914
for this value of
this equation,
we
have, for an
nearly;
0.046,
n the small terms
sensibly vanish,
we
obtain
as our final value

n
The
'date of perihelion
T =
F-i,
April 30"
PROBLEM V.
86.
FI,
passage
.)
therefore,
is,
d
h
April 30 8 .895
of numerical values (F_3 F_ 2

of any function F(T) which is analytic"// //
series
To find an approximate
unknown:
0.04633
X&X 24" =
+ 0.04633
Given a
.
135 -=-2914
algebraic expression for
F (T)
in
terms of the variable argument.
Let us put
T
= T-
(364)
and TAYLOR'S Theorem gives
F(T)
Upon
F'"(),
F(T)
F(t + r)
+ T F'(t)+
-F"(f)
F'"(f)+.
in
(365)
the
(365)
F'(t), F"(t),
=
8
'
which expresses
'
-)'
F(T)
numerical coefficients
+K--
EXAMPLE.
the following
century
O^+Ce-.
-)r<+.
as a rational integral function of

T
r,
with
(366)
known
being the value of the variable argument
counted from the fixed epoch
fifth
expressions for
as given by (175), we obtain
substituting
.
F(t)
t,
as defined
by (364).
From NEWCOMB'S Astronomical Constants we take

table of the mean obliquity of the ecliptic (e) for every
Tear
199
THE THEORY AND PRACTICE OF
200
Therefore,
if
we
shall
have
*
v.
where
T,
NTKIIPOLATION.
put
T!
we
HI,,,
+ *i
_.
I
_,.
-1
/
'
(= T--T^)
we
rf
(366)
r?
the value of the variable argument counted
is
from the assigned epoch 2\

usual methods the values of
tute these in (36(5a),
*"* "*"
Fm
F'm F'm
,
F','n
we compute by
if
Accordingly,
and
the
substi-
shall obtain the expression required.
As
an example, let us express the obliquity (e) as a function of

the time (T,) counted from the epoch 1600.0 in terms of a century as
the unit.
Reverting to the above

t
Whence we
we
table,
1500*
Tj
take
m =
1600"
0.20
find
*
Fm =
23
F'n
29' 28".69
-46".761
F'^
_0".0443
F'J,'
Substituting these values in the formula (366a),
we
+0".01088
obtain the
required expression, namely,
=
87.
23
29' 28".69
-46".76lTj -0".0222r? +0".0018lTj
GEOMETRICAL PROBLEM.
A circular well four feet in diameter
by a horizontal cylindrical shaft whose diameter

Find the volume of the portion of the shaft within the
is
centrally intersected
is
one
foot.
well.
Solution
parallel
to
its
Consider
axis, at
vertical
horizontal
section
of this rectangular section
and
2V.R 2 -a: 2
for its breadth, or height,

h
lamina
of the shaft
distance x from the latter,
having the differential thickness dx. Then,

and r, respectively, we
well and shaft by
or
= 2\l^^
if
we denote
shall
and
the radii of
have for the length

Therefore, the volume of the differential section
dV =
Ihdx
201
is
whence
-* )
2
(r'-x*) dx
I'pon substituting the given values of
and r
in this formula,
it
becomes
V =
tic
This expression belongs to the class of functions known as ellipintegrals, and therefore cannot be integrated directly.
Accordingly,
we proceed
it
will
to evaluate
V by mechanical
quadrature.
For
this
purpose
be convenient to put
x
sin
dx
cos 0<10
whence
and the preceding expression
V =
AVe now tabulate
TT -T-
18) as follows
V becomes
for
fcos
Jo
0\/16-sinW0
F (&) = o>cos
>|lG
(367)
sin
(where
o>
= 10
202
Accordingly,
we
t
take
*<o
85
....
and proceed by formula (259)
thus, observing that J'_
z/"j,
and j;+,, j;^, .... are all zero, and remembering that the factor w
has already been introduced, we find
:
'/L,
and
V
'Fl+t
3.1168 cubic feet
Various other problems and applications of a similar nature

indeed, Astronomy itself presents a large variety of
might be added
But the leading principles of our subject have already been
such.
88.
We
therefore feel confident in

developed, explained, and exemplified.
leaving the student who has thoroughly mastered these principles,
believing him fully capable of solving any further questions or prob-
lems that
may
arise in his practice.
203
EXAMPLES.
Derive the expression for the sum of the cubes of the
Ans.
integers.
1.
first
Find from the following ephemeris the instant when Autumn

commences
that is, the instant when the Sun's right-ascension (a)
2.
equals twelve hours.

