Theory of Observations

UC-NRLF
*C Its IDb
i
fe
Wi "i
iTHEORY OF OBSERYATIONS
LIBRARY
OF THE
UNIVERSITY OF CALIFORNIA.
Class
wm^
THEORY OF
OB SE RVATION
THEORY OF
OBSERVATIONS
BY
T. N.
THIELE
DIRECTOR OF THE COPENHAGEN OBSERVATOHY
-<>>3i5Kc>-.-.
o! rue
Y
f
'>
UNIVERSITY
Of
PUBLISHED HY
CHARLES & EDWIN LAYTON

56
FARRINGDON STREET
LONDON
1903
t4-
COPENHAGEN.
PRINTED BT BIANCO LUNO
CONTENTS
Numbers
The Law of Causality.
I.
of
Page
formulae
1.
Belief in Causality
2.
The Observations and
3.
Errors of Observations
4.
Theoretical and Empirical Science
II.
5.
On
6.
Laws
their Circumstances
Laws
of Errors.
Rei)etitions
Actual Errors and Laws of Presumptive Errors
of
The Law
Large Numbers of Repetitions

Four Different Forms of Laws of Errors
7.
8.
of
III.
Tabular Arrangements.
Frequency and Probability

Repetitions with Qualitative Differences between the Results
Repetitions with Quantitative Differences between the Results
9.
10.
12.
Curves of Actual Errors of Observations in Discontinued Values
13.
14.
Curves
Curves
11.
IV.
16.
17.
18.
1.
28.
14
Laws
of Errors.
Law
15
16
of Errors
19
Problems
19.
20.
10
11
of Errors
Functional
or Exponential
14
Their Determination by Interpolation
The Typical
10
Actual Errors for Rounded Observations

of Presumptive Errors
V.
Curves of Errors.
of
Typical Curves of Errors

Particular Measures of Curves
1.5.
913. The Binomial Functions

14.
Some of the more general Functional Laws
111731
20
of Errors.
Series
21
Numbers
Laws
VI.
of
of Errors expressed by Symmetrical Functions.
formulae
21.
22.
23.
Page
1516. Coefficients of the Equation of Errors. Suras

1724. Half-luvariants
2527. Mean Values, Mean Deviations, Mean Errors
of
Powers
22
24
27
29
Examples
Relations between Functional
VII.
24.
25.
28-
29.
3031.
30
31
83
very general Series by Half-Invariants
Functions of
26.
27.
28.
29.
30.
31.
32.
33.
Functions of
of
Two
3435. Linear Functions
47.
of Errors of Functions of Observations.
35
36
Linear Functions
or
of
More Observations. Bonds

Unbound Observations
37
38
39
Examples
Non-Linear Functions
41
Examples
41
Laws
3846. Laws
37.
Laws
One Single Observation
3233. Half-Invariants
36.
of Errors and Half-Invariants.
Examples
VIII.
Laws
Their Kelations
Mean Value. Approximation to the Typical Form

of Errors of the Mean Deviation and Higher Half-Invariants
Transition between Laws of Actual and of Presumptive Errors. Rules of
41
of Errors of the
44
Prediction
34.
4850. Determination
of the
Law
of
Presumptive Errors when the Presumptive
Mean Value
is
known beforehand
35.
51.
51
of Observations.
Weights
Probable Errors and other Dangerous Notions
36.
52 - 58.
52
Free Functions.
IX.
Conditions of Freedom
53
55
56
Examples
37.
38.
Possibility of regarding certain
5961. Every Function
Two
40.
41.
42.
Bound Observations
of Observations
6266.
of
Two, which belong
to
67.
Complete Sets of Free Functions

Orthogonal Transformation
Schedule of Liberation
58
.59
59
59
60
62
General Remarks about Computation with Observed or Inexactly Given Values
63
Can Laws
44.
The
of Errors
Adjustment.
be Determined by Means of Observations whicli are not Repeti64
tions ?
6872.
.56
Example
43.
46.
Sum
Mutually Free Systems of Functions
X.
45.
as free
can be divided into a
Example
Every Single Observation likewise
39.
47
50
Examples
Principle of Adjustment
Criticism the Method of the Least Squares
;
Adjustment by Correlates and by Elements
67
68
70
XI.
Numbers of
Adjustment by Correlates.
formulae
47.
48.
7381.
8284.
Page
71
General Solution of the Problem

Summary and Special Criticism
72
Schedule for Adjustment by (Correlates
49.
.'lO.
Modifications of this
73
Method
75
75
Examples
XII.
51.
53.
52.
54.
One Equation
Adjustment by Elements.
Normal Equations
8789.
Special Case of Free Normal Equations
90102. Transformation of the General Case into the Preceeding Special Case
103-108. The Minimum Sum of Squares
85 86.
for each Observation.
80
83
84
85
86
Criticism
55.
.56.
Under- Adjustment and Over- Adjustment

Examples
XIII.
Special Auxiliary Methods.
Lawful and Illegitimate Facilities

109110. Adjustment upon Differences from Preceeding Computations
69. Ill 113. How to obtain Approximate Freedom of the Normal Equations
The Method of Fabricated Observations. Example
60.
The Method of Partial Eliminations
61.
Example
62. 114116. Rules for the General Transformation of the System of Elements
77
80
94
94
96
98
57.
58.
.'
99
100
103
105
Example
63. 117-120. The Method of Normal Places
Example
64.
Graphical Adjustment
106
HO
112
XIV. The Theory of Probability.
121123. Relation of Probability to the Laws of Errors by Half-Invariants
66. 124125. Laws of Errors for the Frequency of Repetitions. Obliquity of these Laws
65.
XV. The Formal Theory
of Probability.
Addition and Multiplication of Probabilities
119
121
68.
Examples
Use of the Polynomial Formula
Examples
126.
127 129.
Linear Equations of Differences.
for Probabilities
by Repetitions
XVI.
70.
71.
72.
73.
123
123
Oppermann's Transformation
124
126
Examples
118
67.
69.
116
of Errors
The Determination of Probabilities a Priori and a Posteriori.
Theory and Experience

Determination a Priori
129
130
130-133. Determination a Posteriori and
Example
134-137. The Paradox
its
Mean
Error
132
134
of
Unanimity.
Bayes's Rule
134
Numbers of
XVII.
Mathematical Expectation and
its
Mean Error.
formulae
74.
75.
76.
Mathematical Expectation
138140. Examples
141143. Mean Errors of Mathematical Expectation of Unbound Events
Examples
144-146. Mean En-or of Total Mathematical Expectation of the Same Trial
147.
77.
Examples
The Complete Expression
Page
137
138
139
140
141
142
of the
Mean
Errors
142
I.
We
THE LAW OF CAUSALITY.
assumption that everything tJiuf exists, and everything

that happens, exists or happens as a necessary consequence of a previous state of things.
1.
If
a state of things
Any
difference
by some
start with the
is
repeated in every detail,
believed; in the
with which
it
belief.
will,
this
in
same consequences.
part the same, must be explainable
to exactly the
difference in the other part of the causes.
must be
if
must lead
between the results of causes that are
This assumption, which
our
it
is
may
It
he
is
closely
may
same way
as
we
believe the fundamental assumptions of religion,
and intimately connected.
be denied in theory,
The law
but not in practice.
of causality forces itself upon
Any
of causality.
If
we
But
we could imagine
all
if
no one
in that very question he bears witness
are consistently to deny the law of causality,
observation, and particularly
who
person
watchful enough, catch himself constantly asking himself,
has happened, and not that.
If
be called the law of causality, cannot be proved, but
we must
to
denies
it,
else,
why
the
law
repudiate
all
prediction based on past experience, as useless and misleading.
for
an instant that the same complete combination of causes
could have a definite number of different consequences, however small that number might
and that among these the occurrence
of the actual consequence was, in the old sense

no
observation
would
ever be of any particular value.
Scientific
word, accidental,
observations cannot be reconciled with polytheism. So long as the idea prevailed that the
be,
of the
depended on whether the power of Njord or that of Skade was the
result of
a journey
stronger,
or that victory or defeat in battle depended
on whether Jove had, or had not,
listened to Juno's complaints, so long were even scientists

obliged to consider
below their
it
dignity to consult observations.
But
holds good,
if
the law of causality
and compared with
We
is
acknowledged
to
others, teaches us the laws by
when
which always
correctly
appraised
which God rules the world.
can judge of the far-reaching consequences
which the law of causality was not valid at

which the effects of the law are more or less veiled.
ditions in
be an assumption
then every observation gives us a revelation which,
it
would have,
all,
by
if
there were con-
considering
tiie
cases
in
In inanimate nature the relation of cause and effect
so clear that the effects are
is
determined by observable causes belonging to the condition immediately preceding, so that

within this domain,
the problem,
observed results
may
be solved by a tabular arrangement of the several
means
It is the prerogative of living
we must
ences received, and
cell
When, however,
of interpolation.
beings to hide and covertly
history.
be the only indication present at the
may
a transmitter of the
still
domain look
therefore within this
whole of the past
difference
moment
causes and their
same simple,
effects,
as
in
transmit the influ-
to
the
influencing
causes
the construction of a
single
for
of the observation that the cell
is
which may date from thousands of years back.
operative cause,
In consequence of this the naturalist, the physiologist,

ceptionally attain the
beings are the object
living
becomes more complicated.
of our observations, the case immediately
throughout the
and the transformation of the
according to the causing circumstances,
tables into laws by
definite,
the physician,
can only
ex-
quite
and complete accordance between the observed
can be attained by the physicist and the astronomer within
their domains.
Within the living world, communities, particularly human ones, form a domain
where the conditions of the observations are even more complex and difficult.
Living
For though
beings hide, but the community deceives.
munity
cause and
tries to
oft'
as
change one
either to
effect,
give
its
yet
law
divine or
every
fixity,
by
are constantly broken
tittle of
any
it is
not in the power of the com-
community lays down

and to make it operate
its
own laws
also.
Every community
as a cause; for instance, by passing
and changed.
in the case of communities, represents observations, has
therefore a very difficult task; although the observations are so numerous,
historical
to
light the truths
An
Observation
is
are able from
believe in
we must
and true revelations; the
and propagate.
it
does not
amount
of sensation
an ob-
to
which are or
that
so
In order that
an
than
law
of
the
believe that, in all observations,
difficulty
understand the answer correctly.

hypothesis
re-
their effects.
other
assumption
and on the other hand by
with each other according to the law of causality,
virtue of the law of causality
or
by independently bringing to
a putting together of several results
some represent causes and others
essentially correct
to conceal,
isolated sensation teaches us nothing, for
are supposed to be connected
By
assist in the work,
which the communities want
moving the wrong opinions which these

servation.
we
answer only a very few questions in cases where the intellectual weapons of
and speculative criticism cannot
2.
it
but nevertheless the laws of the community
threats of punishment;
Statistical Science which,
them alone
bond between
really divine law, or to break the
is,
to ask searchingly
observation
causality,
may
it
be
we
enough and
free
from
must include a
get
to
every
perfect
description of
all
the circumstances in the world, at least at the instant preceding that at
which the phenomenon
is
But
observed.
it
clear that this far surpasses
is
Real observations have a
even in the most important cases.
a short statement of the time and place of observation,
we
much
simpler form.
refer to
what
with the observation we, generally, not only disregard everything which
little
which we
the
or no influence, but
call
By giving
known of the
may
be supposed to
we pay attention only to a small selection of circumstances,

we expect, in virtue of a special hypothesis concerning
essential, because
and
of cause
relation
is
and, of the infinite multiplicity of circumstances connected
state of things at the instant;
have
what can be done,
effect,
that
observed phenomenon
the
be
will
effect
of these
circumstances only.
Nay, we are often
to
compelled
disregard
certain
circumstances
as
unessential,
the phenomenon; and we do
this either
though there is no doubt as to their influencing
because we cannot get a sufficient amount of trustworthy information regarding them, or
because
would
it
be
instance in statistical
impracticable to
trace out
their
connection
with
the
For
eftect.
can
observations on mortality, where the age at the time of death
phenomenon, we generally mention the sex as an essential

circumstance, and often give a general statement as to residence in town or country, or as
to occupation. But there are other things as to which we do not get sufficient information:
as the observed
be regarded
he has been more or

even
And
whether
the dead person has lived in straitened or in comfortable circumstances,
whether
less
exposed to infectious disease, etc.
and we must put up with
this,
one or other of these things was the principal cause of death.

analogous cases are frequently met with both in scientific observations and in everyday
if it
is
certain
that
occurrences.
In order to obtain a perfect observation
it is
necessary, moreover, that our sensations
should give us accurate information regarding both the phenomenon and the attendant
circumstances; but all our senses may be said to give us merely approximate descriptions
of any
phenomenon
recognizes
accuracy
rather
than to
no difference which
is,
for the sake
moreover,
often
of convenience.
falls
greatly
measure
short
it
of a
increased
The man who has
accurately.
certain
by
to
the
Even the
finite
use
of
finest of our senses
magnitude.
arbitrary
This lack
round
of
numbers
measure a race-course, may take into
account the odd metres, but certainly not the millimetres, not to mention the microns.
Even our
3. Owing to all this, every actual observation is affected with errors.
best observations are based
upon hypothesis, and often even on an hypothesis that
is
cer-
influence
tainly wrong, namely, that only the circumstances which are regarded as essential,
the phenomenon; and a regard for practicability, expense, and convenience makes us give
approximate estimates instead of the sharpest possible determinations.
Now and
then the observations are affected also by ijross errors which,

1*
although
not introduced into them on purpose, are yet caused by such carelessness or neglect that
and ought
they could have been,
more
often call the
what the word

of
it,
name
have been, avoided.
to
or less unavoidable errors accidetdal.
meant, and what
originally
still
unknown
For accident
element, involved in
accidental, whether a die turns
connected with the throwing, the
man, not even the
chance)
is not,
often lingers in our ordinary acceptation
to
vents us from fully comprehending the connection between

is
(or
happen without any cause, but only a

some relation of cause and effect, which pre-
a capricious power which suffers events

for the
we
contradistinction to these
When we
them.
say that
it
up "six" or "three", we only mean that the circumstances

that no
fall, and the rolling of the die are so manifold
cleverest juggler
and arithmetician united in the same person, can suc-
ceed in controlling or calculating them.

In
really
observations
many
know more
we
We
or less.
many circumstances about which we
reject as unessential
may
be justified in this; but
such a circumstance
if
is
of
importance as a cause, and we arrange the observations with special regard to

we may sometimes observe that the errors of the observations show a regularity which
sufficient
it,
is
The same may be the case if, in computations dealing

we make a wrong supposition as to the operation of some
not found in "accidental" errors.
with the results of observations,
Such
circumstance.
errors are generally called systematic.
be found that every applied science, which
It will
4.
divided into two parts,

one.
(observational)
very
much on
their
theoretical
theoretical
part
of the science
its
and consequences of the hypotheses.
deals
with what we suppose to be accurate
is
the development of the form, connection,
reasonings
But
must change
it
its
hypotheses as soon as
it is
that they are at variance with experience and observation.
The empirical
them
may be
mathematical) part and an empirical
one to the other can be allowed.
determinations, and the object of
clelr
well developed,
Both are absolutely necessary, and the growth of a science depends

No lasting
influencing one another and advancing simultaneously.
divergence or subordination of
The
(speculative or
is
with
the
side of the science procures
theoretical
propositions,
necessary, the hypotheses of the theory.

vations.
that, as
But
it
must not
shown above,
forget
it is
itself
and
induction
By
though
is
and arranges the observations, compares

entitled by means of them to reject, if
it
it
may have
founded on hypotheses.
can deduce laws from the obsera natural inclination to do so
The very form
and especially the selection of the circumstances which are
to
of the observation,
be considered as essential
and taken into account in making the several observations, must not be determined by rule
of thumb, or arbitrarily, but must always be guided by theory.
Subject to this
science should work
it
must
as a rule be considered
somewhat independently
of one
best,
another,
that the two
each
in
its
sides
own
of the
particular
In what follows the empirical side will be treated exclusively, and
way.
on a general plan, investigating not the particular way in which
it
will be treated
statistical,
chemical, phy-
and astronomical observations are made, but the common
sical,
they are
rules according to
which
submitted to computation.
all
LAWS OF EEROES.
II.
observation is supposed to contain information, partly as to the

which
we are particularly interested, partly as to all the circumstances,
phenomenon
connected with it, which are regarded as essential. In comparing several observations, it
b.
Every
in
makes a very great

or
difference,
whether such essential circumstances have remained unchanged,
whether one or several of them have changed between one observation and another.
of the former case, that of repetitions, is far simpler than that of the latter,
The treatment
and
try
is
to
selves
therefore
more
particularly the subject of our investigations;
master also the more
general case in
difficult
its
nevertheless,
we must
simplest forms, which force them-
upon us in most of the empirical sciences.
By
we understand those
repetitions then
observations,
in
which
all
the essential
circumstances remain unchanged, in which therefore the results or phenomena should agree,
if all
the operative causes had been included
more, we can without

that no essential
been no such
among
our essential circumstances.
hesitation treat as repetitions those observations, in which
Further-
we assume
circumstance has changed, but do not know for certain that there has
change.
Strictly
speaking,
this
would furnish an example of observations
with systematic errors; but provided there has been no change in the care with which the
essential circumstances
have been determined or checked,
simpler treatment applicable to the case of repetitions.

missible,
in
if,
it
is
permissible to employ
for instance, the observer
the records of a
during the repetitions has perceived any uncertainty

and
therefore paid greater attention to the following
circumstance,
^
repetitions.
6.
The
special
features
of the observations,
accuracy, depend on causes which have been
some overlooked uncertainty

no speculation can indicate
left
and
in
particular
out as unessential
late
their degree of
circumstances,
or
on
Consequently
and particularities of observations.
These
case of repetitions can this estimate be undertaken
The phrase
'^
in the statement of the essential circumstances.

to us the accuracy
must be estimated by comparison of the observations with each

work.
the
This would not however be per-
of errors
is
directly
used as a general
name
other,
but only in the
and without some preliminary

for
any mathematical expres-
sion representing the distribution of the varying results of

repetitions.
Lmvs of actual errors are such as correspond to repetitions actually carried out.
But observations yet unmade may also be erroneous, and where we have to speak hypothetically about observations,
we
titions,
or
have
to
do with the prediction of results of future repe-
vent any misunderstanding
we then
call this idea "Zfrns
of repetitions
must
some variations
entail
The two
of presumptive errors".
kinds of laws of errors cannot generally be quite the same thing.
number
In order to pre-
are generally obliged to employ the idea of "laws of errors".
Every variation
in the corresponding
the
in
law of errors; and
we compare two laws of actual errors obtained from repetitions of the same kind in
equal number, we almost always observe great differences in every detail. In passing from
if
actual repetitions to future repetitions, such differences at least are to be expected.

over, whilst
any collection of observations,
on examination give us
can be used
its
law of actual
which can
errors,
it
ments of an angle, the
results of our first
while the following repetitions
still
measure-
If, for instance, in repeated
measurements
more frequently, and
second quadrant, and even commenced
series of repetitions that
not every
is
for predictions as to future observations.
More-
at all be regarded as repetitions, will
all fell
within the
first
quadrant,
at last exclusively, fell within the
to pass into the third, it
would evidently be wrong
to predict that the future repetitions would repeat the law of actual errors for the totality
of these observations.
In similar cases the observations must be rejected as bad
or
mis-
conceived, and no law of presumptive errors can be directly based upon them.
7.
Suppose, however, that on comparing repetitions of some observation we have
several times determined the law of actual errors in precisely the

first
small numbers of repetitions,
then larger and
still
larger
then, on comparing these laws of actual errors with one another,
come more
agreements extend successively to
bound
to
numbers of
alike in proportion as the

all
repetitions
same way, employing
numbers
each law.
for
we remark
grow
that they be-
greater,
by
we cannot have any hesitation

possible number of repetitions,
vary with the number of repetitions, then
dictions concerning future observations,
This, however,
is
made under
wholly legitimate only,
essentially the
when
it
is
If
and that the
those details of the law which are not
the law of actual errors, deduced from the largest
at
necessity
in using
for pre-
same circumstances.
to be expected that, if
we could
obtain repetitions in indefinitely increasing numbers, the law of errors would then approach
namely the laiv of presumptive errors itself, and would not oscillate
between several forms, or become altogether or partly indeterminate. (Note the analogy
with the difference between converging and oscillating infinite series).
We must therefore
single definite form,
distinguish between good and bad observations, and only the good ones, that
satisfy the
errors
and
is
those which
above mentioned condition, the Uno of large numbers, yield laws of presumptive
afford a basis for prediction.
As we cannot
repeat a thing indefinitely often,
we can never be
quite certain that
a given method of observation
may
be called good.
laws of actual errors,
on
shall always rely
of concordant repetitions, as suffi-
law of presumptive
ciently accurate approximations to the
we
Nevertheless,
deduced from very large numbers
errors.
And, moreover, the purely hypothetical assumption of the existence of a law of

behaviour of the laws of
presumptive errors may yield some special criteria for the right
number
actual errors, corresponding to the increasing
We
must here notice
and inapplicable, we
that,
to
it
change
why
by such a test proves bad
sometimes by a theoretical
shall nevertheless often be able,
of errors, to find out the reason
and
of prediction.
series of repetitions
the given method of observation
show that
so that the checks will at least
In the case mentioned in the preceding paragraph, for instance, the
time of observation
A4id
light
if
there to be reckoned
is
we do not
among
attain our object, but should fail in
errors,
After
what may be the cause of

error
is
failure,
and,
sceptic
in
is
The
obvious.
may
at
throwing
be said even at this
for
so,
will generally
time give up
catch
may
doing
phenomenon,
the
whole
himself speculating on
he must acknowledge that the
never to be looked for in the objective nature of the conditions, but in an insuffi-
cient development
large
his
remedy
various
we may
failures
repeated
matter in despair; but even the most thorough
good as
has been improved.
it
many attempts
it
of large numbers, as quite inapplicable to any given refractory
be out of the question.
as
means that may be employed,

that absolute abandonment of the law
we have been made acquainted with the
and the various forms taken by the laws of
not
is
the essential circumstances.
upon some phenomenon by means of good observations,
stage, before
criticism
peculiarities in the irregularities of the laws
and sometimes by watching the
of the method,
others,
when a
and establish the
of the repetitions,
conditions necessary to justify their use for purposes
the methods employed.
of
numbers has the character of a
belief.
harmony with human thought that we, sooner
From
There
this
is
in
point
all
of view
external
then
the
conditions
or later, by the use of due sagacity,
law
of
such a
parti-
cularly with regard to the essential subordinate circumstances of the case, will be able to
give the observations such a form that the laws of actual errors, with respect to repetitions
in increasing
numbers,
will
sidered valid as the law of

8.
show an approach towards a definite form, wliich may be conpresumptive errors and used for predictions.
Four different means of representing the law of errors must be described, and
their respective merits considered,
namely:
Tabular arrangements.
Curves of Errors,
Functional
Laws
of Errors,
Symmetric Functions of the Repetitions.

In
comparing these means
of representing the laws of errors,
we must
take into
consideration which of them
is
the easiest to employ, and neither this nor the

description
forms of the laws of errors demands any higher qualification than an elementary
knowledge of mathematics. But we must take into account also, how far the diflerent forms
of the
are calculated to emphasise the important features of the laws of errors,
may
e.
i.
those which
be transferred from the laws of actual errors to the laws of presumptive errors.
more thorough knowledge of mathematics would be
this single point, certainly, a
than that which
may be
expected from the majority of those students
As the
occupy themselves with observations.

the
presupposes
determination
some propositions from the
III.
in its simplest form.
values
of limiting
to
who
are obliged
to
of the law of presumptive errors
numerous approximations,
infinitely
differential calculus would, strictly speaking, be

necessary.
TABULAR ARRANGEMENTS.
In stating the results of
9.
definition
On
desirable
Identical
results
the several repetitions
all
will of course
we
give the lajv of errors
be noted by stating the
number
of
the observations which give them.
The
table of errors,
when arranged,
will state all the various
results
and the
fre-
quency of each of them.
The
table of errors is certainly
quencies of the several results, that

total
number
of repetitions.
law of large numbers,
are,
It
as
is,
improved,
must be the
number
the
relative
denote this transition in our ideas: probability
sumptive
errors,
proportion of the
on the supposition of
sidering
infinitely
number
numerous
the relative frequency
in it the relative fre-
frequencies which,
of observations
constant values of the law of presumptive errors.
the
when we include
the ratio which each absolute frequency bears to the
is
Long
is
approach the
in
a law of pre-
results to the total
number,
There can be no objection to conas
errors
corresponding probability of the law of presumptive errors,
an approximation to the
and the doubt whether the
relative frequency itself is the best approximation that can be got from the results
given repetitions,
It
mentioned
is
rather of theoretical than practical interest.
makes some
if
the
difference
phenomenon
in
is
several
the
usage gives us a special word to
of coincident
of the law of actual
to
increased,
repetitions.
according to
other respects
Compare
of the
73.
as well as in the one just
such that the results of the repetitions show qualitative
differences or only differences of magnitude.
10.
In the former case, in which no transition occurs, but where there are such
abrupt differences that none of the results are more closely connected with one another than
with the
rest,
the tabular form will be the only possible one, in which the law of errors can
9
be given.
This case frequently occurs in statistics and in games of chance,
reason the theory of probabilities, which
is
of observation.
am
it,
and made
of opinion, however,
for this
the form of the theory of observations in which
these cases are particularly taken into consideration,

vious authors have begun with
and
it
demands
All
special attention.
pre-
the basis of the other parts of the science
that
it
is
both safer and easier to keep
it
to
the last.
11.
that there
is
If,
however,
there
is
each other than
all
that are near one another.
have the following form
is
retained,
foot too high
care to
table of the results of firing
at
bring
a
together the results
target
may
for
instance
the left
Central
13
Central
foot too
we must take
1 foot to
between them, or that some results are nearer
the rest, there will be ample opportunity to apply mathematical methods;
and when the tabular form
such a diiference between the results of repetitions,
either a continuous transition
low
Total ...
4
20
134
26
180
10
On
the other hand,
values,
we
select
values,
we
shall
a table arranged
according to the magnitude of the

and
followed
a single middle value, preceded
by nearly equal numbers of
well
fitted
a
which
is
to
quantity
very
represent the whole series of
get
in
if,
repetitions.
If,
we
also
we
while
are thus counting the results arranged according to their magnitude,
take note of those two values with which
part of the total
say 16 per
ct.),
deviations.
If
number, and
we may
we
we
respectively (a) leave the first sixth
is
to
consider these two as indicating the limits between great and small
two values along with the middle one above referred
state these
give a serviceable expression for the law of errors, in a
although rough,
we ought
upon the last sixth part (more exactly
(b) enter
Why
not to be despised.
we ought
way which
to,
we
very convenient, and
is
to select just the middle value
and
the two sixth-part values for this purpose, will appear from the following chapters.
CURVES OF ERRORS.
IV.
12.
Curves of actual errors of repeated observations, each of which we must be
able to express by one real number, are generally

line as the axis of abscissae,
quantities,
we mark
ofl'
constructed as follows.
points corresponding
to
straight
the observed numerical
and at each of these points we draw an ordinate, proportional to the number
We
of the repetitions which gave the result indicated by the abscissa.
hand draw the curve

and regular as
of errors
possible.
the ordinate will be
through the ends of the ordinates, making
0,
but do not really appear,
it
free
smooth
as
the form of the curves of errors
If the observation
is
essentially
among
the repetitions,
on the axis of abscissae.
or the point of the curve falls
case occurs very frequently,
almost discontinuous.
with
then
For quantities and their corresponding abscissae which, from the
nature of the case, might have appeared,
this
On
bound
Where
becomes very tortuous,
to discontinuous
numbers,
for
instance to integers, this cannot be helped.

