Chemistry
Library
QB
145
D72i
1900
THE LIBRARY
OF
THE UNIVERSITY
OF CALIFORNIA
LOS ANGELES
in
m
A TREATISE
ON
PRACTICAL ASTRONOMY,
AS APPLIED TO
GEODESY AND NAVIGATION.
C. L. DOOLITTLE,
Professor of Mathematics and Astronomy, Lehigh University
FOURTH AND REVISED EDITION.
SECOND THOUSAND.
NEW YORK
JOHN WILEY & SONS
LONDON
CHAPMAN & HALL, LTD.
1900.
COPYRIGHT, 1885,
BY C. L. DOOLITTLE.
Braunworth, Munn & Barber
Printers and Binders
Brooklyn, N. Y.
Chemist^
PREFACE.
THE following work is designed as a text-book for univer-
sities and technical schools, and as a manual for the field
astronomer. The author has not sought after originality,
but has attempted to present in a systematic form the most
approved methods in actual use at the present time.
Each subject is developed as fully as the necessities of the
case are likely to require; but as the work is designed to
be a practical one, those methods and developments which
have merely a theoretical or historic interest have been ex-
cluded.
Very complete numerical examples are given illustrative
of all the prominent subjects treated. These have been
selected with care from records of work actually performed,
and will show what may be expected in circumstances ordi-
narily favorable.
Such auxiliary tables as are applicable only to special prob-
lems will be found in the body of the work; those which
have a wider application are printed at the end of the volume.
The universal employment of the method of Least Squares
in work of this kind has led to the publication of an introduc-
tion to the subject for the benefit of those readers who are
not already familiar with it. This introduction develops
the method with special reference to the requirements of
6
45
IV PREFACE.
this particular class of work, and it has not been the design
to make it exhaustive.
For the materials employed original papers and memoirs
have been consulted whenever practicable. The illustrative
examples have been drawn largely from the reports of the
Coast and other government surveys. For most of the exam-
ples of sextant work, as well as for many valuable' sugges-
tions, the author is indebted to his friend and former col-
league Prof. Lewis Boss. Much assistance has also been
derived from the excellent works of Chauvenet, Brunnow,
and Sawitsch.
Fully appreciating the difficulty of eliminating all mis-
takes from a work of this character, the author can only hope
that this one may not prove to be disfigured by an undue
number of such blemishes.
C. L. DOOLITTLE.
BETHLEHEM, PA., May 20, 1885.
CONTENTS.
INTRODUCTION TO THE METHOD OF LEAST SQUARES.
PACK
Errors to which observations are liable i
Axioms 2
The law of distribution of error 3
The curve of probability 5
Determination of the law of error 6
Condition of maximum probability 1 1
The measure of precision 12
The probable error 13
The mean error 15
The mean of the errors 17
Precision of the arithmetical mean 18
Determination of probable error of arithmetical mean 20
Probable error of the sum or difference of two or more quantities 22
Principle of weights 23
Probable error when observations have different weights 26
Comparison of theory with observation 29
Indirect observations 32
Equations of condition Normal equations 35
Observations of unequal weight 36
Arrangement of computation 37
Computation of coefficients by a table of squares 41
Solution of normal equations 43
Proof- formulae 47
Weights and probable errors of the unknown quantises 54
Mean errors of the unknown quantities 65
vi CONTENTS.
INTERPOLATION.
PACK
Notation 7 1
General formulae of interpolation 7 2
Arguments near beginning of table 78
Arguments near end of table 82
Interpolation into the middle 84
Proof of computation.. , 85
Differential coefficients 86
The ephemeris Lunar distances 92
PRACTICAL ASTRONOMY.
CHAPTER I.
THE CELESTIAL SPHERE TRANSFORMATION OF CO-ORDINATES*
Spherical co-ordinates 100
The horizon Altitude Azimuth 102
The equator Declination Hour- angle Right ascension 103
The ecliptic Longitude Latitude 104
Having altitude and azimuth, to find declination and hour-angle 107
Having declination and hour-angle, to find altitude and azimuth 112
To find hour-angle of star in the horizon 114
To find distance between two stars ' 115
CHAPTER II.
PARALLAX REFRACTION DIP OF THE HORIZON.
Definitions 120
To find equatorial horizontal parallax 120
Parallax at any zenith distance 121
Form and dimensions of the ear-th 122
Reduction of the latitude 124
CONTENTS. vii
PAGE
Determination of the earth's radius 127
Parallax in zenith distance and azimuth I3 1
Parallax in right ascension and declination 142
Refraction J 53
Descartes' laws 154
Bessel's formula for refraction 155
Refraction in right ascension and declination 157
Dip of the horizon 160
CHAPTER III.
TIME.
Sidereal time 163
Solar time 164
Inequality of solar days 164
Equation of time 166
Sidereal and mean solar unit 168
To convert mean solar into sidereal time 170
To convert sidereal into mean solar time 172
CHAPTER IV.
ANGULAR MEASUREMENTS THE SEXTANT THE CHRONOMETER AND
CLOCK.
The vernier 174
The reading microscope The micrometer 176
Eccentricity of graduated circles 180
The sextant 183
The prismatic sextant 186
Adjustments of the sextant 188
Method of observing 190
Index error 194
Eccentricity of the sextant 196
The chronometer 207
Comparison of chronometers 208
The clock 209
The chronograph 211
VJii CONTENTS.
CHAPTER V.
DETERMINATION OF TIME AND LATITUDE METHODS ADAPTED TO
THE USE OF THE SEXTANT.
PACK
Determination of time
By a single altitude of the sun 215
By a single altitude of a star 220
Conditions favorable to accuracy 222
Differential formulae 223
Equal altitudes of a star 228
Equal altitudes of the sun 230
Latitude 2 33
By the zenith distance of a star on the meridian 233
By a circumpolar star observed at both upper and lower culmination. . 235
By the altitude of a star observed in any position 236
By circummeridian altitudes 238
Gauss' method of reducing circummeridian altitudes of the sun 247
Correction for rate of chronometer 250
Latitude by Polaris 256
Correction of altitudes for second differences in time 260
Probable error of sextant observation 265
CHAPTER VI.
THE TRANSIT INSTRUMENT.
Description of instrument 269
Value of level 276
Adjustments of instrument 279
Methods of observing 283
Theory of the transit 284
Diurnal aberration 289
Equatorial intervals of threads 291
Reduction of imperfect transits 294
Determination of constants:
The level constant 295
Inequality of pivots 296
The collimation constant 302
The azimuth constant 305
Personal equation 316
Probable error and weight of transit observations 318
Application of the method of least squares 322
CONTENTS. IX
PAGK
Correction for flexure 335
The transit instrument out of the meridian 338
Transits of the sun, moon, and planets 331,
Correction to moon's defective limb 343
The transit instrument in the prime vertical 348
Mathematical theory 352
Errors in the data 356
Reduction to middle or mean thread , 356
Application of least squares to prime vertical transits 372
CHAPTER VII.
DETERMINATION OF LONGITUDE.
By transportation of chronometers 379
By the electric telegraph 388
By the moon 398
By lunar distances 400
By moon culminations 413
By occul tat ions of stars 423
Prediction of an occultation 435
Graphic process of prediction 443
Computation of longitude 444
Correction for refraction and elevation above sea-level 460
Observations of different weights < 474
CHAPTER VIII.
THE ZENITH TKI.ESCOPE.
Description of instrument 478
Adjustment of instrument 481
The observing list 484
Directions for observing .... 48"
Value of micrometer screw 488
Value of micrometer when level is not known 493
General formulae for latitude 501
The corrections for micrometer, level, and refraction 502
Reduction to the meridian : 504
Combination of individual values of the latitude 507
Ya'ue uf micrometer from latitude observations 509
X CONTENTS.
CHAPTER IX.
DETERMINATION OF AZIMUTH.
PAGE
The theodolite .- 521
The signal 523
Selection of stars Method of observing v . 524
Errors of collimation and level 525
Azimuth by a circumpolar star near elongation 526
Correction for diurnal aberration 530
Circumpolar stars at any hour-angle 535
Correction for second differences in the time 537
Conditions favorable to accuracy 542
Azimuth when time is unknown 543
Azimuth determined by transit instrument 546
Circumpolar star at any hour-angle 552
CHAPTER X.
PRECESSION NUTATION ABERRATION PROPER MOTION.
Secular and periodic changes 559
Mean, apparent, and true place of a star 560
Precession 560
Struve and Peters' constants 563
Bessel and Leverrier's constants 564
Precession in longitude and latitude 564
Precession in right ascension and declination 571
Proper motion 578
Expansion into series 583
S'ar catalogues and mean places of stars .'. 590
Nutation 598
Aberration 603
Reduction to apparent place 609
The fictitious year 617
The Tabula Regiomontance 620
Conversion of mean solar into sidereal time 623
LIST OF TABLES . . 626
INTRODUCTION TO THE METHOD OF
LEAST SQUARES.
1. When a quantity is determined by observation, the re-
suit can never be regarded otherwise than as an approxima-
tion to the true value. If a number of measurements of the
same quantity are made with extreme care, no two of the
values obtained will probably agree exactly ; at the same
time none of them will differ very widely from the true one.
There is a limit to the precision of the most refined instru-
ment, even when used by the most skilful observer, and
therefore the determination of a quantity depending on in-
strumental measurement, however carefully made, must be
imperfect. It becomes then a problem of great .practical
importance to determine how the mass of data resulting from
observation shall be combined so as to give the best possible
value of the quantity sought. The theory of probabilities
furnishes the basis for such an investigation.*
2. Observations are liable to errors of three kinds :
First. Constant errors, or those which affect all observa-
* The reader is supposed to be familiar with the theory of probability as de-
veloped in the ordinary text-books on algebra. See, for instance, Davies
Bourdon, edition of 1874, p. 322, or Olney's University Algebra, p. 294.
2 LEAST SQUARES. 3.
tions of a given series alike. These may result from a
variety of causes, such as errors in the instruments used,
personal error of the observer, errors in the constants of re-
fraction, parallax, etc., used in the reduction of observations.
A proper investigation will generally show the magnitude of
such errors, and consequently the necessary corrections at
least the more important ones. We shall suppose the data
to which our discussion applies freed from such errors, as
their investigation does not come within the scope of this
subject.
Second. Mistakes, such as recording the wrong degree in
measuring an angle, or the wrong hour in the clock reading.
When such errors are large they are not likely to give much
trouble, as their true nature appears at once. When they are
small they may prove embarrassing. The present discussion
does not apply to them, and we shall suppose that no undis-
covered mistakes have been made.
Third. Errors which are purely accidental. It is to these
that our present investigation applies.
At first sight it might seem that such purely accidental
errors were entirely outside the sphere of mathematical in-
vestigation, but we shall see that they follow a very definite
(aw, and that theory is verified in an exceedingly satisfactory
manner by observation.
3. We shall assume as the basis of our investigation the
following axioms :
I. If we have a series of direct measurements of a quantity,
all made with equal care, the most probable value of the
quantity will be obtained by taking the arithmetical
mean of the individual measurements.
II. Plus and minus errors will occur with equal frequency.
III. Small errors will occur with greater frequency than
large ones.
4- DISTRIBUTION OF ERRORS. 3
Various attempts have been made to prove the first of
these as a proposition. All such proofs are more or less
unsatisfactory, and for elementary purposes it is more ex-
pedient to assume its truth at once. The "most probable
value" there mentioned must- be understood as the value
which most nearly represents the given data, and from the
evidence furnished by this series of observations alone it is
the best attainable approximation to the true value.
The principles are supposed in all cases to be applied to a
large number of observations; the larger the number the
more closely will the results correspond to the laws assumed.
The Law of Distribution of Error.
4. Let x be a quantity whose value is to be determined
by observation either directly or indirectly.
Let MV M^ MV . . . M m be the individual values obtained.
Then regarding M^ as a determination of the unknown
quantity x, its error will be (M l x). Similarly, (M, x],
(M 3 x), . . . (M m x] will be the errors of the other ob-
served values.
Let us write
(M, ~x} = A, (M, -x) = A... (M m - x) = A m . (i)
Let j, = the probability of the occurrence of the error A l ;
y^ = the probability of the occurrence of the error A^
y m = the probability of the occurrence of the error //,,
Then our second and third axioms assume a law as existing
such that the probability of a given error occurring will be
LEAST SQUARES. % 4.
a function of the magnitude of the error itself. We shall
therefore have the equation
y = vW, ........ (2)
in which A represents any error, and y the probability of its
occurring.
If this reasoning seems obscure, a different application of
the same logic may possibly assist in comprehending it.
Suppose we have a large number of tickets in a lottery-
wheel. Let a definite proportion of them be numbered i,
a certain other proportion respectively 2, 3, etc. Then
the probability of drawing any given number from the wheel
will be a function of the number itself viz.:
Suppose i ticket in every 55 numbered I
2 tickets in every 55 numbered 2
3 tickets in every 55 numbered 3
10 tickets in every 55 numbered 10.
Then every ticket would have one of the numbers, i, 2, 3, 4,
5, 6, 7, 8, 9, 10, and
The probability of drawing a i would be ^;
The probability of drawing a 2 would be ^-;
The probability of drawing a 10 would be ^g-.
Or if k represents any one of the numbers from i to 10 in-
clusive, the probability of drawing a k will be - =/(), or
y f(fy 1S tne equation which represents the probability of
drawing a k.
If now we were ignorant of the relations existing between
the successive numbers i, 2, 3, etc., and the relative number
5-
CURVE OF PROBABILITY.