Date
1898
204
TIIK
THEOKY AND PKACTJCE OF INTERPOLATION.

Year
APPENDIX.
ON THE SYMBOLIC METHOD OF DEVELOPMENT.
While many of the formulae and results in the foregoing
have been derived by somewhat indirect methods, yet the processes employed in every case have involved nothing but purely alge89.
text
braic operations and principles.

For the benefit of such students as
may be interested, we shall

devote a brief space to the more direct and potent form of
development known as the symbolic method. In this our only purpose
is to exhibit the simple manner in which the fundamental formulae of
now
may be deduced ; leaving the student to

the
broader field thus opened by suggestion.
upon
the text
90.
Let us define the symbol of operation
AF(T) = F(T+<a)
enter for himself
by the relation
F(T)
(368)
from which we formulate the following
DEFINITION
duces
the
increment
The
increment
a>
in
upon any function of T profunction which corresponds to the finite
The operation of
the
in the variable T.
relation (368)
may be more
briefly
= F^-F, =
expressed in the form
J'a
where n can have any value. Thus, taking n

the schedule on page 15, we have
(369)
0,
and referring to
(370)
Similarly
'"
- *
CT
Jf1
Tj*
"
(371)
APPENDIX.
206
Thus
it
is
evident that the effect of operating with

upon any t<iliis simply to form the first difference of that function and
lar function
Whence
the succeeding tabular value.
evident that
is
it
we have
AAF =
A A?', = A (J{) =
J['
(372)
A)
AAF. =
J'.'
AA
that the operation of

upon any tabular function
second difference bearing the same subscript. But this
double operation of A may be conveniently characterized by A 8 ;
hence we write
It follows
produces the
=
In like manner,
4'
&*F.
(373)
J'.'
denoting any integer, we have
(374)
A'/;
and,
more
generally,
A'/;
n being a non-integer,
(A A A
times)
F = J?
(375)
Let us now consider the operation of differentiating

with respect to T and multiplying the derivative by &>. Denoting the
operator in this process by D, we then have
91.
D^;
^p
*'
(376)
also
"
(377)
(ODD
times)
F =
n
oi
(()
(378)
<"'-^
The fundamental laws
or principles governing the combination

of symbols of quantity in algebraic operations are the following
92.
APPENDIX.
I.
The
Law, by virtue of which

a (p+q + r) = ap + aq + ar
Distributive
The Commutative Law, expressed by
II.
ab
III.
207
The Index Law, which

ar
the equation
ba
asserts the relation
a'
a r +'
proceed to show that the symbols of operation, A and D, when

itself or with symbols of quantity in the manner
indicated below, also obey these fundamental laws
and hence that,
wherever found in similar combinations, A and D may be treated alge-
We
combined each with
braically precisely as if they ivere themselves mere symbols of quantity.
We
shall first consider the
By
(1).
definition,
symbol A.
we have
A (*+/+ ....) = (F^+f^+ ....)- (*+/+

= (F^-F.) +(/+,-/)+ -...
= A t\ + A/. +
.
which proves the Distributive

(2)
The
Law
for the
factor a being a constant,
symbol A.
we have
= aFn+1 -aF = a(Fa+l -F,) =

n
thus showing that

combines -with constant quantities in accordance
with the Commutative Law.
(3) r
and
denoting positive integers, the relation (375) gives
A 'A'/; = A
(A'/;)
AV/i."
= J
r+<)
B
A.* + 'Fn
or
Therefore, so far as positive integral indices are concerned, the symbol

obeys the Index Law.
Retaining the limitations and the notation used above, similar

results are easily obtained for the operator D, as follows
'.)'.
APPENDIX.
208
df
Daf.
(2)
(~)aF
"r+<
(^L^l
.F.,
L-
o,
aD^,
D-+-.P,
within the limitations imposed

These relations prove that
the symbol D obeys the fundamental laws of algebraic combination.
To
94.
of
a limited extent
of
and D.
D" D~2
1
definitions,
Now
....
the
necessary to consider negative powers