13.
If the
observation
evitable loss of accuracy,
made
limit for each observation, a

to
in
is
either of necessity or arbitrarily, in spite of
round numbers, so that
somewhat
it
different construction
gives a lower
of the
some in-
and a higher
curve of errors ought
be applied, viz. such a one, that the area included between the curve of error, the axis
of abscissae, and the ordinates of the limits,
But
within these limits.

degree of accuracy
in this
involved
in
way
the
free sides
proportional to the frequency of repetitions
use of round
can be made by laying down rectangles

trapezoids with their
is
the curve of errors
between
may depend
numbers.
This
very
construction
the bounding ordinates,
approximately parallel
to the
much on
or
the
of areas
still
better,
tangents of the curve.
If the
11
limiting round numbers are equidistant,
the
mean
heights of the trapezoids or rectangles

In this case a preliminary
are directly proportional to the frequencies of repetition.
struction of curve-points can be

It
a very
is
common
made
as in 12,
custom,
and may often be used as
con-
sufficient.
but one not to be recommended, to draw a broken
between the observed points instead of a curve.
line
There can be no doubt that the curve of
14.
errors, has the
cases
this
critical drawbacic,
form
errors, as a
for
and were not the said uncertainty
advantage of perspicuity,
would perhaps be
sufficient.
Moreover,
it
the law of
so
in
is
in
many
practice
quite possible, and not very difficult, to pass from the curve of actual errors to one which
may
hold good for presumptive errors
though, certainly, this transition cannot be founded
upon any positive theory, but depends on
siiill,
which may be acquired by working at good
examples, but must be practised judiciously.
According
law of large numbers we must expect that, when we draw curves
to the
of actual errors according to relative frequency,
for
numerous
series of repetitions,
first
based upon small numbers, afterwards redrawn every time as we get more and more repetitions,
the
curves,
furnished with
which
at
peaks and
first
valleys,
constantly
will
simpler and more smooth, so that at
we cannot
of observations,
other.
We may
changed their forms and were plentifully
gradually
last,
become more
when we have
each
a very large but
distinguish the successive figures
other,
as also
finite
number
we have drawn from one an-
thus directly construct curves of errors, which
of curves of presumptive errors,
like
may
be approved as pictures
but in order to do so millions of repetitions, rather than
thousands, are certainly required.

If
from curves of actual errors
to the curve of
presumptive
partly by an estimate of
for
errors,
how
for
we must
small numbers
are to
draw conclusions
great irregularities
we may expect
for the
in a curve of actual errors
form
of regular
curves
we must suppose that the curves of presumptive errors will be very

we must get some practice, but this is easy and interesting.
as
as
guess, but at the same time support our guess,
the.given number, partly by developing our feeling
that sort,
we
of
regular.
In both respects
Without
actual errors,
we
large deviations
feeling tied
that
in question,
cause
to the particular
points that
why
greater
determined the curve of
approach them, and especially not allow many
on the same side to come together.
deviations (the reason
those
down
shall nevertheless try to
We
can generally regard
large
mentioned in the chapter on the Theory of Probabilities)

as compared with the absolute frequency of the result
will be
errors,
than the square root of that number (more exactly ]//t
frequency of the result,
as
n the number
of all repetitions).
times as great as this ought not always to be avoided,
^~
,
where
But even deviations two

and
Ave
may
be satisfied,
one third of the deviations of the determining points must be called large.
We
/;
is
the
or three
if
may
only
use
12
the word 'adjustment" (graphical) to express the operation by which a curve of presumptive
errors
determined.
is
The adjustment
(Comp. 64).
have approached too near to some imaginary

curve of actual errors, then the curve
Our second
is
ideal,
is
but
an over- adjustment,
called
we have kept
if
too
the
to
close
said to be tinder-adjusted.
guide, the regularity of the curve of errors, is as an aesthetical notion
somewhat vague kind. The continuity of the curve is an essential condition, but
not sufficient. The regularity here is of a somewhat diflerent kind from that seen
of a
is
we
if
the
of simple,
examples
The curves
acquainted.
circumstances
essential
indefinitely large.
it
in
continuous curves with which students more especially become
we would never
of errors get a peculiar stamp, because
of the
observation
so
absurdly
Nor would we without necessity
retain
form
select the
deviations could become
the
that
of
which
observation
might bring about discontinuity. It follows that to the abscissae which indicate very large
The curve of errors must have
deviations, must correspond rapidly decreasing ordinates.
the axis of abscissae as an asymptote, both to the right and the
All frequency being
left.
positive, where the curve of errors deviates from the axis of abscissae,
keep on the positive side of the

of a bow,
It
latter.
with the axis of abscissae
for
must
the
string.
the
must exclusively
or less get the appearance
In order to train
we recommend
apprehension of this sort of regularity,
more
therefore
it
the
&
of figs. 2
study
for
eye
3,
the
which
represent curves of errors of typical forms, exponential and binomial (comp. the next chapter,
are drawn from
p. 16, seqq.), and a comparison of them with figures which, like Nr. 1,
actual observations without any adjustment.
The
best
way
to acquire practice in
that no student ought to neglect
the law of presumptive
errors
it,
may
may
be
drawing curves of
errors,
be to select a series
considered
as
which
so important
is
for
which
before
us in
of observations,
known, and which
is
tabular form.
We
tions;
commence by drawing
curves of actual errors for the whole series of observa-
then for tolerably large groups of the same, and lastly
random and each containing only a few

besides
the
curve of actual errors,
scale, so that the abscissae are
observations.
On
for
each
small groups
another one of the presumptive errors,
common, and
taken at
drawing we draw
also,
on the same
the ordinates indicate relative frequencies in
proportion to the same unit of length for the total number.
The proportions ought
to be
chosen so that the whole part of the axis of abscissae which deviates sensibly from the
curve, is between 2 and 5 times as long as the largest ordinate of the purve.
Prepared by the study of the differences between the curves, we

to the construction of curves of presumptive errors immediately
of the curve
which correspond
to the observed frequencies.
not consider ourselves obliged to
In
pass
on at
last
from the scattered points

this construction
we must
reproduce the curve of presumptive errors which we may
13
know beforehand; our

of a curve
which
is
way
is
to represent the observations as nearly as possible by
means
as smooth and regular as that curve.
The following
this
task
table of
as an exercise.
500
results, got
by a game of patience,
may
be treated in
14
The equation
next chapter).
of the curve is:
= 2-0228 + 0-0030(a' 11) 0-6885( ll)2+0-01515(j; 11)3 0-001678(^ 11)

Log2/
15.
By
the study of
many
curves of presumptive errors, and especially such as
represent ideal functional laws of errors,

exists a typical
form of curves of
with this form
Familiarity
But we must not expect

it
deviates
which
is
to get the impression
particularly
to get it realised in all cases.
considerably
distinguished by symmetry.
For this reason
it
may become
we
so glaring that
are forced by
scarcely be any doubt that

is
we have not a
to say, different
this conclusion.
If,
not
for
the
extremes,
so
that
the
curve of
series of repetitions
proper,
but a combination
methods of observation have been used and the
results
In such cases we cannot expect that the law of large numbers will
mixed up together.
force,
observations are
the
to
two or several smaller curves of errors beside one another, there can
errors is divided into
that
it
abscissae there are intervals which
are as free from finite ordinates as the space beyond
remain in
have considered
errors,
instance, between the extreme values of repetitions
of several;
errors.
curves, an example taken from real observa-
Deviation from the typical form does not indicate that

good.
of presumptive
errors.
presumptive
But
there
that
which in consequence of its marked want of symmetry

from the typical form.
Fig. 4 shows this last mentioned law of
skew curve of
of a
errors,
fail
useful for the construction of curves
is
to give, alongside of the typical
important
tions
we cannot
and we had
them by tracing out the
we cannot
retain
which distinguish the groups of the
series,
better, therefore, reject
essential circumstances
such observations,
if
but have been overlooked.

16.
When
a curve of presumptive errors
of the ordinate for any given abscissa; so far then
means of the curve of

ness.
is
drawn, we can measure the magnitude
we know
the law of errors perfectly, by
but certainly in the tabular form only, with
errors,
Whether we can advance
further depends on, whether
we
all its
copious-
succeed in interpolating in
the table so found, and particularly on, whether we can, either from the table or direct from
the curve of errors, by measurement obtain a comparatively small
which
By
or by
we
number
of constants, by
to determine the special peculiarities of the curve.
interpolating, by
means
of Newton's formula, the logarithms of the frequencies,
drawing the curves of errors with
the logarithms
of the frequencies
as
ordinates,
succeed, as above mentioned, in giving the curve the form of a parabola of low
often
(and always even) degree.

Still easier is it to
make use
of the circumstance that fairly typical curves of errors
show a single maximum ordinate, and an

curve for a short
maximum
distance
is
inflexion
almost rectilinear.
point and of the points of inflexion,
we
on each side of
By measuring
it,
near
which the
the co-ordinates
of the
shall get data sufficient to enable us to
15
draw a curve of
errors which, as a rule, will deviate very little
from the original.
With
however, holds good only of the curves of presumptive errors.
cannot operate in this way, and the transition from the latter
meantime
to
depend on the
Laws
eye's sense of beauty.
of errors
be represented
may
is
expressing the
This method only
results.
such a way that the frequency of
in
stated as a mathematical function of the number, or numbers,
the results of repetitions
from that of curves of errors
differs
circumstance that the curve which represents the errors has been replaced by
tical
formula; the relationship
methods, to maintain a
is
so close that
is
it
difficult,
principal instrument.
ideally
errors
of
make computations with
of these two
of errors
the
is
mathematical speculation; we start from the properties
Its source is
the typical functional law
the
in
mathema-
between them.
strict distinction
which are considered essential in
its
when we speak
In former works on the theory of observations the functional law
to
we
the actual ones
former seems in the
tlie
FUNCTIONAL LAWS OF ERROES.
V.
17.
to
All this,
From
good observations.
deduced; and then
is
observations in
order
to
it
these the
remains
to
formula for
determine
how
obtain the most favourable or most
probable results.
Such investigations have been carried through with a high degree of refinement;
but
it
must be regretted that
The study
way the real state of things is constantly disregarded.

errors and the functional forms of laws of actual errors
in this
of the curves of actual
have consequently been too much neglected.
The representation
of functional laws of errors, whether laws of actual errors or laws
must
of presumptive errors founded on these,

of repetitions,
study the cases in which the arguments
constant
differences,
argument,
counter
is
is
and
the
considered as the functional
that
we cannot
for infinite
one
is
e.
this table.
the results
function,
law of
of the
which
errors.
We may
repetitions)
the
gives
Here the only
directly employ the usual Newtonian

is
here be content to
proceed by
frequency
difficulty
arguments must be
to interpolate,
rate, increase
0.
We
much
faster
the
we en-
an integral algebraic one, and gives infinite values
must
not the frequency

for
of
formula of interpolation,
arguments, whether positive or negative, whereas here the frequency
out to be inapplicable;
any
(i.
interpolated
as this supposes that the function
infinite
necessarily begin with a table of the results
and be founded on interpolation of
therefore
employ some
but
its reciprocal,
itself, y,
will often
become
infinite for finite
artifice,
of these
and an obvious
This, however, turns
arguments, and
than any integral function of low degree.
will,
at
16
we have
But, as
it
of the frequency,
to the logarithm
when we apply
already said, the interpolation generally succeeds,
Log y
assuming that
-\-
^ cx^
hx
-\-
gx"",
-\-
where the function on the right side begins with the lowest powers of the argument x,
and ends with an even power whose coefficient </ must be negative. Without this latter
condition the computed frequency,
would again become
0,
and
its
infinitely great for
logarithm
oo
like
these frequencies of the interpolation,
few of which
,?;
That the observed frequency is often

does no harm.
Of course we must leave out
-j- co
them by very small
or replace
may become necessary
select
to
As a
finite
rule
frequencies,
possible
to
succeed by this means. In order to represent a given law of actual errors in this way,
we
it
arbitrarily.
must, according to the rule of interpolation, determine the coefficients
number must be
we have
g,
whose
at least as large as that of the various results of repetitions with
which
deal.
to
Here
also
This determination, of course,
is
we may suppose that the law
of presumptive errors
of the actual errors.
pressed by a small
18.
of
And though
number
it
log y
can be ex-
this
supposition,
of x,
must, at any rate, be tried before any other.
these, the simplest case,
Among
simpler than that
is
course, does not imply that
of
this,
a, b, c,
a troublesome business.
of terms containing the lowest powers
so obvious that
is
nevertheless,
is
it
of the second degree
Log
2/
namely that in which Log y
a-\-bx
gives us the typical form for the functional
a function
is
cx^,
law of
errors,
and
for the curve of errors,
or
with other constants
he-^\)
/dO-""n-^)
=--
(2)
where
^
The function has

as unit
the
for
^+T + n2 + rf:3 + --- =
therefore no other constants than those which
frequencies
/(,
and as zero
and unit n
corresponding typical curve of errors has therefore in
The
2-71828.
all
functional form of the typical law of errors
for
the
be interpreted
may
observed
values;
the
essentials a fixed form.
has applications
in
mathematics
which are almost as important as those of the exponential, logarithmic, and trigonometrical
functions.
In the theory of observations
its
importance
been over-estimated by some writers, and though
is
many good
so
great
that,
it
has
observations show presumptive
as well as actual laws of errors that are not typical, yet every student
perfectly familiar with its properties.
though
must make himself
17
Expanding the index we get
so that the general function resolves itself into a product of three factors, the first of
is
which
constant, the second an ordinary exponential function, while the third remains a typical
functional law of errors.
recommended.
Long usage reduces
this
form to
but this
"';
In the majority of its purely mathematical applications
unless (as in the whole theory of observations) the factor
form cannot be
e-''
is
preferable,
in the index is to be preferred
on account of the resulting simplification of most of the derived formulae.
The
differential
Dê^lU)
e~^\n) with regard
coefficients of
-(,j;^
to
are
3n''a;)e~TW
(4)
Dê"
2^1
X>8g-
The law
numbers)
is
(^)
2x^-\-3- bn*x) e~~iu}

3x*
coefficients
+3
5w*
Sa;^
(products of odd
of D'e
2\n)
2Vn/ and
3 5n^) e'î^)'
numbers
and
binomial
can be got from a comparison
of the two identical series for equation
m)''
(3),
one being the Taylor
the two exponential series with
m^ and
as
can also be induced from the differential equation
nîy+i^
Inversely,
we obtain
xip
x'-ip
x^f
x\
x^<p
a;Y
(p
the numerical
oti^
n- 12 (aê -^ bn^
the other the product of e
arguments.
The general expression
obvious.
It
(x'^
numerical
of the
of the coefficients to
series,
= w- "
a;
for the
n^Df
='
/)+'
+ +
(
1)
D^<p
0.
products of the typical law of errors by powers of x
= n^D^f
= n^D^f 3n*D(p
= n^D*(p + Gw^-DY + 3wV
= m1Z)55p 10wZ)Y 15w*-^f
= n'^DY + \bn^''D*<p + '^bn^D'^f + 15nV
= 2Vn/
coefficients
-\-
n^tp
being the same as above
of these
(5)
strated by the identical equation n-^x''+^<p
By means
formulse
every
(4).
This proposition can be demon-
= D{xTip)-{-rx''-^f.
product of any integral rational function by

3
18
exponential functions and functional typical laws of errors can be reduced to the form
K<p
c^f-
^-^Df
where
<p
and thus they can
+ 11 i>V - ^ ^V +
easily be differentiated
IX
my
e~'i\n'l
(6)
and integrated.
Every quadrature of this form
can be reduced to
f\ (x) e
where fi(x) and

can be
solved
f.j{x)
vir)
-|-
f^ (x) J er iV^rrj dx,
are integral rational functions;
numerically by aid
functional law of errors,
jy
e"^'\
of the
thus a very large class of problems
following table
of the
together with the table of
its
typical
or
exponential
integral \rjdz.
19
Here
jy,
dH
'^\,
-r^
each of them, the same for positive and negative values
are,
of z\ the other columns of the table change signs with z.
The
means
interpolations are easily worked out by
of Taylor's theorem
= ^ + S-C + J3-C' + 5S-C- + Ag-C- + ...
..=,
(7)
and
The
shows that the frequency is

>, for which
typical form for the functional law of errors (2)
and that
always positive,
the frequency has
its
it
arranges
maximum
value
ij
itself
symmetrically about the value x
x^m^n the frequency
For
h.
curve of errors are the
is
z/
=A
The corresponding points

points
between the curve of errors and the axis of abscissae, reckoned from the middle to
in
the
will be
nh 0"85562
2-50663, only nh 0-39769
nh
and as the whole area
given in 11, as to the limit
creases
e~~i'^
is
toward
= 0,
if
only
falls
from
ct.)
all
which
is
a;
the other
of the inflexions,
either
between the great and small
In
z>b.
almost
all
the foundation
= m 4^
is
w,
nh l/2;r
consequently
of the
rule,
errors.
of the
applications
practical
theory of observations
Theoretically this superior assymptotical character of the function
expressed in the important theorem that, for
but also
one asymptote to
outside
The area
of inflexion.
shows how rapidly the function of the typical law of errors de-
table
zero.
it
(more exactly 16 per
not quite that sixth part
The above
of
0-60653.
2;
^j^ oo
not only e"i'^
itself
is
differential coefficients; -and that, furthermore, all products of this function
its
by every algebraic integral function and by every exponential function, and
the differential
all
quotients of these products, are equal to zero.
In consequence of this theorem, the integral \e

as
the
sum
of equidistant values
without any correction.
This
of e
If the
1
sum
into
for very large intervals the error
of one unit the
numbers taken out
curve of errors
is
1.
V2n can
be
computed
by the interval of the arguments

is
not
quite correct,
the
an integral being only semiconvergent in
can be easily stated, but as
of our table are not sufficient to
show
far
as
intervals
this error.
to give relative frequency directly, the total area
nhV27r; h consequently ought to be put
Problem
dz
simple method of computation
underlying series for conversion of a

this case;
2'' multiplied
2'^
must be
"^^
Prove that every product of typical laws of errors in the functional
1 /x m\2
he''~2\'~ir) ,with
form
How
do the constants
h,
same independent variable x, is itself a

m, and n change in such a multiplication?
the
typical law of errors.
20
Problem
the
mean
2,
How
small
are
Problem
3.
To
errors exceeding 2, 3, or
4 times
find the values of the definite integrals

Sr
Answer: sn+i
\x''e
2\
dx.
J 00
and
$2,
(2
1) 9r"+'\/27c.
Nearly related to the typical or exponential law of errors in functional form
are the binomial functions,
power
of
on the supposition of the typical law of errors?
error,
19.
the frequencies
which are known from the
of a binomial, regarded as a function of the
coefficients of the terms of the
number x
of the term.
"
21
any integral values of
w, r,
and
Mi)fin-t{r)
which means
we
consider
when
that,
the trinomial
+ b)-\- c)"
to be ((a
it
{a
(11)
Mr)-fin-r(t),
+ c)"
is
developed,
it
is
whether
indifferent
+ (6 + c))".
or (a
For fractional values of the argument
binomial function
x, the
j3{x)
can be taken
in an infinity of different ways, for instance by

.
sin;:a;
This formula results from a direct application of Lagrange's method of interpolation, and
more general formula
leads by (10) to the
^,,_
~
l'2., .n
'^"'^^
sin^ra;
.^,
^
~^tor'
(l-x){2-x)...{n^x)
'
This species of binomial function may be considered the simplest possible, and has
some importance
pure mathematics; but as an expression of frequencies of observed
in
values, or as a law of errors,
it
inadmissible because,
is
for
>n
or
it
negative,
gives
negative values alternating with positive values periodically.
is
may
however,
This,
when X
integral,
ô(^)
As
be remedied.
we can put
/9o {x)
and
has no other values than
1,
for instance
sin Tzx \^
'
TIX
by (10) then
sin^
nx
(13)
l)^
^.(^)
form
is
are ==
tial
[^.
(x\f
+
'
(x 2)V
?r^
Here the values of the binomial function are constantly positive or 0.

cumbersome and although for x
co the function and its principal
0, this
property
is
lost here,
when we multiply by
But
this
coefficients
integral algebraic or by exponen-
functions.
These unfavourable circumstances detract greatly from the merits of the binomial
functions as expressions for continuous laws of errors.
When, on
the contrary, the observations correspond only to integral values of the
argument, the original binomial functions are most valuable means for treating them.
Pn(x)
= 0,
if j;
>w
or negative, is then of great importance.
But
this case
must be
That
referred
to special investigations.
20.
To represent non-typical laws
of errors
the .choice between at least three different plans:
in
functional form
we have now
22
the formula (1) or
1)
(,a+IJX+yXi+...-XX^''
the products of integral algebraic functions by a typical function or (6)
2)
l/I m\'
I-
y
3)
sum
of several typical functions
y
This account of the more prominent
among
the functional forms,
disposal for the representation of laws of errors,
we can even
instruments, by means of which
adapted
for the representation of
laws
required in theoretical speculations
of
prove that
may
in
As
to the
errors.
We
If
is
our
certainly possess good
do not want
upon laws of errors; nor
at
find
general
series
forms for the series,
the exact representation of
anything,
we have
too
many forms
their value correctly.
important transition from laws of actual errors to those of presumptive

form of the law leaves us quite uncertain.
errors, the functional
series is too irregular,
We
which we have
we
more than one form
the actual frequencies more than reasonably difficult.
and too few means of estimating
(14)
l^lhê-^ri^l
and cannot in the
The convergency
of the
least be foreseen.
ask in vain for a fixed rule, by which
we can
select the
most important and
trustworthy forms with limited numbers of constants, to be used in predictions. And even
if we should have decided to use only the typical form by the laws of presumptive errors,
method by which we can compute its constants. The answer, that the
"adjustment" of the law of errors must be made by the "method of least squares", may
not be given till we have attained a satisfactory proof of that method; and the attempts
we
still
lack a
that have been

I
think,
all
made
to
deduce
it
by speculations on the functional laws of errors must,
be regarded as failures.
VI.
LAWS OF ERRORS
EXPEESSED BY SYMMETRICAL FUNCTIONS.

21.
curve of errors
All
in a functional
constants
or, generally,
of the several results
of a law of numerical
of the repetitions,
i.
e.
errors,
errors,
functions
every general
property of a
must be symmetrical functions
which are not altered by inter-
all the values found by the repetitions

no
the same essential circumstances,
interchanging whatever can have any
changing two or more of the

correspond to
law of
influence on the law of errors.
results.
For,
as
of the
Conversely, any symmetrical function of the values
23
And we must
observations will represent some property or other of the law of errors.

able
to
whole law of errors
the
express
itself
by
every
such collection
of
be
symmetrical
functions, by which every property of the law of errors can be expressed as unambiguously
as by the very values found by the repetitions.
We
have such a collection in the coefficients of that equation of the " degree,
whose roots are the n observed

equation,
For
values.
we know these
if
we get an unambiguous determination
titions, i.e. the
law of
of
But other collections
errors.
coefficients,
them has each of the degrees
1,
. .
same requirements; the

integral, and that one
also fulfil the
and
essential thing is that the n symmetrical functions are rational
of
and solve the
the values resulting from the repe-
all
and that none of them can be deduced
from
the others.
The
With
collection of this sort
that
is
Oj,
0.^,
easiest to
is
compute,
sums of
the
the powers.
the observed values
O3, ... o
we have
So
s>
+ ..^Yol =n
o'^,+ol
o^-\-o,^..ô
<-\-o\+
s,
0";
+ 0^ +
<>
-^
(15)
oi;
and the fractions
may
also be
only important to reduce the

value oi 0^
,o; for
if
employed as an expression
observations to
the diiferences between
suitable
the
for
the law
of errors;
it
is
which must be an average
zero
observations
are
small,
as
compared
with their differences from the average, then
may become
practically identical,
and therefore unable
to express
more than one property
of the law of errors.
From
a well
1
known theorem
+ UiW +
ai(u'^
which are identical with regard

Sr
of the theory of symmetrical functions, the equations
=
^
(1
Oi<y) (1
whose roots are the n observations; and
(1
OnW)
^'^
g^lOg(l O^W)
to every value of
can be computed without ambiguity,
<î<")
if
10,
we know
vice
versa,
we
the
by
learn that the

coefficients
sum
Ur
differentiating
of the powers
of the
the
/-^^"
last
equation,
equation
f^,
^1
24
with regard to m, and equating the coefficients we get
tti
2a2
S]
a^s^
-\-
s..
(16)
na
-[-
a_i s,
-^"-t
'<'>
from which the coefficients are unambiguously and very easily computed, when the
are directly calculated.