5
of tickets so marked, we could, by drawing a sufficiently large
number of tickets from the wheel, determine it, at least ap-
proximately. In this case we have to determine the proba-
bility of a given event occurring, viz., that of drawing a
ticket marked with any given number k. In the above prob-
lem we have also to discuss the probability of a certain event
occurring, viz., that of the appearance of any given error A
in any one of our observations taken at random.
The Curve of Probability.
5. In the equation y =
(A -f- i), y-i = (p(A -\- 2), and / 3 = q>(A -f- 3).
If now the limits between which the errors of our series he extend to 10",
we see that the probability _y, will differ but little from y 3 , and the sum of all the
probabilities j', -f-Js -J- y 3 will differ but little from jy, or
8 Ay =
(A?) -\- log
(M m - x]
~ --- =
d log y(J/, .r) ^ log
d(M~- x] d(M~-
This equation gives the means of determining x as soon
as the form of the function q> is known, and this can best be
determined bv considering a particular case. As this func-
tion is strictly general, if we have once determined its form
in a special case the result will be applicable to all cases.
We have assumed as an axiom that in the case of direct
measurement of the quantity sought the most probable value
will be the arithmetical mean of the individual measurements.
This principle will furnish the basis for investigating the
form of the function (p.
In case of direct measurement we have for the unknown
quantity
6. DETERMINATION OF THE LAW OF ERRORS. Q
I
which may be written
(M l -x} + (M,-x)+...+(M m -x)=o. . (8)
Equation (6) may be written
("-*) +<".->
Comparing equations (8) and (9), we see that since the
quantities (M l x}, (M z x), etc., are independent of each
other, these equations can only be satisfied when the coeffi-
cients of (M l x), (M^ x), etc., in (9) are respectively equal
to the same constant quantity. We have therefore
;- *) _ = .
d\ no- m( M r\
= k. (10)
(M, - x) d(M, -~xj ~ (M, - x)d(M,-^}
d log- cp( M m x)
(M m x) d(M m x)
Writing for (M x] in general A, we have
d log cp( A) = kAdA,
and, by integration, log tp(A) = %kA* -\- log c,
c being the constant of integration,
or (p(A) = ce&* ........ (11)
From axiom III. it appears that as A increases this quan-
tity must diminish, and this requires the exponent of e to be
10 LEAST SQUARES. 7.
negative. As J 2 cannot be negative, it follows that k must
be so. Writing therefore \k = If, our equation becomes
(12)
7. Let us now consider the constant of integration c. This
may be determined by substituting the value of 9>(J) in (4),
giving us
a special form of the integral known as the gamma function.
For the purpose of integrating the expression, place nA = /.
Then dA = , and we have
As / in this expression is involved only in the quadratic
form, we evidently have
-*dt = 2e-^dt = 2A
(in which we write the integral equal to A for convenience).
In the definite integral jf e~ fl dt the value will be the same
if we write another symbol instead of t. Therefore
Multiplying both members of this equation by f e~ fl dt, we
have
8. CONDITION OF MAXIM I'M PROBABILITY. I I
In the second member of this equation write v = tu,
dv tdu. Then
/^-* a d+)
' "" )//= "icT+vy
which between the given limits becomes -f ^
Therefore
A* _ L /"" '-*-*! (.3)
In this equation the constant h will require further con-
sideration ; hut if we assign any arbitrary value, as unity, to
h we can readily construct the locus of the equation. It will
at once appear that the general form will be that shown on
page 5.
Condition of Maximum Probability.
8. Substituting in equation 5) the values of
etc., from (13), it becomes
12 LEAST SQUARES. 9.
From this equation we see that P will increase in value as
the exponent of e diminishes, or P will be a maximum when
A* -f- A* -{-... -J- A^ is a minimum, thus giving us the im-
portant principle
The most probable value of the unknown quantity is that which
makes the sum of the squares of the residual errors a minimum.
From this principle comes the name Method of Least Squares.
The Measure of Precision.
9. Let us now consider the constant h.
Substituting in equation (3) the value of (A"},
(A") of the value A",
etc., and in all
+ J, + 4 + . . . dm. = 2mcp(A'}A' -f
+ 2mcp(d'")A r " + etc.
From the definition of the mean error e we shall have
_ 2mcp(A')A" -f 2tn(p(A"}A"* -f 2w(/r')J ///2 -f etc.
Expressing this by an integral, by the same method of rea-
soning as was used in deriving equation (3) we have
This equation expresses a relation between e and h. To
effect the integration, let as before hA = t. Then dA = -y-,
and we have,
Va
Integrating this by parts by placing n = t and dv = e- fl tdt>
and substituting in J udv = uv J vdu, we find
which readily gives e* = -y- a (20)
13- THE MEAN OF THE ERRORS. I/
Substituting the value of h from (19), we have
= I.4826r; )
r = .6745, 1 ....... (2I)
From these r is readily computed when we know e, and vice
versa.
7/fc Mean of the Errors.
13. Another quantity which is much used as an auxiliary
for computing r is The Mean of the Errors. This must not
be confused with the mean error. It is thus defined :
The Mean of the Errors is the arithmetical mean of the differ-
ent errors all taken with the positive sign.
Let ?/ = the mean of the errors. Then to determine the
relation between // and r we proceed in a manner similar to
that followed in the previous section. As before, let
d', A" , A'" , etc. = the individual errors.
'\
J
Combining equations (27) and (24), we readily find
. = ,.2533 -4=; r .= 0.8453
Ll J .
^ ( r,7 o> ' (28)
In these expressions [+ z/] represents the sum of the residuals
all taken with the positive sign.
These simple formulas (27) and (28) are of great practical
value. When the number of observations is not large the
values given by (27) will be a little more accurate than those
* From what precedes we see that this assumption would be rigorously true if
the number of observations were infinite.
22 LEAST SQUARES. 1 6.
by (28), but when the number is large (28) will be sufficiently
accurate for practical purposes, and the facility with which
they are applied is something in their favor.
Probable Error of the Sum or Difference of Two or More
Observed Quantities.
16. Let us next suppose the unknown quantity x, instead
of being directly observed, to be the sum or difference of two
or more quantities whose values are obtained by direct
measurement ; viz. :
Let x y, y in which y, and y^ are independent of each
other and whose values are directly observed.
Let the individual errors of observation be
For 7,, J/, J/', . . . A\
Forj,, 4', 4",... A.
The errors of the individual determinations of x will then be
(j/ J/), (j," z/;o, . . . (4 m A m );
and if is the mean error of a determination of x, we shall
have
m? = (J/ j a y 4. (j/' 4") + . . . + (4 * j,*)'.
Expanding and making use of the symbol for summation,
we' = [J,J,] 2[J,J a ] + [J 9 J 9 ],
Let f, and e t = the mean errors of a measurement of y l and
/, respectively. Then since, for reasons before explained,
I/- PRINCIPLE OF WEIGHTS. 2 j
the middle term ([J^]) may be regarded as vanishing in
comparison with [A^J an< 3 \_^^^\, we shall have
m? me* -f- w^ 2 ,
or * = V^~+~C (29)
In a manner precisely similar we may extend the method
to the sum or difference of any number of observed quanti-
ties, so that in general if we have x = y^ y^ . . . _ y mt
the mean errors being respectively , e,, . . . m , we shall
have
= *V + < J + V + + *' = nl. (30)
Suppose next that we have x = a^y^ a^y^ . . . a m y m ,
in which a v a v . . . cr m are constants. If, as before, e,, f a , . . .
f m are the mean errors of y )> y m , then the mean errors
of a t y t , a^y v . . . a m y m will be respectively a v a t e v . . . a m m ,
and the mean error of x
2 C = vtv]. . (31)
Principle of Weights.
17. In the foregoing we have assumed all the observations
considered to be equally trustworthy, or, as it is expressed
technically, of equal weight. As will readily be seen, we
shall frequently have occasion to combine observations of
different weights. It is therefore important to ascertain
how to treat them, so that each shall have its proper influ-
ence in determining the result.
Confining our discussion for the present to the case of a
directly observed quantity, the most elementary form of the
24 LEAST SQUARES. 1 7-
problem will be that where the quantities combined are them-
selves the arithmetical means of several observations of the
weight unity. Thus, suppose the quantity x to be deter-
mined from m such observations ; the most probable value
of x' will then be
From a second, third, etc., series of m", m'", etc., observa-
tions we have respectively
_
Combining all these individual values, we have for the
most probable value of .*-
' + '" + +
The value of x will not be affected if we multiply both nu-
merator and denominator of this fraction by any constant a ;
viz.,
_ am' x' -J- am"x" -J- am'"x'" -}-... / \#
;' -j- am" -\- am'" -j- . . . ' ' \3 )
I/. PRINCIPLE OF WEIGHTS. 2$
in which we may regard am', am" , etc., as the respective
weights of x', x", etc. a may be integral or fractional.
From this we see that the weights are simply relative quan-
tities and are in no case to be regarded as absolute.
From the foregoing we have the following practical rule :
When observations are to be combined to which different weights
are to be ascribed, the most probable value of the unknown quantity
will be obtained by multiplying each observation by its weight,
and dividing the sum of the products by the sum of the
weights.
It is clear that the difference of weights may result from
a variety of causes other than the simple one considered
above ; as, for instance, one series of observations may be
made with a more accurate instrument than another, or by a
more skilled observer. Thus, for example, it may be the
case that ten measurements made by one observer will have
as much value as twenty made by another. If the weight of
an observation of the first series be unity, one of the second
would only be entitled to a weight of one half ; or more gen-
erally,
Letting/ = the weight of an observation of the sepond series,
Then 2/ = the weight of an observation of the first series.
If then we have a series .* x n x v etc., of observations of
the weights A A A etc " an ^ consequently
A+A+A + .--
as the most probable value of x, it is evident that, whatever
may have been the cause of this difference of weight, we may
consider each value x t , x^ etc., as derived from /,,/,, etc., in-
dividual observations of the weight unity. Let
2 6 LEAST SQUARES. 1 8-
_ t he mean error of an observation of the weight unity ;
etc., the mean errors of # x v etc.
The whole number of observations being equal to /, + A
_|_^ s _j_ . . . [^] observations of the weight unity or of the
mean error e, we have for the mean error of*, from (25),
nfl
(34)
77^ Probable Error when Observations have Different Weights.
18. The mean taken according to weights, as in equation
(32) or (32)*, is sometimes called the General Mean. In order
to derive the formula for the probable error in this case, let,
as before, 8 be the error of the general mean x n \ viz., x x = 6.
Then, the notation being as before, we have
J i = v t S, A^ = v^ 3, 4, = v t 6, etc.
The error A l belongs to x l and therefore appears /, times ;
The error A^ belongs to x^ and therefore appears / 2 times;
Therefore [/^J] = [pvu] - 2\_pv\d + [p]d\
For the same reason as in previous cases \_pv\ may be dis-
regarded as being inappreciable in comparison with the other
terms, when we have
1 8. OBSERVATIONS OF DIFFERENT WEIGHTS.
Substituting for d the mean error of x from (34), we- have
= \pvv\ + 6*.
Now, as *! is equivalent to/ : observations of weight, unity,
there will be the equivalent of /, errors equal to J, ; and s l
being the mean error of x^ we shall have
Whence from (33),
Similarly,
And m being the whole number of quantities, or observa-
tions, Xv x t , etc., we have
Our equation therefore becomes zc a = [/^ -j- **, from which
and from (34),
and from (21),
(35)
(m - I)' J
; in these formulae is the number of individual observations,
or quantities, * x v etc., and must not be mistaken for the
sum of the weights.
It will be evident upon a careful comparison of these ex-
pressions with the formulae (27) that we should have reached
28 LEAST SQUARES. IQ.
the same result by multiplying each quantity * x etc., by
the square root of its weight, and then proceeding exactly as
we have previously done with observations of equal weight.
We have therefore established the following rule which we
may apply in combining observations of different weights :
First reduce all observations to a common unit of weight by
multiplying each by the square root of its weight, then combine
them precisely as if they had originally been of equal weight.
For examples of the application of the formulae see pages
515 and 516.
General Remarks.
19. We have hitherto considered only those cases where
the unknown quantity is derived in the simplest manner from
observation, viz., by direct measurement or by the sum or
difference of directly measured quantities.
Before proceeding to the more complex cases a few general
remarks may not be out of place.
Equation (13), which represents the law of distribution of
error, and on which the subsequent discussion is based, rests
upon two hypotheses neither of which is ever fully realized
in practice, viz., that the number of observations is infinite,
and that they are entirely free from constant errors, i.e.,
errors which affect all alike. The formulas deduced when
applied to the cases which actually arise can give us only
approximate results, although they will be the best attainable
approximations from the given data. This is particularly to
be borne in mind when the number of observations is small.
The probable errors in such cases are apt to be entirely illu-
sory, and in general are only reliable when the number of
observations is large enough to exhibit approximately the
law of distribution of error derived from the hypothesis of
an infinite series of observations.
20. COMPARISON WITH OBSERVATION. 29
The second hypothesis mentioned above, viz., that con-
stant errors do not exist in our data, can never be fully realized,
and this fact is often the source of great annoyance and un-
certainty in combining observations taken under different
conditions. Such errors arise from a variety of causes, some
easy to investigate and others not at all so. It is of very
frequent occurrence that a result derived from a single series
of observations will give a small probable error, and yet differ
widely from that derived from a second series to all appear-
ances equally good. It sometimes happens that computers
who are puzzled by such occurrences attribute the difficulty
to faults in the method, the truth being that they are due to
the presence of a class of errors with which the method does
not profess to deal.