2
and
meaning and use of A" A"
it is
are easily understood
thus, from the foregoing
we^have
A('^)
where 'FH
F.
defined as in the schedule on page 134. Then, in analogy

with the usual mode of expressing inverse functions, we may write
is
Whence we have
= F
(379)
which shows (1) that the operation of AA~'( = A) leaves the subject function unaltered, and (2) that negative jmivers of A al#o O/K-;/
the Index Law.
The
relation
A-'JF;
may be taken
the definition
as
'Fn
(380)
of
the operator
A"
.....
1
.
Similarly,
we
have
A-
/;
"Fn
A-
^,
'"/'
(381)
Again, consider the relation
D^ =
,g
which, from the point of view above taken,
F =
n
D-'y
(382)
may be
written
(383)
APPENDIX.
209
Then we have
= DF =
DD-H>
(384)
whence we see that negative powers of D likewise follow the Index

Lftn-.
we
Moreover, from equation (382),
dFn =
or
obtain
and therefore
F =
or 1 fr
which, with (383), gives
D-'y
It
follows that
More
tlinl
the
specifically
operation of
(385)
-jTwtJ
D"
is
equivalent to
Operating upon any function with
function with respect to
In like manner
an integration.
D"
integrates
divides the resulting integral by w.
and
we have
D'-Fn
m-^rjlT*
==
(386)
and so on.
Having thus dejined and explained the use of the symbols

A~ A" A, A, A
and D~ D" D, D, D
and having shown that these symbols may in general be combined
95.
of operation,
algebraically as if they were merely symbols of quantity, we now proceed to derive the fundamental relations of the text, as originally
proposed.
The theorem of
the change in sign of the odd .orders of

caused by inverting a given series of functions is easily
To this end, let us suppose that j;/ of the direct or given
proved.
becomes
series,
Then, since
[J[ ] when that series has been inverted.
9(>.
differences
r)
we have
-AF = F,-FW =
t
'Whence, regarding
(-A)^.
..
[,/;-],
-A
as operator,
(-A)^.
==
[//;"],
it
follows that
.
(-A)^ =
[//n
and therefore
[J<"]
(-A)'F,
which establishes Theorem
III.
= (-1/A^ =
(
(-1)'/*|"
(387)
APPENDIX.
210
97.
By
definition,
we have
A/-;
= F.H-
/;,
hence
where
e is
system of logarithms.
the base of the natural
We
have,
therefore,
1
which
is
98.
A =
(388)
the fundamental relation between
and D.
From, (388), we get
A = e-l =
D2
D8
D4
....
D+^+^+^+
(389)
and hence, by involution,
A =
A =
D2
D8
A'
D'
+
+
+
D8 +
D +
| D'
+iD + ....
|D +JD +.
'
D4
,* 4
(3f
2
1) D'+
(390)
These expressions are equivalent to the formulae (21).

Again, from the last of (390), we derive
2
that
D' +2
F.
is
J<"
where
for
efficients
brevity
of
i+1
,
w'F^
is
a^'^Fi'
a^FV
+ ....
we have written a,, a.,, ....

in (390).
D<+2 ....
Whence,
,
which
we have
the algebraic statement of
Theorem V.
(391)
to denote
if
F (T)
the co-
EE aT<
211
APPENDIX.
99.
Expressing the relation (388) in logarithmic form, we get

D
log.(l
+ A) =
A-+-V
whence
D
From
-A +
= A - A +|iA
= A - i}A"+ J A 3
D-
(392)
....
these relations the formulae (45)
(393)
or the equivalent group (165)
immediately follow.
We
next consider the question of reducing the tabular in19.

Since in the preceding
terval from a> to m<a, as discussed in
definitions of A and D the magnitude of the interval is arbitrary, we
100.
have here only to denote by 9 and d the corresponding symbols in

the reduced series
evidently the same relations will then exist between d and d as were found above for A and D. Thus we obtain
;
and
since,
= mL
= m
mD
(394)
by (388 j, we have
1
we must have
also
1
Whence we
+A =
+d =
e*
(395)
find
1-)-
m(m
-\
1)
m(ni
-\
!)(?
2)
A + ....
and therefore
i(i
1)
m(m
!)('
2)
(396)
which are equivalent to the relations expressed