22.
able collection
But from the sums of powers we can easily compute also another serviceof symmetrical functions, which for brevity we shall call the half-invariants.
these can be defined as
Starting from the suras of powers Sr,
Ûj, yug, //a,
by the
equation
M,
+
^^
f-l _2 _1_ f-i -.3
s,e\^
^0
so^f
power of
T,
we
7'
-t-
the
get each
first
S3
term of (17) as
Taking the logarithms of
e'''^
e''^^
Shrt and
expressed as a function of
So(A:i
i-
12
|2
13
13
(17)
r.
+ t^' + t^' + --
'
^'
II
jl
which we suppose identical with regard to

As Sr == IV, this can be written
By developing
+ ...6'"'^
(18)
equating the coefficients of each

...fir-
//j
(19)
+%!/'.+/'?)
(17)
we get
... =
f.+|.'+|^ +
i.g(.
s,
i^jr
So
So r*
+ i^V+i;i + ->
,
(20)
and hence
s.
(s,s
s^):sj
(S3sj-3s,s..9
(s, sj
The general law

through the equations
- 4s,
of the
s, sj
relation
(21)
2s;):.9j
3s', si
+ Us,
between the
fi
s] s
and
- 6s;)
s is
sj
more
easily
understood
25
S4
where the numerical
(22)
/^l
+ %2
'^2
+ 3/^3
+/.4
''l
coefficients, are those of the binomial theorem.
be demonstrated by differentiation of (17) with regard to
s,+fr
|r^+|r3+...
being satisfied for
all
''O
+ |r'^ +
+
(/.,+|^-r+|r^
equations
can
the resulting equation
r,
...)(.s-+^r
...)(23)
values of r by (22).
These half-invariants possess several remarkable properties.
SqC
affects only
îj
==e
li
}?.
consequently any transformation

0, ...o,
Tliese
in
the
From
(18)
we
get
+...-{- e
(24)
0'
-|- c,
any change of the
same manner, but leaves fi^, /ig,
zero of all observations

... unaltered;
fi,,
any
change of the unit of all observations can be compensated by the reciprocal change of the
unit of
r,
and becomes therefore
Not only the
indifferent to /igr", /jIsT^,
^n
"0
'0
but also the half-invariants

of remaining
We
ratios
'"n
have the property which
is
so important
in
a law
of errors,
unchanged when the whole series of repetitions is repeated unchanged.

have seen that the typical character of a law of errors reveals itself in the
elegant functional form
Now we
shall see that it is fully as easy to recognize the typical laws of errors
of their half-invariants.
^2
n'^.
Here the criterion
is
that
/^,.
=0
This remarkable proposition has originally led
me
to every other system of symmetrical functions; it is easily

if
m for the
We begin by
we take
while
/x^
=m
and
to prefer the half-invariants
demonstrated by means of
(5),
zero of the observations.
forming the sums of powers
quency of an observed x
we
r>3,
if
by means
is
proportional to
(p
(x)
.Sr
of that
e~TV^^/
law of errors where the

;
as this
law
is
get
(+<
iX^(p{x)dx.
fre-
continuous
26
For every
"
(.+
\D'"(p(x)-dx
J
Z>'-V(cc)
(5)
=
=
(22)
Now
3, 18).
we remark
that S2r
that s^r+i
s^
(compare problem
Z)'-i^(
c)
0,
00
consequently we learn from
we use
we have
differential coefficient D"'(p(x)
0,
but
l-3-n*Sg
1
3 -5
W^Sn
the half-invariants can be found by (22) or by (17).
1) 82,-2;
(2r
w''^
= fîSo
=
Si
fi^So fiiSi
2^2S, =//,,s,^+ /ijSo
S3
%2S2 = /^l3 + S/igS, + /i,Su
4
= /i,s, + 6/i3S2 + 4//4S1 +

Sj
4^^283
5//2S4 =
+ 10/Za3 + lO/iiSa + 5/i5Si +/iSo
Sg
= 0.
^n^ and
the solution
==^3 =^^ =
Sj
/is.s
/î>^5
we
see that
By
is
(17)
we
fi^
îj
=0
=0
=0
=
==
get
Equating the
if
If
then writing for (22)
coefficients of
z''
we
get here also
/ij
== w,
ii^^=
n'^,
//,
> 3.
If
we wish
to demonstrate this important proposition without
change of the
zero,
is somewhat diffi(3) whose general demonstration

we can commence by the lemma that, for each integral and positive value of r, and
for r ^^ 0, we have for the typical law of errors
and without the use of the equations

cult,
also
Sr-,-1
The function
if
(p(x)
==
n'^
we now between these
xr er 'iK"^)
ax
get
is
WAV
+ rw^
Sr_i.
equal to zero both for
x =<x. and
limits integrate its differential equation
ijH^
we
{rnâf-'{x
\
^
in)x'')e-^hr),
/
= Sr+i + msr + r n^ Sri_,
for
00;
27
where
x''e
If
we now from
21
dx.
00
(22) subtract, term by term, the equations

"1
28
we
get,
-S'
and as
/ii
=^ -^
/,)'
The computation
(21),
because s^
the
in
of
and of
As
/i,
y^^^j
/'2-
T^
will
*//!
often be easier than
must frequently be computed with more
latter
is
according
by the equation
figures.
There
often to be preferred to either of these methods of
have no influence at
to (24),
as zero a convenient, round number,
c,
very near
^/,,
all
on
is
com-
same increase of every
a change in the zero of the ohservationts involves the

it will,
[i^,
by this formula
//.,
however a middle course, which

putation.
2s,
^._,.
We
and by reference
select therefore
to
this
zero
the
observed values are transformed to
0\
When
s\
we have
and
u,
s\
'J,
sums
indicate the
Oj
0,
still
... 0
OnC.
and
//',
=//i
c,
then
and
'^
2'(o
have
f,
of the transformed observations,
'^
We
to
nJ
(26)
c)'^
(l^-cY-
mention a theorem concerning the mean deviation, which, though
not useful for computation,
is
useful for the comprehension and further development of the
The square of the mean deviation ^2 i^ equal to the sum of squares of the difference
between each observed value and each of the others, divided by twice the square of the
idea:
number. The said squares are:
''
,
(0,
Ol)^
OJS
(0,
o)2,
(Ol
(O.^
0,)-,
(O5;
0.^2^
Oi)2,
02)^
(0
(02
o)2,
(o
developing each of these by the formula
column
separately,
we
find the
sums
(0,,,
(0
On)''
o)^ = Om 2omOB +
and
first
adding each
29
2s, +
2s,o, +
sy,
sy,
o^
and the sum of these
o,
2s,o +
2s,s, + s,s^ =
ss,
a,
s.
s,
{s^s,
s\),
consequently,
l'l\or-o,r
The mean deviation
greater than the least,
is
the values of repetitions from the
(27)
^2slfi.,.
less
than the greatest of the deviations of
mean number, and
less
than Vh of the greatest deviation
between two observed values.
As
to the higher half-invariants it
may
here be enough to state that they indicate
various sorts of deviations from the typical form.

the //2r+i
being
from
different
zero,
positive or negative values of //4r,
peaked
or
Skew curves
flattened
of errors are indicated by
(divided)
forms respectively by
and inversely by /ur+2-
But we have
For these higher half-invariants we shall propose no special names.

already
introduced double names "relative frequency" and "probability" in order to accen-
tuate the distinction between, the
and the same we ought
laws of actual errors and those of presumptive errors,
to do for the half-invariants.
In
what follows we
the half-invariants in laAvs of presumptive errors by the signs ir instead of
be reserved
from
laws
for
of
laws of actual errors, particularly when we shall

actual
explained later on,
errors
the
name mean
î, for actual as well as for
"mean
error",
those
to
of laws of presumptive errors.
mean
infinite
to
For special reasons,
to
//,
be
and
Thus,
if
= s
Lim=,(//.J
error.
In speculations upon ideal laws of errors,
tinuous or to relate
will
transition
presumptive means; but instead of "mean deviation" we say
when we speak
called the square of the
which
/tr
of the
value can be used without confusion both for
X.^
is
of presumptive ones.
treat
shall indicate
when the laws
numbers of observations,
this
are supposed to be
distinction
is
of
con-
course
insignificant.
Examples:
1.
Professor Jul.
Thomson found
for the constant of a calorimeter, in
with pure water, in seven repetitions, the values

2649, 2647, 2645, 2653, 2653, 2646, 2649.
If
we
take here 2650 as zero,
-],
we read the observations

-.3,
as
-5, +3, +3, -4,
-1
experiments
30
so that
s'
7,
consequently
fi^
s',
= 8,
2650
70
The mean
deviation
is
"no", which counts 0.
What
then
is
Answer: u,
have voted
^'
^'
for a
what values
Out of
VI
,
motion
.,
^
^
~\-
either "yes",
is
the
repetitions
in
+ b-\-e
{(a -f c) (a
c against
^'^
ab
inn (n
-r~~
1),
m)
.-c^
'
two of the numbers
""
a) {ab
rr-^
which a voters
(0),
and examine
6-)
-\-
Sea -{-be
c)
(2a
b'')
'
(^+ Ff^P
c)2
+ 6 + 4(a cY) (a+ 6 +
when one
b + 2c) -{-6{a
n}*
is
unanimous, the double condition fji^=fi^

is for, another sixth against, while two
sixth of the votes

If
to
is
fi
.j
yix
a, b,
and
c,
be == 0,
and consequently
without a being ==
^, //g
=
^=/4 =;U5
0,
but
are
/i.^,
e,
/iu
1,
0.
What
0, 0, 0, 0,
b-
b (a-\-
e)
which does not disappear unless

0.
symmetrical and almost typical
repetitions give the quite
Answer:
VIL
4tnn-\-n^)
tnn(m^
Ua =^
''^^
fi.i
n "no".
the
approximation to the typical form.
=^Q,
"yes",
while b have not voted
+ ica + be
(a-f 6
thirds do not give their votes.
//j
have given
or
1,
{a^bêy
only satisfied
4.
which counts
errors, in half-invariants, of a voting in
c give the nearest
c
'
..
Disregarding the case when the vote

is
3.
mn
==
(-|- ]),
and
for a, b,
2649.
8\2
70;
the expression for the law of errors in half-invariants ?
Determine the law of
3.
consequently
1 =
s',
an alternative experiment the result
In
2.
and
hiAv
of
errors,
are the observed values?
+1.
RELATIONS BETWEEN FUNCTIONAL LAWS OF ERROES

AND HALF-INVARIANTS.
24.
The
multiplicity of forms of the laws of errors
makes
it
impossible to write
For though these forms are of very different

none
of
them
can be considered as absolutely superior to the others. The functional
value,
form which has been universally employed hitherto, and by the most prominent writers, has
a Theory of Observations in a short manner.
in
my
opinion proved insufficient.
I shall
here endeavour to replace
it
by the half-invariants.
31
But even
if
should succeed in this endeavour,
am
sure
that not only the functional
laws of errors, but even the curves of errors and the tables of frequency are too important
and natural
put completely aside without detriment.
to be
Moreover, in proposing a new plan for this theory,
have
as precisely and completely as possible its relation to the old and

I
therefore consider
it
felt it
my
duty to explain
commonly known methods.
a matter of great importance that even the half-invariants, in their
very definition, present a natural transition
to
the frequencies
e' +...
and
to the functional
law
of errors.
If in the equation (18)
ne\l
some
of the
o,'s
\l
are exactly repeated,
be counted not once but as often as
it
o,
is
of course
is
repeated.
+ "
understood that the term
e"''^
must
Consequently, this definition of the
half-invariants may, without any change of sense, be written
I^(Oi)-e~\i''^^'^ "^F""
"*"
i:^
(0,-)
e".-^
(28)
where the frequencies y{oi) are given in the form of the functional law of
continuous laws of errors the definition must be written
eii
we know
if
Thus,
half-invariants
^12
+18
.U(o)eio
the functional law of errors and
may
be found.
If,
Example
1.
"
(.+
\
(f
(o)
do
V/2ff
then
it
the integrations, the
may
be possible also
(o).
*/ ''~"''Y
InJu and
\f{o)e''^do
=
=
and consequently
we can perform
X,-,
zhAe
Ehe'"*'
2ii,^
+ -r'\e--A
do
n,
For
(29)
Let ^(o) be a sum of typical functional laws of errors,
_
then
U(o)e'-do,
we know the
inversely,
to determine the functional law of errors
if
errors.
do,
32
aid of the formulse (19) that express
By
difficult
mi,
iii,
tions
as functions of the ^
compute the principal half-invariants. The inverse problem,

and hi by means of given half-invariants is very difficult, as it
to
high degree, even
of a
sum
only a
if
(or
fi.)
it is
not
compute the
to
results in equa-
two typical functional laws of errors
of
is
in question.
2.
Example
What
are the half-invariants of a pure binomial law of errors?
we
observation r being repeated /9(r) times,
2et^+t^'^'t^'
+-
^(0)
The
write
+ /5(l)e^ + ..- + /?()e"'" =
(l+^T.
consequently
(/î-|-)r+^^r^+^|r='+.--
= '^log cos
Here the right hand side of the equation can be developed by the aid of Bernoullian
numbers into a
series containing only the even
Hi
and
"2
^/jr+i
powers of r, consequently
(r>0)
0,
further
n
A'a
^"4
Example
'
^^4
3.
this
^"
What
(the complete terms of (p
From
we obtain by
are
'
/ê
= ^n
>
/^8
17
Tg"
'
</)")?
Here
differentiation with regard to r
^ nqe-^
we
^^
31
^ ^'
the half-invariants of a complete binomial law of errors
by further differentiation
putting r
/^lo
get
np
^2
npq
{p+if
npq(pq)
(p+qf
33
^'
lp+W\\p+ql
npq
4.
Example
but this
is
b''
==
is
ip
+ qr
ba''.
is given by its half-invariants

forming
Determine the several observations and their frequencies.
side of the equation (18) is
.Spe-'M +6e'"''
V-
-iê^'"''
This law of errors
fp(r).
j7-
Thus the observed values
side of (18).
ra
hand
left
8py( j>-g )\
((p-iV
-Wp+q)
law of presumptive errors
a geometrical progression, ^r
Here the
IP+W'
b^
e^'"'
-|3
+ ...l
and has
and the
are 0, a, 2a, 3a, ...

is
also the form of the right

relative frequency of
nearly related to the binomial law, which can be
considered as a product of two factors of this kind,
rjnr
^r
T--r
\nr
\r
It
is
perhaps
the
to
superior
^BnivWd"--.
'^^
|m
'
law
binomial
as
representative
of
some skew laws
of errors.
Example
order
that
are
5.
while
0,
law of errors has the peculiarity that

all
even
half-invariants
are
equal to each other,
.(0)--(i + (f)+(|)' +
Example
irrational
values
6.
Xir
of
odd
Show
2a.
the table of a function,
/2r+l
25.
As
0,
/,,
replaced by
r2'
"4
in
whose computation fractions under
for the
J
have
T2U' ^8
5'5'2'
most general functional form of a continuous law of errors we have
(6)
(p(x)
-)
Determine the half-invariants of the law of presumptive errors
in
been rejected and those over
where
half-invariants
the observations must be integral numbers, and that for the relative frequencies
all
proposed
all
IX m\^
= e~~i\^l
34
Now
is
it,
any ambiguity as functions of the coefficients
By
(29)
we can
a very remarkable thing that
we
and vice
versa.
get
+
se[L'
fc,,
express the half-invariants without
''
'^'
Y^(k,<p(o)-^^^D,p{o)
+ ^-^D^f{o)...)do,
|2
= nk^]/2i:. By means of the lemma

= e^ ((/)'-' ^(o) rD'-^<p(o) + + { ry-^
D<-(p{o)do
where SQ
..
'ê"''
which
<p
{o))
any f{o) by differentiating with

have in this particular case, where <p (o) and every D' <p (o) is
0,
is
demonstrated
easily
for
C"^"
l/o-m\2
Y^ D' e-2\-ir) do
1/0 !\2
C"*""
{ zYY
*
00
-^[-ir]
do
(- rfw
-^
(xY
regard
if
\e'''
to
o = i
oc
(f[o)do,
we
only,
,,1',^
l/2;re'"^+T^.
00
Consequently, the relation between the half-invariants on one side and the coefficients k
of the general functional law of errors on the other, is
A;oeTl'+^''-'!l"'+-=(A:+|-r
If
we
write here
>(',
==>!,
and
}!
.,
= k^ m^,
constants by the other can, according to (17), be

substitute only in these the
It will
to
A;,-
for the
s,-,
be seen that the constants
and
)!
and
which they belong, are generally superfluous.
may
+ |r^+|r...)- +
),
for
It is easiest
In this case
we
get
to
put
m =
X^
and
A-,
=0,
k^
k,
k,
k.
The law
=0,
k.^
n"^
=kf^X^,
We
errors
This superfluity in our transformation
and the constants must be
and most natural
of
/i.
and the special typical law of
m,
set
the formulae (19) and (21).
be useful in special cases for reasons of convergency, but in general
sidered a source of vagueness,
(30)
the computation of one
made by
or
=
k^
it
must be con-
fixed arbitrarily.
X.^.
^kX^,
k. =- kX^,
and further
^ kAK +
= k,{X, + 3bX,X,)
= kAX, + b6X,X, + 3bX])
'ÔXl)
of the coefficients is explained by writing the right side of equation (30)
35
by half-invariants in this manner the explicit I'orm
Expressed
of
equation
(6)
is
+
+
l+^((^-^)'-3^(x-/,))
1
(31)
>i.
LAWS OF ERROES OF FUNCTIONS OF OBSERVATIONS.
VIII.
There
26.
relating to
have the
to
nothing inconsistent with our definitions in speaking of laws of errors
is
any group of quantities which, though not obtained by repeated observations,
like property,
namely, that repeated estimations of a single thing give
errors of one kind or other,
expressions
for
such
requires, besides this,
stances,
prima
The various forms
valid.
facie equally
owing
multiple and slightly differing results which are
to
rise,
of laws of actual errors are indeed only
and the transition
multiplicity;
only that the multiplicity
and that the values must be mutually independent
to
It
all.
is,
Compare
24,
consequently,
not
Example
others
unknown circum-
in that sense that

in
summary
law of presumptive errors
the
caused by fixed but
is
circumstances have connected some repetitions to
common
to
a manner
none of the
which cannot be
6.
define the law of errors for a function
difficult to
single observation.
Provided only that the function
observed values Oj,
o.^
is
univocal,
we can from each
... o determine the corresponding value of the function,
of one
of the
and
f{o^), f{0,), ...f(On)

will then be the series of repetitions
in
law of errors of the function, and can be
the
treated quite like observations.
With
Oi
are different,
we can have
be added together.
those forms of laws
to
respect, however,
idea of frequency (probability)
we must make one
=f
/"(o,)
and
(Oi),
little
in this
Here, however, we need only
of errors
which make use of the
Even though o,- and

case the frequencies must evidently
reservation.
just mention this,
and remark that the
laws of errors when expressed by half-invariants or other symmetrical functions are not
influenced by
it.
Otherwise the frequency

probability.
The ordinates
discontinuous values;
in the functional
but
law of
the same
is
of the
curves
the abscissa
errors.
In
o,-
for
/"(o,)
as
for
o,-,
and therefore
also
the
of errors are not changed by observations with

is
replaced by
/"(o,),
and likewise the argument
continuous functions, on the other hand,
areas between corresponding ordinates which
must remain unchanged.

5*
it
is
the
36
In the form of symmetrical functions the law of errors of functions of observations
be computed, and not only
may
fore
compute,
for
symmetrical functions of the
know
if
the several observed values, and can there-
all
many and important
In
latter.
of the powers
s'
The
s'g
= s^
Sy, s',
functions by half-invariants.
It is
sums
sufficient
if
the
we
if
f{o)
= o^
for
then the
of the powers s^ of the observations,
= s^,
s'
is
it
at last
etc.
here a proposition as to laws of errors of the linear
is
principal thing
cases
For instance,
from these.
directly
of the squares are also
m = 2m
only constantly
27.
and
we can compute the symmetrical
as
the symmetrical functions of the observations,
functions of the functions
sums
when we know
each of them, the corresponding value of the function,
almost self-evident that
if
/ii
o'
ao-\-h
a//
(32)
etc.
;j.'r
a>r (>!)
For the linear functions can always be considered as produced by the change of
both zero and unity of the observations (Compare
However
to
(24)).
special the linear function ao -\-b
get on with the formula
circumstance, the very same
may
be,
That we can succeed
(32).
we always
this
in
in practice
is
owing
to
manage
a
happy
as, in numerical solutions of the problems of exact mathematics,
about that we are but rarely, in the neighbourhood of equal roots, compelled to
employ the formulae for the solution of other equations than those of the first degree.
brings
it
Here we are favoured by the

to be small, so small
for
to
fact
that
we may suppose
speak more exactly
the
that
errors
we may
each series of observations Oj, Oj, ... o assign a number
c,
in
good observations
generally in repetitions
so
near
them
all
that
the squares and products and higher powers of the differences

0,
without any perceptible error

i. e.,
may
0,0^
be
left
0,
out of consideration in computing the function:
these differences are treated like differentials.
The
differential calculus gives a definite
method, in such circumstances, for transforming any function f(o) into a linear one
m=
The law
of errors then
(^
f(c)
+ f'{c)-(o-c).
becomes
ifio))
f(c]
+r
(C) {[I
(0)
c)
/-(/i
(0))
37
But
functions
also by quite
elementary means and easy
others of linear form.
into
J_
c-{-(o
and the law of errors
is
If
_
c)

io
28.
With
that
we should
statements which severally
have,
may
for
It is
still
speak of
we must
we may also,
closely what we
quantities
define
more
each of the observed quantities
in,
which was out
and
law of errors
o',
a series of
be looked upon as repetitions:
Om
Oj,
Now
this is not sufficient.

o',
more observed
o'.
it
makes a
difference
there are or are not such as are
if,
common
among
the special
to observations of the
We
want a technical expression for this. Here it is not appropriate only

observations which are, respectively, dependent on one another or independent;
different series.
we
'
necessary for the idea of the
o'n o\,
to
X
''-
For then another consideration comes
0,,
But here
--îfÂo)-o)
of laws of errors, only
speak
circumstances by o and
c'
^-M-
(I)
of the question in the simpler case.

/'(o, o')
c)-'
respect to functions of two or
are to understand by repetitions.
of
we may often transform

then we write
then
in case of repetitions,
_ J_ _ J_ _
c-jo -c)
c^'
artifices
instance f{o]
for
are led to mistake the partial dependence of observations for the functional dependence
of exact
shall
quantities.
repetitions of different
propose
to
designate
observations by the word
these
interdependences
particular
of
"bond", which presumably cannot cause
any misunderstanding.
Among
the repetitions of a single observation, no other bonds
such as equally bind

of the method.
fair
all
the repetitions together, and consequently belong to the pecularities
But while,
repetitions of
must be found than
for instance, several
pieces
cast
in
the
same mould may be
one another, and likewise one dimension measured once on each piece,
two or more dimensions measured on the same piece must generally be supposed to be
bound together. And thus there may easily exist bonds which, by community in a circumstance, as here the particularities in the several castings, bind some or all the repetitions of a series each to its repetition of another observation; and if observations thus
connected are to enter into the same calculation, we must generally take these bonds into
account. This, as a rule, can only be done by proposing a theory or hypothesis as to the
38
mathematical dependence
between
the
the
treatment
right
under
falls
and
observed objects
and whether the number which expresses
this is
their
common
known from observation
methods of adjustment which
those
circumstance,
or quite
unknown,
be mentioned
will
later on.
then in a few special cases only that we can determine laws of errors for
It is
functions of two or more observed quantities, in ways analogous

single observation and
If
the
f{o, o\ o",
its
observations
o,
are repeated
.),
any
Oi
Oi,
f{Oi,
for
other
.
To do
separately.
29.
observations
f{o, o\ o",
etc.,
and
If,
o,
the others
o',
common
and
compute
this a
to
of a
o,,
o't,
each
for
value
calculation
...
o",
no bond at
is
may
it
number
and
will
function y,
same way
as
between the repetitions of the
all
which we must
the equally
all
valid
to
try to reduce
values
for
o',
oi
we
shall in this case
for o,
o',
for o",
com-
o" ...
for
determine the law of errors for the
be able to compute
might be laid down;
sufficient
the simple repetitions of one
here easily become too great a task to
of propositions
be almost
of
j'th
but otherwise
i,
of the
of the
at all of the special nature of the bonds.
this is the principal case to
But while
quantity.
which teaches us
but one of them
us in what follows,
sum
viz.,
that
of the observed quantities
o'.
If the
f{o)
same
the
the
into
of errors can be determined for this, in just the
each of the numerous combinations,
of special importance
and
enter
to
herein combine every observed value for o with every one
.),
Concerning
is
each
law of errors by means of the laws of errors
y's
are
then we must, in order to represent
single observed
pute y
for
we need no knowledge
so
...
which
.,
circumstance,
on the contrary, there

o",
such values of y must be treated analogously
all
for
bonds,
such a way that, in general,
we can
and laws
.),
o"
o',
in
repetition are connected by a
without
what holds good
to
functions.
and
for
law of errors
<p(o')
for
o',
is
in
given
then
it
is
the
form of relative frequencies or probabilities,
obvious that the product ^(o)^(o') must be the fre-
quency of the special sum. o-j-o'.

In the calculus of probabilities
some cases
of
bound observations
we
shall consider this form
will find their solution
to the treatment of the said case with half-invariants.

If
occurs with the observed values

0,, 0,,
and
o'
Om
with
o',,
then by the
mn
o',,
repetitions of the operation
o;,,
0=ô-{-o' we
here
we
more
closely,
and there
shall confine ourselves
39
0,
"',.
+ o;,
0,
0m-\-0\, 0m-\-0\,
Indicating by
where
Mr
the half-invariants of the
+ o;,
o,
and
fi'^
7+
T2
of
o',
we
e!i
0^.
= o-|-o',
sum
and w are the numbers of repetitions of o and
the half-invariants of
and
0,
o'.
we
get by (18)
Consequently,
if
nr represent
get
ell
eli
Jl
li
finally
(34)
-Mr ==
/ir
H'r
Employing the equation (17) instead of (18) we can also obtain fairly simple
expressions for the sums of powers of [o-ô') analogous to the binomial formula. But the
extreme simplicity of (34) renders the half-invariants unrivalled as the most suitable symmetrical functions and the most powerful instrument of the theory of observations.
More
generally, for every linear function of observations not connected by
any bond,
= a + bo + CO' -{-...do'",
we
obtain in the same
manner and by
M,{o)
(32)
= a+
MAo)^
b;t,
b';,,
+ cy,^... + dy:'
c//,
d^';'
(35)
ea.
MAo)
+^>:
r>l.
When
we
the errors of observation are sufficiently small,
be able to give the most different
functions
a linear form.
In
shall also here
consequence of
generally
this,
the
propositions (34) and (35) acquire an almost universal importance, and afford nearly the
whole necessary foundation for the theory of the laws of errors of functions.
Example
1.
Determine the square of the mean
order of equidistant tabular values,
mean
error for every value being
between which there

l^.
error
is
for
differences
of the
n^"^
no bond, the square of the
40
Example
viz. the time,
t,
AîJ*)
>?2(o, 403+602 4o,+o)
;rj.)
/î^;
1.1.10.14
2
"*
the observation of a meridional transit
2.
By
when a
But
as
it
t -\-
p {p
when the
f cosec p.
The mean
Example
in
= polar
sin
tity, the time
distance),
very meridian
quantities,
from the
f,
be assumed that the time and the distance are
may
not connected by a bond, and as the speed of the star
known value
we observe two
behind a thread, and the distance,
star is covered
meridian at that instant.
70^2
4n-2
we always
constant and proportional to the
is
state the observation by the
one quan-
error is
3.
making marks on
scale is constructed by
such a way that the square of the mean error on each interval
To measure the distance between two
objects,
it
is
we determine
at regular
=
the
intervals,
/j-
distance
of each
object from the nearest mark, the square of the mean error of this observation being
How
is
great
the marks
mean
the
error in a
measurement, by
we use?
X, (length)
Example
good
(^2
1)
4.
Two
proportion to the distance,
how
is
the
Ki^)
5.
>?,-!- 2/;
mean
a',.
The
their rectangular co-ordinates.
great
which there are n intervals between
points are supposed to be determined by bond-free and
measurements of
Example
passed, which we compute by the formula
is
equally
errors being small
error in the distance
in
J?
2.
Under the same suppositions, what
is
the
mean
error in the inclina-
tion to the a;-axis?
Example
6.
Having three points
in a plane
determined in the same manner by
their rectangular co-ordinates (a;,,yj, (xj,?/,), {x^,^/^), find the

at the point (d7,,yj)
Ân
Ji, J^,
J;i
J^
mean
+ Jl + J]
--.
ji J,
being the sides of the triangle; Jj opposite to (^,,2/i)-
error
of the
angle
41
7
Examples
trianglff
and
\ (d\
to transform
pensable
2/^
of
resulting from
combination
the
where a or b ox d may be
difficulties.
computing the halfTo do so it seems indis-
of those of the arguments.
of all
sum
Sr indicates the
both
o'^
is
of the
its
and
oC")'',
sum
the
and
it
repetitions
being
0), s^P
Sr
Of
+ A'\.
J'
(
for
laws of errors into the form of systems of suras of powers.
the
of terms of the standard form Eko'^
if
method
possible to indicate a
it is
be integral and rational,
/"(o, o', ...0"")
Thus
k, (quadrangle)
Al);
no general expression for the law of errors can be found.
means
invariants of the function by
oW.
for integral rational functions
Nevertheless, even in this case
sums
of the areas
Non-linear functions of more than one argument present very great
30.
determinations
and a plane quadrangle.