The remedy for this difficulty is to vary as much as pos-
sible the conditions under which the observations are made,
and in a manner calculated to eliminate as far as possible
those constant errors which cannot be investigated.
Comparison of Theory witJi Observation,
20. The test of theory is its agreement with observed facts.
We may in this manner test the truth of the law which we
have derived for the distribution of errors.
We have the probability that an error falls between the
limits a expressed by the equation
In accordance with the theory of probabilities, / here is a
fraction which expresses the ratio of the number of errors
3 o LEAST SQUARES. 2O.
between a to the whole number. If then the number of
observations is m, the number of errors between a will be
To test the law expressed by this formula we have only to
compute the probable error of the series of observations under
consideration by (27) or (28), and then h by (19). The value
of the integral will then be obtained from Table I., and we
shall be in possession of everything necessary for comparing
the number of errors between any two limits as indicated by
this formula with the number shown by the series of observa-
tions. Many such comparisons have been made, and always
with satisfactory results, when the number of observations
compared has been large. A perfect agreement is of course
not to be looked for, as our formula has been derived on the
theory of an infinite number of observations ; and further, we
are not in possession of the true errors for comparison with
the formula, but the residuals instead, which will always differ
from the errors unless we are in possession of the absolutely
true value of the unknown quantity-
As an illustration of the above the following tabular state-
ment gives the result of a comparison with theory of the
errors of the observed right ascensions of Sirius and Altair.
The example is given by Bessel in the Fundament a Astrono-
mice.
In a series of 470 observations by Brad lev the probable
error of a single observation wns found to be r = o".2637,
whence h = 1.80865. Therefore for the number of errors less
than ".i the argument of Table I. will be t = hA = .180865.
With this argument we find for the integral .20188, which
multiplied by 470, the entire number of errors, gives 95 as
20.
COMPARISON WITH OBSERVATION.
the number of errors less than " .\. In a manner similar to
this the following results' were found :
Between
No. of Errors
by Theory.
No. of Errors
by Experience.
o" o and o". I
95
94
o.i and o".2
89
88
o '.2 and o .3
78
73
o".3 and o .4
64
58
o".4 and o .5
50
5i
o".5 and o' .6
36
36
o".6 and o .7
24
26
o".7 and o' .8
15
14
o".8 and o' .9
9
10
o".g and i' .0
5
7
over i' .0
5
8
This agreement is very satisfactory, but here, as in other
similar examples, the larger errors occur a little more
frequently than theory would indicate.
This is probably due to the fact that (unconsciously, per-
haps) every observer will occasionally let an observation pass
which is not up to the average standard of accuracy. Small
mistakes will sometimes occur, also, which are not of sufficient
magnitude to attract attention. A consideration of the matter
has led to attempts on the part of Peirce of Harvard College
and Stone of England to establish criteria for the rejection
of such doubtful observations. On the other hand it has been
proposed to overcome the difficulty by determining a system
of weights which should give those observations which show
large discrepancies less influence than those showing small
ones.
This branch of the subject, however, is beyond the scope
of the present work. It is an exceedingly delicate matter
to deal with, and from its nature is probably incapable of a
mathematical treatment which shall be entirely satisfactory.
Every computer occasionally feels compelled to reject
3 2 LEA S T SQ UA RES. 21.
observations. This should always be done with extreme cau-
tion. As for the criteria for this purpose hitherto proposed,
probably the most that can be said in their favor is that their
use insures a uniformity in the matter, thus leaving nothing
to the individual caprice of the computer.
Indirect Observations.
21. We have now investigated the simplest case of the
determination of the unknown quantity by observation, viz.,
that when the quantity to be determined is measured directly.
In the more general form of the problem the unknown
quantities are connected with the observed quantities by an
equation of the form
f(x,y,z, . . .) -M,
M being given by observation, and x,y,z, etc., being the un-
known quantities. This general form includes the case which
we have previously investigated, where there was only one
unknown quantity. Each observation furnishes an equation
of this form ; therefore a number of observations equal to that
of the unknown quantities will completely determine their
value.
This would leave nothing to be desired if the observations
were perfect ; but owing to the errors to which they are liable,
the values of x, y, z, etc., will be more reliable the greater
the number of observations on which they depend. If now
we have four unknown quantities, x, y, 2, and w, four observa-
tions will give us four equations from which the values of the
unknown quantities may be determined. If we have more
than four equations, we may determine values of the unknown
quantities by combining any four of them. As the equations
depend on observations more or less erroneous, we. should
thus obtain a variety of values for x, y, z, and w, all of them
probably in error to some extent.
21. IN DIRE C T OBSER VA TIONS. 3 3
The problem then is this : Of all possible systems of values
of the unknown quantities, to find that which most accurately
represents all of the observations.
We shall confine ourselves to the consideration of linear
equations; and as the problems in which we shall be more
particularly interested do not give rise to equations of more
than four unknown quantities, we shall limit our discussion to
that number. It will be obvious, however, that it can be
extended to any number.
Suppose we have the following system of equations :
a \ x ~\~ b^y -\- c^z -\- d^w n^,
a** + b, y + cjs + d,w = n 3 ; } - - (3 6 )
in which x, y, z, and w are unknown quantities, a, b, c, d,
etc., are coefficients given by theory, and ;/ a , etc., are
quantities given by observation.
If now the data were perfect we should obtain the same
values of x, y, 2, and w by combining any four of these
equations. Owing, however, to the errors of observation to
which #,, n^ etc., are subject, it is not probable that a substitu-
tion of the true values of x, y, z, and w (if we knew them)
would exactly satisfy anv one of the equations.
Let z',, v a , v 3 , etc., be the residuals obtained by substituting
in equations (36) for x, y, z, and w their approximate values
such that the following equations will be rigorously satisfied :
a^x -\- 6, y -f- c^s -\- d^<,v = ,
a^x -f- b n y -\- c^z -f- d^w = n^
a 3 x -f- b 3 y + c^z -|- d 3 w = , >,. f (37)
54
LEAST SQUARES.
21.
Now the most probable values of our unknown quantities
will be those which make the sum of the squares of these
residuals a minimum, viz.,
v>* + < + < + etc. = f(x, y, z, w)
(38)
must be a minimum.
In these equations x,y, z, and w are supposed independent,
therefore the differential coefficients with reference to each
variable must separately be equal to zero to satisfy the
conditions of a minimum. That is,
d\yv\ _
~dx~ '
d\vv\ _
dy
o,
d_\vv\ _
~dz~ ~
dw
Writing out these expressions in full, we have the following :
dv, dv^ dv, \ ,
i
dy
dz\
, ,
dy dy
dv. . dv* . dv.
v, -r 1 + ^ -H + ^s -r 1
1 dw ' dw ' dw
= o .
(39)
x, y, z, and w being independent, we have from -(37),
dv,
dv,
dv,
a a) etc. ;
dx ~ a "
dx
dx
dv,
dv -> h
dv,
b etc. ;
dy = ' *
dy - - *
dy
dv, _
&s = ~ C
dv,
dz
= - c v etc. ;
dv, _ d
dw
dw ~ "
dv,
dw
= d,, etc. ;
21. INDIRECT OBSERVATIONS. 35
by means of which values equations (39) become
^+w+^t;;;=o;[; (4o)
Substituting for v lt v etc., their values from (37), we have
for the first of these
-= O.
afi^y -{- a^c^z -(- a^d^w a^n,
ajj + a,d,w - a,n,
The second of (40) becomes
and similarly for the remaining equations. Using Gauss'
symbols of summation, we have therefore
{ad\x + \ab~\y + \ac~\z + \ad}w = [an] ; -|
\ab~\x + \bb\y -j- \bc\z +
[ac]x -\- \bc\y + \cc\z +
These are called Normal Equations, and the values of the
unknown quantities obtained by solving them will be the
system of values which makes the sum of the squares of the
residuals v^ v etc., a minimum, and therefore the most prob-
able system of values. Equations (36) are called Equations oj
36 LEAST SQUARES. 22.
Condition, or Observation Equations. An inspection of (41)
gives us the following rule for solving a series of equations
of condition :
Multiply each equation by the coefficient of x in that equation,
then add together the resulting equations for a new equation,
then multiply each equation by the coefficient of y in that equation,
and, as before, form the sum of the resulting equations. Continue
the process zvith the coefficients of each of the unknown quantities.
The number of resulting Normal Equations will be equal to that
of the unknown quantities, and t/ie values of the unknown quanti-
ties deduced therefrom will be the most probable T a lues.
It must be borne in mind that this process supposes the
number of equations of condition to be greater than that of
the unknown quantities. If it is less, this process will give
us a number of equations equal to that of the Quantities to be
determined, but they will be indeterminate none the less than
the original equations were, as can be easily shown.
Observations of Unequal Weight.
22. In deriving the normal equations from the equations
of condition, we have regarded the latter as of equal weight.
In the more general case the weights will be unequal.
In the equation a,x + b, y -f- cj& -f- dju if we suppose,
as in (33), that /, represents the weight of an observation,
viz., of that , is the mean error of and e the mean error
of an observation of weight unity, we have
Multiplying the above equation by Vp^, we have
+ b, Vp,y + c, Vp,z + d, Vp,w = , Vp,, (42)
23. ARRANGEMENT OF COMPUTATION. 37
an equation in which the mean error of the absolute term
#1 ^A i s f a "d tne we ight unity. In the same manner we
multiply each equation by the square root of its weight, thus
reducing them all to the same unit of weight, when we pro-
ceed precisely as before in forming the normal equations.
Computation of the Coefficients.
23. The method of forming the normal equations is now
fully explained; the work of computation, however, is some-
what laborious, especially when the number of equations of
condition is large. It will therefore be important to arrange
the work so that the numerous multiplications and additions
may be performed with the least liability to error, and so
that convenient checks may be applied for insuring accuracy
in the results. The multiplications may be performed by
logarithms, in which case a four-place table will give the
necessary degree of precision, or Crelle's multiplication-table
may be employed with advantage.* We shall also show
how to perform the multiplications by the use of a table of
squares.
Convenient proof-formulas may be derived as follows: Let
the sum of all the coefficients entering into each equation be
formed in succession, and represent them by s with the proper
subscript. Thus :
a, + b, + c, + d, - a = s, .
* Dr. A. L. Crelle's " Rechentafeln vvelche alles multipliciren und dividiren mit
Zahlen unter Tausend" (Berlin, 1869).
38 LEAST SQUARES. 24.
Multiplying these sums by their respective a, b, c, etc., in
succession, and adding the products, we shall have the follow-
ing equations for checking the accuracy of the coefficients of
the normal equations :
+ [off] + M + M - [an] = [] ;
[W] - [to] =
[*] - M -
[dd} - \dn\ =
[off] + [dff] + [fo] + [W] - [to] = M ;
M + M + M + [*] - M -
This requires the computation of the additional terms [as],
\bs~\, . . . and the agreement must come within the limit of
error of the computation. These additional terms will be
further useful for checking the accuracy of the solution of
the normal equations, as will afterwards appear.
24. If it should happen that the coefficients of one unknown
quantity in the equations of condition were much larger than
those of another, considerable discrepancies might exist in
the agreement of the proof-formulae with the sums of the co-
efficients. It will generally be necessary practically to limit
the computation to a certain number of decimals, when the
products of the large quantities may introduce errors into
the last places, where the products of the small quantities
introduce none.
This difficulty is overcome by substituting for the unknown
quantities other quantities which will make the coefficients
of the same order of magnitude throughout. This is con-
veniently accomplished by selecting the largest coefficient
with which an unknown quantity is affected and dividing
each of the coefficients of this quantity by it. Thus, let
, A y, & be the largest coefficients of the quantities x,y, z, w,
respectively, which occur in the equations of condition, and
let v be the largest of the series of known quantities t ,
25. ARRANGEMENT OF COMPUTATION. 39
n 3 , . . . Then we may place the equations of condition in the
following form :
(*) + (W + ?) + () = ' ;
where the unknown quantities are (ax), (Py\ and the
values obtained in solving the equations will be in terms of
i'. The equations will be made homogeneous by this pro-
cess before beginning the work of forming the normal equa-
tions. The sums s lt s t , . . . will be most convenient for the
purpose to which they are applied, if they are formed from
these homogeneous equations.
For the kind of problems which we shall have occasion to
solve in the following pages there will seldom be a system-
atic difference in the magnitudes of the coefficients of the
different unknown quantities of importance enough to render
this operation necessary. In cases, however, where there is
a marked difference in this respect it will be advisable to
incur the slight additional labor involved, and in some cases
it becomes a matter of considerable importance.
25. The formation of the normal equations with the accom-
panying proof-formulas will therefore require the computa-
tion of the following quantities:
[aa] [ad] [ac] [ad] [an] [as] ;
[bb] [be] [bd] [bn] [bs] ;
M [cd] [en] [];
[dn][ds^
[nn] [ns] .
40 LEAS 7' SQUARES. 25.
The latter will be employed for checking the final compu-
tation, as will be shown hereafter. As will be seen, there are
twenty of these quantities required in a series of four equa-
tions. In general the number will be* - i,
where n is the number of unknown quantities.
Let a sheet of paper be ruled with a number of vertical
columns represented by the above formula. In the first
horizontal line will be the symbols of the products written in
the columns below, viz., [aa], \_ab~\, . . . and in the last line the
sums of the products. If the results are correct the proof-
equations (44) must be satisfied. The algebraic signs of the
various products will demand special attention, as they form
a very fruitful source of error.
If the application of the proof-formulas is postponed until
the conclusion of this part of the computation, the position
of an error is often shown at once, since each sum, with the
exception of the sum of the squares, is found in two different
proof-equations. If two of the proof-form ulae fail to be
satisfied, while the others prove true, the error is in the term
common to both ; while if only one equation fails to be satis-
fied, the error is in the quadratic term.