101.
may
in
(64).
The equation
be written in the form
F,
(397)
212
APPENDIX.
Hence
is
the binomial 1
raise
to
-f-
may be
defined as an operator whose effect

r
subscript of the subject function. AV hence
by unity the
we have
(1
(1
+ A)^ =
+ A)'/; =
+ A)/\ = I\
+
(l
A)F = F
(l
I
\
and generally
(1
We
+ A)"/; = Fn
(399)
therefore obtain
F =
or
""
tl
which
""
to
^*Q
|*>
to
i"
the fundamental formula of interpolation due to
is
We
102.
now
find
convenient to introduce a
it
operation, which, ^from its similarity and relation to A,

nate V: this operator is defined by the equation
VF = Fi-F^ =
t
From
this relation
we
at
XEWTON.
new symbol of
we shall desig-
j;_,
(401)
once derive
(402)
whence
it
appears that the operation of

of order r which
upon any tabular function

upon the upward inwhereas the successive opera-
produces the difference
falls
clined diagonal through that function ;

produce, as already shown, those
tions of
differences falling upon

the downward diagonal line. Moreover, from the complete similarity
likewise
of character of these two operators, it is obvious that
follows the fundamental laws of algebraic combination.
The
we
relation
between
and
is
easily
*_,
found
thus,
from (401),
obtain
(1-V)/;
also,
(403)
from (397), we have

i
F<
(404)
APPENDIX.
Whence we
213
find
._,
= F
and therefore
1- V =
+ A)-
(l
(405)
which gives
log(l-V)
-log (l + A)
(406)
Again, combining (388) and (405), we obtain
- V =
As
103.
us
derive
the
'
e-
(407)
an immediate application of the preceding relations, let

formula (75).
By means of (388), equation (399)
becomes
whence, changing the sign of
we
n,
find
Therefore
r
which
is
NEWTON'S Formula
for
==343 +
backward
----
(408)
interpolation, as given
by
(75).
104.
Formula (66) of the
text
is
deduced by means of
easily
the identity
A =
Thus we
A'/-;
find
(l
+ A)'
-'(! + A)'
whence, by (399), we obtain
jm
"0
which
is
-*
F -iF
lr i-l T
f
4l
^ '~ F
^(3
1
t
1 t'- 2 )
- '^~ X
|2
the same as equation (66).
F
-
"
|;j"
'
I
'
'
APPENDIX.
214
We
now pass to the derivation of the fundamental formD

of
mechanical
ulae
log (1 -|- A), we have
quadrature. Since
105.
Whence,
+
+
(Â-^- ^-- ^-
interpreting the
\~'
.
member according
first
/;,
to
(385), and
the
term A-'F, as in (380), we find
This
is
the fundamental
formula (a) of (250)
quadrature, from which the

To obtain (I) of (250) in-
of
relation
at once derived.
is
we have only
,
//,',_,, X_ 2 , d',"-*,
volving the differences
employ the relation (406), and the above development becomes
log (i
"
(V-
+ A)!-'/; =
- f-
8
,
|-io g (i-v)r'^,
4
,
6
,
V- & V - A V 2
A-
to
9
o
V - ,81 J* V*4
T-Jo
the interpretation of which gives
^J'^-^J'^- tfv J^
-rf e 4^~,ffl v Jl*-
(411)
agreeing with formula (6) of (250).

106.
Similarly,
Now
the
first
we
obtain for the second integration
pair of terms
in
the
right-hand
member may be
written
(A-
+ A"
/;
= A-
(l
+ A)^,, = A-
2
/-,, +1
and therefore the preceding expression becomes
from which (324) immediately follows.
= "Fn+l
APPENDIX.
215
Again, we find
4
,
= (V-'-V-'+T'.v-oioV'-iioV'-^lisV'-a^V 5
Transforming the
first
two terms of the
last
(413)
)*'
expression,
we
ob-
tain
V- - V2
F,
= V-
(1
- V)
f\
= V-
+ A)- -^,.
1
(1
Xow, because
raises by unity the subscript

the operation of 1 -jof the subject function (101), it follows that the operation of
1
diminishes that subscript by one unit.
Accordingly, we
(l-j-A)"
have
and hence the
relation
(413) gives
"^+1 + '^- J hÛ- J U^-3- 5 Sn^^-Tiili^-51
(414)
which is equivalent to the formula (326). These expressions complete