X, (triangle)
Even
Find the mean errors in
8.
ksa
powers 0'' can be written as
.s'j
s^P
(including
the
1.
the
mean
M,
already takes the
value
SkSa
Jlf,
=n^n\
s't
. .
sP.
and generally Sr
s^s'^,
consequently
cumbersome form
Example 2. f]xpress
mean deviation
observed without bonds.
exactly
by the half-invariants of the co-ordinates the mean
of the square of the distance r^
=x''-\-y^,
if
x and y
are
Here
ôir'')
So(^)o(/)
^4 i^) o (y)
= ss\
oo'
and
value and the
(''
Here S^
of
powers of the function 0, we get
/'>
Determine the mean value and mean deviation of the product
of two observations without bonds.
cases
powers of the repetitions
'>
course, this operation is only practicable in the very simplest cases.
Example
sum
every such term the
for
of all
If
+ ^s^
{x) s^ (y)
+ So {x) s,
[y)
and
//2
31.
in
/U
W+
fi, (y)
+ 2 {x)y +
+ îMAy)lÂy) + 2(/., (ii)r +
4//3 i^)fî (^)
(/i,
The most important application
(o,
itself.
+02+...0)
4/., (y) {fz, (y)Y.
of proposition
mination of the law of errors of the mean value

/'I
4fi, (x) ((,,{x))^
(35)
is
certainly the deter-
The mean value
known
half-invariants
>ij,
k^,
..
43
repeating observations and parts of observations,
so
that you can directly form
and
values which should be substituted for the observed results;

cially in the case of observations of a novel character,
this
to be
is
or with peculiarities
the
mean
done espe-
which lead us
doubt whether the law of errors will be typical.
to
This remarkable property
though with
the
less certainty,
coefficient of
of errors
the laws of errors of
if
for
any
but also,
only,
provided only
not so great relatively to the corresponding deviation
is
throws
it
however, not to the mean
peculiar,
any linear function of several observations,
to
any single term
from the typical form that

seen that,
is
the other terms into the shade.
all
the observations
all
o,
o',
...
of their linear functions will be typical too.
if
it
(35)
are typical,
o<'")
And
From
is
the law
the laws of errors
are not typical, then that of the linear function will deviate relatively less than any of the
observations
o,
o',
... Om-
To avoid unnecessary complication we represent two terms of the linear function

simply by o and o'. The deviation from the typical zero, which appears in the r*-^ halfinvariants (>
2), measured by the corresponding power of the mean error, will be less
>
= o-{-o'
for
than for the most discrepant of the terms o .and
The inequation
Ar
says only that,

it
is
if
Ar
and
the laws of errors for
the law of errors for
0'
more
where
briefly
is
positive, r
When we
r*"
it is
evident that
B\
introduce a positive quantity
>
(U -\- ly
(5
{U-\-ir.
Remembering
that
(c/r-f
1)'',
f/'
and
^
a;
-f
U-yf
j;-
>
+
if
2,
U-\-U-'
Consequently
(T
>
ir
XU
U, so that
> R\
it
is
x>0,
easily
we
demonstrated that (T
ir =S (K
-\-
If
get by the binomial formula
+ 2'-2>{U^+
-{-
form,
this involves
(!)'
T>
> 2.
deviate unequally from the typical
But
that deviates most.
&^
or
o'.
C/"V.
+ 1)'
or
>
44
and
^^
(>ir
K
but this
is
the proposition
^,
we have
+ V _
+ r,v
^'r
for
asserted,
(ir(0)r^
(A.(O)r'
the extension to
any number of terms
causes no difficulty.
But
must more
thus becomes
if it
or less
general law
that
law of errors of linear functions
the
approach the typical form, the same must hold good also of
all
mode-
rately complex observations, such as those whose errors arise from a considerable number
of sources.
The expression "source
of errors" is
employed
to indicate
circumstances which
undeniably influence the result, but which we have been obliged to pass over as unessential.
If
we imagined
these circumstances transferred to the class of essential circumstances, and
substantiated by subordinate observations, that which
is
now counted an
observation would
occur as a function, into which the subordinate observations enter as independent variables;
and as we may assume, in the case of good observations, that the influence of each single
source of errors
typical form
is
which
this function
small,
its
may
be regarded as linear.
law of errors would thus show,
if
we knew
The approximation
to
the laws
of
of errors
lost, simply because we, by passing them over as unessenmust consider the sources of error in the compound observation as unknown. More-
the sources of error, cannot be

tial,
we may take
over,
it
for
every such
granted that, in systematically arranged observations,
source of error as might dominate the rest will be the object of special investigation
if
necessary, will be included
The
calculations.
result then is
and,
removed by corrective
that great deviations from the typical form of the law of
among
the essential circumstances
or
errors are rare in practice.
32.
It is of interest, of course, also to
for the determinations of /j^
acquire knowledge of the laws of errors
and the higher half-invariants as functions of a given number
of repeated observations.
But though the symmetry

Here the method indicated in 30 must be applied.
of these functions and the identity of the laws of presumptive errors for Oj, o.^,
Om
afford very essential simplifications, still that method is too difficult. Not even for /x^ have
.
In my "Almindelig lagttagelseslcere", Kobenhavn

the general law of errors.
have published tables up to the eighth degree of products of the sums of powers
discovered
1889,
Sp s^
I
.
.,
cable.
expressed by suras of terms of the form o\o'\o"*; these are here directly appli-
In
W.
Fiedler: "Elemente der
Formen", Leipzig 1862,

difficult,
up
to the
10* degree
will
be found.
because they require the preliminary transformation
Up of the rational
Hirsch,
tables
neueren Geometrie und der Algebra der bindren
equations 21.
and Cayley has
There are such tables also in
given others
in
Their
of the Sp to
use
the
is
more
coefficients
the Algebra by Meyer
the Philosophical Transactions
1857 (Vol. 147,
45
have computed the four principal half-invariants of n^
p. 489).
mkîn.^)
m^/l.,{fi.,)
m'^l.ifx.,}
^h
= (w l)H, + 2m {m 1) X]
= (m - l]H^ + r2m {m - 1)H,L, + 4 (m - 1) (m - 2)^J +
+ 8ot2(w<
= {m l)^;^ + 24m (m
+ 32m (m - (w 2)^5/3 +
+ 8m (m (4m2 - 9m + Q)r, + 144m2 (m l^XJ] +
+ 96m2 (m 1) Ûl^-z 4' 48m (m 1) A^
(m
l)yi5
m';},(/i,J
(38)
l)'^
l)/l6>i.,
1)
(>w
Here
is
the
number
of repetitions.
Of
/ia
and
only the
tn'n,{/xs)
mnAf^;)
pt^
mean
mean
values and the
errors have been found:
= (m l)(m 2)^3,
-=
{ni-\y'(}n-2)U,+9m{m-l){m-2y^(XJ,+r,)+
+ 6m
'^
1) (m 2)^^
(m
(39)
and
mUi (/i J
m';._,
(^ J
=
=
6m + 6)X, 6m (m AJ
(m 1)2 (w2 6m + 6)Ha +
6m + 6) (2m'' Ibin +
+
+ 8m (m
+ 18m(m l)(m 2)(m 4)(m''-6m + 6)/i5/i3+

+ 2m (m (17m* 204m3 + 852m 1404m 828)>i;
+ 24m2 (m (3m 38m + 150m 138)^4 +
+ 144m''' (m (m 2) (m 4) (m 5)
+
(m
1)
(m2
1)
15)/le-<2
1) (m'''
1)
-j-
1)
Further
m*X,{/i,)
m^/ii (//g)
mnîli-,)
J>1.,
24m(m l)(m2 6m + 24)/i*.
know
=
=
(40)
-|
1',
/I
1)
4-
only that
(m l)(m 2)((m2-12m+12)>lf, 60m/i2yi3},
30^=^ + 150m'' 240m + 120)/i6

(m
36m + 36)/ii
30m (m
(7m2
60m(m 1) (m 2) (3m 8) ^J
l)(m 6)/l^
60m''(m
( l)(m 2)(m* 60m3+420m'' 720m + 360)^,
630m (m
2) 8m + ^)k^X.>
210m(m l)(m 2)(7m 48>-|-60)/i^yi3
1260m2(m l)(m 2)(m 10)yi3/i^
(m
(41)
1)
1)
1) (>
/}.,
(42)
(jw^
(43)
MU,{fts)
l)(Hi 126)6 + ]806w 8400w'' + 16800;2-15120w + 5040)/(

56w (w 1) (31w -^540/w'' -f 2340w'^ 3600w + 1800)-i6yij
1680w(m 1)(ot 2)(3ot=* 40/2-fl20> OGMs/ia
l)(49w* 720)'' + 3168m 5400>M + 3240)/i5
70m(OT
150^2 + 576w 540)/i,
840?w2 (m
(7w
2) 18?w +
10080''' (m
1) (m
l)(m'' 30^4-90);?;.
840wi(7
(44)
(m
1)
/ij
40)yi5;..,
(/''
Some
Î's of products of the
teristics as the
/x,^,
and
/i^,
present in general the same charac-
/i,
The most prominent
above formulae.
1) It is easily explained that
X^
of these characteristics are:
only to be found in the equation î(/i)
is
= î;
indeed no other half-invariant than the mean value can depend on the zero of the obser-
In
vations.
my
computations this characteristic property has afforded a system of multiple
checks of the correctness of the above results.

2) All
of
mean
Xi(/ir) are functions of the
0*
degree with regard to m,
all
squares
and generally each ^(/ir) is a function

with
accordance
the law of large numbers.
perfect
errors ^^(/Jtr) are of the (1)^' degree,
of the (1
3)
r>l.
mean
s)ti>
The
If r
is
degree, in
m 1
factor
appears universally as a necessary factor of ^(/ir),
an odd number, even the factor
even number, this factor
more Xs with odd
is
constantly found in every
No
indices.
m2
and,
appears,
term that
is
if
likewise,
if
only
is
an
multiplied by one or
obliquity of the law of errors can occur unless at least three
repetitions are under consideration.
(m
4)
Many
2)
(w
If,
indicate
particulars
... [ni
r)
these
functions
compounds
supposing the presumed law of errors to be typical, we put X^
then some further inductions can be made.
kUh) T
e
\1
^l(M
I
^'\1
T2
to the squares of
2>l
r\l=^
+-=h--f^j2
mean
errors of
of
fir
we
factorials
= ^4 ^... = 0,
In this case the law of errors of
As
as
and powers of m.
fi^
may
be
1'+
=U(o)e^do.
00
get under the
same supposition:
(45)
47
This proposition
for the
at our disposal
mean
the relative
the index
ju,
is
large
enough
fir
pis
number
tlien
errors,
are by no
to give us /i,
can be only approximately indicated;
guessed,
we have
computation of a law of actual
errors of ^/j, //j
If
r.
If
very great interest.
is of
it
ni of repetitions
be
will
seen
that
means uniform, but increase with

and
precisely
fi^
then
fairly well,
/i 3
and the higher half-invariants are only
and
be
to
the repetitions are not counted by thousands or millions.
if
As
numerical coefficients in
all
as the coefficients
1,
2, 6,
and 24 of
yij
(nr) increase with r, almost in the
we must presume
^^,
that the law
same degree
of increasing
uncertainty of the half-invariants has a general character.
We
have hitherto been justified in speaking of the principal half-invariants as the
complete collection of the fir's or

of the
m half-invariants
m repetitions.
first
errors for
We
now
;ir's
with the lowest indices, considering a complete series
to be necessary to
an unambiguous determination of a law of
accept that principle as a system of relative rank
with increasing uncertainty and consequently with
of
the half-invariants
importance of the half-
decreasing
invariants with higher indices.
We
instance
if
(pitch and
need scarcely say that there are some special exceptions to this
= ^'
/i^
toss),
then
as in alternative experiments with equal chances for
).^[ii^)
is
reduced to
Now we can undertake
33.
X\,
to solve the
which
problem
actual
is
not a mathematical one, but
state of a finite
presumed repetitions
predict, by
oracles
means
is
number
it
of observations
an evident
trespass;
is
to
and
For
only of the (2)'"' order.
main problem of the theory of obser-
vations, the transition from laws of actual errors to those

this
is
rule.
and against
of presumptive
Indeed
errors.
eminently practical. To reason from the

the law
it
is
governing infinitely
mere attempt
at
numerous
prophecy
to
of a law of presumptive errors, the results of future observations.
The struggle for life, however, compels us to consult the oracles. But the modern
must be scientific; particularly when they are asked about numbers and quantities,
mathematical science does not renounce
its
right of criticism.
We
claim
that
confusion
ambiguous use of words must be carefully avoided; and the necessary

must be restrained to the acceptation of fixed principles, which must agree
of ideas and every

act of will
with the law of large numbers.

It is
hardly possible to propose more satisfactory principles than the following:
The mean value of all available repetitions can he taken directly, without any
change, as an approximation to the presumptive mean.
If only one observation without repetition is known, it must itself, consequently,
be considered an approximation to the presumptive
The
mean
value.
solitary value of any symmetrical and univocal function of repeated observations
48
must
in the
same way,
as an isolated observation,
this function, for instance /ir
be considered
;,
presumptive mean of
^, (fir)-
Thus, from the equations 3741, we get by
tlie
repetitions:
/ii
,2
^ ^
{w l)(w 2)^''
(47)
m"
{f^^+j^^mf'')
[m
as to X^,
P.7,
/?8
it
is
1) (w,
2) (j2
to
preferable
\2m + 12)
use the
principalli/ by
we predict
+,^^1/^2/^3);
equations 42
Inversely, if the presumptive law of errors
of any theory or hypothesis,
(/^'^
is
44
iinown in
themselves, putting only
this
way, or by adoption
the future observations, or functions of observations,
computing their presumptive mean values. These predictions however, though
univocal, are never to be considered as exact values, but only as the
and most impor-
first
tant terms of laws of errors.

If necessary,
useful,
did
if
functions
to
mean
errors
of observations
the single observations.
by
nay
not release the questioner from thinking and from

is
a difference in the manner.
and higher
the presumed
These supplements
nor
responsibility,
half-
law of
oracles
do the modern
If the crossing of a desert is
calculated to
20 days, with a mean error of one day, then you would be very unwise, to be
you provided
as of living.
for exactly
20 days; by
Even with provisions
so doing
for 21
you can carry with you provisions for

Your life must be at stake to make you
if
be
often
may
The ancient
necessary, for the correct interpretation of the prediction.
ones; yet there

last
predictions with the
for the predicted
computed
which itself belongs
invariants,
errors,
we complete our
sure,
you incur as great a probability of dying
days the journey
is
evidently
2325
days, the undertaking
set out
with provisions
may
for only
But
dangerous.
17
be reasonable.
daysor
less.
In addition to the uncertainty provided against by the presumptive law of error,

the prediction
When
this
may
be vitiated by the uncertainty of the data of the presumptive law
law has resulted from purely theoretical speculation,
calculate its uncertainty.

left to
choose between
its
It
may
be quite exact,
admission and
its
or
partially
or
rejection, as long as
by repeated observations has given us a corresponding law
can be improved on.
it
is
always impossible to
absolutely
no
itself.
trial of
of actual errors,
false,
we
are
the prediction
by
which
it
49
If the
errors,
we
law of presumptive errors has been computed by means of a law of actual

according to (37),
can,
employ the values
^j,
and the number
...
/{g,
of
In this case the complete half-invari-
actual observations for the determination of ^riî)-
ants of a predicted single observation are given analogously to the law of errors of the
sum
of two bondless observations by
Xr
is
Though we can
more difficult, or
far
in the
Xrifli)-
same way compute the uncertainties
rather impossible, to
make
of k^,
and
X.^,
X^,
it
use of these results for the improvement
of general predictions.
Of the higher half-invariants we can very seldom, if ever, get so much as a rough
The same reasons that cause this
estimate by the method of laws of actual errors.
difficulty,
render
it
matter
of
less
to
importance
In determining X^ and
2.
As
exclusively
3.
to
upon
the
Employ the
From what
X^
upwards,
as
rely
indications obtainable by actual observed values for the intermediate
is
when you have
the choice between the theories in
said above of the properties of the typical
that no other theory can fairly rival
is
must be:
theoretical considerations.
half-invariants as far as possible
It is
with high indices, say from
half-invariants
determination.
precise
almost entirely upon the actual observed values.
rely
X,^,
any
laws of presumptive errors
Therefore the general rule of the formation of good

1.
obtain
it
in the multiplicity
law of errors,
and importance of
it is
(2).
evident
its applications.
not only constantly applied when X^, X^, and ^5 are proved to be very small, but
In these
used almost universally as long as the deviations are not very conspicuous.
cases also great efforts
will
be
made
to
reduce the observations
modifying the methods or by substituting means of

non-typical single observations.
by the
difficulty of establishing
The preference
many
the
for
to
the typical
form by
observed values instead
typical observations
an entirely correct method of adjustment
is
it
of the
intensified
(see the following
chapters) of observations which are not typical.
In those particular cases where X^ or X^ or

theoretical considerations (proposition 2 above) as to
not to result in putting the latter

1899,
p. 140,
0.
As shown
^^5
/^
in
cannot be regarded as small, the
^^^ higher half-invariants ought

g ^^^
" Videnskabernes Selskabs
Oversigter"
such laws of errors correspond to divergent series or imply the existence of
imaginary observations.
The
coefficients
kr
of the functional
law of errors (equation

7
(6))
50
have this merit in preference to the half-invariants, that no term implies the existence
of any other.
This series
* (X)
where ^{x)
k,
<p{x)-^ D^,p [X)
g/i^
+ ^ DV [X) - *| D^<p {X) + ^ D^^{x)
(the direct expression (31) is found p. 35), is therefore
and in every prediction of future
'..
recommended
The functional form
as perhaps the best general expression for non-typical laws of errors.

of the law of errors has here,
results,
the
advantage of
showing directly the probabilities of the different possible values.
The skew and other non-typical laws
of errors
seem
have some very interesting
to
applications to biological observations, especially to the variations of species.
The
scientific
treatment of such variations seems altogether to require a methodical use of the notion of
laws of errors.
Mr. K. Pearson has given a
series of skillful
computations of biological and
other similar laws of errors (Contributions to the Math. Theory of Evolution, Phil. Trans.
V. 186, p. 343).
Here he makes very interesting
efforts to
develop the refractory binomial
But there are evidently

problems, and the above formula
functions into a basis for the treatment of skew laws of errors.
no natural links between these functions and the biological

(31) will prove to be easier
and more powerful instruments.
or discontinuous laws of errors,
Example
1.
From
more
refined
In
cases
of very
abnormal
methods of adjustment are required.
the 500 experiments given in
14
are
to
be
calculated
the
presumptive half-invariants up to X^ and by (31) the frequencies of the special events out
of a
number
4-708, Xi
of Sg
= 3-895,
= 500
and Xi
new
repetitions.
= 26-946.
You
find
l^
= 11-86,
X^
=4-1647,
X-^
comparison of the computed frequencies with
the observed ones gives:

Frequency
Events
will
51
52
we must expect
Consequently, for any isolated observed value
io-^,io)r
35.
that
X,(o).
In the following chapters, and in almost
(50)
all practical
we
applications,
shall
work only with the typical law of errors as our constant supposition. This gives simplicity
and clearness, and thus
may be recommended as a short statement of the law of
= Xi
errors,
a^b
indicating a result of an observation found directly or indirectly by
and
tation with observations,
By
= Vî
expressing the
the "weights" of observations
mean
the squares of the
errors,
mean
we understand numbers
consequently
-p.
same method,
When
Vr
is
as
the
the weight of the partial
here, according
mean
value
the
itir,
to
,^
represent the weights
mean
abscissa of any single body, r

vations,
to the
according
value
miVi+m^v^
...
must be
+ mrVr
,^^.
analogous to the formula for the abscissa of the centre of gravity,
is
when we
have been obtained by
(37),
total
itself
formula
to the
computed according
which
numbers
latter
compu-
result.
inversely proportional to
The idea presents
speak of the means of various numbers of observed values which

the
same
error of the
its
We
loeight.
if
nir
is
the
speak also of the weights of single obser-
above definition, and particularly
in
cases
where we can only
estimate the relative goodness of several observations in comparison to the trustworthiness

of the
means
of various
numbers
The phrase probable
of equally
good observations.
error, which we
observers, is for several reasons objectionable.
still
It
find frequently
employed by authors and
can be used only with typical or at any
and indicates then the magnitude of errors for which the

of smaller and larger errors are both equal to \.
The simultaneous use of
rate symmetrical laws of errors,

probabilities
"mean
and "probable error" causes confusion, and

that must be abandoned, as it is less commonly applicable, and as
the ideas
in
the
cases
0-67451/17,
of
error"
the
while
typical
on
the
law
other
of
hand
independent of that of the probable

one,
still
errors
the
error.
it
can only be computed
computation
of
the
mean
error
error as
is
quite
As errors which are larger than the probable

and the
as
the
mean
rarer "large" ones.
probable error tempts us constantly to overvalue the degree of accuracy
very
evidently the latter
hy the previously computed mean
frequently occur, this idea is not so well adapted
a limit between the frequent "small" errors
it is
error to serve as
The use
we have
of the
attained.
More dangerous still is another confusion which now and then occurs, when the
expression mean error is used in the sense of the average error of the observed values
according to
except
their
when we
are
numerical values without regard
to
certain of a law of typical errors,
the
signs.
This gives no sense,
and with such a one
this
mean
53
error" is
idea
K^
that
is
we
of a square
The only reason which may be advanced in defence of the use of this
some little computations, viz. some squarings and the extraction
which, however, we rarely need work out with more than three signi-
are spared
root,
ficant figures.
FREE FUNCTIONS.
IX.
The foregoing propositions concerning the laws
36.
especially of linear functions
mean
value and
its
mean
correct
consider mutually unbound, another thing which
matics.
For
Oj
is
observe the abbreviated notation

.
o,
that
we
LAH)
While the
circumstances
=
=
r,^,(o,) -f
r'J,(o,)
R"
[r"o],
+ r^,{On) =
+ rUAOn) =
that
there
(52)
[r-îo)].
(53)
also resemble the observations
even though
they
derived
are
from
"'
unbound observations.
employed in the computation of both R'
these results will generally be
aB'
the
bound
to each other.
+ bB" =
[(>'
-f-
[r'o]
If
and
This, however, does not
bB".
We
can,
at
any
a function of the unbound observations,
+ br")
(54)
o].
some importance for the treatment of those cases in which

observations are bound.
They must be treated then just like results, and we
This possibility
must
in
can be bonds between them, and, unfortunately, there can be
rate, represent the function of the results directly as
the single
== [roj
[r^,{o)]
prevent us from computing a law of errors, for instance for aB'
'
ViO^ -\-...ro
results of computations with observed quantities, taken singly, have laws
only some observations have been
a linear function of unbound observations only,
is
same way as the observations, they
between "results",
bonds
quite foreign to exact mathe-
have demonstrated the rules of computation (35)

î(-B)
of errors in the
is
only upon the supposition that the result
More-
apprehension of what observations
we may
it
The
from exact mathematics.
differs
error, just like the single observation.
must be founded on a
the computation
over,
an exact quantity, but always a law of errors which can be
result, particularly, is not

its
of functions
form the basis of the theory of computation with observed
values, a theory which in several important things
represented by
of errors
try to represent
is
of
them
as functions of the circumstances
which they have in common,
and which must be given instead of them as original observations. This may be difficult
to do, but as a principle it must be possible, and functions of bound observations must
therefore always have laws of errors as well as others;
only, in general,
to compute these laws of errors correctly simply by means of the laws
it
is
not possible
of errors of the
54
we cannot,
observations only, just as
by means
compute the law of
in general,
of the laws of errors for
aE -\-bR"
errors for
and E".
In example 5, 29, we found the mean error in the determination of a direction

between two points, which were given by bond-free and equally good (/.^{x) == Xîy) ==
2
measurements of their rectangular co-ordinates, viz.: Xi{Ii)
and then, in example
-^^,
we determined
It
seems
an
the angle
obvious
in a triangle
= X^_{R')-\-k^(R") = -^ +7p2where J,
and J" are the
J',
then
conclusion
^"*
whose points were determined
that,
*'î^ i^ "^*'
as
correct
sides of the triangle.
R'
R'\
the solution
The cause
in the
it
remarkable then that, when
is
With
unbound
equally good
is
1)
6,
same way.
we must have kîV)
is ^ j(
F)
of this
is,
~-^,2-
R"
Tj
of course,
the co-ordinates of the angular point enter into both directions and bind R' and
But
tliat
together.
a right angle, the solutions are identical.
observations, Oq, Oj, o^, and O3,
2o, + 0o) =
^2(02
^2(03202+0,) =
we
get
6-ij(o)
6^2(0),
but
= 20Xi{o),
= (03 2oj + o,) (Oj 2o, + 0o),
''2(03-302+30,-0,)
03 3o2+3o, Ofl
although
according to which
we
should expect to find
i?2(03
But
3o2+3o, Oo)
;2(08 203+0,)
+ /2(o22o, + 0o) =
12^2(0).
on the other hand, we combine the two functions
if,
R'
where ki(R')
= b?>Xi{p)
aR'-\-bR", then,
(53a2
O0+60,
and
;2(-K")
curiously
+ 14J2)^3(o) =
402
enough,
aU^(R')
-\-
14>i2(o),
we
R"
and
2oi+3o2 O3,
and from
as
get
the
this
compute
correct
result
ij for any function
XîaR' -\-bR")
bn,{R").
Gauss's general prohibition against regarding results of computations

especially
those of mean errors
from the same observations as analogous to unbound observations,
has long hampered the development of the theory of observations.
To
discovered
simplify
Oppermann
that
our
and, somewhat later, to
the prohibition
calculations.
We
is
not absolute,
must
therefore
Helmert
is
due the honour of having
but that wide exceptions
enable
us to
study thoroughly the conditions on which
actually existing bonds may be harmless.
Let Oi,...o be mutually unbound observations with known laws of errors,

,Jg(oj),
of typical form.
Let two general, linear functions of them be

[po]
[qo]
= p,o^ +...+;>o
= g,o, +... + ô.
^1(0,),
55
For these then we know the laws of errors
X,{qo-\
[qk,{o)l
L,[<i6\
[2%(o)l,
Xr\qo\
'
<^''
O/
>-
F = a\^po]-[-b[qo], the correct computation