Before proceeding further it is recommended that the
reader refer to the example found on page 329. The num-
ber of observation equations is twelve, each of which has
been multiplied by the square root of its weight. The num-
ber of unknown quantities is three, the coefficients of which
have no systematic difference in magnitude of sufficient
importance to require the application of the process for
rendering them homogeneous. The formation of the
normal equations is found on page 330. The number of
* It is the sum of a series of terms in arithmjtical progression minus i; num-
ber of terms = (n -\- 2); first term = i; last term = (-)- 2).
26. ARRANGEMENT OF COMPUTATION. 41
unknown quantities being three, we require by the formula
just given fourteen columns. It will be observed that the
proof-formulae are perfectly verified, as they should be in
this case, no decimal terms having been neglected.
Computation of the Coefficients by a Table of Squares.
26. By whatever method the multiplications are performed
a table of squares will be found very convenient for the
quadratic terms. Terms of the form [ab] may also be com-
puted with such a table, as will appear from the following.
We have a,b } = {(<*, -f- b^ - a? - b?\\
The quadratic terms [aa], [bb], . . . will be computed in any
case, so there will only be required in addition the terms of
the form [(a -[- &f~\. In case of four unknown quantities we
shall require the following quadratic terms :
[(a + Vf\ \(a + ;) 2 ] [(a + d}^ [(a - )] ;
(46)
[dd~] \(d n
[ss] [].
The last two will be employed in checking this and the sub-
sequent computation. Thus for the case of four unknown
quantities we have sixteen terms of the above form, or, in
general, < + -H + *) + ,.
LEAST SQUARES.
26.
The equations having been multiplied by the square roots
of their respective weights, and the coefficients made homo-
geneous if necessary, the computation will be carried out as
shown in the following scheme :
[*]
[*]
[(<* + *)]
In order to derive a convenient proof-formula we square
both members of equations (43) and add
[>] + 3 {[aa] +
+
+
f
+ []} -
(47)
For an example of the application of the above method
the reader will turn to page 334, where the normal equations
are computed from the equations of condition before re-
ferred to. This method possesses some advantages over
that by direct multiplication: the most important of these is
in the fact that the liability to error in algebraic signs is for
the most part avoided. Care being taken in forming the sums
(a -f- b\ (a -(- c\ etc., no further attention need be given to
the algebraic signs until the coefficients of the normal equa-
tions are completed.
27. SOLUTION OF NORMAL EQUATIONS. 43
Solution of the Normal Equations.
27. In the solution of the normal equations the work should
be arranged so that it may be conveniently reviewed for
detecting errors in case such exist, and so that proof-formulae
may be applied at the various stages of progress.
The order in which the unknown quantities are determined
is generally indifferent except in the case where the nature
of the problem is such that one or more of them cannot be
determined with accuracy from the equations. We may
know in advance that we have a case of this kind, or it may
be discovered in solving the equations.
It will be shown hereafter that the weight of any unknown
quantity will be determined by arranging the solution in such
a way that this quantity is determined first. The weight will
then be represented by its coefficient in the last equation from
which the others have been eliminated. If now this coefficient
is very small it shows that this quantity cannot be well
determined without additional data, and the solution must
then be arranged so that the uncertainty in this quantity will
have the least effect on the others. In case a preliminary
computation shows that the weight of any unknown quantity
is very small, the elimination will be repeated in such a way
that this quantity is first determined. The values of the
others will then be expressed in terms of this one. If then
at any time additional data become available for determining
this quantity, or if it is known from any other source, the
other quantities become known also.
As such cases will seldom occur in the problems with
which we shall have to deal, it will not be necessary to enter
more fully into the matter at present.
28. In the elimination it will be convenient to employ the
method of substitution, using a form of notation proposed by
44 LEAST SQUARES. 28.
Gauss. In developing the formulas, we shall suppose as before
the number of unknown quantities to be four. It will be a
simple matter to extend or abridge them in case of a greater
or less number.
The equations to be solved are
\aa~\x + \ab~\y + \_ac\z -f \ad~\w = [an] n
\ab\x + \bb\y + \bc\z + \bd~\w = \bn~\ , I
\ac\x + \bc\y -f \cc\z + \cd~\w = \cn\ ; f '
\_ad^x+ \bd~\y + \cd]z + \_dd~\w = \_dn\.\
From the first of these we have
x= \an\ _ \ab\ M \ad\ ,
\aa\ [oaf [aa] [aa]
which value being substituted in the remaining three equa-
tions, we shall have x eliminated. The first of the resulting
equations will be
and similarly for the remaining two.
Let us now write
f ~Vl - \bb i] ; \bd} - f^[^] = \bd i] ; }
}- (49)
28. SOLUTION OF NORMAL EQUATIONS. 45
and for the coefficients of the second equation,
M
(49)
Similarly for the third,
Mrwi _ r^
Our three equations then become
\bb \\y + [^ i\z + [^i]w = [^ i
\bc \\y + \cc i> + [^ i]w = [ i] ; . . . (50)
\bd\\y + [^i> + [^i]w = \dn i
In these the same symmetry of notation is preserved as in
the normal equations, and it can easily be shown that the
terms \bb i], \cc i], and \dd i], which have the quadratic form,
will always be positive.
From the first of (50) we have
r^i] \bd\ \bdi\
y - ~
This is to be substituted in the second and third, and the fol-
lowing auxiliary coefficients computed :
46 LEAST SQUARES. 28.
which process gives us the following equations :
\cc 2\z + \cd 2\w = [en 2] ;
\cd2~\z -\- \dd2\w = \dn2\.
From the first of these,
Substituting this in the second, and writing
\dd2\ - [^2] = \dd& \_dn 2] -
we have \dd $\w = [dn 3] ; ...... (54)
from which W=[ ....... (55)
z, y, and x can now readily be found by substituting succes-
sively in (53), (51), and (48).
The first equation in each of (41), (50), (52), and (54) are
called elimination equations, and are here brought together
for convenience of reference :
\ad\x + \ab~\y + \ac~\z + \_ad\w = [an] ; ^
\bb \\y + \bc i> + \bd i]w = \bn i] ; I
\cc 2\s + \cd2~\w - \cn 2] ; [
\ddi\w =
This is all that will be strictly necessary in case the weights
and probable errors of the unknown quantities are not re-
quired.
29. PROOF-FORMULAE. 47
Proof -Formulas.
29. Convenient proof-formulas for checking the accuracy
of the successive auxiliary coefficients may be derived from
the summation terms [/], [fa], ... of equations (44).
Referring to these formulae, let us write
Substituting for [fa] and \as\ their values, this expression
may be written in the form
Therefore, writing for the quantities in the brackets their
values, we have
[fa i] = \bb i] + \bc i] + \bd i] - \bn i],
a formula by which the accuracy of the coefficients in the
second member can be tested, and which requires the addi-
tional auxiliary quantity \bs i].
Proceeding in a similar manner, we shall require for check-
ing the computation at the end of the first stage of the eli-
mination the following auxiliary quantities :
[fa I] = [fa] -
; !.= -
48 LEAST SQUARES. 30..
when we shall have the following proof-equations :
\bs i] = \bb i] + \bc i] + \bd i] - \bn i] ; \
\cs i] = \bc i] + \cc i] + \fd i] - [en i] ; V . (57)
[ds i] = \bd\\ + [fl] -f [>tf i] - [dn i] . )
In the same manner we have, for checking the next step in
the operation,
\cs 2] = [cs i] - I[fo i] ; [ds 2] = [_ds i] - [^ ] [fo i]:
, g ,
and finally, [_ds 3] = [ds 2] - [ 2] ;
[^3]= [^3]- [<*3].. ..... (59)
The agreement of these two values of [ds 3] must be within
the limits of error of the computation, and it furnishes a very
accurate control over the accuracy of the computation up to
this point.
30. After the values of x t y, z, w have been determined, a
most thorough proof of the accuracy of the entire computa-
tion is obtained by means of the residuals, v lt v . . . obtained
by substituting these values of x, y, z, w in the equations of
condition, (37), p. 33, viz. :
-f- c^ -\- d^w n l
-f cjs + djv n,= v^,
(37)
3- PA' OOF FORMULAE. 49
Multiplying these equations by v^ v v 3 , . . . in order,
adding, and writing, in accordance with the notation em-
ployed,
we have
[/;] \av\x \bv\y [cv~\z \dv~\w = \vv\\
but by equations 140),
[av] = o, [bv] = o, \cv\ = O, \dv] = o.
Therefore \irJ\ = \vv\ (60)
Now multiply equations (37) by #., n v n 3 . . . in order, and
add, viz. :
[] j/27/];tr \bn\y \_cn~\s \ciii\w =[nv] = [vv\. (61)
By means of this equation \vv\ may also be computed as
soon as x, y, z, w become known. But we have
_[*] \aV\ \ac\ \ad-\
S M M r M M
Let this value be substituted in (61), and write
r -. \an\ r -. r -.
^ "M =
also write [^ i], [en i], etc., for their values, when we have
[ i] \bn \\y \cn \\z \dn \}w \v~J\.
Let the same process be carried on for eliminating y, z, and
50 LEAST SQUARES. ^31.
iv in succession from this and the resulting equations. We
shall have in all the following auxiliary quantities to com-
pute :
\nn I] = [] - |gj[>] ; \nn 2] = \nn i] - ^j^bn i] ;
Om 3] = [2] - [^|][2]; [ 4 ] = [ 3] - f^]^ 3]-
Either of the following equations will then give the value of
[mi] \an~\x \bn~\y - \cri\z - \_dri\w \vv~\\ -^
\nn i] \bn \\y \cn i\z [dn i~\w = [vv] ; \
\nn 2] \cn2\z \dn 2\w \vv] ; \- (62)
[3] - [dn$\w= \_vv~]', '
\nn 4] = [vv] .
Only the last of these will generally be used.
31. The value of [nn4] [vv] can be derived from the
summation quantities [ns], [ns i], etc., with very little addi-
tional labor. We have
[ns] = [an] + \bn\ + \cn\ + \dn\ - [].
Let us write [ns i] = [ns] - [ ^M,
and substitute in this expression for [its'] and [as] their values,
when it may be placed in the following form :
32- ARRANGEMENT OF COMPUTATION. 5 1
or what is the same thing,
[ns i] = \bn i] + \cn i] + \_dn i] - \nn i].
Proceeding in a similar manner to form in succession the
following auxiliary quantities, we have the series of equations
by which the accuracy of the quantities \bn i], \cn i], . . .
\nn 4] may be verified :
3] -
\ns i] \bn i] -f [en i] -|- \dn i] [ i] ;
[5 2] = [r 2] -f- [^ 2] [# 2] ;
(63)
Only the last of these equations will generally be required.
Form of Computation.
32. In computing the various auxiliary quantities which
occur in the solution of a series of normal equations, the work
should be arranged so that it may be carried through from
beginning to end in a systematic manner in order to keep a
general oversight of the results at the various stages of prog-
ress, and to apply conveniently the proof-formulas. This will
be the more important the greater the number of unknown
quantities. The following scheme will be found to answer
these requirements.
It will generally be found expedient to make the computa-
tion by the use of logarithms, but in some cases the computer
may prefer to perform the multiplications and divisions by
the aid of Crelle's table. In the following scheme we have
52 LEAST SQUARES. 3 2 -
supposed logarithms used. A sheet of paper is first ruled
with vertical columns, the number of which is greater by two
than that of the unknowa quantities. In the first horizontal
line will be written in order the coefficients which are com-
bined with a, viz., [aa~], [at], . . . [an], [as], and immediately
below these their logarithms. Attention is directed to this
line by means of the letter E in the margin, as it is the first
of the elimination equations (56), and will be used for deter-
mining x after/, z t and w become known.
In the third line are the coefficients \bb\ [be], . . . [fa], so
placed that the letters combined with b fall in the same verti-
cal column with the same letters combined with a, viz., [be]
under [ac], \bd] under [ad], etc.
In the fourth line of the first column is now written
log p-J the value of which, as well as those of all the quan-
tities in this column, must be carefully verified, as an error
in this factor may not be detected by the proof-formula.
The log ^ ^ is now written on the lower edge of a card
and added in succession to the logarithms of [aft], [ac], . . .
[as], and as each addition is performed the natural number is
taken from the logarithmic table and written in the place in-
dicated in the scheme. With a little practice the computer
will be able to make this addition mentally, and take from
the table the corresponding number without writing down
this logarithm. Thus we shall have
|>#] written under [//];
t .Sac] written under [bc\;
32.
ARRANGEMENT OF COMPUTATION.
53
i.r&i i
[Mx]
log \bb i]
[/;<:
log [
log [a
log \cd 2]
[rfrfl]
if* 3
log L^ i]
<:*a]
log [. 2]
U.
rf]
log [rfrf 3]
.0^3,
log
[*]
[ftn]
log [** i]
"
IIP.
IV.
v {i L
X'.
Prcof-Equati
-[!]
Practically only those proof-equations which are distinguished by an accent will ordinarily
be employed. The lines marked by an E in the margin give the logarithms of the coefficients
of the elimination equations. The logarithms marked * must be carefully verified, since aa
wor in one of these may escape detection by the proof-equation.
For the application to a numerical example see page 331.
54 LEAST SQUARES. 33.
and by subtraction,
\bbi\\bci\,\bdi\, \bni\\bsi\.
These are the coefficients of the second elimination equation,
and will be used for determining y after z and w have become
known. The I in the margin refers to the proof-formula
by which the values of these quantities will be verified.