the fundamental relations of mechanical quadrature.
written
(.*
and therefore th
from which (324) immt
TABLES.
TABLE
218
FOU
COEFFICIENTS
BINOMIAL
I.
XEWTON'S INTERPOLATING COEFFICIENTS.
TABLE
FOU
COEFFICIENTS
UINO.MIAL
I.
XEWTON'S INTERPOLATING COEFFICIENTS.
219
TABLE
220
FOB
COEFFICIENTS
STIRLING'S
II.
STIRLING'S INTERPOLATING COEFFICIENTS.
TABLE
e
r.
's.
^
I)
B
o
O
X
n
P
BO
II.
STIRLING'S INTERPOLATING COEFFICIENTS.
TABLE
222
FOH
COKFFICIENTS
HK*1KI.'S
III.
BESSEL'S INTERPOLATING COEFFICIENT-.
TABLE
FOK
COEFFICIENTS
-î:i.'s
I'.i
III.
BESSEL'S INTERPOLATING COEFFICIENTS.
223
TABLE IV.
224:
FOB
COKFFICIENTS
XEWTON'S COEFFICIENTS FOR F'(T}.
TABLE IY.
von
COEFFICIENTS
XEWTON'S COEFFICIENTS FOR F\T).
225
TABLET.
226
Fon
COEFFICIENTS
STIRLING'S COEFFICIENTS FOR
F'(T}.
TABLE V.
FOU
COEFFICIENTS
STIRLING'S COEFFICIENTS FOR F'(T).
227
228
FOR
COEFFICIENTS
TABLE VI.
BESSEL'S COEFFICIENTS FOR F'(T).
TABLE YI.
FOB
COEFFICIENTS
BESSEL'S COEFFICIEKTS FOR
F\T}.
229
230
TABLE
VII.
TABLE VII.
s
231
TABLE VIII.
232
COEFFICIENTS FOR COMPUTING

Fn = F no, (FJ + \ a + B/3 + Ty).
4-
BIBLIOGRAPHY.
LIST OF THE PRINCIPAL PAPERS, MEMOIRES, ETC., UPON THE SUBJECTS OF INTERPOLATION
AND MECHANICAL QUADRATURE.
Astrand
Viei-teljahrsschrift der
(J. J.).
Astronomischen Gesellsehaft, Vol.
(1875),
279.
p.
Baillaud
(B.).
Bienayme
Annales de 1'Observatoire de Toulouse, Vol.
B.I.
XVIII
Liouville, Journal de Mathematiques, Vol.
(Jules).
Boole (George).
Brassinne (E.).
II, p.
(1853), p. 299.
Treatise on the Calculus of Finite Differences, Chapters II and III.
Liouville, Journal de Mathematiques, Vol.
Brunnow (F.). Lehrbuch

Cauchy (Augustin). (a)
XI
(1846), p. 177.
der Spharischen Astronomic, p. 18.
(c)
Liouville, Journal de Mathematiques, Vol. II (1837), p. 193.

XI (1840), p. 775.
Vol.
Comptes Kendus,
Vol.
Ibid.,
(1844), p. 1183.
(d)
Connaissance des Temps, 1852 (Additions),
(b)
XIX
Chauvenet (Wm.).
Davis
(C. H.).
129.
Spherical and Practical Astronomy, Vol.
The Mathematical Monthly (Cambridge,
Doolittle (C. L.).
Encke
p.
(J. F.).
Treatise on Practical Astronomy,
(a) Berliner
(b)
Ibid., 1837, p. 251.
(c)
Ibid., 1852, p. 330.
(d)
Ibid., 1862, p. 313.
p. 79.
Mass.), Vol. II (1860), p. 276.

p. 70.
Astronomisches Jahrbuch, 1830,
The Mathematical Monthly (Cambridge,
Ferrel (William).
I,
p.
265.
Mass.), Vol. Ill (1861),
p. 377.
Gauss (Carl
Grunert
F.).
(J. A.),
Werke, Vol. Ill, p. 265.

Archiv der Mathematik und Physik, Vol.
(a)
Ibid., Vol.
(b)
XX (1853),
XIV (1850),
p. 225.
p. 361.
Hansen
(P. A.),
(a) Abhandlungen der Koniglich Sachsischen Gesellsehaft der Wissenschaften (Leipzig), Vol. XI (1865), p. 505.
Tables de la Lune, p. 68.
(b)
Herr und Tinter.

Jacobi
(1>)
(C. G. J.).
Ibid., Vol.
Klinkerfues (W.).
Lehrbuch der Spharischen Astronomic, Chapter
II.
(a) Crelle's Journal, Vol. I (1826), p. 301.
XXX (1846),
p. 127.
Theoretische Astronomic (2d edition, 1899), pp. 67 and 490.

233
BIBLIOGRAPHY.
234
Kliigel's Mathematisches Worterbuch.
See article "Einschalten,"
wliicli
includes
brief history of the subject.
() (Euvres, Vol. V, p. 663.