=
of F
{{ap-\-bq)o\ will further give
=
X,{F) = (ap, + hq^)XAo^)^...^(apn-\-Hn)kAon)
of the law of
For a general function of these,

errors by
means
Xr{F)
[po]
'^
X.,{po\
and
2ab
mean
bound
to
values can be computed unconditionally, as
unbound observations, and that the law of
[qo\ were
in
a difference,
is
XîF)
the
term
containing
[po] and
if
the
[50] were not
consequently
[pqXîo)]
PlqxX.^{o^)
+2V/,;.,(o)
the functions [^0] and [go] can indeed be treated in
upon
as
ought not to be found in the formula,
remains typical.
errors
one another.
When
for the
r>2.
for
Only in the square of the mean error there

factor
(56)
+ i'Â^Lô] + 2a.[i;5 XM}
appears then, both that the
It
if
=
=
law of errors
this
for every linear function
supposition.
We
consequently, the formula
respects like
is
(57)
unbound observations,
found correctly determined also
such functions mutually "free functions", and for such,
call
mean
for the
XApo]a
all
them
of
[_qo}b)
error
a'lj^n.m
+ b-'VqH.ip)]
(58)
holds good.
If
this formula holds
If
two functions are mutually
good
for
one set of
finite values of
a and
b,
it
holds good
for all.
and
each of them
free,
is
said to be "free of the other",
inversely.
Example
1.
The sum and
2.
When
difference
of
two equally good, unbound observations
are mutually free.
Example
the
co-ordinates
of a point are observed with equal accuracy
and without any bonds, any transformed rectangular co-ordinates

mutually
free
the
same
will be
free.
Example
is
for
of
every
3.
The sum
diff'erence
function of such differences.
or
the
mean
value of equally
good, unbound
between two of these, and generally also
observations
free of every (linear)
56
Example
4.
The
differences between one observation
bound observations cannot be mutually
Example
and two other arbitrary, un-
free.
5.
Linear functions of unbound observations, which are
6.
Functions with a constant proportion cannot be mutually
all different,
are
free.
always
Example
37.
In accordance with
propositions must hold good

stances:
it
is
what we have now seen
all
of free functions, corresponding
which are influenced by the same circum-
also of observations
not necessary to respect
free.
connecting bonds;
it
is
possible that actually
bound observations may be regarded as free. The conditions on which this may be the
case, must be sought, as in (57), by means of the mean errors caused by each circumstance
and the
coefficients
particularly
by which the circumstance influences the several observations.
Note
Jf two observations are supposed to be connected by one single circumstance which

in common, such a bond must not be left out of consideration, but is to be
have
they
respected.
in the
same
If,
same
to
Likewise,
if
there are several bonds, each of which influences both observations
direction.
on the other hand, some common circumstances influence the observations in the
and
direction, others in opposite directions,
work as
treating
them
38.
linear,
or
if,
forcibly as the other, the observations
at
as
unbound
is
at
any rate
less
Assuming that the functions
moreover, one class must be supposed
may
possibly be free, and the danger of
than in the other cases.
which we
of
shall speak in the folloving are
any rate may be regarded as linear when expanded by Taylor's formula,

may reject squares and products of the deviations
because the errors are so small that we

of the
observations
from fixed values; and assuming that the observations o,, ...
which
all
given,
we can now demonstrate
the functions depend, are unbound, and that the values of ^.^Cî)
Out of the
total
system of
of
o,
on
^'(On) are
functions
all
= p,Oi +
n observations we can
partial system containing all those,
m<n
a series of important propositions.
[po]
of the given
... -\-pnO
arbitrarily select
partial
systems of functions, each
which can be represented as functions of a number of
mutually independent functions, representative of the system,

[ao]
fliO,
+ anO
[do]
== diOi
+ do
which no one can be expressed as a function of the others.
the existence
of other functions
which are
free
We
can then demonstrate
of every function belonging to the partial
57
system
[go]
g ô I
-\-
[ga X2]^0 ...
[ao] .... [do].
by
represented
...-{- c/nOn
[ffd/i^]
=0.
of
free
IS
For
It
[{xa
. . .
-\-
[g(xa-^...+zd)?.2]
into a
sum
of the
function
Any
we
system
[po]
[po]
^ z[gdL,] =
. .
now
can
equations
for
equations
0.
one single way be resolved
in
[p'o], this proposition
is
free of the partial
[pa^j]
x[aa ki]
[pdX^]
x[ad X2]
number
-\-
-\-
(59)
...
We
[p'dX^]
0,
all
that
find
z[da X^]
(GO)
of the
the determination of the latter,
be written
may
+ {x [ao] + ... + Z [do]}.
[p'o]
from which we determine the coefficients x

equal to the
function
... '}'2[do],
By means of the conditions of freedom, [p'aL^]

concerns the unknown function [p'o] can be eliminated.
is
consequence
the
[do], while the other belongs to this system.
addendum
call the free
of
free of every function of the partial
x[ao]-j-
x[cia ^^]
such
that
prove
functions of the same observations, one of which
of two
system represented by [ao]

If
total
zd) o]
-\-
to
in
[do]
must be
[go]
system,
because
[tto]
if so,
sufficient
is
-\-
z[dd X2]
The number
z unambiguously.
unknown
because,
-\-
of these
quantities, and they must be sufficient
according to a well known proposition
from
the theory of determinants, the determinant of the coefficients
[dak^]
[aaXi],
[adk^],
is
sum
positive, being a
tions
[ao]
XîOr)
dr,
[ddki]
of squares,
X(0s)
d,
and cannot be
0,
our supposition
contrary to
could,
[do]
unless at least one of the func-
be
represented as
function
of
the others.
From
the values of
a;
...
[p'o]
If
belongs
[po]
termination of
identically [p'o]
to
2 thus found, we find likewise
[po]
x[ao]
the partial
z expresses
its
system
coefficients
z[do].
represented
in
by
(61)
[ao] .... [do]
that system only,
and
then
0.
the
de-
we
get
58
But
we take [po] out
if
from zero and

[ao]
. . .
free of that partial system.
must produce
[qo]
[fi?o],
Let [po]
functions
[ao]
[c?o]
[po]
[qo]
belongs to the partial system of

free function as [po].
independent of one another and of the
functions,
we then
if
in
[ro] be
If
manner the very same
in this
and
find [p'o] out of [po]
out of [ro] as the free
[r'o]
[ao]...[do], the n functions [ao]... [t^] and
functional parts in respect to
may
of the partial system, then (61) gives us [p'o] as different
[jP'o]
['"'o]
be the representative functions of the total system of the functions of Oi...o, because
no relation a[p'o]-\-
a [po]
-|-
4- S[ro]
... -\-
we employ
possible; for by (61)
is
n [ao] -\-...-^<p [do] ==
-\-
tative character of [po]

If
d[r'o]
[p'o]
and [ao]
[i-o\
.
..
[r'o] or other
[go]
of the partial set [ao]
all free
system [ao]
in contradiction to the
n
.
m
[ko]
can be free of [ao]
[do], as representative functions of another partial system
must be
free of every function of the partial
No
).
other function
[do] than those belonging to the system [go]
Thus
should have more than n independent functions of the n variables o,
we can
presumed represen-
mutually independent functions
(Compare the introduction to this
[rfo]
result in a relation
might
[do].
of Oj...o, then every function of this system

.
it
[ko]
.
of
Oi...0n
otherwise
we
o.
selecting arbitrarily a partial system of functions of the observations Oj ...o
with reference to given squares of mean errors
^.^(0^)
X.,(on)
distribute
the linear homogeneous functions of these observations into three divisions:

1)
the given partial system [ao]
2) the partial system of functions

3) all the rest, of
compounded
which
[go]
is
[j/o]
[ko],
which are
of functions is
a reciprocal
first
is
always in only one way
If the
[ao]
and
system to one of the second.
first partial
property.
[ko] were selected arbitrarily instead of the
would be found
free of the former,
proved that every such function
by addition of one function of the
The freedom
.
it
[do],
second partial system
[do], then only this latter
as the free functions in 2); the composition of every function in 3)
would
remain the same.

Example.
Determine the parts of Oi-\-o.^, Og-)-o<, Oj-j-o^, and 02-\-o^, which
are free of o,-|-03 and o^-j-o^,
on the supposition that
exact and unbound.
Answer:
^ {o^
+ 0, O3 oJ,
all
etc.
4 observations are equally
59
39. Like
other functions of the observations o^.-.o,,, each of these observed
all
values, for instance o,,
\ao]
from
[do], the other
= d'i+o'i
Oi
to the
system of [go]
follows, generally, that [po]
system of [ao]
[c/o"]
two quantities, one
of
to the partial
o't
...[ko"]
exist
o'a
belonging to the system of
o<
which
[ko],
is free
is free
relations [ao']
and [po"] evidently belongs

of this. Accordingly there must
[do']
likewise
this
be carried
to
operation
can be split up,
an
in
the utmost.
to
through
It
is
[ao]
the representatives
[c'o]
[d'o]
which were
all
the
all
till
Such a
[bo],
to
according
[do],
analogous way,
separate the functions
[c'o]
[d'o]
functions are mutually
we
as functions
of
tions is analogous to the theory of the transformation
example
2),
and
is
call
When we
we
what
are to use in
errors,
of
. .
[d" o],
it
a
= {r:;^
f
'
MJ^+
because
shall
into free func-
(comp.
and determinants
enter into
here
follows.
the
complete
This
is
very easily done, for
we have
+ ^^drr^
^'M
(62)
^[d^d'k^Y['^""l}
demonstrated by substitution in the equations for the direct transformation
[ab'X^]
[ad" X^]
[ao]
aiOi+
[d o]
fi?,o,-f
=-
...[b'd^L,]
like
d".
\[aaX^]^
is
set
often important to be able to undertake the
is
opposite transformation back to the observations.
which
quite
co-ordinates
rectangular
have transformed the unbound observations Oi...o into
set of free functions [ao], [b'o]
every single
a complete
the observations,
treated in several text-books of the higher algebra
propositions only, which
them; and their mutual freedom
under the name of the theory of the orthogonal substitutions.

those
free,
computations of laws of
all
unbound observations, and thus wholly replace the original observations.

41. The mathematical theory of the transformations of observations
36,
see,
number
smaller
sufficient to enable us to express all
is
we
could, for instance,
select
(37),
enough
to
would then represent the suband in

within the partial system [ao]
[do]
[bo],
can be treated, by
effect that they
system
partial
representative
functions of these observations,
all
has the
collection of representative functions
Their number
of free functions.
and
[co]
one
of
[do]
free of [ao]
free of [ao]
rest.
[ho],
way we may continue
one of
and from the others
system of functions,
this
[ao]
except tvhen
easy
We
however, that also the partial systems of functions can be split up.
among
between o'[...o'n.
0,
into three or niore free quantities, is of no consequence for the following,
imagine
But
of this.
= [po"]-\-['po'],
That the functions of observations
40.
system which
[do], [po'] to the
between the w functions o\.

relations
sum
the
is
-\-anOn
+do,
0.
8*
60
As
[ao]
former,
the original observations, considered as functions of the transformed observations
[d^d],
we
must be mutually
find by
just as well as the latter are free functions of the
free,
mean
computing the squares of the
expresses the formal condition that
and the equation that
(o,)
two of the most remarkable properties
free of ot,
is
o,
errors X^
of the orthogonal substitutions:
"'
'
and
=
If all observations
is
i-^
mean
-^
(63)
(64)
sum
which gives
also
a more elegant
of the squares of the thus reduced observations
not changed by any (orthogonal) transformation into a complete set of free functions.
We
have
which, pursuant to the equations (63) and (64),
sums
is
r^
reduction
error,
the preceding equations, the
all
and functions are stated with their respective mean error as
unity, or are divided by their
form to
+---
is
easily
demonstrated by working out the
As
of the squares in the numerators on the left side of the equation.
identical, the same proposition holds good also,
o^...o and n
this equation
of the differences between
corresponding to them Vi...,
variables
arbitrarily selected
for instance,
corresponding differences between the values of the functions.
Also here
and of the
is
(66)
42.
the easiest
must
For the practical computation of a complete set of
way
to bring forward the functions of such a set
select a sufficient
number
taken into consideration.
The function
transformed
r^ and
,
set.
[ao\,
of functions
and
fix
free functions it will be
one by one.
the order
in
which
which
By multiplying
is
the
first in
the selected
this list, is
we
be
now, unchanged, taken into the
function by suitable
constants
subtracting the products from the remaining functions [6o] in the
Of these then the one which
into the transformed set.
to
For a moment we can imagine this order to be arbitrary.
according to 38, from each of these separate the addendum which

function.
In this case
these are
This function
is
is
is
list,
we
can,
free of the selected
founded on function Nr. 2 on the
multiplied in the
of the forin
list is
taken
same way and subtracted from
61
the
still
remaining functions, so that they give up the addenda which are free of both the
and so on. The following schedule shows the course of the operation
selected functions,
for the case
Fnnctions
4.
62
.
[bbk\-[ab^] [baX] [aaA]

:
=
=
[ccX]--[avA]-[caXy.[aaX]
=
[dcX]-[acX]-[daX]:[aaX]
[b'b'X],
[ftc-yi]
=
[cbXl -[ab/].[caX] [aa^
= [d'b'^,
[dbi]-~[abk]-[daJ\:[aaX]
:
[c'b'X],
[c'c'X]-[b'c'^
- [acX]
[baX\ : [aaJi]
[db'X\
- [6'c'yi] '[d'b'X]
[d'c>X]
[c'c'X],
[crf/i]-[ac?/l].[ca/i] :[aa>i]
the
sums
importance
X^[ao]
be seen, there
will
the products
of
as
they
= [aaA],
X^lb'o]
properly so
the
squares
[b'b'X],
i^[(f'o]
= [c"c"X],
to be carried out, in
ones. (Because
errors
JS
3''<'
degree.
squares
= [d"'d"'X]
for
each of
of
special
are
transformed functions,
[d"'d"'X].
is
known
and with
/(^(oi)
Ooi
VJo
1,
Sums
+0oj -flOg +O0, +0oj
0-110
1-210
0-1
1
i-J'o,
_4
3-3
6-4
12
17
35
35
35
12
35
"35
T
"Y
_1
7
FJ303-i#O3
3
"3
"35
12
17
35
3 6
12
36
1
36
'35
_1
i
5
1
Fj03-4J^03+fFJ03-iJ^03
be an
10
With symbols
we have then:
Coefficients
36^
to
transformation into free functions
such a way that the higher diiferences are selected before the lower
Function
"3
the
of the
and k^[d"'o]
The
"35
12
35
2
[c'd'X]
[b'b'X]
[d"d'X] : [c'V'i]
certainly, J* etc., possibly, represent equations of condition).
for the diiferences,
0,
of
computation
The function tabulated
of a table with equidistant arguments.
integral algebraic one, not exceeding the

is
mean
- [c"d"X]
of double
The sums
called.
the
[b'd'X]
[b'b'Ji]
of
[<^b'X\
Five equally good, unbound observations Oj, Oj, O3, o^,and O5 represent
Example.
values
are
[d'd'X\
a check by means
is
[bak] [aa?.]
[d'c'i],
[d"d"X]
As
[M/l]
=
[ddX]-[adX]-[daX]:[aaX] =
= [c"c"Xl [c'd'X]-[b'd'X]
= [d'd"X]
[b'b'^
= [d"c"X\, [d'd'X\-[b'd'^.[d'b'X]
= [d"d"X]
[b'b'A]
- [ad?.]
[b'c'Ji],
of the Products
63
The complete
of
set
observations
free
and
the
squares
of
mean
their
errors
are thus:
(0)=
(1)
(2)
o,
=
=
+ Jô,-^y*o,
Vjo,-iJô,+l(Vdô,-U'o,)
Jô.^^J'o,
(3)=
(4)=
VJô.-Uô,
J^03
Through
be sufficient
for
this
= UOi + 02+0, + 0, + 0,),

= ,V(-2o^-o, + o, + 2o,),
= i(2o,~o^-2o,-~o, + 2o,),
= H-Oi + 2o2-2o,+05),
= Oi-4o,+6o3-4o, + o,
/,(0)
>i,(l)
;.,(2)
^,(3)
^.,(4)
and the preceding chapter we have got a basis which
computations with observations and, in a wider sense,
for
=
=
=
=
=
1
Jq
f
|
70
will generally
computations
with numerical values which are not given in exact form, but only by their laws of errors.
We
compute the law of errors for a given, linear function of reciobservations whose laws of presumptive errors we know.
By this we can
can, in the
free
procally
first
place,
solve all problems in
which there
more or
less
data,
problem.
When
exact
than
not given a greater number of observations, and other
of the
independent unknown values of the
reciprocally
means of the exact mathematics, have expressed

a function of the given observations, and when we have
we, in such cases, by the
unknown numbers
each of the
is
as
then we_ can,
succeeded in bringing these functions into a linear form,

the laws of errors for each of the
by
(35),
compute
unknown numbers.
Such a solution of a problem may be looked upon as a transformation, by which

in other ways given values are transformed into functions, each corre-
n observed or
to its particular value
sponding
It lies often
the solution of the problem
known
quantities, perhaps
unknowns
among
unknown
the independent,
values of the problem.
near thus to look upon the solution of a problem as a transformation,

is
not the end but only the means of determining other un-
many
of the problem.
when
Thus,
other,
which are
for instance,
all
explicit functions of the
we compute
independent
the 6 elements of the orbit of
a planet by the rectascensions and declinations corresponding to 3 times, not precisely as
our end, but in order thereby to be able to compute ephemerides of the future places of
the planet.
is
only
But while the
limited
validity
when we want
to
of this
determine
view
the
is
absolute
in
exact
mathematics,
presumptive laws of errors
it
Of sought
Only the mean values, sought

and with these the general linear
functions by the given laws of errors for the observations.

as well as given, can be treated just as exact quantities,
transformation of n given into u sought numbers,
with altogether n^ arbitrary constants,
remains valid, as also the employment of the found mean numbers as independent variables
in the
mean value
If
we want
of the explicit functions.

also correctly to determine the
mean
And
to
transformation than that into free functions.
if,
we may employ no other

extent, we may choose the
errors,
some
64
independent unknowns of the problem as we please, we
often
may
succeed in
carrying
unknown
through the treatment of a problem by transformation into free functions; for an
number may be chosen quite arbitrarily

unknowns looses, as a function of the
Altogether are \n[n-\-l) of the
But
free
w""
unknown can
get
an
factor.
arbitrary
mean
the
errors
for
matter how
the several unknowns, no
many
there
can be computed only in such a way that each of the sought numbers are directly
be,
expressed as a linear function of the observations.
laws of errors of the observations are not typical,

X-i
still
coefficients of these transformations arbitrary.
n"-
the problem does not admit of any solution through a transformation into
if
functions,
may
and each of the following
coefficients,
observations, only an arbitrary coefficient in com-
even the
the preceding one;
parison to
in all its
and the higher half-invariants
The same holds good

and we are
to
also
examine how
when the
it
with
is
in the laws of errors of the sought functions.
importance, nay a privileged position as the only legitimate proceeding,
Still greater
gets the transformation into a complete set of free functions in the over-determined problems,
which are rejected as self-contradictory in exact mathematics. When we have a collection

of observations whose number is greater than the number of the independent unknowns
of the
problem,
then the question will be to
standpoint of the observations.
We
determine laws of actual errors from the
must mediate between the observations that contradict
one another, in order to determine their mean numbers, and the discrepancies themselves
must be employed
repetitions,
to
determine their mean deviations,
the discrepancies conceal themselves behind
and require transformations

which, as the problem
the
instance,
Besides,
etc.
sum
is
of the
those of the
does not contribute,
for
All
detection.
their
But
angles
must come
of a
plane triangle,
of the problem,
forth by
we have not
to do with
the
changes of the circumstances
the
functions
over-determined, have theoretically
unknowns
as
of the
necessary
must be
selected
to the determination of
the transformation
observations
values ,
for
as
special
for
use.
which the theory
by which the problem
is
to
be solved.
As we
moment
shall see in the following chapters on
here that
we transform
with mutually free observations, and must not
formed functions
in
Adjustment,
into a system of free functions.
various ways must come
itself
it
becomes of essential
The transformation begins
introduce any bond, because the trans-
forth as observations
which determine laws
of actual errors.
X.
4.3.
ADJUSTMENT.
Pursuing the plan indicated
tion of laws of errors in
in
5 we
now
proceed to treat the determina-
some of the cases of observations made under varying
or different
65
But here we must be content with very small results. The

hardly ever be solved. The necessary equations must be taken from
circumstances.
essential
general problem will
the totality of the hypotheses or theories which express
say their half-invariants
of the
as functions of the varying or wholly different circumstances
Without great
observations.
the terms of each law of error
all
equations can be reduced considerably,
if
the multiplicity of these theoretical
however,
regret,
we suppose
all
the laws of errors to be exclusively
of the typical form.
For each observation we need then only two theoretical equations, one representing
its
presumptive mean value
the other the square of
ii(o,),
But the
tions of the essential circumstances.
other
unknown
mean
its
equations will
theoretical
as
error X^(Oi),
generally
func-
contain
the arbitrary constants of the theory, and these must be elimi-
quantities,
nated or determined together with the laws of errors.
The complexity
is
still
great enough
to require a further reduction.
We
must, preliminarily at
by theory, or at least their mutual
ratios,
solution by the indirect proceeding.
used in the
far
the weights.
mean
If
errors to be given directly
the
not,
we
word "criticism"
concerning the
Hypothetical assumptions
shall try to expose beside the several
But even
further
if
we
Moreover,
suppose
will
it
X,^ {Oi)
are
solutions,
confine our theoretical equations
the
using for brevity
the
and other operations connected with them.
for these
the presumptive
to
and the arbitrary unknown quantities of the theory, the solutions
we
problems require a
approximation and checked and corrected by special operations which, as
first
as possible,
events, suppose the
all
theoretical
be
to
equations
or
linear
means
^1(0,)
will only be possible if
reducible
generally be necessary to regard as exactly given
many
this
to
form.
quantities really
found by observation, on the supposition only that the corresponding mean errors will be
small enough to render such irregularity inoffensive.
In the solution of such
functions of observations
problems we must rely on the found propositions about
with exactly given
In the theoretical equations of
coefficients.
each problem sets of such functions will present themselves, some functions appearing as
given, others as required.
are,
for
now
The
observations,
the given observed values
among
instance,
the
unknown
0,,
now
independent variables of these functions,
as
means
the presumptive
quantities
required
for
the
^1(0,);
exact
the latter are,
satisfaction
of the
theoretical equations.
What
is
said here provisionally
about the problems that will be treated in the
following, can be illustrated by the simplest case
same observation, resulting

retical equations
the forms
==
(discussed above) of
in the observed values Oi, ... o.
If
we
repetitions of the
here write the theo-
without introducing any unnecessary unknown quantities, they will show
.î(o,)
Xîot)
or,
generally,
/i,[a(o,
oj)].
But these equations are

9
66
evidently not sufficient for
determination of any
of the observed values into any function [a{0i
tion
functions being useful only to check
we
0t)],
these
which
-\-
differences Oi
ot
and by
o,
-\-
this
we
mean being equal
we thus
If
may
^(Oi
be
the
to
sum
the
of the observed values.
oi
that two sorts of special cases
be solved immediately.
that
also
only
+ "),
mean
to the actual
first
It
theory
fully
...
it
o,
may
is
of
which
occur, in
be that the theoretical equations concern-
may
ing the observations leave some of the observations, for instance o^,
may
is
oj.)]
The only w* functions
as representatives.
consider a general series of
the greatest importance to notice
our problem
find the sought determination by

^l{Oi)
the presumptive
of the
values
must evidently be proportional
can be free of this partial system,

.
another
if
But we must remember what
by our criticism.
X.^ (Oj)
erroneous
ijnow about free functions: that the whole system of these functions [a(o,
a partial system, with
0,
which they only give
î(Oi),
The sought common mean cannot be formed by the introduc-
found beforehand.
is
>i,(Oifc)
ttie
determines
certain
of
others
quite untouched;
the observations,
it
for
instance o.
In the former case, that

observation o,,
it
is
is
when none
of all the theories in any
way concern the
evident that the observed value o, must be approved unconditionally.
any mean value found by repetitions, but

stands quite isolated, it must be accepted as the mean A,(Oi) in its law of presumptive
errors, and the corresponding square of the mean error ?..^{0i) must then be taken,
Even though
this observation does not represent
unchanged, from the assumed investigations of the method of observation.

a quantity that
If, in the latter case, o is an observation which directly concerns
can be determined theoretically
then
it
must
the
is,
in
as such,
all
same way
further
X (o)
(for
instance the
quite superfluous
as
sum
of the angles of a rectilinear triangle),
long as the theory
computations be replaced by the theoretically given
must be replaced by
and then
maintained,
is
zero, as the square of
the
mean
and
value;
it
in
on
the
test
the
error
errorless theoretical value.
The only
possible
meaning of such superfluous observations must be
correctness of the theory for approbation
we
or
rejection
(a third
result
is
to
impossible
when
are dealing with any real theory or hypothesis), or to be used in the criticism.
In such a test
which we
it
must be assumed that the
will call u, is identical with the
for o, consequently, that
M =
/ii(o),
and
mean
theoretical value corresponding to
value in the law
of presumptive
the condition of an affirmative result
obtained from the square of the deviation, (o
m)2
in
o,
errors
must be
comparison with XîOn).
The
67
equation
at
(o
m)'^
/ig
need not be exactly
(o)
but the approximation must
satisfied,
any rate be so close that we may expect to find ^2(^) coming out as the mean of
numerous observed values

44.
If
then
all
of (o
34.
Compare
u)^.
the observations o^ ... o
But generally
cases, the matter is simple enough.
under one or the other of these two
fall
the
observations
for the
by theoretical equations of condition which, separately, are insufficient

of the single ones.
in
Then the question
is
whether we can transform the
such a way that a clear separation between
can
the
will be
o,
connected
determination
series of observations
two opposite relations to the theory
made, so that some of the transformed functions of the observations, which must
be
be mutually free in order to be treated as unbound observations, become quite independent

while the
of the theory,
rest
are
dependent on
entirely
This can be done, and the
it.
computation with observations in consequence of these principles,
is
what we mean by
the word "adjustment".
For as every theory can be
fully expressed
by a certain number, n
m,
of theoretical
equations which give the exact values of the same number of mutually independent linear
functions, and as we are able, as we have seen, from every observation or linear function
of the observations, in one single way, to separate a function which
of these just
named
theoretically given functions,
system, represented by
functions, this system
is free
and independent
and which must thus enter into another
must include
all
those functions of the ob-
servations which are independent of the theory and cannot be determined by
it.
flach of the
thus mutually separated systems can be imagined to be represented, the theoretical system by
m,
the non -theoretical or empirical system by

all
represent
upon
observations and
all linear
mutually
functions of the same,
The two systems can be separated
in a single
way
tation of each partial system, by free functions, can occur in

It
its
is
due and the observations
retical system,
each with
the idea of the adjustment,
its
by means
theirs, in
and particularly the n
only,
unbound obser-
although the represen-
many ways.
of this
transformation,
to
give
the
such a way that every function of the theo-
free representatives of the
same, are exchanged,
theoretically given value, which, pursuant to the theory, is free of error.
the other hand, every function of the empiric system and, particularly,
tives
which together
and which may be looked
as a complete, transformed system of free functions, consequently as
vations.
theory
free functions,
remain unchanged as the observations determine them.
its
On
free representa-
Every general function of
the n observations \do\ and, particularly, the observations themselves are during the adjust-
ment
two univocally determined addenda: the theoretical function \d'o\ which

should have a fixed value D\ and the non-theoretical one \d"d]. The former \d'o\ is by
split
into
the adjustment changed into
D' and made
errorless, the latter is not
changed at
result of the adjustment, D'-\-\d"d\, is called the adjusted value of the function,
all.
The
and may
68
be indicated as
M,...M.
the adjusted
[du],
The forms
values
of the observations
themselves being written
of the functions are not broken, as the distributive

principle f{x-[-t/)
f{^)-\-f{y) liolds good of every homogeneous linear function.
The determination
mean
of laws
values
adjusted value
of errors
mean
the
is
of the
adjusted values
by repetitions.
is
analogous to
the formation of the
For theoretically determined functions the
value on the very law of presumptive errors; for the functions
we have the extreme opposite
that are free of the whole theory,
mean
limiting case,
values
represented by an isolated, single observation. In general the adjusted values \du\ are ana-
logous to actual
mean
values by a more or less numerous series of repetitions.