It will not be necessary to proceed farther with this ex-
planation, as a reference to the scheme in connection with
the formulae for the auxiliary quantities will show clearly the
process. The elimination being completed, the quantities
[4] and [nsj] are computed as shown in the scheme, the
agreement of which with each other and with [vv], obtained
by substituting the values of x,y, z, w in the equations of
condition, furnishes a most thorough proof of the accuracy
of the entire computation.
Weights of 'the Most Probable Values of the Unknown Quantities.
33. In case of a single unknown quantity determined by
direct observation, the computation of the weight of the
arithmetical mean was found to be very simple. In the case
under consideration, where the equations to be solved con-
tain several unknown quantities, the difficulty is greatly
augmented.
In our equations of condition we have supposed the quanti-
ties observed to be , etc. We have already shown that
if the resulting equations of condition are not of equal weight,
they may be made so by multiplying each by the square
root of its respective weight. We shall therefore in investi-
gating the weights of the unknown quantities assume the
weight of each observation to be unity.
33- WEIGHTS OF UNKNOWN QUANTITIES. 55
Let p x ,p u ,p z ,p w , be the weights of x,y,z, and w respectively;
e x , s y , z , w , their mean errors.
Let be the mean error of an observation.
As all of our equations are linear, it is evident that if the
elimination of the three unknown quantities x, y, and z be
completely carried out, the resulting equation will give w as
a linear function of ;/ # 2 , ;/ 3 , etc. Similarly, if x, y, and w be
eliminated, we shall have z expressed as a linear function of
the same quantities, and so of each of the others.
We may therefore write
x or 1 i -f a^ n , -f ay* 3 + etc.n
jiS+S+Sstf (64)
w = S 1 n 1 -f- tfjW, -|- ] ;
\aV\x + [M] J + \bc\z ,.
\_ac\x + \bc\y + M*
Let us now assume the following system of equations :
\_ad\Q + [>0] 0' -f \ac-\Q" + 0^]0 //7 = P n
/7 \bd-\Q" = o ; I
\cd-\Q'" = o ;
These equations will be possible, as there are four unknown
quantities, Q, Q', Q", and Q'", and four equations for determin-
ing their values; further, as the equations are of the first de-
gree there will only be one system of values for Q, Q', etc.
Now let the normal equations be multiplied by Q, Q', Q",
and Q'", in their respective orders, and the resulting equations
added. Then in consequence of (67) in the resulting equations
the coefficients of x,y, and z will be zero, and that of w unity.
Therefore we shall have
w = \_aii\Q + \bn~\Q' + \cn\Q" + [_dii\Q" r . . (68)
We shall now show that Q" = [#tf], and is therefore the
reciprocal of the weight of w.
Let us expand the quantities contained in the brackets,
equation (68), and compare the results with the last of
equations (64). We thus find the following values of d it # 2 ,
etc.:
. . . (6 9 )
J
34- WEIGHTS OF UNKNOWN QUANTITIES. 57
Multiplying each of these by its a and then adding, then
multiplying each by its b, c\ and d successively and adding,
we have by (67) the following equations :
= o
+ 4A + ' =
[cd] = o ; f " "
[dd~\ i . J
Now let each of (69) be multiplied by its d and the results
added. Then by (70) we have
*A + *A + *A + = [**] = <2'". Q- E. D. (71)
The solution of equations (67) therefore determines the
weight of w. In a precisely similar manner the weight of
each of the unknown quantities may be determined. Thus,
to determine the weight of x, we write for the second mem-
ber of the first of (67) unity instead of zero, and write zero
for the absolute term of each remaining equation. The re-
sulting value of Q will be the reciprocal of the weight of jr.
This process is simple enough in theory, but its application
is laborious, as we must solve equations (67) separately for
the weight of each unknown quantity. This does not involve
so great an amount of labor as may at first appear, as much
of the computation will already have been performed in the
solution of the normal equations. It is easy, however, to
derive a process which will generally be much more con-
venient. It is as follows :
34. In the solution of equations (41) by successive substitu-
tions we found for the final equations in w see (56)
We shall now show that the coefficient \dd^\ = -^777, and
is therefore the weight of w.
58 LEAST SQUARES. 54-
For this purpose let us write equations (41) as follows :
\ad\x -f- \ab~\y + \ac~\z + \_ad~\w [an] = A ;
\ab\x -f \bb\y + [&]* + [&/]w - \bii\ = B ;
-w - [en] = C;
- ^ = D.
Let us now suppose the equations solved by means of the
auxiliaries Q, Q', Q", and Q'", determined from (67), when we
shall have
w = \an\Q + \bn\Q + \cn\Q" + \dn\Q"
+ AQ + BQ' + CQ" + DQ>". (72)
This will now be the same value of w as before obtained, if
we make A==C=D = o.
Let us now suppose the equations solved, as before, by
substitution. Since in this process no new terms in D are
introduced, the coefficient of D will not be changed in the
final equation for w, and we shall have
[dd$\w = \dn 3] -f D + terms in A, B, and C;
from which w = f^3 + ^-- -f terms in A, B, and C.
Now it is evident that the coefficients of A, B, C, and D must
be the same in this equation as in the value before obtained,
equation (72). Therefore
Q- E. D.
We therefore see that we can obtain the values of the un-
known quantities from equations (41), and at the same time
their respective weights, by arranging the elimination so that
35- WEIGHTS OF UNKNOWN QUANTITIES. 59
each in succession shall come out last. The coefficient of the
unknown quantity in the final equation will be its weight.
35. In solving a system of four equations like the above
it is best to proceed as follows: Let zv be determined, as
above, by substitution in the order x, y, z. We then have
w with its weight from
[dd^]w = [dn 3].
Equations (56) then give successively z, y, and x.
. Let now the elimination be performed in the opposite order,
viz., w, z,y, when we have x with its weight from the equa-
tion
\aa 3> = [an 3],
[aa 3] being the weight of x.
This value of x must agree with the former value within
the limits of error of the computation, thus furnishing a con-
venient check to the accuracy of the computation.
For the weight of y and z we need not repeat the elimina-
tion, but proceed as follows :
Let us suppose the elimination performed in the order x,
y, w, z. We shall then have the same auxiliary coefficients
as in the first case, as far as those indicated by the numerals
i and 2, and equations (52) will be the same as before ; but
as the elimination will now be performed in the order w, z,
instead of z, w, we write them
\dd2~\w + \cd2~\z = [dn2] ;
\cd2~\W + [cc 2\z = [en 2] .
From the first of these,
- fr** 2 3 _ \fd2~\
~ \dd2~\ \dd2~f*
60 LEAST SQUARES. 35.
Substituting- this in the second gives us for the coefficient
of.*
But we have \dd$\ = \dd*\ -
From these two equations we find
And in a similar manner,
We therefore have the following precepts and formulae
for computing the weights in the case of four normal equa-
tions :
First, perform the elimination in the order x, y, s, w,
then p w = \_dd-$\ ;
-(73)
Second, perform the elimination in the order a/, z,y,x,
then p x = [a a 3] ;
WEIGHTS OF UNKNOWN QUANTITIES. 6l
The formulae for the auxiliary coefficients for the second
elimination may be derived from those for the first by simply
interchanging the letters a and d and b and c. The process
is so simple that it will be unnecessary to write them out in
full.
Other Expressions for the Weights.
36. When the equations have been solved, as already ex-
plained, and the various checks applied, so that the computer
is convinced that the results obtained are reliable, it may be
undesirable to repeat the elimination merely for determining
the weights of the first and second unknown quantities. We
may derive convenient expressions for computing tne weights
in this case, as follows :
Suppose four solutions of the equations to be carried
through so that each unknown quantity 7 in turn is first deter-
mined, the order of the others remaining the same : we should
then have each unknown quantity with its weight completely
determined, as we have already seen. The solution of the
equations for which we have given the complete formulas is
in the order d, c, b, a, where we have written the coefficients
instead of the unknown quantities. Tf now we substitute the
values of w, z, and y in the third, second, and first of equations
(56) in order, we have finally the expression for^r, which will
be a fraction with the denominator
[aa] \bbi\ \cc2\
In the four solutions which we have supposed made, the un-
known quantities last determined will be in succession x,x,x,
62 LEAST SQUARES. 36.
y, and the denominators of the expressions for their values will
be as follows :
\_aa\\bb i\ c \_dd2\ e O
\aa\\cci\\_dd2\\bb
\bb\ a \cc i] a [dd2\ a \a
where the subscripts show which unknown quantity is first
determined in each solution. As the elimination is performed
by successive substitutions, no new factors being introduced,
it follows that these expressions are equal to each other re-
spectively.
It is evident that when the order of the elimination is
changed so that a different quantity is first determined, the
order of the others remaining the same as before, the values
of the auxiliary coefficients \bb i], [cc2], etc., which do not
contain the coefficient of this quantity will remain as before.
Suppose, as above, the unknown quantities to be determined
in the order d, c, b, a. Now let a second solution be made in
the order c, d, b, a; then all of the auxiliary coefficients as
far as those designated by the numerals i and 2 will remain
as before. In a third solution following the order b, d, c, a,
the coefficients designated by the numeral i will have the
same values as in the first case ; while in a fourth determina-
tion in the order a, d, c, b, they will all differ from the first
series of values.
Thus indicating by the subscripts only those coefficients
which have values different from those given by the first
elimination, we have the following equations:
\aa\ \Mi] [CC2-] \ddj\ = \ad\ \bb i] \dd*\ [0:3];
M \bb I] \cc2\ \ddj\ = \ad\ \cc i] [dd2\ [W 3 ];
\ad\ \bb I] \cc2\ \ddi\ = [W] [cc i] \dd2\ [aa 3].
36.
WEIGHTS OF UNKNOWN QUANTITIES.
We already have the weight of w. The weights of z, y, and
x are given by these last equations, viz. :
- - (74)
In applying these formulae the following additional auxiliary
coefficients must be computed :
j>]-to-.],
fc i] a - M
(75)
In case of three unknown quantities the formulae become
- (76)
where [^ i] a has the value given above.
64 LEAST SQUARES.
37. An elegant expression for the weights is obtained by
making use of the determinant notation. Thus, referring to
the normal equations (41),
(2'", the reciprocal of the weight of w, given by equations (67),
is the same as the value of w obtained from the above equa-
tion by making [ari\ = [bti] \cn\ o and \dn\ = i.
Therefore writing A for the complete determinant which
forms the denominator of the above expression, D'" for the
partial determinant formed by dropping the last horizontal
line and last vertical column, D" for the partial determi-
nant formed by dropping the third horizontal line and third
vertical column, and similarly D' and D for the other two,
we have
A = 4;
(77)
A number of other forms may be derived for the weights,
all of which involve about the same numerical operations as
the above. In certain special cases different forms may be
more convenient, but for our immediate purposes it will not
be necessary to develop the subject further.
It may readily be seen fram what precedes that the rela-
tive weights of the unknown quantities may be derived, even
when the number of observations does not exceed the num-
ber of unknown quantities. No probable errors, however,
can be determined in this case.
3- MEAN ERRORS OF UNKNOWN QUANTJ7^IES. 65
Mean Errors of the Unknown Quantities.
38. For determining the mean and probable error of an
unknown quantity nothing further is required except the ex-
pression for the mean error of an observation. It is supposed
that the equations of condition have been reduced to the
common unit of weight by multiplying each equation when
necessary by the square root of its weight.
The values of x.y, z, and w, as deduced above, are the most
probable values as deduced from the given data. When
substituted in the equations of condition the residuals
?/,, z' 2 , v 3 , etc., will not be the true errors unless the derived
values x, y, z, and w are absolutely the true values, a condi-
tion not likely to be realized.
Let (x -\- 6x\ (y -\- dy), (z -)- 8z), (w -j- dzv) be the true values ;
A^ A v ^ 3 , ... 4 m , the true errors.
We shall then have two systems of equations, as follows :
a^x -\- b } y -\- c^z -\- d^v n 1 =
a,x + b,y + cj + djv , =
= v . [ (78)
-*,= -2f,K79)
Let us multiply each of equations (78) by its v and add the
resulting equations. Then by (40) the coefficients of x, j, z,
and w will vanish, giving us the relation before derived,
66 LEAST SQUARES. 38.
Proceeding in the same manner with (79), we find
\vn\ = [vA] ........ (81)
Therefore \vA\ = \vu\ ........ (82)
In order to obtain an expression for the sum of the squares
of the true errors, viz., \_AA\ in terms of the sum of the
squares of the residuals [vv], let us first multiply each of
equations (78) by its A and add the resulting equations;
secondly, let us multiply each of (79) by its A and add in
like manner. The results are as follows :
\aA~\x + \bA\y +\cA\z + \_dA\w - \nA\ = - \yA~\ = - [vv~]
[aA} (x + 8x) + [_bA] (y+6y) + \cA] (z + fcr)
+ \dA\ (w + dw) \nA\ = -
Subtracting the first of these from the second, we obtain
= \vv\ - \aA~\Sx - \bl\Sy - [cA]dz - [d^Sw. (83)
If we could now assume dx, 8y, 6s, awrl 8w to vanish, we
should obtain, since m? = \4A~\ by definition,
This will give us a close approximation to the true value of
e when m is large.
For a more accurate determination of s we must endeavor
to find approximate values of (ad~\$x, \bA~\dy, etc. The true
values are beyond our reach, but principles already estab-
lished give us a means of approximation.