Vol.
Ibid.,
VII, p. 535.
Lagrange
(b)
Laplace
(J. L.)-
Mecanique Celeste, Vol. IV, pp. 204-207.
(P. S.).
Le Verrier
(U.
Annales de 1'Observatoire de Paris (Memoires), Vol.1, pp.121,
J.).
129, 151, 154.
Loomis
Practical Astronomy, p. 202.
(Elias).
Maurice
Connaissance des Temps, 1847 (Additions),
(Fred.).
British Association Report, Vol.
Merrifleld (C. W.).
Newcomb
Newton
Radau
p.
321.
Principia,
Book
III,
Lemma
p. 56.
V.
Crelle's Journal, Vol. II (1827), p. 252.
Olivier (Louis).
Oppolzer
L (1880),
Logarithmic and Other Mathematical Tables,
(Simon).
(Isaac).
p. 181.
Lehrbuch zur Bahnbestimmung, Vol.
(T. R.).
(Rodolphe).
(a)
Liouville, Journal de
and 596.
II, pp. 1
Mathematiques, 3d Series, Vol. VI
(1880), p. 283.
Bulletin Astronomique, Vol.
(ft).
VIII (1891), pp. 273, 325, 376, 425.

"
See article Interpolation."
Rees's Cyclopedia.
Sawitsch
(A.).
volume
Tisserand
(b)
(c)
Abriss der Practischen Astronomic, Vol.11,
p.
416;
or see
the one
edition, p. 818.
(F.).
(a)
Ibid., Vol.
Comptes Rendus, Vol. LXVIII (1869),
LXX (1870),
Traite de Mecanique Celeste, Vol. IV, Chapters
Valentiner (W.).
Handworterbuch der Astronomic, Vol.
Watson (James C.).

Weddle (Thomas).
p.
1101.
p. 678.
and XI.
II, pp.
41 and 618.
Theoretical Astronomy, pp. 112, 335, 435.
Cambridge and Dublin Mathematical Journal, Vol. IX (1854),
p. 79.
68*

Rice (The Theory and Practice of Interpolation) (1899)

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Rice (The Theory and Practice of Interpolation) (1899)

Diunggah oleh

Hak Cipta:

Format Tersedia

THE

THEOKY AND PRACTICE

MECHANICAL QUADRATURE, AND OTHER IMPORTANT PROBLEMS

WITH THE REQUISITE TABLES.

ASSISTANT IN THE OFFICE OF THE AMERICAN EPHEMERIS,

IN preparing the following

originality, either of subject matter

author has attempted no marked

been written of Interpolation, Quadratures,

Indeed, sufficient has hitherto

Yet of the numerous contributions

to these allied subjects,

a consequence the author has

repeatedly felt the need of a work which would

exclusive of other matter

a simple, practical, yet comprehensive discus-

concerning Differences, Interpolation, Tabular Differ-

and Mechanical Quadrature;

But while viewing the matter

in this practical sense, the writer

and developments are

vidual methods which in

the customary forms of analysis.

have been employed in preference to

previous writers for valuable

KNCKE, LOOMIS, NEWCOMB, and SAWITSCH as most valuable sources

WALLIS, MOUTON, COTES,

have purposely been omitted because

present century, however, the author believes that

have been included, and trusts

works of BOOLE, CHAUVENET,

desires especially to mention the

detected will be regarded merely as oversights.

Special care has been given to

preparation and printing

a considerable cost of labor,

a few of these tables have appeared before, several are here

thanks and appreOffice, and to Mr. M. E.

Naval Observatory, for much valuable service and many useful

suggestions received during the

Mr. E. C. RUEBSAM, of the Nautical Almanac

preparation of this treatise.

simple justice even demands

commendation of the publishers, whose uniform courtesy, accuracy and

D.C., December, 1899.

General remarks concerning tabular functions and the construction of

Effect of inverting a given series.

Differences of two combined series.

Irregularities in the differences of functions

Detection of accidental errors by differences,

General properties of differences.

which only one function requires correction,

Remarkable formal relation between the expressions for

Converse of the foregoing proposition.

which are not mathemati-

any rational integral function of the Mth degree

and rhnrNumerical illus-

Change of the argument interval from <a to m<a

u>F'(t), <a"F"(t), etc., in

Practical result of the foregoing investigation.

Expression of any difference in terms of tabular functions,

Expression of any tabular function in terms of

reduction of tabular interval,