For while
==
X^[d"o\, we have Xîdu]
X^{D')^ Li[d"u]
k.[d"o], consequently
X^(\do\)
smaller
than k^ {do].
X^\d'o}
The
mean
the weight of the

45.
By
ratio
-fj^
is
analogous to
number
the
of the repetitions or
value.
"criticism"
we mean the
the
trial of
or theoretical
hypothetical
made in the adjustment, with respect to the mean errors of

new
determinations
of the mean errors, analogous to the determinations
observations;
by the square of the mean deviations, n^, will, eventually also fall under this. The basis of
the criticism must be taken from a comparison of the observed and the
adjusted values,
suppositions, which have been
the
[du].
for instance the differences [do]
According
to the principle of
34 we must expect
the square of such a difference, on an average, to agree with the square of the correspon-
ding
X-i
mean
error,
X^ [du],
[do]
[du])
X^ ([do]
we
get
X^ ([do]
but
as
[du]) =
[do]
[du] =
X^ [do]
X,
[d'o]
[dti]
D,
and
X,i[d'o]
(68)
which, by way of parenthesis, shows that the observed and the adjusted values of the same
function or observation cannot in general be mutually
([do]-[du] )^
We
free.
ought then
to
have
^^^>
X^[do]-X^[du]
on the average; and
the
number
for
sum
of terms of this form
of the terms, nota bene,
if
we must
expect the
mean
to
approach
there are no bonds between the functions [rfo] [rfj<];
but in general such bonds will be present, produced
by
the adjustment or by the selec-
tion of the functions.

It is
no help
if
we
select the original
consequently form sums such as

(o
and unbound observations themselves, and
u)^
{X,(o)-XAu)y
for after the
adjustment and
free functions
to
the
system
its
such as Oi...o.
change of the mean
errors,
Only one single choice
of the mutually free functions which,
in
is
u^ ... u are
immediately
the
not generally
safe, viz., to stick
adjustment, have themselves
69
the
represented
the
observations:
theoreticallly given
And
both before and after the adjustment.
the
as
differences of the
functions identically vanish, the criticism must be based upon the
= A,.
[ao\
- B'Y
[6'm]
to the theoretically free functions
{[ao]-AY
2 [ao]
the
'
>lj
[6'o]
series,
put down
to be
ought
us the advantage that
mean
may
=2
for the
We
the latter,
-Ay
the adjustment,
ratios
is
0.
for
cit.
u^
are only
its
same
men-
M)'^
^-A")
no decided contradiction between putting down
only, while
U^
we can
^2 [du]
(o)
Xi
we
(0
m)'
by the diminution of the denomi-
(u)
get no certainty for

in
it.
these two
sums
call "scales", viz. scales for
More generally we
of squares, conse^
measuring the influence
call
the scale for the function [do].
(72)
[do]
function- or
The theory has then
the scale sinks to
theory
loc.
special set of
the right side in this expression with the above
^2 (m)
/?2
scale
the deviations of the several
(71)
(O
between the corresponding terms
X^ (M)
A^[du] =
concerning the free functions.
m^Vn m.
k,{0)
of the adjustment on the single observation
the
we can
consequently,
which we dare not approve on account of the bonds produced by
then there
quently
If
series,
we know, being such a
{[b'o]-B')^
nators, can get the value n; only
I, (o)
whole
k^ (M)
at the smaller value
The
according to
satisfied;
as the expected value of each
have consequently
we compare the sum on
^2 (o)
1,
base the criticism on
[b'b'X^-]
tioned
on
.v by giving the functions which are free of the theory the
values as the observations.
If
[ft'i'yij]
be compared to the observations like v^ ...v
distinguished from v^
([g o]
(70)
'
to be strictly
error
of the proposition (66)
we can
observations from their adjusted values,

values as
([b'o-\-Fl_
m^'\/2(n m).
put down the expected value as n

But now we can make use
offers
of which
term of the
free
last-mentioned
m terms corresponding
[aak^]
w
m.
must be expected to be
Of course we must not expect this equation
sum
which
of the series
([ao\-AY
/ij
the second equation (46) the square of the
It
= B'
.. [6'o]
(\b'o\
[aw]
>lj
Only of these we know that they are
determines by the observations.
the adjustment
and the
functions
lowest limit
has had no influence at
observation
entirely
all;
0,
we
has
its
greatest
possible
value,
decided the result of the adjustment.

get just the reverse k2[du]
the whole determination
is
/{^[do],
i.
viz.
1,
But
if
e.
the
based on the accidental
70
and
value of the observation,
observations in
for
this
we
case
jo
u)^
get
Even though the scale has a finite, but very small value it will be inadmissible to deWe understand now, therefore, the
1.
pend on the value of such a term becoming
sum
of the
superiority
M)2
{o
n as a bearer of the summary
We may
part
also very well,
of the criticism,
the
to
sum
of the squares
m,
criticism.
on principle, sharpen the demand
for
adjustment on
(0
whole sum of the squares
that not only the
so
approach the value n

several
L, {0)
AA)
1^2 {0)
(0
the squares
of
=
- M)
m)21
the
must
hip)
from
but also partial sums, extracted
the
same, or even
its
must approach certain values. Only, they are not to be added up as

So much
units, but must be sums of the scales of the corresponding terms,
terms,
numbers of
we may
trust to the
may
ciously applied,
The sum
sum
of the squares
/--^
that this principle,
be considered as fully justified.

(o
of the
squares
M)2
possesses
h [0]
an interesting property which
other authors have used as the basis of the adjustment, under the
The above sum
the least squares".
value that
'
when judi-
name
all
of "the method of
of the squares gets by the adjustment the least possible
can get for values v^
. . .
i?
which
satisfy the conditions of the theory.
X, {0)
The proposition
is
that
the
[c"o^^
(66) concerning the free functions

[c"m],
observations,
= \d"'u\
\d"'o']
consequently
for all the free functions
by
just
shows that the condition of
putting
for
this
minimum
which are determined by
each v the corresponding
adjusted
value u.
The carrying out
46.
form
in
which the theory
is
of adjustments depends of course to a high degree on the
The
given.
those of the
general
theoretical equations will generally include
unknown
observations and, beside these, some
equations, which we just want
form, however,
is
unpractical,
are
eliminated,
so
that
it
is
to
and may
determine through the adjustment.

also
some
number than
This
be transformed through the
easily
We
always go back to one or the other of

easy to handle: either, we assume that all the elements
usual mathematical processes of elimination.
two extreme forms which
quantities, elements, in smaller
the theory
is
given as above assumed by n
linear equations
of condition with theoretically given coefficients and values,
adjustment by correlates; or,

an equation for each observation, consequently no equations of condition
between several observations. This is easily attained by making the number of the elements
we manage
as large
the
to get
{^m)
name
as
may
of elements.
be necessary:
This
sort
we may
for instance give
of adjustment
is
called
some values
of observations
adjustment by elements.
We
71
two forms in the following chapters XI and XII,
shall discuss these
whose rules
correlates
when
deduce.
it is easiest to
we
In practice
the adjustment by
prefer adjustment by correlates
by elements when
nearly as large as n, adjustment
is
first
is
small.
XL ADJUSTMENT BY COREELATES.
We
47.
suppose we have ascertained that the whole theory
^ A, ...
= C,
is
expressed in the
where the adjusted values u of the n observations

are the only unknown quantities; we prefer in doubtful cases to have too many equations
equations
[au]
[cu']
rather than too few, and occasionally a supernumerary equation to check the computation.
The
first
thing the adjustment by correlates then requires
made
corresponding to these equations, are
Let
[ao].,
by
this operation,
by
the
indicate the n
.. \c"o\
and
the adjustment
system of free
this
free
mutually free functions which we have got
[d"'o],
[^"o]
It is finally
according to (62), the
and according
(compare also (63))
accomplished by transforming back from
For this inverse transformation,
equations are:
= ^^^
hioi)
representatives of the empiric
then principally made by introducing the theoretical values
is
functions.
modified functions to the adjusted observations.
to (35)
that the functions \ao\ ... [coj,
is
one another by the schedule in 42.
beside these, imagine the system of free functions completed
let us,
other arbitrarily selected functions
functions;
into
free of
[ao]
>i,
+ ^jJ^ x,[g''o^]
(74)
dr
c"'
.^
gf
\,,,
As
we
the adjustment influences only the
have, because [au]
{n ^ ^ +
-
and
= A,
[c"u]
= C",
r"
+ ,,
w*
first
and Xilû]
terms of each of these equations,
= ... = l^ \c"u\ =
il'"
C"
+ Y:J^,'r~^ \d"'o\ +
Q"
0,
1
[g'of.Xîo,)
(75)
72
Consequently
[ao\-A
\c"o]-C"\
and
^.(o.)-^.(.)
Thus
for the
+ "- + [o-?Q7j) =
^:Nl[aai;]
computation of
all
the
observations,
each of these,
we need but use

in
coefficients
(78)
the diiferences between the observed and adjusted values
mean
of the several observations and the squares of their
the whole adjustment,
^^(''-")-
the
for
and the mean errors of the several
the values
theoretically
and thereby indirectly
errors,
given
and the two values of
functions,
namely, the theoretical value, and the value which the observations would
give them.
The
factors in the expression for
_
^"~"
o,-
<,,
m]~A
_
-
^'"
[c>'o]-C"
[aaX.y
which are common
its
The
name.
way
to all the observations, are called correlates,
and have given the method
improved values of the observations are computed in the easiest
adjusted,
by the formula
M.
By
sum
0.
X, (Oi){aiKa
+ c7K,n}.
(79)
writing the equation (78)
and summing up
the
'
[c"c"A,\
for all values of i
from
1 to n,
we demonstrate the
proposition concerning
of the scales discussed in the preceding chapter, viz.
_;,()
X,(o)
48.
respect
to
It
deserves to
the symbol
/ig.
all these equations are homogeneous with

makes no change at all in the results of the
the scales, if our assumed knowledge of the mean errors
be noticed that
Therefore
adjustment or the computation of
in the several observations has failed

if
only
the proportionality
is
by a wrong estimate of the unity of the mean errors,
preserved;
relative weights of the observations.
equations of the criticism
it
we can
adjust
The homogeneousness
correctly
is
if
we know
not broken
till
only the
we reach the
73
{[ao]-A)^
+ ([c"o]-C")^
(82)
_
It
the
h{o)
that
follows
[o-iif
L
correctness
\{aKa
c"K,..)HM^
= n-m \/2{n-m)
in this form, the
criticism
of the
hypothetical
the series of observations
is
unity
The
has originally been quite unknown.
"summary criticism", can only be used to try

mean errors, or to determine this if it
of the
special criticism, on the other hand,
can,
where
divided into groups, give fuller information through the sums
of squares
2^^^ =^(l-f;4),
taken
errors for
We
each group.
for
may,
for instance, test or
(83)
determine the unities of the mean
one group by means of observations of angles,
for
another
by measurements of
distances, etc.
The
ought
to
criticism
to be small,
has
also
other
means
at its disposal.
whose mean
particularly those
Thus the
differences (0
errors have been small,
u)
and they ought
change their signs in such a way that approximately

(84)
k^
(o.)
such series of observations as are
for natural or accidentally selected groups, especially for
nearly repetitions, the essential circumstances having varied very

If,
essential
little.
ultimately, the observations can be arranged systematically, either according to
circumstances or to such as are considered inessential, we must expect frequent
n.
and irregular changes of the signs of
If not,
we
are to suspect the observations of
systematical errors, the theory proving to be insufficient.

49.
It will
not be superfluous
to
what has been said
present in
the
form of a schedule of the
working out of the free

functions. We suppose then that, among 4 unbound observations o,, 0^, O3, and o^, with
the squares on their mean errors -^2(01)1 '<2(''2)i îi'^s)^ ^^^ Â"^*)^ there exist relations
adjustment by correlates
here,
also as to the
which can be expressed by the three theoretical equations
The schedule
aû^
-\-
aû^
^î'"'3
"=
^4**4
-S
a,M,
-|-
\hu\
=
=
6j,
+ 62M2 +
\cu'\
== C.
r-iM, +C.^M2 +C3?/a +C.Û^
\au\
is
aîi^
-\-
then as follows:
10
74
The given
A
0,
/},(,
Free functions
a,
ir
c'
K
K
0,
ÂOt)
0,-M,
M,
;,(o,-M,)
[60]
^2(0,)
c'
[co]
\h'o\
=3
[c"o]
[c'o]
[b'b'k]
ib'c'X]
[caX\ [cbX] [ccX\
[C'6'yi]
[C'c'/i]
[caX\
K,
\b'o]
[c'6'yi]
By
\_c'b'X\
[aaX]
{b'b'X\
A
[ao]
[6'o]-g
[aaX]
ib'b'k]
[c'c'K]
the adjustment properly so called
Ui
summary
KliaaX^-]
= C,
C'-r'B
e'{
c'i-r'b'i
[c'o]-r'[6'o]
[c"o]
{ccX\-r\ca^
[c"c"X]
[c'c:x\-rVh'x\
we compute
n%
-\-c'iKci<)Xi{o,)
c"^
i/n
+ K\,
[b'b'k^-\
+ ifV-
we ought
[c'-c'-yi^]
LM
3l/6.
A, \hu\
B, and
compute [r^]
and
it
is
useful
to
Moreover
m,, m^, m^,
u^.
further to
with the values we have found for
add a superfluous
C"
criticism
In order to get a check
[cm]
of:
== {aiKa^b'iKt,
I
for the
[cVX]
c'i
\c.bX\-p\caX]
for proof
and summary criticism
C = C-rA
= d yai
= [c6] Y \a6\
Ydo\
= B-Â
= biÛi
= [H-/?M
=
Sum
[c"o]-C"
Kc" ==
computed by means
Oi
and
[c"c">i]
'
free functions are
b'i
{o,-uy:X,{o,)
"
'
[aa^]
B'
as proof.
Criticism
[baX] [bbJi] [bcX]
= [baX\
The
1-;,{m,)
c'
[aaX\ [abX\ [acX\
Con-elates
>!,(m,)
4
[ao]
/?
Scales
Adjusted values
C"
theoretical equation, for instance [(a-{-b-{-c)u\
= A-\-B-\-C,
through the
75
computation of the
free
which
functions,
correct
is
such a superfluity leads to
if
only
identical results.
50.
It
is
deficiency in the adjustment by correlates
that
employed as an intermediate link in a computation that goes beyond

as
good
The method
it.
We
the criticism on the same, but no farther.
are often in
want
the adjusted values
of
with determinations of the mean errors of certain functions of the observations;
must be made
such problems the adjustment by correlates
solve
to
The simplest course
is,
after
think, immediately
is
and
the determination of the adjusted values of the several observations
as
far
cannot well be
it
in order
in a modified form.
drawing up the theoretical equations of
condition to annex the whole series of the functions that are to be examined, for instance
.
[rfo],
. .
so
and include them in the computation of the
[eo],
must take
order
the
that
theoretical
the
of
the
functions;
The functions
finished.
necessary to
mention
mean
And
error.
In doing so we
free functions.
not to mix up the theoretically and the empirically determined functions,
care
[d"'o],
it
must
operation
.
|
eô]
unconditionally
made
not
are
others
free
the
give
which are separated from these
the
precedence to
the treatment of these
till
it
is
is
quite
scarcely
remain unchanged by the adjustment both in value and in
at last the adjusted functions [du],
B\
are determined as linear functions of A,
C",
.[eu], by retrograde transformation,"
[d"'o],
[eô].
Example 1. In a plane triangle each angle has been measured several times, all
measurements being made according to the same method, bondfree and with the same
(unknown) mean error:
for
angle
has been found 70
jd
1)
A
C
The adjusted values
measurement
<>
ti
50
.1
60
0'
5" as the
2.
(Comp. example
of
6 measurements
"
10
2"
..
15
the angles are then 70, 50, and 60,
for
and
0-5, 0-3,
the
*'
mean
error for single
0-2.
Five equidistant tabular values,
42.)
12, 19, 29,
have been obtained by taking approximate round values from an exact table, from
which reason their mean errors are

successive hypotheses that
all
dilierence is exactly
be used for
schedule,
first
all
the
\/-^.
The adjustment
the table belongs to a function
and the hypothesis of the second degree
two
mean number
V/SOO == 17"3, the scales
Example
41, 55,
t)
0'
is
of the
2, in the following schedule
term
corresponds to
the
first
performed under the

2^,
and
1"'
degree,
by the special hypothesis that the
varied
marked
four modifications of the problem, so that in

first
is
3"",
2''
The same schedule may

(or).
the sums to the right in the
modification
only,
and the sum of the
terms to the second modification:

10*
76
^2(0)
12
19
29
41
55
12
1
1
1
12
1
12
1
1
J*
77
The deviations
are evidently too large {o
is
1'4,
+2'4,
due to the use of round numbers; the sum of the squares
to be
220-8 instead of 3
is
2-2, 1-0, +2-2)
also
1/6,
consequently, no doubt, an over-adjustment.
The
for
M,
and
nowhere reach
-m
29-2
19-4
-0-4, -0-2,
0-4,
and may consequently be due
a,
the squares
4-8 instead of
well.
very
agrees
(3-4)2, (4.4)2^
(5.4)2^
Example
(6-4)2,
^nd
constant
Indeed,
41-0
11-8
9-8
7-8
deviations o
also
T^'
and J*
0,
2,
gives
differences:
its
11-6
The
J^=0,
special adjustment of the second degree,
13-8
0-0,
to
+0-2
the use of round
subtraction
from
of 0*04
from which the example
(7-4)^,
numbers; the sum of
_
3 1/6
Between 4 points on a straight
3.
54-8
line the
would lead
m,
to
taken.
is
6 distances
"si
are measured with equal exactness without bonds.
we
notice that every scale
millimeter
scale,
which
is
|.
It
is
recommended
after
displaced
By adjustment we
the
work the example by a
actually to
measurement
of
find for instance
each distance in order to
avoid bonds.
XII.
51.
correlates
most frequent
for
by
Though
as well
as
cases,
ADJUSTMENT BY ELEMENTS.
every
problem
by elements,
in
adjustment
be
may
the difficulty in so doing
is
solved
where the number of equations of condition
adjustment by elements, and this
is
therefore
employed
in
both ways, by
often very different.

is
The
large, are best suited
far oftener
than adjustment
correlates.
The adjustment by elements

tion is represented by one equation
linear functions of
unknown
requires the theory in such a form that each observa-
which expresses the mean value
values, the ''elements", x, y,
z:
/i
(o)
explicitely
as
78
^liOi)
P,x-\-q^y-\-
... -{-r^z
(85)
where
the
p, q, ...
are
theoretically
observations
All
given.
are
supposed
to
be
unbound.
The problem
is
then
first
to
determine
the
adjusted
values
which each of these equations (85) which we

observations", gives the adjusted value u of the observation.
X,
tj,
...
z,
after
Constantly assuming that ^^(0)
is
known
system (85) deduce the following normal equations:

pkx{0)
.
h{o)
for
of
these
elements
call 'êquations
for the
each observation, we can from the
79
(X
everywhere used
we
add,
but
for /is(o));
if
we multiply
get
\(hp-\-kq-\-
these again respectively by
lry'
/*,
k,
and
0,
Âo)
that
is
hpi
so
the normal
that not only
consequently,
But
A^2,
4-
Irt
equations, but the very equations
be written with
all
m1
or a smaller
number
for the observations can,
of elements.
the system of functions represented by the normal equations
further,
The
is
free
we can
get by eliminating the
elements x, y, ... z from the equations of the observations (85).
But elimination of an
of every one of the conditions of the theory.
element, say for instance x,
leads to
the
latter
functions
p,ii,(Oi)
jOiA,(o,),
and among the
must be found the functions from which not only x but

eliminated, and consequently the conditional equations of the
linear functions of these
elements are
other
But
it
is
all
the
theory.
easily seen that the functions
and
PiXi{ot)pkX,(Oi)
^,{0)
The
are mutually free.

cularly
latter is the left
aimed at the element x;
coefficient of
element;
it
it
is
hand
side of the normal equation
has thus been proved to be free of
therefore in the adjustment be
we have been
is
parti-
formed by multiplying the equations (85) by the
PP
and has the sum of the squares
in each,
which
all
as
the
coefficient
of this
the conditions of the theory, and must
computed by the directly observed values, for which reason
able in the equations (86) to rewrite the function
po
as
In the same
Âo)
way we prove that
the other normal
all
the elimination from (85) of
its
equations are free of the theory, each through
particularly prominent element.
While, in the adjustment
by correlates, we exclusively made use of the equations and functions of the theory, we
put
cally
all
these aside in the adjustment by elements, in order to work only with the empiri-
determined functions which the normal equations represent.
The
of the elements
coefficients
normal equations
in the
The
it is
the diagonal line,
in
element to which the equation particularly
whose equation
is
it
the
it
will
be seen,
for
easy to state.
coefficients
each of them
as
and each of them has a significance
arranged in a remarkably symmetrical manner,

the problem which
are,
square
occurs.
of the
We
have
mean
which are respectively multiplied
refers,
error
are as
sums
for that
l,{o)
the
of squares all positive,
and
function of the observations in
for instance
P.
by
= h po
J
80
The
qp
in
ys
equation.
coefficients outside the diagonal line are identical in pairs, the coefficient of x,
particular equation,
They show immediately
to be mutually free;
if
we must have
52.
these
sums
special
is
for
the same as the coefficient of y,
if
If
now
in
a;'s
some of the functions
instance x's
pq
Âo)
function
should happen
is
to be free of
ys
function
0.
the elements have been selected in such a convenient
of the products vanish,
particular
way
that
and the normal equations consequently appear
all
in the
form
pp
(87)
81
in
many
first
special cases radical changes can lead to very beautiful solutions (see 62).
thing, however,
is
that the elements are eliminated one by one,
way
computation of them
method which may
to secure a
is
in such a
be selected
is
prepared,
and moreover, constantly,
in
and
this
The
must
so that the later
such a way that freedom
attained.
This can,
we commence
if
x
and which
viz.
PP
remembered that we are not allowed
X can then be
k^{^)
for the
the other
'
(90)
computation of x,
normal equations; but
to multiply the latter by any factor.
it
must be
The equation
x-\-fy-{-
..
-\-(oz
PP
(91)
PP
X
The functions
in the other equations
(q-<pp)6
(rcDp)
X
po
become, by this means, not only independent of x but also free of
or of $,
p_
q^ -jpp
The equations which

same
multiplied
X vanishes when the
that
is
written
f
where
po
subtracted from
products are respectively
particularly refers to x.
+ pr
T
y+
put aside to be used later on
will be
by such factors,
instance by eliminating the element x, be attained
for
The normal equation which
in the following way.
for
be always applied;
in
hko)
pq
X
PP
0,
etc.
a double sense have been freed from
characteristic functional form as the
normal equations had.
for
If
x, get exactly the
we
write
(92)
30 that the equations for the observations
ftf+5.'y+
we not only
get,
as
we
Wo]
X
see at once,
become
.+r\z =
u.,
82
but also
83
With
the equations
elements
these
the adjusted
for
become
and
+ rrc =
pp
we want
tions of
several
observations
(99)
M,.
errors
q'q'
+1^
p,'
If
mean
for the squares of their
of the
values
+ .---+rf
(100)
^2(M,)-
compute adjusted values and mean errors for the original elements or functhe same, the means of so doing is given by the equations of transformation
to
+ fy-\-
or by (90), the first equation (95)
and the
only the original elements
...
x, y,
+(oz
4-
in
(o'z
being identical with (101).
last of (98),
For not
z are easily computed by these, but also the coef-
ficients in the inverse transformation
=
=
a;
y
z
Now,
operations
and
we
if
is
get an expression for
mean
X^(F)
for
of
9i
54.
In
that
5',
we
the theory
the
summary
sum
(102)
...
/,
then by obvious numerical
get
we want
6'^
the
the
Ît fPi^
qq
c^^
computation of
equations
^^(m,) for
= *Pi
r"
which
is
observations,
computing
their
î
<-'?;;
of
the adjustment by
elements we must not overlook the
minimum sum
of squares for the benefit of
criticism as well as for checking our computation.
(0-
many
of observations,
r" are quite superfluous for the coefficients of (95).
proposition concerning the computation of the
the
...+ ^C
it,
transforming
coefficients by (92), or
we remember
error
4-7-C
c.
a^
special criticism
we may take advantage
but
,+
a given linear function of x,
for the square of its
If
...
f+3j-)-
to
approach the value
only the "elements, without being obliged to
know
n m,,
We
are able to
compute
as
soon as we have found
the adjusted
values for the separate

11*
84
And
observations.
ment, but
for
this
any values whatever of the elements.
formed elements,
f,
must be computed by
From
this
we
computation can be performed, not only
...
jy,,
for the observations
get
61-2
Âo)\
we here
The values
(99)
{o-vf
If
Ci-
substitue for
i-r-
for
q'O
'?!
the legitimate adjust-
It is easiest to
show
this for trans-
corresponding to these
85
for the several
formed values of the coefficients in the equations
The sums
always be dispensed with.
may
observations,
nearly
and the sums of the products.
of the squares,
/i
must be transformed; but they
Mj
transformed
the
first
of
by their help we find values and mean errors
transformations, and
the
r'.
indirectly
also
the computation
for
Only
r^.
q'.
we cannot escape
criticism,
special
but
ones,
the
The adjusted values
ones.
original
for the elements,
the observations can, consequently, also be computed without any knowledge
M of
are in themselves sufficient for the determination of
the
of
^^ (m.)
considerable
often
^2 (")
consequently for a
work which
is
necessary
for
the purpose.
For the summary criticism by
we have
dispense with
seen,
We
of the elements.
the case
is
= n m^V2{nm),
]{ot
hio)
0,-,
Mi,
minimum sum
computed much more sharply, and always with much greater

and /^.^(o), than by the formulae (105), (106), and (107).
Add
scales
but that the very deviations
0,
m,,
when they
when
of squares
certainty, directly
on
to this, that the special criticism does not exclusively rest

"
means
ought, however, to restrict the work of adjustment so far only,
generally
by
even, as
the after-computation of the several observations by
either very difficult or of slight importance, for this
is
we can
/i^
(m,)
and the
are arranged according
/j (0)
to the
the
more
or less essential
Systematical errors, especially inaccuracies
criticism.
theories,
errors;
will
2"
know, even be used

56.
havn 1879, and

the
of signs, will
frequent changes
form of a
multiplied
Crelle's
series
which)
errors
Journal
how we ought
vol. 94),
factor,
i.
an
or
to try to
and
e.
J. P.
may, we
where the theory gives the observations in

(infinite)
number
of terms, each term being
adjustment
gets the special task of indicating
by
how many
Formula (107)
this.
= m^ l/2(w w).
J
errors
Gram, Udjevningsraekker, Kjaben-
ro
h (0)
mere absence of
improve the defective theory.
element, and where consequently
the criticism
means of doing
(o-u)
[
hypotheses
which might remain hidden by the check
terms of the series we are to include in the adjustment.
nishes us with the
in
and such progression in the
with an indeterminate
elements must be employed,

(or
defects
some circumstance,
adjustment (compare Dr.
unknown
by an
as a function of
disclose
ill
to indicate
series of
By
or
themselves in the surest and easiest way by the progression of the
betray
regular variation in
according to
circumstances of the observations, are even a main point in
fur-
For the
terms in the
series,
which
here indicated by
is
an element, consequently to one of the terms of the
we take
unity;
all
into
this,
series of
adjustment.
consequently, becomes more favourable
<
the terms for which
If
1.
we
retain
out
terms for which
ro
>
we make
if
by about a
we
leave out
any terms which essentially
under this rule, the adjustment becomes an under-adjustment;

leave
For each term
equation of criticism is diminished
the right side of the
the result of the criticism,
correspond, each of them, to
2",
if,
fall
on the other hand, we
ourselves
guilty
of
an over-
upon
as
mutually
adjustment.
Example 1. The five-place logarithms in

unbound observations for which the mean error
The
place.
"observations",
table
is
are
constantly
looked
l/J^
of the fifth
log 795, log 796, log 797, log 798, log 799,
decimal
log 800, log 801,
log 802, log 803, log 804, and log 805, are to be adjusted as an integral function of the
second degree
log (800 -^t)
x'-{- y't
we
In order to reckon with small integral numbers,
2-90309
+ 0-00054
t,
zt^.
-\-
subtract before the adjustment
both from the observations and from the formulae.
Taking 0-00001 as
our unity, we have then the equations for the observations:
2 =
2 =
=
=
=
1
1
=
=
1
1
1
From
this
we
get
36
420
=
=
540=
is
by + 2bz
4y-{-16z
x 3/4- 9z
X 2y
iz
x
-{-
a;
1/+
'^z
a;
x-{-\y+ \z
= x-^2y-\= + 3y+
a;
=: X -^
iy
-\-
4z
9z
I6z
= + 5y + 252.
a:
156, and the normal equations
The element y
132a;
0y+
0a;+1320y+
1320a;
0/
1320^
Oz
23496s.
consequently immediately free of x and z, but the latter must be
made
87
free of
it
one another, which
from the third.
is
done by multiplying the
The transformation
substituted for x, and
we have
first
equation by 10 and subtracting
into free functions then only requires $
= x -j-lOz
88
Both the checks agree: the sum of squares

of the scales
11
is
12
is
x 0-7836 =
sum
9-40, and the
3.
ought to be noticed that the adjustment gives very accurate results throughout
with the exception of the beginning and the end. The
greater part of the interval,
It
the
exactness, however,
is
2.
Example
is
supposed
Co-|-Ci cos
r+s,
theory, consequently, has the form
The
to be the object of observation.

o
We
not greatest in the middle, but near the 1" and the 3'* quarter.
finite, periodic function of one single essential circumstance, an angle F,
sin F-j-Cj cos
2F+ s^sin 2F+
assume that there are n unbound, equally exact observations
whose
Show
difference is constant
and
-,
n
F=
for instance for
0,
for
a series of values of F,
60M20M80, 240, 300.
that the normal equations are here originally free, and that they admit of an exceedingly
simple computation of each isolated term of the periodic series.