Multiplying each of equations (79) by its a, and adding,
we have
\_ad\x + \aV\ y + \ac~\z + \_ad~\w - [an] \ _ _ f ,-,
-f \_ad\Sx + \ab\Sy + \ac\dz + \ad\1w I
38. MEAN ERRORS OF UNKNOWN QUANTITIES. 6?
Comparing this with (41), we see that the first line is equal
to zero.
Multiplying each equation of (79) by its b and adding,
then in a similar manner by its c and d and adding, we have
finally
\aa~\dx + [aV]dy + [ac\3z + \_ad~\dw = - [>J] n
[oft]6* + \bb~\6y + \bc\6z + [&/]tfw = - \bA\ [
[ac\Sx + [bc}Sy -f [><;] [z/z/], and the terms following [w] in (83)
must be positive.
Let us now perform the indicated multiplication in (86).
Confining ourselves to the last equation, since the form is
the same for all, we can indicate the result as follows :
- \dA-\dw = 4*^,4+ 4V.4 f 4 we shall have
-^ (88)
38- MEAN ERRORS OF L\\'KXOW.\' QUANTITIES. "69
With the values of p x , p v , p z , and p w computed by (73), we
have finally
and the probable errors of ;r, y, z, and w will be obtained by
multiplying these respectively by .6745.
We have now developed the subject as far as is necessary
for our purposes. A complete example of the solution of a
series of equations with three unknown quantities, together
with the determination of their respective weights and
probable errors, will be found in connection with article
(191) of this volume.
INTERPOLATION.
39. In the Nautical Almanac are given various quantities,
such as the right ascension and declination of the sun, moon,
and planets, places of fixed stars, etc., which are functions of
the time. This is assumed as the independent variable, or
argument as it is termed by astronomers. The ephemeris
gives a series of values of the function corresponding to
equidistant values of the argument. In case of the moon,
which moves rapidly, the position is given at intervals of one
hour; the place of the sun is given at intervals of twenty -four
hours ; while the apparent places of the fixed stars vary so
slowly that ten-day intervals are sufficiently small. When
any of these quantities are required for a given time, this
time will generally fall between two of the dates of the ephe-
meris seldom coinciding with one of them ; the required
value must then be found by interpolation.
Interpolation in general is the process by -which, having given
a series of numerical values of any function of a quantity (or argu-
ment], the value of the function for any other value of the argu-
ment may be deduced without knowing the analytical form of the
function.
We shall consider the subject more in detail than will be
necessary for the simple purpose of using the ephemeris,
on account of its importance in other directions.
In what follows we shall suppose the values of the function
given for equidistant values of the argument, which will
always be the case practically. Also the intervals must be
39 INTERPOLATION, GENERAL FORMULA. /I
small enough, so that the function will be continuous between
consecutive values of the argument.
Let w = the interval of the argument.
. . . (7--/3W), (T-2w\ (T-w\ (T\ (T+w), (T+2w\
(T+3w), . . . = the values of the argument.
The notation for the arguments, functions, and successive
differences will be shown by the following scheme :
Argu- ist ad 3d 4th 5th
ment. Function. Difference. Difference. Difference Difference. Difference.
T
The notation shows at once where each quantity belongs
in the scheme. The first differences are forrned by subtract-
ing each function from the quantity immediately following
it, the argument being the arithmetical mean of the arguments
of the two functions. Similarly the second differences are
formed by subtracting each quantity in the column of first
differences from the one immediately below it, and so on for
the successive orders of differences. It will be observed that
the even orders of differences, /", /"', etc., fall in the same
horizontal lines with the functions themselves, and have the
same arguments, while the odd orders, /', f", etc., fall be-
tween those lines. The even differences all have integral argu-
ments, and the odd differences fractional arguments.
The arithmetical mean of two consecutive differences is
indicated by writing it as a function of the intermediate
argument. For example :
f\ T) = %[f\ T-\w
72 INTEKPOLA TION. 40.
40. Suppose now we set out from the function whose argu-
ment is T. Evidently,
) +f"(T
Proceeding in this manner, we readily discover the law of
the series; viz., the coefficients are those of the binomial
formula, and each successive function,/',/", etc., is on the
horizontal line drawn under the one which immediately pre-
cedes it. Thus we have the general formula
nw) =f(T) + nf(T+ $w) + " f"(T+ w)
+ (90
If we assign integral values to n we obtain the tabular
values, viz.,f(T-{- w),f(T-\- 2w\ etc.; but the formula is not
used for this purpose, but for interpolating between the
tabular values, in which case n is fractional and must be ex-
pressed in terms of the interval of argument w as the unit.
41. A more convenient form may be given to this expres-
sion (91), as follows : We have
(T+ *0 ;
2w) =f iv (T)+ 2f\T+ %w) +f\T) +f vii (T+
4 ! - INTERPOLATION, GENERAL FORMULA. 73
Substituting these values in (91) and reducing, we readily
obtain
f(T+ w) =f(T) + nf'(T+ i
The law of the series is obvious ; viz., a factor is added to the
numerator of each succeeding coefficient alternately after
and before the other factors, the last factor of the denomi-
nator being the same as the order of differences. The succes-
sive differences are taken alternately below and above the
horizontal line drawn immediately below the function from
which we set out.
Formula (92) will be used for interpolating forward. For
interpolating backward a better form may be derived by
writing ior f\T -\- \w], f'"(T -\- \w], . . . their values in terms
Changing ;/ at the same time into n, since the formula is
to be used for interpolating backwards,, we readily find
f(T- nw} =f(T) - nf(T- *,) + *!L=J->f(T)
- i- 2 t
1.2.3.4
74 INTERPOLA TION. 42.
42. In applying (92) and (93) it will be more convenient
to write them as follows :
=f(T] + n f'(T+&) + F- f"(T)
(92)j
-rnu] = f(T) - n
5
(93),
In (92), and (93), each difference is used to correct the one of
the next lower order immediately preceding it, and the quanti-
ties to be multiplied will generally be small. In interpolating
a value of the function corresponding to a value of the argu-
ment between Tand (T + w), we use (92), and set out from
f(T}. If the argument is between (T -}- %w) and (T-\-w),
we use (93), and set out from f(T-\- w).
When the interpolation is carried to any given order of
differences, as the fifth, it is a little more accurate to take the
arithmetical mean of the last differences, which fall immedi-
ately above and below the horizontal line drawn in the vicinity
of the required function. Thus the last term of (92), and
(93), would bef v (T\
43. For the quantities tabulated in the American Ephe-
meris it will only be necessary to carry the interpolation to
second differences ; but for computing ephemerides or tables
44- INTERPOLA TION, EXAMPLE. 75
of any continuous function, much labor is saved by comput-
ing the quantity directly for a comparatively few dates and
supplying the intermediate values by interpolation. If the
function is of such a character that some order of differences,
as the third, fourth, or any other, vanishes, this gives exact
values for the interpolated quantities, and in fact the process
may then be used for computing values of the function for
any value whatever of the argument. It is on this principle
that "tabulating engines" are constructed.
44. As an example of the application of (90), (92),, and (93),,
we take from the American Ephemeris the following values
of the moon's right ascension for intervals of 12 hours:
1883,
July /=* /' /" /'" /* /"
3 d, o h 5' 4*5' i5 S .'6S
29 39.05
I2 h 6 14 54.73 - 27.08
29 11.97 6.91
4th, o h 6 44 6.70 33.99 + 2.01
28 37.98 4.90 .06
I2 h 7 12 44.68 - 38.89 -f- 1.95
27 59.09 - 2.95 - .01
5th, o h 7 40 43-77 -41-84 + 1.94
27 17.25 i.oi .16
I2 h 8 8 1.02 -42.85 + 1.78
26 34.40 -f .77 - .33
6th, o h 8 34 35.42 - 42.08 + 1.45
25 52.32 +2.22 -.33
I2 h 9 O 27.74 39.86 -+- 1. 12
25 12.46 +3-34
7th, o h 9 25 40.20 - 36.52
24 35-94
i2 h 9 50 16.14
76 INTERPOLA T1OX. 44.
Example i. As an example of the application of (92),, let
us interpolate the moon's right ascension for 1883, July 5th,
4 1 '-
Since the interval of the argument w is here I2 h , we have
in this case nw = 4 h , or n -fa = |. Setting out from July
5th, o h , we have
ATC/) = .01
f v = .040
/" = + I-94Q
Corrected, f iv = -f- 1.900
VU*+--- - 792
f" I.OIO
Corrected, f" = - 1.802
^y- 1 !/'" + =- .801
/ ;/ 41.840
Corrected,/" = 42.641 <<
/" + -=+ 14-214
Corrected,/' =27 m 3i 8 .464
f= a =
1883, July 5th, 4 11 , a = 7 h 49 u '54 8 .26
This value agrees exactly with that found in the American
Ephemeris for 1883 (see page 115).
44- INTERPOLATION, EXAMPLE.
Example 2. Let us now apply (93), to determine the moon's
right ascension, July 5th, 2o h . Here we set out from July 6.
As before, n = %,f v (T) = .33.
+v
5 f
f iv
Corrected,/' 11
154
= + 1-450
= + 1.604
. . = + .668
= + -770
= + 1-438
= - -639
= 42.080
4 r
Corrected, /'"
^ r
Corrected, /"
tnl J ^
= - 42.7 J 9
T \ " tr\
Corrected,/' =26 n '2o s .i6o
/= = 8^^4^
1883, July 5th, 20 h a 8 h 25 m 48 9 .70
The algebraic signs of the various corrections are deter-
mined without difficulty, as follows: If a horizontal line be
drawn in the table of functions and differences (p. 75) in the
vicinity of the given argument (in the first of the above
examples immediately below 5 d o h ), the successive differences
required will fall alternately below and above this line.
78 INTEKPOLA TlOtf. 45-
Beginning with/" 1 ' we determine the correction to/" 1 ", which
is to be applied so as to bring the value nearer to that imme-
diately below the line. In this case/"'" = -j- 1.94; that which
immediately follows is + 1.78 ; therefore the correction must
be subtracted from 1.94, giving the corrected f iv = 1.90.
The value of f" is i.oi ; the value immediately above
the line is 2.95. The first must be corrected so as to
bring it nearer the latter, giving in this case the corrected
f'"= 1.802, and so on for each difference in succession.
That is,
When the quantity is the horizontal line >
the correction so as to bring it in the direction of the one in
the same vertical column immediately it.
Special Cases.
45. Whenever (92), or (93), can be applied, nothing more
will be necessary ; they require, however, a knowledge of
the value of the function for several dates both before and
after those between which the interpolation is made. It is
sometimes necessary to interpolate between values of the
function near the beginning or end of the table: as, for in-
stance, we might require from the tabular values of the
moon's right ascension, given on page 75, to determine the
value between the dates July 3d, o h , and 3d, I2 h , or between
7th, o h , and 7th, I2 h . In either of these cases the series of
differences terminates with f'\ so the above formulae will
only give the value to first differences inclusive.
We shall consider the two cases separately.
46. First. For arguments near the beginning of the table.
As before, calling the arguments between which it is re-
quired to interpolate the function, T and T -f- w, we may
apply formula (91), setting out from f(T).
46. INTERPOLA TION, SPECIAL CASES. 79
If the argument for which the value of the function is re-
quired is nearer T-\- w than T, it will be a little simpler to
set out from T-}- w and interpolate backwards. In this case
the formula requires the following modification:
Changing n into n, we have
f(T- nw)=f(T)-nf'(T + *w) + - f"( T + w)
_ n(n +0( + 2) ( + 3) (n + 4)^,- , 5 ,
1.2.3.4.5
From the manner of forming the successive functions, we
have
' (T+\w-)=f(T-%w)+f"(T)
f"(T + iw) = /'"( 7*+tH-/"( r +
/ - ( T + 2 w) = /iv( 7- +
/' ( 7- + f w) = /
Substituting these values in the above and reducing, we
have
RT- nw} = f(T] - nf'(T- &) + ( f"(T)
(g^jQ^^-f^l) ( + 2)
1.2.3.4
_ (H - I> ( +0( + 2) (n +
1.2.3.4.5
80 1XTERPOLA TIOX. 46.
For greater convenience in the application, (91) and (941
may now be written as follows :
S(T+ w) = f(T) + n /'(7-+ia>) + - L f"(T+ w)
f(T-nw) =./!T) + | -f'(T~^) + "-=^- \f"(T)
. . (95) ,
Example 3. Required the moon's right ascension, 1883,
July 3d, 4 h . Referring to the series of values (Art. 44), we
have for this case uw = 4'' ; /. n = .
f v = - -06
H -r = = + -044
f _j_ 2.010
Corrected,/' 1 ' = -j- 2.054
Corrected, y" 7 = - 8.279
46- INTERPOLATION, SPECIAL CASES. 8 1
^ !/-...=+ 4 . 599
/" = -JT-o ;8_
Corrected,/'' = 22.481
^T^ {/"= + 7494
/' =29-39^.050
Corrected,/' =29 m 46 s .544
{/'...= 9 m 55 s -5i5
/= or = 5"45 m i5 s .68o
1883, July 3d, 4", or = 5 h 55 m uM95
Example 4. Required the moon's right ascension, 1883,
July 3d, 8 h . In this case \ve use formula (95),, since the
argument is nearer I2 h than o h . n \.