Example 3. Determine the abscissae for 4 points on a straight line whose mutual
distances are measured equally exactly, and are unbound.
Example
and 60).
Example 4. Three unbound observations must, according
3,
elements, so that
=
=
0,
0,
The
jcS
Aj(o,)
xy,
,^5,(05,)
most simply by the
of both elements, the
known a
for
common method
2'"'
degree,
we
o.,
where the middle equation has
a2) + 26(02 6) =
ab) + 26(03 =
2a(o^
2^
+
,
.,
approximate values
= +
2af
6f
+ ^'y
2h7j,
(4a''
6'^)
rj,
be produced
= +
double weight.
2a(o,
consequently not free of
may
get the equations of the observations:
ah =
O2
h^ =
still
This
of presupposing
We
oâ^
is
1
-1
b for y, and considering the corrections f and; tj

^
a
therefore put a;
f < ^"^^ 2/
'J-
x and
be the elements of the adjustment.
Rejecting terms of the
=
=
depend on two
to theory,
the linear form.

theory, therefore, does not give us equations of
in several ways,
to
(Cmp. Adjustment by Correlates,
The normal equations
+ 2i)f+2a6)y
2ah f + (46 + 2a2)
jy
are:
but we find
62(52o
2a.6os
+ a2o8)
,
'
(Svfppa^{bô,2abo^âHs)
''2(^)
+ 26^
4(a2 +
a'
62)2
,,
2a
+ i
89
For the adjusted value u^ of the middle observation we have
(a^
If
+ fe^M, =
abô,+{a'-{-b*)o,+a'^bo,,
we had transformed the elements (comp.

f =
jy
^,{u,)
i-
^^J^-
62) by putting
aC ^0
='b^-\-au,
or
we should have obtained
free
i(i
+ + o,
normal equations
+ 2a60j + 6*03) 2(a2 + 62)2 _

=
2( a6o, + (a2_62)o2 + aJo3)
i{a^J^b^)î:
2(a2oi
If
we had placed
squares as a
The conditions
of
minimum
5,1/
^ -^j^ =
rfmin
The
a and b
immediately that
(o,-a^)(o,-b^)
OjOg
Still better is
it
to
0^ =
introduce
{o^
s^
bô^
is
not very
a^-\-b'^,
difficult.
We
see for instance
aô.^.
s^)a-\-o,J) =
('-H^)
must then be
A
0.
by which the equations become
0,
s*-sHo,-ô,)-\-o,o,-ol
s^
of
{0,-abf
consequently,
two values of
sum
j.\ /.
2abo.^-{-
oâ-\-(o3s^)b
If the errors in 0,, 0^,
of the
== min.
i,ow =
ab)a->r(o.^b')b
7\
{o^
solution with respect to
or
principle
are:
d min
b'')û.
might have been founded on
- )2 + 2 (02 - ab)^ + (03 - *")'
(0,
absolute confidence in the adjusting
a solution
minimum,
2(a''
=('-i^)+";-
and O3 are not
large, oô.^
0-^
must be small; one of the
small, the other nearly equal to Oi-^-o.^; only the latter can
be used.
12
90
Further,
91
which give the simple
21ogx
results
logo.
^,(logx)
(^,3^j log-^
^^.(J^^.).
This solution agrees only approximately with the preceding one.

for a
for
moment
that, in this
way, we might do without the supposition of approximate values
the elements, but this
from
far
is
the
repeat
^b
y
it
is
are small in proportion to
o,-
errors
and
/,
agree,
and we must
get the theoretically
necessary, but not a sufficient condition, that x
only
mean
of the
squares
as also
Unless the exactness of the observations
small.
are
errors of
then
the
till
adjustment
And
For the sake of the weights we
being the case.
must, with the same care, demand that a and x,
values.
might seem
It
o,
is
correct
also so great that the
and
mean
the laws of errors of the logarithms cannot
itself,
be considered typical at the same time as those of the observations themselves.
=2,
the radius
y^
/,
of four
^ 20,
and without bonds: Xj
errors
and x^
The co-ordinates
5.
Example
mean
because
obtain
we cannot use the common form
+ {y-b)' =
of the observations
Xi
-\-
the equations are not linear

of the
of
elements,
we
Ayi
These equations
know
â
1'^!
for
yt
first
we must
each point, writing for instance

b
-\-
r sin
we must work by
which the
In order to
adjustment by elements,
F,-
successive corrections
A a, A
approximate system can be obtained
For the theoretical corrections
Axt
and
Ayt
6,
by
of the
= A a Ar
= A6 +A
cos Vi
-{-
>
sin
F,-
A Vi
+A
F.-
for the observations lead us to
r sin
F,-
r cos
F,-.
a system of seven normal equations.
makes
it
possible
for
us immediately to discover
the artifice.
that every transformation of equally well observed rectangular co-ordinates results
in free functions.
The
radial
and the tangential corrections
A Xi cos Vi A yt sin
A Xi sin Vi A cos
-\-
and
3,
the "method of partial elimination" ( 61) these are not difficult to solve, but here the
simplicity of the problem
We
get by differentiation of the above equations
Axi
By
x^
r\
necessary for
r cos F,
ordinary computation from 3 points.

co-ordinates
= 18;
y^
of the equations
add a supplementary element, or parameter, F,
Ar, AF,
x^^ 16,
embraces more than one observed quantity besides the elements.
the separation
As
a circle are observed with equal
in
In the adjustment for the co-ordinates a and b of the centre and
4.
(x-a)^
it
points
= 10;
y.
y,-
F,-
F,-
=A
=A
w,-
<<
12*
92
can,
quantities,
mean
here be taken directly for the
consequently,
values
of corrections of observed
and as only the four equations
= A a sin A 6 cos F, A Fj
<,
contain the four corrections
F,-
F,
of the parameters, they can be legitimately reserved for
the successive corrections of the elements.
In this way
A i = A a cos F,+ A i sin F^ + A r

with equal mean errors,
^.în)
kî^)
kîiy),
are the "equations for the observations"
of this adjustment, and give the three normal equations:
In
+ Ab [cos F sin F] + A r [cos F]
= A a [cos F sin F] + A 6 [sin^ F]

= A a [cos F]
Ab [sin F]
the
case
fourth point
6 = 10;
= A a [cos^ F]
[A n cos F]
[A n sin F]
[Aw]
lie
+ A r [sin F]
+Ar
-\-
special
on the
under consideration, we
circle
with r
= 10,
easily see
that the
4.
first,
second, and
whose centre has the co-ordinates a
= 10
and
the parameters are consequently:
Fi
00'0,
F^
537'8,
Fg
F,=
and
1350'0,
21652'2.
= 2'9290 and y3 = 17-0710,

A = and A 3 = 0-1005;
For the third point the computed co-ordinates are: X3
= +0-0710 and A^g = 0-0710,

= and A = The "equations the
differences A
= 0-0000
1-0000 A a + 0-0000 A 6 + 1-0000 A
0-0000
0-6000 A a + 0-8000 A 6 + 1-0000 A r =
0-7071 A a + 0-7071 A 6 + 1-0000 A r = 0-1005
0-8000 A a 0-6000 A & + -0000 A r = 0-0000.
A
consequently,
all
other
0,
<3
a;3
for
0.
yt
a;,-
observations" are
The normal equations
are:
For cliecking
2-5000 A o + 0-4600 A +
A = + 0-0710
0-4600 A a + 1-5000 A 6 + 0-9071 A r = 0-0710
= 0-0929 A a + 0-9071 A 6 + 4-0000 A r = 0-1005.
6
E
By
Ar
elimination of
we
0-0929
get
A a + 0-4390 A 6
B = 0-4390 A a + 1-2943 A 6
eliminating A b
A = +2-3490 A a
+ 0-0733
= 0-0482
==
2-4978
and by
From
jK,
B, and
Aa
The checks
+0-0896.
we compute
+0-0381
A6
0-0501,
and
Ar =
0-01465.
are found by substitution of these in the several equations.
The 4 equations
93
Aw,:
for the observations give the following adjusted values of
= +0-0234, A M2 = 0-0319, A ^3= 0-0770, and

u)
=
of squares
consequently
(here
Aw,
Ami
the
(o
sum
(8
=
For
this,
(0-0234)2
is
7)/} 2)
+ (0-0235)^ + (O-OISI)^ =
(0-0319)2
= 00151;
0-00235.
by the equation (108), we get
0-01010
0-00271 0-00356 0-00147 =
0-00236
as the final check of the adjustment.
The 4 equations
AF,
for
AF5,
-4-17'2,
give us
tt
AF3
+20'8,
Thus, by addition of the found corrections
F,
we have
the
017'2,
F,
5328'6,
whole system of elements
definitive values.
In both cases
Ax
0-0232
+0-0191
+0-0166
0-0123
The sum
F3
9-9499,
and
V,
approximation,
21630'6,
+0-0002
+0-0002
+0-0257
0-0166
+0-0320
0-0234
0-0001
0-0090
+0-0152
comp.,
0-0000.
agrees with the above value,
hypothesis
may have been
approximation by means of our
will try the next
are:
0-0000
+ (A y)^] = 0-00236,
first
values of the co-
At
0-0232
this
they are not the
if
differences, obs.
An
Ay
of the squares, [(Aa;)2
who
13457'1,
the next
for
which indicates that the approximation of

Indeed, the students
10-0381,
the approximate values,
formidw, the resulting
to
= 21'6.
AF,
we must compute by them the adjusted
ordinates, according to the exact

Point
9-98535,
-2'9, and
sufficient.
final differences,
will, in this case, find only small corrections.
From
the equations A, B, and R, which express the free elements by the original
bound elements,
A a, Ab, Ar,
we
easily
compute the equations
for the inverse
trans-
formation:
A a = 0-4257 A
A6 = 0-1444 -^ + 0-7726- B
A r = 0-0228 A 0-1752 B + 0-25
By
these,
any function of the elements
function of the free functions A, B, and
for
E.
a given parameter can be expressed as a linear
R; and by
yi^C^)
= 2-3490
>i
5,
X^(B)
= l-2di?, X^,
94
and ^2
(J?)
= 4^2,
the co-ordinates
X^(x)
'i2(y)
the
mean
x and y
Thus the squares
error is easily found.
of the
mean
errors of
are
= {2-3490 0-4257 + 0-0228 cos F)" + 1-2943 (-0-1752 cos F)

+4(0-25 cos Vf}l^
= {2-3490( 0-1444 + 0-0228sinF)'' + l-2943( 0-7726 0-1752 sin F)" +4(0-25 sin
here very
Only the value k^ = 0-00236, found by the summary criticism,
(
F)''}>i2
is
uncertain.
SPECIAL AUXILIARY METHODS.
XIIL
We
57.
have often occasion
adjustment by elements; and

shorten
it
tations
lie
as
much
to
means which
as possible, even by
near enough to tempt the
computation, but not so
much
use the method of least squares, particularly
sometimes requires so much work that we must
this
many who
are not quite lawful.
are soon tired by a
Several tempsomewhat lengthened
by looking for subtleties and short cuts.
the method was formerly considered the best solution
try to
among
And
as,
moreover,
other more or less good
not the only one that was justified under the given supposition,
it
is
no wonder that
it
has come to be used in many modifications which must be regarded as unsafe or wrong.
After what we have seen of the difference between free and bound functions, it will be
understood
stand out
values.
that
much more
as to
And
of transgressions
the consequences
some
the
in
clearly
mean
errors
extent justly more
against
the method of least squares
of the results than in their adjusted
importance
is
attached to getting tolerably
correct values computed for the elements, than to getting a correct idea of the uncertainty,
the lax morals with respect to adjustments have taken the form of an assertion to the
effect
we
that
especially if
can,
we take
within
this
care that a
minimum.
This, of course,
by stating
all
is
domain, do almost as we
without any great harm,
of squares, either the correct one or another, becomes a
sum
wrong.
the artifices which
like,
even
In a text-book
experienced
we should do more harm than good
computers have allowed themselves
employ, under special circumstances and in face of particularly great

a few auxiliary methods will be mentioned here,
so,
when simple caution

58.
When
is
which are
difficulties.
to
Only
either quite correct or nearly
observed.
methodic adjustment was
first
employed, large numbers of figures
were used in the computations (logarithms with 7 decimal places), and people often comas an unavoidable evil, when
plained of the great labour this caused; but it was regarded
the elements were to be determined with tolerable exactness.
however, to get on by means of a
much
simpler apparatus,
We
if
can very often manage,
we do not seek something
95
During the adjustment properly
which cannot be determined.
work with three
able to
fied:
both
we
the elements
respects
Often
formation.
figures.
seek
can be
it
must be small and
difficult
free of
enough
The condition
we ought
satis-
and in
time by appropriate trans-
we
that
itself
be
to
two conditions are
one another, or nearly so;
to protect oneself in
which would have made the work easy.
artifices
freedom of the elements.
so called,
this ideal presupposes that
only through the adjustment
is
it
But
learn
know
to
the
This applies particularly to the mutual
of their smallness
is
satisfied, if
we everywhere
use
same preparatory computation as is necessary when the theory is not of linear form.
By such means as are used in the exact mathematics, or by a provisional, more
the
allowable adjustment,
or less
a set of values
v^
several elements
we
corresponding to
get,
the
several
observations
o,
o,
.Vn which are computed by means of the values x^^

z^ of the
.z, and which, while they satisfy all the conditions of the theory with
.
any rate considerable exactness, nowhere show any great deviation from the
and x
?,
corresponding observed value. It is then these deviations o,
x^... which are
or
perfect
at
made
the object of the
observations and elements themselves
adjustment, instead of the
we know, they have mean
with which,
error
When
common.
in
in
a non-linear theory
the equations between the adjusted observation and the elements are of the general form
Ui
F{x,
.z),
they are changed into
'-^'=(f)(^-^)+--by means of the terms of the

If the
approximation.
first
degree in Taylor's series,
or by
(109)
some other method
of
equations are linear

Ui
we
+ (f)(^-ô)
= PiX +
i-iZ
have, without any change, for the deviations:

Ui
No
tions
0,
Vi =
Xa)^
'Piix
...
show only two
far as that goes,
putation, and
significant
figures
ô)-
(110)
whose devia-
and then computation by 3 figures
it
a considerable extra-work in the preparatory
certainly requires
must not be overlooked that computations with an exactness
decimal places will often be necessary in this part

the utmost care as a function of
mean
errors, so that
a;
of the deviations
o,-
z,,
lest
is,
as
is
especially
Vi
ought
to be
of
com-
many
computed with
any uncertainty in this computation should
we dare not put
This' additional work, however,

list
(^
sufficient for the needs of the adjustment.
The method
The
''
special luck is necessary to find sets of values, ?;,,... a;^, ... ^n,
Vi
increase the
X^
{ov)
k^
(o).
not quite wasted, even when the theory
is linear.
will, by easy estimates, graphic construction, or directly
96
by the eye, with tolerable
slips of the pen,
observations,
ment.
certainty
The preliminary
lead
etc.,
observations
equations,
to rely too confidently
linear form or into equations
equations ,
F(x,
save a whole adjustment;
w,-,
after the solution of the
terms
higher
and
provisional,
in
for
o,-
vt
Where
it
must be repeated
agree with the values through Ui
On
which ^^
the adjustment
the values
until
Vi
(o,)
is
generally
of
,-,
justify the rejection of
to
may
by
got
is
as
be regarded
only
direct
as
computation,
known
till
we
get a sufficient
where the observations represent
only as functions
unknown
of the
probabilities,
quantities which
to give us.
is
The form
59.
the actual
possible,
in the linear equations of adjustment.
This, for instance, is the rule
the adjustment itself
elements,
If not,
the whole the adjustment ought to be repeated frequently
approximation.
for
is
employed, and with the same degree of accuracy

way only can we see whether the approximate system
series.
Taylor's
normal
on the transformation of the equations into
elements and values has been so near to the final result as
the
the
be
to
In this
as in the computation of ,.
of
of deviations
ought
z)
may
of
the adjust-
to get into
under the criticism after the adjustment.
falls
In computing the adjusted values, particularly
we ought not
gross errors in the series
which must not be allowed
rejection of such
ultimate rejection, however,
to the discovery of
and in particular the selection of its system of

by purely mathematical considerations as to the
of the theory,
determined
rule
and only exceptionally by that freedom between the elements

which is wanted for the adjustment. On the other hand it will generally be impossible
to arrange the adjustment in such a way that the free elements with which it ends, can
elegance
all
of the
formulae,
be of direct, theoretical interest.
mentioned
reasons
in
the
only the theory pays so
middle course, however,
always desirable, for the
is
foregoing paragraph, and very frequently
much
respect to the adjustments that
it
it
is
also possible, if
avoids setting up, in the
same system, elements between which we may expect beforehand that strong bonds
exist.
will
Thus, in systems of elements of the orbits of planets, the length of the nodes and
the distance of the perihelion from the node ought not both to be introduced as elements;
for
a positive change in the former
will,
in consequence of the frequent, small angles of
inclination, nearly always entail an almost equally large negative
a theory says that the observation
ought not to be written

negative, but
is
u=p-\-qt,
= r'\-q{t
tQ),
unless
where
t^
is
all
observations.
and
this
line
becoming numerically small in
two corresponding
t,
latter.
If
the formula
the fs are small, some positive, and others
an average of the parameters corresponding to
we succeed, in this way,

can be known by the coefficients of all
the
change in the
a linear function of a single parameter,
If
comparison
in avoiding all strongly operating bonds,
the normal equations outside the diagonal
with the mean proportional between the
coefficients in the diagonal line,
then
we have
at any rate attained so
97
much
that
we need not use
the
many more decimal
the adjustment
for
calculations
always be sufficient when the elements are originally

and not during the adjustment are first to be transformed into freedom
which
places than about the 3,
mutually
in
free,
will
with painful accuracy in the transformation operations.

If,
by careful selection
in
products [pq]^)
of the
elements,
numerical value exceeds about
we even
of the
sum
get so far that no
mean
of the
proportional between the
or in many cases only ^ of these amounts,

corresponding sums of squares V[pp] [qq]
then we may consider the bonds between the elements insignificant. The normal equations
,
themselves
then be used to determine the law of error for the elements; we compute
may
provisionally a first approximation by putting all the small
sums
of products
0,
and
in
by substituting the sums of the products

and the values of the elements as found in the first approximation. For instance:
we
the second approximation
[po]
correct
the [j9o]'s
[pq\ y^
....
- [pr\ z, =
(111)
[pp\ x^
while
=
As the
errors
in
these
,s,
t:{M-M-...-M}.
determinations
are
of
the
second order,
it
will not,
themselves are small deviations from a provisional computation, be necessary to
if
the
o's
make any
further approximations.
Even
the
if
bonds between the elements, which are stated
of the products, are stronger,

tion.
If
we can
get
new
sums of the products
the
we can sometimes
get
if
we can
some
tions
has
plete
attain
single ones,
may
made
year,
observations, which are just such functions of the elements that
taken into consideration, we will of
will vanish if they are also
freedom
we cannot only
facilitate
hesitate to use this
means;
If,
especially as such
for
instance,
meteorological observations at some fixed station for a

shall not hesitate in the adjustment,
giving them the weight
And
of the elements by rejecting from a long series of observations
we do not
leave out the overlapping observations, or to
')
it,
but also increase the theoretical value and clearness of the result.
very well be employed in the criticism.
we
terms of the sums
them untied without any transforma-
course put off the adjustment until, by introducing them into

the computation,
in
unused observa-
an arctic expedition
little
more than a com-
by means of periodical functions,
make use
of the
means
of the double values,
of single observations.
In what follows we write, for the sake of hrevity, [pq\ for
to
[^J.
13
60.
sin
of course the fabrication of observations
Though
which an applied science can commit, there

class
important
of cases,
in
in general, the greatest
is,
a rather numerous and
exists, nevertheless,
which we both can and ought
method which just
to use a
depends on the fabrication of such observations as might bring about the freedom of the
As a warning, however,
theoretical elements.
against misuse
give
it
a harsh name:
the
method of fabricated observations.

If,
we consider the problem which has
for instance,
adjustment, both by correlates and by elements,
served us as an example in the
the determination of the abscissae for
viz.
4 points whose 6 mutual distances have been measured by equally good, bondfree observa-
we can
tions,
now
scarcely after the
"12
+ 0,3
-|-
given indications look at the normal equations,
Oj4
"12 + + "24
0,3 O23 + O34
''23
Oj,
Oj,
= Sxj 1x2 laâ 14

= lXi-\-Sx^lx.^lx^
= la;, laj2+3a-3
la?,
O3,
la;,
i^x^-\-6x^,
lajj
without immediately feeling the want of a further observation
=
which,
if
we imagine
distance ^,(o)
it
^ Xr
to
a;,,
la;,
4-
Iscg
have the same weight

will give
+ 1^4
4" 1^3
as
each of the measurements of
by addition to the others, but without specifying the
value of 0,
+ 0,2 + 0,3 + 0,, =

0,2 + 028 + =
0,3 023+03, ==
0, 02, O3, =
4
and consequently determine
What
in
this
all
of the
abscissae
4a;2
4a;8
4a;,,
4 abscissae as mutually free and with fourfold weight.
and other cases
entitles
minateness in the original problem of adjustment
any
4x,
by means
of
the
us
distances
to
fabricate
observations
between the points.
such problems in exact mathematics we get simpler, more symmetrical,

tions by introducing values
the theory of observation.

only
while
mean
values,
but also
is
indeter-
here, the impossibility of determining
When we
and
treat
easier solu-
which can only be determined arbitrarily; and so it is also in

But the arbitrariness gets here a greater extent, because not
mean
we can always make use
errors
of a
the complete or partial liberation
must be introduced
fabricated observation
of the
elements,
for greater convenience.
And
in indeterminate problems for
we must here
carefully
demonstrate,
by criticism in each case, that the fabrication we have used has not changed anything
which was
really
determined without
it.
99
In
the
the
above
values
adjusted
example,
for
the
this is seen in the first place by
Xr Xs,
distances
and
then
disappearing from
O's
by
own
adjusted
adjustment will consequently

other determinations.
the
determine
neither
The mean
nor
it
get any
show the same and, moreover,
errors
The
influence on
in such a
the
way that
has been brought into the computation
becomes independent of whether
criterion
let
value,
= 0.
determined as the sum x^-\-X2^x^'\-x^, and leading only to the identity
all
an indeterminate number or with an arbitrary value,

The scale for
as before, we have for Oj XîO) == 1.
as
also generally a sufficient proof of our right to use the
for,
is
after the
adjustment as well
consequently
method
and
0,
this is
of fabricated observations.
The method of partial eliminations. When the number of elements is

becomes a very considerable task to transform the normal equations and eliminate
61.
large,
it
The
the elements.
difficulty is nearly proportional
the square of that number.
to
Long
numerous that adjustment by correlates could be
indicated, a correct adjustment by elements can become practically impossible. The special
criticism is quite out of the question, the summary criticism can scarcely be suggested, and
before
the elements would
the very elimination
become
must be made
so
easier at
the elements enter into the expressions for
any
some
price.
then happens that some of
If it
of the observations only, and not at all in
the others, then there can be no doubt that the expedient which ought
is
the partial elimination (before
observations concerning them.
we form the normal
These observations
first
to be
employed
equations) of such elements from the
will by this
means be replaced by
and they
functions of two observations or more, which will generally be bound;
certain
will be
and more dangerous degree the fewer elements we have eliminated. By

we may, consequently, imperil the whole ensuing adjustment, the foundation
which, we know, is unbound or free observations as functions of its elements.
If now it must be granted that the difficulties can become so great that we cannot
so in a higher
this proceeding
of
on an absolute prohibition against illegitimate elimination, we must on the other
insist
hand emphatically warn against every elimination which is not performed through free
functions, and much the more so, as it is quite possible, in a great many cases in which
abuses have taken place, to remain within the strictly legitimate limits of the free functions,
by the use of "the method of partial eliminations".