- = -06
f"> = + 2.01
Corrected,/'" = + 2.05
Corrected,/"' =+ 8.082
/ 7/ = 27.080
Corrected,/" = 23.488
82 INTERPOLA TION. 4/.
n I
" = + 7-829
Corrected,/' = 29'"3i s .22i
{ /'...= 9'" 50/407
/ = a = 6 h 14'" 54 s . 730
1883, July 3d, 8 h , = 6" 5'" 4 S . 3 2 3
47. Second. Arguments near the end of the table.
Proceeding in a manner precisely similar to that of the
previous article, we readily obtain the formulae
f(T -f nw) = f(T)
-_
i .2.3
_ |w) . (97)
3) ( _ 4} (T _^ ( }
I .2.3.4.5
I second i ^ tnese a ppli es ^ or interpolating in the
47- INTERPOLATION, SPECIAL CASES. 83
j ^c^eaS j .
direction in which the argument . The above
may be written as follows :
f(T+nw) =f(T) + n {/'(T + Ja,) + ^=-' \f"(T)
f"(T-
- m) = f(T)+n -f'(T- *,) + - f"(T- w)
+ " | -/'"( r-f ,) + ^ | /( T - zw)
. (9 8,)
Example 5. Required the moon's right ascension, 1883,
July ;th, 4 h .
/" = ~ 36-52 ; /' = 24 35.94; /=
Substituting in (98) as above, we find
Example 6. Required the moon's right ascension, 1883,
July ;th, 8 h .
By substituting the numerical values in formula (98^ we
find for this case
a = 9" 4 2 m 7 8 . 9 7.
It will be observed that in the application of formulae (95),
(95)i' (98), and (98), the algebraic signs of the various correc-
84 I. \TERPOLATIOX.
tions may be determined in a manner entirely similar to that
explained in connection with formulas (92), and (93),. (See
Art. 44-)
Interpolation into the Middle.
48. When the function is to be interpolated for a value of
the argument half way between two consecutive dates of the
table, this is called interpolation into the middle.
For this case either 192), or (93), may be used, but a more
convenient formula is obtained as follows. Write in place
of n in (92) :
yr T + to) = A T) + \f\ T + to) + ^, */"( T)
to)
Then in (93) let = J, and set out from ( T + of) :
S(T+ to) =f(T+ w) - i/'
Taking the mean of these equations, obscrvinjr in the result-
ing equation that the coefficients of the odd differences,
/',/'", etc., vanish, and writing
' )} =-[f(T+
49- PROOF OF COMPUTATION. 85
\ - $f"(T+*w)+TfoF(T+tw)
- roW~ l \ T + i^) + . . . (99)
') }}} (99),
Example 7. Let it be required to determine the moon's
right ascension, 1883, July 5th, 6 h . We must interpolate into
the middle between July 5th, o h , and July 5th, 12''.
/* = -f 1.860
- A/" - - -349
/" = - 42.345
Corrected,/'' 42.694
-\\f- .. = + 5-337
Therefore 1883, July 5th, 6 h ,tf=7 h 54 m 27 8 .73
Proof of Computation.
49. The method of differences furnishes a very convenient
check on the accuracy of a computation, when, for a series
of values of an argument succeeding each other at regular
intervals, a series of values of any function have been com-
puted. Suppose an erroneous value of one of these quanti-
ties, f(T) -(- x, has been obtained, x being the error. The
functions, with the respective differences, would then be as
follows :
(7->+6,
r+I, - ,
86 INTERPOLATION. 50.
Thus the errors in the function has increased to >x in the
fourth difference, the greatest deviation being in the horizon-
tal line where the erroneous value of the function is found.
Suppose, for example, an error of 5 s had been made in
computing one of the values of the moon's right ascension
given in Art. 44. The scheme of differences would then
be as follows:
July
f=a
f
f"
f"
/"*'
n. m
s.
3d,
o"
545
15.68
29
39-05
I2 h
6 14
54-73
27.08
2 9
11.97
- 1.91
4th,
O h
644
6.70
- 28.99
- 17-99
28
42.98
- 19.90
I2 h
7 12
49.68
- 48.89
+ 3I-95
27
54-09
+ 12.05
5th,
O h
740
43-77
- 36.84
1 8.06
27
17.25
- 6.01
I2 h
8 8
1.02
- 42.85
26
34-40
6th,
O h
834
35-42
We see at once without going further than second differ-
ences that the value for July 4th, I2 h , is erroneous.
Differential Coefficients.
50. When we have a series of numerical values of a func-
tion, corresponding to equidistant values of the argument,
we may compute the numerical values of the differential co-
efficients from the tabular differences as follows: Either
form of the interpolation formula is arranged according to
ascending powers of n. The function f(T-\- nw) expanded
by Taylor's formula, and the differential coefficients, com-
pared with the coefficients of the different powers of n in the
above expansions, give at once values of these quantities.
5- DlFl-EREXTIAL COEFFICIENTS. 87
The most rapid convergence, and consequently the best
formulae, will be obtained by introducing into formula (92)
the arithmetical means of the odd differences situated above
and below the horizontal line drawn through the function
irom which we set out, using the notation for the arithmeti-
cal mean given on page 71.
From the manner of forming the differences we readily see
/' (T + &>) = f (T) + if"(T) ;
f"(T+&) = f'"(T} + if(T).
These values being substituted in (92), we readily derive
, - / m ,
1.2.3 .-
(n + 2) (n + i) n (n - i) (n - 2)
1.2.3.4-5
Arranging this according to ascending powers of n, it be-
comes
88 INTERPOLA TION. 5 I.
Expanding the function by Taylor's formula,
dj nW d*f ,M
"TtfT 4 i.2.3.4~ t ~^7" 1.2.3.4.5^
Comparing- the coefficients of like powers of n in these two
series, we have the following values for the differential co-
efficients :
51. Formulas (ioi) will not apply to values of the function
near the beginning or end of the table. We obtain formulae
for these special cases by comparing formulas (91) and (97),
respectively arranged according to ascending powers of n
with Taylor's formula. We thus obtain
For arguments near the beginning of table :
%w) - \f"( T+w) + */"'( T+ f
51- DIFFERENTIAL COEFFICIENTS.
For arguments near end of table :
J
Example 8. Let it be required to compute the numerical
values of the differential coefficients of the moon's right
da d*oi
ascension with respect to the time, -Jr f ~dT i ' ' * or l88 3>
July 5th, o h .
In substituting the numerical values in (ioi), u>,f',f" . . .
must all be expressed in the same unit. It will be convenient
to express them in seconds.
From the numerical values given on page 75 we have
= ~ - 000 458;
= - - 000 20 -
Therefore - = + .038391 ;
= ~ -000972.
This value of - may be regarded as the fractional part of
90 INTERPOLATION. 52.
a second which the moon's right ascension increases in one
second of time at the instant July 5th, o' 1 . In the hourly
ephemeris of the moon given in the Nautical Almanac there
is given in connection with the moon's right ascension the
"difference for one minute," which is simply the value of
the differential coefficient multiplied by 60 ; i.e., we may sup-
pose the a in -p^ to be expressed in seconds, and the T in
minutes. Thus we have for the example above the " differ-
ence for one minute" = 2 S .3O346. So in connection with the
solar ephemeris there is given the sun's hourly motion in
right ascension, which is the value of -^multiplied by 60x60.
The hourly motion in declination is expressed in seconds of
arc.
. 52. By means of these differential coefficients as given in
the ephemeris, the second differences are taken into account
in the interpolation in a very simple manner*, for we have to
second differences inclusive
W = f ' (T + ^ ~ * ///(r) ;
The difference of these expressions is
and
f(T+nw} =
Thus we have only to correct the value of the first differen-
tial coefficient by adding to it algebraically the product of
53- DIFFERENTIAL COEFFICIENTS. 9!
the difference of two consecutive values by one half the in-
terval n. We then use the corrected differential coefficient,
as we should do if the first differences were constant.
Example 9. Required the sun's right ascension and decli-
nation, 1883, July 4th, 4 h , Bethlehem mean time.
As the longitude of Bethlehem from Washington is
6 m 4O s .2, the corresponding Washington time is 3 h 53 m i9 s .8
= July 4th, 3 h .8888 = July 4.162.
From the solar ephemeris for the meridian of Washington
we then find :
Date. Ci. Hourly Motion. $. Hourly Motion.
July 4.0 6 h 53 ra 33 8 .7 9 io s . 3 o7 2252 / 5i // .i - i3".i 9
July 5.0 6 h 57 m 4i s .02 io s .294 22 47' 22".7 - I4".i8
d*a n
w -^ . - = .013 x .162 = .00105
Corrected hourly motion = io s -3o6
10.306 X 3 h .889 = 40 s .o8
Required a = &$4 m i3*.Sf.
w* -^ . * = .99 X .162 as .080
Corrected hourly motion I3 S .27
13.27 X 3 h .8S9 = 51" 61
Required d = 22 $i' $9".$.
53. If values of the differential coefficients are required
for values of the argument between the dates of the table,
we may derive the necessary formulas by differentiating the
function developed by Taylor's formula (100), viz.:
(103)
',
f'(T+\w}=A, f"(T) = A".
In which A" will be the difference between two consecutive
values of A.
54- THE EPHEMERIS. 95
Then
3" -*
22 6
and formula (92), becomes D' D -f- (-^ - g ^"].
Let \A r ^ 2f"J = [z/] = corrected tabular difference ;
(2 - /'^j [fi] -
Then we may assume
- g <2"J = [?] with sufficient accuracy, (105)
in which Q" is the difference between two consecutive val-
ues of Q. (Q and A are inverse functions one of the other,
but the algebraic sign of the correction need give no
trouble.)
It will be a little more accurate if we take for Q" the
arithmetical mean of the differences between Q and both the
preceding and following values found in the table.
Example 11. Required the distance between the centre of
the moon and Fomalhaut, 1883, July 2Oth, ig h 20 5 8 , Gh.
M. T.
From the ephemeris,
July 20th, 15" Q
July 20th, i8 h D 32 41' 20" (5
= ^536 Q,, _ _]_ 2II
= -4747
July 2oth, 21" Z>3i4i' o" Q
= . 4995 ^' = + ^3
Then / = i
h 20 m 5 s = i" 3347
[Q] = -4683
J' = o 27' 14". 5
Mean Q"
= 230
log t 3-6817
D< = 32 14' 5"- 5
If we had neglected the second differences in this example
we should have found A' = o 26' 51", which can only be
96 IN TERP OLA 7 'ION. 55-
considered a rough approximation. If the interpolation be
extended to third differences, we find A' = 27' 13". 8. This
differs from the first value by a quantity which will be of
very little importance in practical cases.
To Find the Greenwich Time Corresponding to a Given Lunar
Distance.
55. First. We may interpolate the time directly from the
ephemeris, neglecting- the second differences; then with the
time so found as a first approximation we deduce the cor-
rected proportional logarithm [<2], and repeat the computa-
tion.
/ being the required quantity, either (104) or (104), give
the first approximation, viz.,
log / = log A' -f PLA, ..... (106)
or PLt = PL A' - PL A (106),
Then with this value of / we determine the corrected pro-
portional logarithm [<2] by (105), and repeat the computation.
Example 12. 1883, July 2oth: determine the Gh. M. T.
when the distance between the moon's centre and Fomal-
haut was 32 14' 5".5.
4536
We find from the ephemeris that on July aoth, i8 h D = 32 41' 20" PL .4747
Given value of D 1 = 32 14' 5". 5 -4995
log A' 3.2134 Therefore A' = 27' 14". 5
PLA = .4747
log t = 3 6881 Approximated = i 1 ' 2i m i6 9
By (105), - _5 being the latitude of the place.
Transformation of Co-ordinates.
64. PROBLEM I. Having given the altitude and azimuth of
any star, to find the corresponding declination and hour-angle.
Let us refer the star's position to a system of rectangular
co-ordinates in which the horizon shall be the plane of XY,
the positive axis of X being directed to the south point, the
positive axis of F to the west point, and the positive axis of
Z to the zenith.
Then will x, y, z = the rectangular co-ordinates of the star;
4, //, a = the polar co-ordinates of the star;
'A being the distance or radius vector.
We then have* x = A cos h cos a; \
y = A cos h sin a\ > ( IIQ )
z = A sin h. J
*See Davies' Analytical Geometry, edition cf 1869, p. 302; or any other
work on analytical geometry of three dimensions.
IOS PRACTICAL ASTRONOMY. 64.
Let the star now be referred to the equator as the funda-
mental plane, the positive axis of X being directed to the
south point of the equator, the positive axis of Fto the west
point, and the positive axis of Z to the north pole.
Let now x' , y', z' be the rectangular co-ordinates;
A, d, t be the polar co-ordinates.
We then have x' = A cos S cos /; \
y' A cos $ sin /; v (i 1 1)
z' = A sin 3. }
The problem now requires these values of x ' , y r , and z' to
be expressed in terms of x, j, and z. We observe that the
axes of Fare the same in both systems; that the axes of X
and Z make the angle 90 q> with those of X' and Z'.
We therefore require the formulas for transformation of co-
ordinates from one rectangular system to another having the
same origin, viz.:
x' = x cos (90 9?) -f z sin (90 ) + z cos (90 cp)\
or x' = x sin cp -f- z cos cp; \
y' = y; [ . (112)
z' x cos q> -j- z sin (p. )
Substituting in (112) the values of x,y, and z from (no),
and oix',y', and z' from (in), dropping at the same time
the factor A which is common to every term, we have
cos 3 cos t = cos h cos a sin cp -f- sin h cos
-(- sin h sin (p. }
64. TRANSFORMATION OF CO-ORDINATES. 1 09
These equations express the required relation, but they
are not in convenient form for logarithmic computation; be-
sides, the required quantities d and t are given in terms of
their sines and cosines.
It is always best, when practicable, to determine an angle
in terms of its tangent. The tangent varies rapidly for all
angles great or small, and consequently if a small error from
any cause exists in the tangent it will have but little effect
on the value of the angle. On the other hand, if the value
of the angle is near 90 or 270 and is given in terms of its
sine, this function will vary slowly with the angle, and a
small error in the sine will produce a large error in the
angle. The same is true of the cosine for angles near o or
1 80. If the angle is near 90 or 270 it may be determined
with accuracy from its cosine, or if near o or 180 it may be
accurately determined from its sine. In any case it can be
determined with accuracy from its tangent.
For the purpose of effecting the required transformation in
(113), let us introduce the auxiliary equations
sin h n cos A'; ) (HA)
cos h cos a = n sin N. )
This will be possible, for we have the two arbitrary quan-
tities n and N, and the two equations (114) for determining
them. Substituting these values in (113), we have
cos S cos t = n sin N sin q> 4- cos N cos
N)- t \
cos S sin t cos h sin a; I (115)
sin S = n sin A 7 " cos
an d when tan is ; third, for
the species of t let us examine the equation
cos 8 sin t = cos h sin a.
Cos & and cos h will always be -|- therefore the species of t
will be the same as that of a.
As an example of the application of these formulas, take
the following:
Latitude of Sayre Observatory = (p = 40 36' 23".9;
Sun's altitude = h = 47 15' i8".3;
Azimuth = a = 80 23' 4".47;
Required tf and t. The computation is as follows :
N} = 9.7929304
tan t = .0252996 cos t = 9.8364670 cos = 9.8364670
tan S = 9.6293974 cos d = 9.9637894
9.8002564
sin N _ cos h cos a
~( IT]v) = 9-2542495 (proof) cog s CQS - f = 9-2542496
112 PRACTICAL ASTRONOMY. 65.
65. PROBLEM II. Having given the declination and hour-
angle of any star, to determine the altitude and azimuth. This
is the converse of the preceding problem. In this case we require
the values of x, y, z in terms of the values of x' , y', z' .
Our formulas (112) for transformation then become
x = x' sin q> z' cos <>; \
y=y f ; [ . . . . (120)
z = x' cos cp -\- z sin q> . )
Substituting in these the values of x, y, z, x',_y', z' , from (i 10)
and (i 1 1), dropping at the same time the common factor J,
we have
cos h cos a = cos 8 cos /sin q> sin 8 cos
,
90 S =1 p, and 90 h = z. The angles are t,
i So a, and q, the angle at the star, called the
parallactic angle. When any three of these quan-
tities are given, the determination of any other
part is merely a question of trigonometry.
1 1 4 PR A C TIC A L AS '2 'A' ONOMY
COROLLARY. To find the hour-angle oj a star when in the
horizon, or at the time of rising or setting.
When the star is in the horizon the altitude, h, is zero, and
the last of equations (121) becomes
cos S cos t cos fp -f- sin 8 sin q> = o,
sin 3 sin (p
or cos*= ---^V 7^= - tan tf tan ?>. . . (122)
From this equation we may determine / ; but, as before re-
marked, it is better to determine the angle from its tangent.
For this purpose first add both members of (122) to unity,
then subtract both members from unity, and we have
cos 8 cos q> sin 3 sin cp
i -4- cos t --
cos d cos (p
cos d cos cp -4- sin 3 sin cp
I cos /= - ^-f- -- -;
cos o cos cp
COS (cp + (?)
or 2 cos \t - ^-- ~~ ;
cos tp cos 3 '
. cos Q S)
2 sm *' = '
Dividing the second of these by the first and extracting the
square root,
At the time of rising the lower sign will be used ; at the
time of setting, the upper. This formula may be used to
compute the time of sunrise and sunset at any place whose
latitude is known. For example, let it be required to com-
pute the apparent time of sunrise at Bethlehem on the morn-
ing of July 4th, 1 88 1.
67. ANGULAR DISTANCE BETWEEN TWO STARS. 11$
From the Nautical Almanac, page 329, we find for the
sun's decimation = 22 52' oi'V
The latitude y = 40 36' 23".9.
(p _ s 17 44' 22".9 cos = 9.9788425
cp -j- 6 = 63 28' 24' '.9 cos = 9.6499288
tan a ^= -3289137
= 55 35' 52". 5 tan \t = .1644569,,
t in n'45".o
t = 7 h 24 m 47 s .
It being sunrise, t is minus. If we subtract this quantity
from I2 h the time when the sun is on the meridian we
have for the apparent time of sunrise
4" 35 m 13 s -
This differs from the ordinary or mean time by an amount
equal to the equation of time, as will be explained hereafter.
(See Art. 92.)
67. PROBLEM III. Required the distance between two stars
whose right ascensions and declinations are known.
The two stars and the pole will form the vertices of a tri-
angle of which the sides will be 90 d, 90 #', and d, the
required distance. The angle opposite d will
be a' a.
a and a' are the right ascensions of the stars.
S and S' are the declinations.
In the triangle two sides and the included
angle are given; the third side is required.
Il6 PRACTICAL ASTRONOMY.
We can apply equations (121) to this case by writing
(compare Figs. 3 and 4)
h = go - d\
cp = 6'
t = a' a;
a = 180 - B.
Thus we have
sin d cos B = sin d cos d' cos 6 sin 6' cos (a' <*) ; j
sin d sin B = cos 8 sin (a 1 ) ; v (124)
cos = sin d sin ' -|- cos 8 cos = - 9 23' i6".7
/T = - 89 42' i".8g
^= 89 52' 55". 5
tanA'= .7677470 0560^= .0062374
cos^'-Jr) = 7.7182360
cos d = 7.3134726
Il8 PRACTICAL ASTRONOMY. 6/.
Applying formulas (IV), to the solution of the same prob-
lem, we have the following:
a' a = 85 52' 37". 35 cos = 8.8567115 cos = 8.8567115
6 = 22 50' 2l"-9 COt = .3755415 COS = 9.9645407
JV = 9 41' 14". 8 tan = 9.2322530 8.8212522
d' = 9 23' i6".7
Ar+S' = o 17' 58".!
B = 66 48' 40". 8
d= 8 9 5 2'55".5
tan (a' a) = 1.1421632
sin N = 9.2260154
cos (IV + d') = 9.9999940 cot (AT + 5') = 2.2817621
factor = 9.2260214
tan B = 0.3681846 cos .#=: 9.5952317 cos = 9.5952317
tan d = 2.6865304 sin = 9.9999991
9.5952308
proof 9.2260214
sin N
= 9>226 214
CHAPTER II.
PARALLAX. REFRACTION. DIP OF THE HORIZON.
68. The same star may be observed from points on the
surface of the earth separated from each other by several
thousand miles. If the distance to the star is so great that
the diameter of the earth is inappreciable in comparison, it
will appear in the same part of the heavens from whatever
part of the earth it is seen. If, however, the diameter of the
earth bears an appreciable ratio to the distance of the object,
then when the observer's position changes there will be an
apparent change in the place of the star. This difference in
position is called parallax.
It is customary in dealing with bodies which have an ap-
preciable parallax to reduce all positions to the earth's cen-
tre. Thus the places of the sun, moon, and planets, which
we find given in the ephemeris, are the places as they would
appear to an observer at the centre of the earth. This which
we are considering is \\iQdiurnalparallax. With the subject
of annual parallax, which depends upon the position of the
earth in its orbit, we have at present nothing to do. It may
be remarked that on account of the great distances of the
fixed stars their diurnal parallax is in all cases inappreciable.
It is only necessary to consider it in connection with the
bodies of the solar system.
I2O PRACTICAL ASTRONOMY.
Definitions.
69. THE GEOCENTRIC POSITION of a body is its position as
seen from the earth's centre.
THE APPARENT* or OBSERVED POSITION is its place as seen
from a point on the earth's surface.
THE PARALLAX is the difference between the geocentric and the
observed place.
It may also be defined as the angle at the bod)' formed by
two lines drawn to the centre of the earth and the place of
observation respectively.
THE HORIZONTAL PARALLAX is the parallax when the star is
seen in the horizon.
THE EQUATORIAL HORIZONTAL PARALLAX is the parallax
when seen in the horizon from a point on the eartlis equator.
It may also be defined as the angle at the body subtended
by the equatorial radius of the earth.
70. PROBLEM I. To find the equatorial horizontal parallax
of a star at a given distance from the earth's centre.
Let n = the equatorial horizontal parallax = PSC;
a = the equatorial radius of the earth = PC;
A star's distance from the earth's centre = SC.
Then from the figure we have
SHITTY ~; .... . ; . . (125)
*The terms apparent place and true place are to be considered simply as
relative terms. When dealing with parallax we speak of the true place as the
place when corrected for parallax. So when speaking of refraction the appar-
ent place is the place affected by refraction, and the true place is the place cor-
rected for refraction, but it may still require corrections for parallax and a va-
riety of other things. When dealing with the places of the fixed stars we use
the term apparent place in a still different sense, as we shall see hereafter.
PARALLAX.
121
s being the place of the star,/ a point on the surface of the
earth, and c being the centre.
For astronomical purposes the mean distance of the earth
from the sun is regarded as the unit of measure. Then for
the sun we have
A = i;
sin n = a ,
(126)
71. PROBLEM II. To find the parallax of a star at any
zenitJi distance, the earth being regarded as a sphere.
In the figure, s represents the place of the star, z the zenith,
E the centre of the earth,/ a point on the surface.
Let
z' = the observed zenith
distance;
z = geocentric zenith dis-
tance;
/ = parallax = PSE;
a = radius of earth = PE\
A = distance of star = SE.
From the triangle SEP
we have
A : a sin z' : sin p.
FIG. 6.
From which
or, from (125),
sin / = j- sin z';
sin = sin it sin
(127)
(128)
/ and it will generally be very small ; hence for most pur-
poses we may write
p = n sin z' . (129)
122 PRACTICAL ASTRONOMY. 72.
The foregoing solution is only an approximation, the earth
not being a sphere as we have there regarded it. For many
purposes this is sufficiently exact, while for others, particu-
larly where the moon is considered, it is not so. A more
rigorous solution requires us to consider the true form of the
earth.
Form and Dimensions of the Earth.
72. The earth is in form approximately an ellipsoid of rev-
olution, the deviations from the exact geometrical figure be-
ing so small as to be inappreciable for our purposes.
The dimensions of the ellipsoid as given by Bessel are as
follows:
Equatorial radius A = 3962.8025 miles;
Polar radius B = 3949.5557 miles;
Eccentricity of meridian e = .08169683;
log e 8.9122052.
Many other determinations of these quantities have been
made, differing more or less from the above, but these are
still in more general use than any others.
Definitions.
73. THE GEOGRAPHICAL LATITUDE of a point on the earth's
surface is the angle made with the plane of the equator by a
normal to the surface at this point.
THE GEOCENTRIC LATITUDE is the angle formed with the
plane of the equator by a line joining the point with the
earth's centre.
THE ASTRONOMICAL LATITUDE is the angle formed with the
plane of the equator by a plumb-line at the given point.
73-
THE DEDUCTION OF THE LATITUDE.
I2 3
If the earth were a true ellipsoid and perfectly homoge-
neous, the geographical and astronomical latitude would
always be the same. Practically, however, the plumb-line
frequently deviates from the normal by very appreciable
amounts. This deviation is always small, but in mountainous
countries, as the Alps and Caucasus, deviations have been
observed as great as 29". Unless otherwise stated, when
speaking of latitude the astronomical latitude is to be under-
stood. We shall also assume for present purposes that
it coincides in value with the geographical latitude.
Let the annexed figure z
represent a section cut
from the earth's surface
by a plane passing through
its axis. This section will
be an ellipse. Let K be
any point on the surf ace, E)
P and P the north and
south poles respectively.
Then HH' will represent
the horizon of the point K.
FIG. 7 .
Let p = CK = radius of the earth for latitude KO'E'\
(p KO'E' = geographical latitude of point K;
' may be readily computed for any
given value of
g>') in the form of a series. For
this purpose we make use of Moivre's formulae, viz.:*
* As some readers may not be familiar with these very useful formulae, we
give their derivation.
Developing = e* by Maclaurin's formula, we have
r"= i+--f -f -4- ^ . etc.;. . . (a)
^ I~l.t^ 1.2.3^1.2.3.4
also, cos .r = i - + - a . etc - ....... (*)
sin JT = JT ---- 1 -- . etc
1.2.3 1.2.3.4.5
74- THE REDUCTION OF THE LATITUDE. 12$
2 COS X =
2 V i sin x =
V^l tan * =
(135)
Writing tan q>' = p tan ?> where / = ~j. substituting for
tan -f- 2 I/ i . \q* sin 4^
-f- 2 V i -J^ 3 sin 6^>, etc.,
or q>' (p = q sin 2 i".i6 sin 49?. . (138)
To Determine p.
75. x and y being the co-ordinates of the point K, we have
..... (139)
(130)
tan q>' = = - tan cp ...... (134)
Combining (130) and (134), eliminating j, we have
or x\\ + tan ^ tan ') = ^ 2 .
Combining this with (139) and (134) to eliminate x, we find
-f tan
cp' I cos q> . .
^^ = ^ V cos ^' cos (>' - ^)' l
The computation of p from (140) is very simple, but it
may be rendered much more so by developing p, or log p
128 PRACTICAL ASTRONOMY. ;6.
into a series. For this purpose we shall regard A the
equatorial radius as unity, when we have
secy = I + ^ tan V = cos>4-| 4 sin>
p ~ i + tan ' ~ & ' ~~J?
i + - tan 3 (p cos" cp -f sm>
B' B*
Let us write . - = i - g ; - -, = i - e\
we have
Taking the logarithms of both members,
2 log p = log (i g' 1 sinV) log (i e* sin 2 ^)-
Developing the second member by the logarithmic formula,
2 lo