This
Oy
is
connected with the cases, in which some of the observations, for instance
Om, according to the theory, depend on certain elements, for instance x,
do not occur in the theoretical expression for any other of the observations.
then, by the formation of the normal equations to separate o^
observations.
diately
We
perform
.y,
of x,
is
o^ as a special series of
begin by forming the partial normal equations for this, and then
the elimination
which
Our object
imme-
y from them, without taking into consideration
whether these equations alone would be sufficient
for a
determination of the other elements.

13*
100
As soon
as
y are eliminated, the process of elimination
is
The trans-
suspended.
formed equations containing these elements (which now represent functions that are
all observations, and functions which
depend only on the remaining elements z,
.
put aside
we come back
till
to the determination ot
y.
The other
free of
m),
are
partially transformed
normal equations, originating in the group o^

Om, are on the other hand to be added,
term by term, to the normal equations for the elements z,-... u, formed out of the remain.
ing observations, before the process of elimination
That
elements x
x'
imagine
P^r then
==
0,
this
all
is
proceeding
quite
the
inserted instead ot
sums
jc'.
continued for these elements.
becomes
legitimate
y transformed into the elements

y'
is
/',
evident
which are
we imagine
if
free oi
u,
the
and then
in the original equations for the observations.
of products with the coefficients ot a^
y' will identically
become
and the sums of squares and sums of products for the separated part of the observaaddenda in the coefficients of the normal equations (compare (57)), come outi
tions will, as
immediately, with the same values as
As an example we may
of 3 points on a straight line.
now
The mode
of observation
scale several times along the straight line,
the unaided eye

of them.
the transformed normal equations.
treat the following series of
The readings
for
measurements of the position
as follows.
and then each time read
either the places of all the points
same row
is
each point are found
in
We
apply a millimeter
oif
by inspection with
against the scale or the places of two

its
separate column, and those on the
same position of the scale. (Considered as absolute abscissabelong

observations such observations are bound by the position of the zero by every laying
down
of
to
the scale;
the
but these bonds are evidently loosened by our taking up the position
against the scale of an arbitrarily selected fixed origin yr as an element beside the abscissae
1,
x^, Xg of the three
101
As the
theoretical equation for the i"" observation in the
and every observation,

to use the
method
therefore, is a function of only
consequently to form
normal equations
for
two elements, there
we choose
If
of partial elimination.
column has the form
s""
first
is
every reason
to eliminate the ys,
each of the 11 rows.
Where
we have
only two points
are observed these normal equations get the form

Or
O,
Or
for three points the
=
=
2yi
/i
-\-
Ojj
Of these equations those
+ O3
==
Ol
03
Xs
-\-
+ Xr
form of the normal equations

Oi
Xr
is
+ î +
+ î
3/i
y-'
^^
+ ^3
-\-Xs.
!/i
referring to the
ij,
have given the eliminated
free
elements
stated above to the right of the observations after the perpendicular.
equations from
subtracting these
By
in the cases
the corresponding
0,
i{0r + 0s)=
^X,
= \Xr + \Xs
\{0r-\-0,)
and in cases where there are 3 points

o.,
By forming the sum

others,
of the equations,
we consequently
'
of these
get,
how
\Xr
ia;., |a;3
1^1 + 1^2 \î
hî \^2 + i^-6-
differences
often
|aJi
for
each column, and counting, on the
each element occurs with one other or with
37"71
+ 206-69
'Ix.-'ix^-lx^
= -^x^^-^x^
= -'{x.-'lx^ + '^x,.
-g-a;^
The case is here simple enough to be solved by a fabricated observation.

most advantageous form found, when its existence is given?
Answer:
two
get the ultimate normal equations:
-168'^8=
its
we
i(Oi + + 03) =
02 3(01 + 02 + ^3) =
M01 + O2 + O3) =
03
Oi
side
equations
where there are 2 points:

Or
right
other
g^ = ^-f-|l +
|l,
weight
23712
How
is
102
after
which we get the normal equations:
168-98
114
U9
37-71
+ 206-69
*i
114
= ^x.
=
consequently,
mm.
Xi
From
these
= 25-38
we now compute
/i
2/2
2/6
We
differences,
for
=
=
=
=
=
==
x^
4-77
and
which
it
is
= + 21-24.
the ys:
33-72
0,
33-27
32-295
2/7
25-945
15-485 0.
,
8-845
=
=
=
y,o=
enough
56-80
-0,
58-27 0,
35-01
45-55 0,
2/1,= 44-29-0.
need not here state the adjusted values

of
^3
to say that their
for the several observations,
sum vanishes both
each column; their squares, on the other hand, will be found to be:
I
nor their
for each row and
103
As
to
functions are
weights of
easy to
a;,
special
criticism
it
is
the
weights
the
of
eliminated
free
2 and 3 times the weight of the single observation, while the

759
-, and ^, very
x^, and x^ after the adjustment become respectively H4'
scales
X^(u)
1
1
'
Xi(o)
as
wliere
respectively
compute the
With 759
liere,
common denominator we
most natural groups:

I
1
2
3
4
5
6
7
9
10
11
"
Weight
after the
adjustment
find for the several scales
and the sums
of
their
104
functions of the original observations, need not always be so iirmly connected with one another
as in the ordinary method. If we, in a suitable way, take advantage of regularity in the obser-
and thereby are
vations,
then the determination of the
But
in our way.
in
a transformation which sets the normal equations
able, to find
order
free,
several elements will scarcely throw any material obstacles
out any
to find
special
transformations,
we must know the
general form of the changes of the normal equations resulting from transformation of the
homogeneous linear funtions of them whatever.

unbound observations in terms of the original elements
original elements into such as are any

If the
the
for
equations
have been
+ iiV + r^z,
Oi-^ Vi^
the normal equations will be:
=
=
[qo]
\ro\
And
if
we wish
to substitute
new
{qp\x+{qq^y
+ \qr]z
[rp]x-j- [rg'Jy
-|-
elements,
f,
;j,
and
[r/]0.
(^,
for the old ones,
we make use
substitutions in which the original elements are represented as functions of the
new
of
ones,
therefore
== /.,f
2/
+ A:,, + Z,C
(114)
[
The equations
Oi
for the observations
then have the form
(jo./ii+?<^2+'<-^3)f+{j'.-î+?.-^2+'.^3)'?
The new normal equations may be formed from

the equation which specially refers to
+ qh^-Jrrh;)
[{phi
o]
[(ph^-i-qh^
+ (i't^+9'J2 + n^3)C-
these, but the
(115)
form becomes very cumbrous,
$ being
+rh^)^
+ [{ph^ + qh^^^rhs) (Mi + â + 'â)] V +

+ [iphi + qhi + rh^) {pi, + +
ql.,
rl^)\ C-
The computation ought not to be performed according to the expressions for the coefficients
which come out when we get rid of the round brackets under the signs of summation [ ].
But
it
is
easy to
give
the rule of the computation with full clearness.
equations are first treated
exactly as if
they were equations
for
The
old normal
for
expressed by the new elements, consequently by multiplication,

and
by columns, by A,, h^,
h^ and addition; by multiplication by k,, k^, and kg and
addition; and by multiplication by î, l^, and l^ and succeeding addition. Thereby, certainly,
X,
we
y,
and
get the
z, respectively;
new normal
equations, but
still
with preservation of the old elements:
105
=
=
[{pki-{-qk^-{-rks)p]x
The second
part of the operation
Upki-^qk^-\-rks)o]
[{ph
+ ql2-i-rl.^)o]
new elements
+ [{pk,+qk^^rk^)q]y + [{pk,-^qk^-\-rk^)r]3
[{ph+ ql2-i-rls)p]x-^[{pl,-^
must
therefore consist in the substitution of the
the original ones in the rigiit hand sides of these equations.
for
to find the coefficients of f
\{\16}
ql^+rl.,)q]y^[{pli-i-qL,-\-rl.,)r]z
In
order
and C' we must therefore here again multiply the sums of
tj,
now by rows, by
the products,
'11
'3
'2*
and add them up.

It
Example.
happens pretty
that there
linear measures,
so that for instance the

cients,
is
irrespective
Already P. A.
into the
mean
the
two
in
terms [po]
pointed
values
on
the
hand
left
k{xr-{-Xm-r) and their half diiferences dr
^^^
Xmr =^
Of.
(*j.
Sr
Ctr
and the transformation of the normal equations
[ao]
[60]
{co\
[rfo]
+ {do\
=
=
=
=
4a;
3a;
2a;
la;
+
+
3?/
6/
42/
2/
the equations.
}^(xr
new ones have
a;,_,.).
[bo\ -^{co-\
bx
+ \0y + 10^ + bu =
[o]
.3a;
{bo\
[co] =
la;+
Sy
bz
l2
- 3m =
2y
2z
I2/
\u
In
the form
consequently, performed just by forming

If the
normal equations are
+ + 1m
+ 40 + 2M
+ 62 + 3m
+ 30 + 4m
22:
+ bu = lOî^ + lOî^
5a;
[f/o]
of
as follows:
--=
[ao\
is,
differences of the original coefficients.
is
sides
this indicates a transformation of the elements
out that
Xf
the procedure
for
normal equation corresponding to x^-r', of course,
as the
observed
this case therefore the equations for the old elements by the
sums and
of scales
investigations
symmetry between the elements, two and two, Xr and Xm^r,

refers to -Xr, has the same coeffi-
Hansen
Sr
instance
normal equation which specially
only in inverted order,

of
for
often,
10 ^^^
+ 20 ^^
6^^+2^^
2
"^^^+4
14
106
As
in
this
as two mutually
differences,
The
63.
titions,
the
in
we always succeed
example,
is
the general
to
when
case
mean
determination of the mean values and the
us to neglect their products and squares.
there are varying circumstances in

is
then reduced to the

Before an adjust-
observations, which are

in the cases
night,
to
to allow
know how
to act in
such cases.
they form
<?,
mean time
of observation
t,
mean
right ascension a,
and consider a and 8 the spherical co-ordinates of the
star at
t.
With
the obvious extensions
most important device

forming mean
this
is
normal place method, the

Such observations whose essential circum-
what
in practical adjustment.
stances have "small" variations,
called the
is
before the adjustment, brought into a
are,
that the law which
connects
place,
by
the observations and
circumstances, holds good also, without any change, with respect to their
Much
normal
values both for the observed values themselves and for each of their essential
circumstances, and on the supposition
is,
speaking
strictly
which only approximate
astronomers have observed the place of a planet or a comet several times
and a mean declination

the time
mere repe-
are
has not been necessary to await the systematic
It
development of the theory of observations to
same
the observations
where the variations of circumstances have been small enough
repetitions, viz. those
in the
from the halt
values
errors of the observations.
ment, therefore, we not only take the means of any

repetitions, but we also save a good deal of work
When
when
the fact that the whole adjustment
to
owing
mean
the
separating
great simplification that results
contradistinction
observations,
in
systems of functions of the observations.
free
trouble
may
mean
be spared by employing the normal place method.
values.
The question
whether we lose thereby in exactness, and then how much.
We
shall
first
consider the
tions being:
?.^{o)
With
the weights v
we form
a-{-bx~\-
z,
-\-
the
equation
for
are
linear
the observa-
de.
the normal equations:
[vxo]
=
=
[vzo]
[vo]
unbound observations o
where the
case
functions of the varying essential circumstances x,
a[v]-\-b[vx]+
a [v.t]
-\-
b [vx^]
-|-
^[^'Z'^]
... -ird[vz]
'^
[vxz]
(117)
\
(118)
If
the whole
series
of
a\yz^ +
observations
corresponding to the circumstances
is
+ (^[f^*].
gathered into a single normal
Z, and with the weight F,
we
place,
shall have:
0,
107
VX
=
=
=
VZ
V
VO
and
[V]
[vo]
[vx\
[vz\
as
= a + bX-^....-J^dZ,
this
normal place
exhaust the normal equation (117) corresponding to the constant
will
mean value and mean
term, both with respect to
we
equations free of (117),
0)(z Z)] =
[v{o
for the determination of the
if
the whole series
small
equations (118a)
are
small of the
order.
term
is
least
-^dZ
elements b
determination.
quantities
normal
squares,
if
we make the other normal
-{-d[v{xX){zZ)]
second
up the
split
places
then the normal
if
the
forming
series,
equation
now
lost completely
xX
and z
Z are
each part a normal
for
the observations
according to the
to
corresponding
place method; and besides
the normal place method
a)
Certainly, the coefficients of these
order,
instead of
(118
+d[v(2 Zf]
and these determinations are
d,
of the
exhausted by the normal
still
a-\-bX-\-
elements b
these
adjusting
method of the
b[v(xX)(z Z)]-{-
on the other hand, we
If,
and
place,
blvixXy]-^
gathered into a single normal place.
is
first
But
error.
by the correct method of least squares:
get,
0)(xX)] =
lv(o
(117a)
the
constant
this determination of
also aflords a determination of the other
d, in such a way, however, that we suffer a loss of the weights for their
This loss can become great, nay
way that does not
suitable selection of
suit
the
purpose;
but
it
total,
if
the normal places are selected in
can be made
rather
insignificant
by
normal places in not too small a number.
Let us suppose, in order to simplify matters, that the observations have only one
variable
essential
circumstance x, of which their mean values are linear functions,
sequently
>li(o)
and that the
x's are
con-
a-\-bx,
uniformly distributed within the utmost limits, x^ and
a;,
we then
let
each
normal place encompass an equally large part of this interval, and we shall find then, this
being the most favourable case, with n normal places, that the weight on the adjusted value of
the element b becomes 1
11
if by a correct adjustment by elements the corresponding
weight
made
is
still
taken as unity.
smaller,
if
The
loss is thus,
at
any
rate,
not very great.
And
it
can be
the distribution of the essential circumstance of the observations
is
108
uneven, and
we can
if
normal place
get a
everywhere where
the
become
observations
particularly frequent, while empty spaces separate the normal places from each other.
The
case
analogous also when the observations are
is
or a few essential circumstances, but the function
For
it
when
possible also to form normal places in these cases;
is
still
functions of a single
of a higher degree, or transcendental.
is
and we can do
so not only
the variations of the circumstances can be directly treated as infinitely small within
each normal place, which case by Taylor's theorem
falls
y =
have at our disposal a provisional approximate formula,
deviation from this, o
with the
f{po),
For
if
we
and have calculated the
y, of every observation (considering the deviations as observations
circumstances and
essential
within the given rule.
mean
errors of the original observations),
then we
can use mean numbers of deviations for reciprocally adjacent circumstances as corrections
which, added to the corresponding values from the approximate formula, give the normal
values.
that
Further,
the
is
it
required here only that no normal place
within
deviations
is
made
so comprehensive
not remain linear functions
do
limits
its
the
of
essential
circumstances.
Also here part of the correctness

loss
is,
is
and
lost,
it
is
how much. The
difficult to say
under equal circumstances, smaller, the more normal places we form. With twice
as many normal places as the number of the unknown elements of the
(or three times)
problem,
will rarely
it
become
perceptible.
With due regard
and the distribution of the weights we can reduce

between the normal places.
distribution
suitable
observations
are
of their
sufficient to
know
we have
do with
to
transformations
of
essential
using empty spaces as boundaries
it,
normal places
of the
also
As
circumstances.
to the essential circumstances
to
depends on what function the

this,
it
however,
the behaviour of the integral algebraic functions, as
we
is,
as
s^
generally,
rule,
when
functions which are essentially different from these, will try through
the variables
to
get
back
them
to
and
certain
to
functions which
resemble them in this respect.
We
need only consider the cases in which we have only one variable essential
circumstance, of which the

,-th
degree.
0, and
We
their
to
value of the observation
circumstances, x,
and weights,
v,
determine r-fl
to
substitutive
together with the essential circumstances, X, and weights, V, belonging
them, in such a way that they treated according
will give the
an algebraic function of the
is
are able then, on any supposition as to the distribution of the observations,
essential
observations, 0,
mean
same
results as the larger

[0V-]
number
to the
method
of actual observations.
of
the least squares
The conditions
are:
OoVo+ ...+OrVr
j
(119)
[OX'V]
XlOo Fo
+ TrOrVr
109
and
[,]
[x^-'v]
x':v,
Vr
(120)
...+xrvr.
These Sr-\-2 equations are not quite sufficient
We
unknowns.
The elimination
roots
Xq,
of the
Xr
When
positive.
remove the
are
the
Vs
difficulty in the best
real
all
are
distributed
where in
that they
to
-\-l,
may be
this interval;
if,
given x's
found, we can compute,
for instance, the essential
way by adding the equation:
the
if
quantities,
Oq, ... Or, by means of two systems of r-f

If,
determination of the 3r-l-3
(and O's) then leads to an equation of the
roots
in the interval from
for the
r+l
degree,
whose
have been real and the v's
first
V^, ... Vr and afterwards
linear equations with r-\-l
uniinowns.
circumstances of the actual observations are contained
and
looiied
if
the observations are so numerous and so equally

as continuous with constant
upon
further, the
sum
of the weights
the substitutive observations will be symmetrical around
0,
2;
mean
error every-
then the distribution of
and, for functions of the lowest
degrees, be
000
o{ V
2-000
'
-577, + 577
'
1-000,
(X
1^000
= -775,
000,
+ 775
-556,
889,
-556
'
-861, ^340, + -340, + 861
'
-348,
\x
= -906, -538,
237,
(X
652,
479,
-652,
348
000,
+ ^538,
906
569,
479,
237
= -932, -661, 239, + ^239, +
'
661,
932
361,
171
'
171,
361,
468,
949, 742, - 406,
468,
000,
6
129,
If,
280,
382,
418,
+ -742, + ^949
406,
280,
382,
in another example, the distribution of the observations
is,
lilcewise,
^129
continuous,
but the weights within the element dx proportional to e^^^ consequently symmetrical with
maximum
by
a;
0,
then
practical interest, will be
the distribution
for
the lowest degrees,
the only ones of any
no
X=
^\f=
-000
jX
=
I
X=
V =
fJt
^ '^
\V
^
~
**
loss of exactness.
the
least
number
required
is
-000,
+1-732
-333,
1-333,
-333
+2-334
-908,
-092
2-857,
-1-356,
-000,
+1-356,
+2-857
-023,
-444,
1-067,
-444,
-023
-3-324,
-1-889,
--617,
-617,
+1-889,
+3-324
-005,
-177,
-818,
-818,
-177,
-006
3-750,
2-367,
1-154,
-000,
+1-154,
+2-367,
3-750
-001,
-062,
-480,
-914,
-480,
-062,
-001
now
able
to represent these substitutive observations as
able also, by the use of such tables in analogous cases, to prevent any
would be possible
It
we had but
not possible.
values
of
normal places,
to
entirely to evade the application of the
form
the
Certainly,
essential
method of
such qualified normal places in just the same
we can obtain normal
circumstance,
twice, others not at
places
but we cannot by
give them the weight which each
of
all.
them ought
By taking
be determined.
to
corresponding
This,
the
to
a simple formation
to have,
of
without employing
into consideration
how much
extreme normal places from this reason must lose in weight, compared to the sub-
stitutive
observations,
can amount
1;
but for r
essential
1049 to 1171.
tion
0,
we have:
first
how many per

our examples we find
estimate
of
2 we lose 15, for r
r 21 p.
the worst
cent the loss, in

the loss to be 0,
3 we lose 19,
for r
for
case,
and
4 we lose 20, and
c.
unbound observations, equally good, ^j(o)

correspond
j-^,
circumstance whose values are distributed as the prime numbers p from
x as the essential circumstance of the observaTaking (^1105): 100
Example.
an
we can
In the
to.
for greater values of
to
-742,
-908,
some of the observations

the
-742,
-092,
adjustment formula contains elements that are
the
mean numbers
'
squares;
as
however,
-1-732,
'
we were
we should be
then
1-000
-2-334,
=
=
=
jX
\
If
1-000,
'
(Z==
i
+1-000
'
=
=
=
'
1-000,
X =
/-
2-000
Eighteen
Ill
112
Now, by application
of the
normal places instead of the original observations, we
obtain on the same suppositions the normal equations:
6-72 =
_ 2-84 =
216-00 a
-54=
1-57 =
By the
-2-80
get:
1-20 b + 29-45 c + 1-87 d

+ 29-45 6+ l-87c+7-93a!
+
1-87 a +
29-45 a
1-87
ft
7-93
ft
1-20
ft
+
+
7-93
c+
1-14 c
1-14
rf
+ 2-45 d.
free equations:
6-72
we
1.20 a
=
=
216-00
+ 29-45 c+ 1-87
(^
113
and
abscissa are, respectively, the observed value
ment
is
drawing
from
irregularities
and going as near as possible
free
The smoothness
observation.
or extremely intricate
when,
of a
we succeed
to
when
relatively best
we must
for instance,
the
satisfies
two conditions
the several points of
this process plays the part of the theory,
curve in
the
of
a matter of course that
is
it
curve which
performed
of being
and
circumstance; and the adjust-
its essential
by free-hand
the theory
is
unknown
confine ourselves to requiring that the
phenomenon must be continuous within the observed region, or be a single valued function.
But also such a theoretical condition as, for instance, the one that the law of dependence
must be
of an integral, rational form,
ment,
the operator has had practice in the drawing of parabolas of higher degrees.
if
be successfully represented by graphical adjust-
may
And
we have seen that also such functional forms as have the rapid approximation to an asymptote
which the curves of error demand,
As
for
the approximation
ment implies that

ordinates
so
make
the
near
the points as
typical form
only that they are
our
to
figure
all
each
care that
take
side of
If
for
cannot pay
number
the
not only in
it,
its
we know the weights
much
or laws of errors.
errors
In
of course,
regard,
errors,
we
the presupposed
to
but something
may
be attained,
as to the exactness of the several observations, or
a graphical
the
observations,
there
can be only a single point in
adjustment, however,
same number
whole extent, but also as
of the
the
of similar deviations.
we draw has
curve
intersect
of the several observations before
to be considered equally good,
observation.
must
be indicated visibly on the drawing-paper and used
we know nothing whatever
If
exactness
and other properties of the law of
particularly with regard to the
mean
required by the several
is
as to the
We
only, the curve
not necessary;
is
adjustment, this ought to
in the graphical adjustment.
several observed points, the idea of the adjust-
the
to
a perfect identity
we know anything
If, after all,
within the province of the graphical adjustment.
lie
they
we can and ought
of obserted points on each
far as possible for arbitrary divisions.
may
be indicated on the drawing, and
observations with the weight n count w-fold.

In
contradistinction
only of bonds between
down on
errors.
and of
to this
it
worth while to remark that, with the exception
observations, represented
the paper of adjustment almost
all
by different points,
it
is
possible
to lay
desirable information about the several laws of
Around each point whose co-ordinates represent the mean values of an observation
its essential circumstance, a curve, the curve of mean errors, may be drawn in
such a way that a real intersection of

tion
is
less
it
with any curve of adjustment indicates a devia-
than the mean error resulting from the combination of the mean errors of the
observed value and that of
its
essential circumstance, if this
is
also found by observation,
while a passing over or under indicates a deviation exceeding the
drawings furnished with such indications enable us to
make
mean
error.
Evidently,
very good adjustments.

15
114
laws of errors both for the observation and for
If the
mean
then the curve of
errors is an ellipse with the observed points in its centre.
further, there are no bonds between the observation and
If,
mean
the ellipse of
errors has its axes parallel to the ordinate
lengths are double the respective
distant
the
mean
the
by
mean
the
well
as
errors
and the
errors
is
reduced
to
mean
the
values
the ordinate,
In
from the central point of observation.
are
continuous curves
circumstance of the observation,
their
jjnown to be free
is
the two points on
of illustrating the laws of errors

as
and
abscissa,
errors.
of the observation
error
means
special cases other
mean
of the
ellipse
mean
circumstance, then
its
circumstance of the observation, the abscissa,
If the essential
of errors,
circunastance are typical,
its
be used.
may
for instance,
If,
continuous functions of the

for
mean
the
errors
may
essential
be drawn on
the adjustment paper.
The
and
errors
graphical adjustment are
independence of a definitely formulated theory.
we can
of the adjusted curve

of the
of the
principal advantages
its
get improved observations
circumstance or abscissa as
we wish, and we can
its
indication of gross
By measuring
the ordinates
corresponding to as
them
select
as
we
many
values
please within
But these adjusted observations are strongly bound together,

and we have no indication whatever of their mean errors. Consequently, no other adjust-
the
limits
of
the
drawing.
ment can be based immediately upon
On
with
the other
hand,
interpolations, both
means we can remedy

place
too
much
the results of a graphical adjustment.
graphical adjustment
can be
advantageously combined
very
preceding and following, and we shall see later on that by this
its
defects,
confidence
in
particularly
its
limited
the observations, and too
accuracy and
little
its
in the theory,
tendency to
i.
e.
to give
an under-adjustment.
By drawing we
frequently insufficient.
attain an exactness of only 3 or 4 significant figures,
The
drawing must be chosen in such a way that the

but then the dimensions may easily become so large that
no paper can contain the drawing.
is
of a
This
is
straight
And
line.
is
full
grasp of the figure,
lost
if the line
by the
flat
line,
which
a practical hint, founded upon experience.
can judge of the smoothness of other curves
then the exactness

is
In order to give the eye a
whole course show only small deviations from the straight
taken as the axis of abscissae.
eye
it
its
is
scale of the
errors of observations are visible;
the latter must in
and that
also,
The
but not by far so well as of that
forms a large angle with the axis of the abscissae,

intersections with the ordinates. Therefore, as a rule,
not the original observations that are marked on the paper when we make a graphical
adjustment, but only their differences from values found by a preceding interpolation.
we must allow |
mean errors. It
the
surpass
In order to avoid an under-adjustment,

curve from the observation-points to
of the deviations of the

is
further essential that
115
the
said
based
is
interpolation
on
graphical adjustment has been made,
number
minimum number
a
it
is
and
of observed data;
after the
safe to try another interpolation using a smaller
new
of the adjusted values as the base of a
interpolation and a repeated graphical
adjustment.
If
of a graphical adjustment are required only in the form of a table
the results
representing the adjusted observations as a function of the circumstance as argument, this

table
also
the interpolated
values.
be based on an interpolation between relatively few measured values,
to
ought
values
being
table of exclusively
When we
It
with the corresponding measured
by comparison
measured values
will
show too irregular
diiferences.
have corrected these values by measuring the ordinates in a curve of
may
graphical adjustment, they

places.
checked
has been said,
be employed instead of the observations as a sort of normal
however, and
it
deserves to be repeated, that they
adjusted by means of the method of the least squares,

called. But we can very well use both sorts of normal
for the computation of the
unknown elements
of the
like the
must not be
normal places properly so
places, in a just sufficient
problem, according to
number,
the rules of
exact mathematics.
That we do not know
their weights,
and that there are bonds between them,
will not
The very circumstance that even distant

made to influence each normal place, is an
here injure the graphically determined normal places.
observations by the construction of the curve are
advantage.
It is
not necessary here to suffer any loss of exactness, as by the other normal
places, which, as they are to be represented as
mean numbers, cannot at the same time be

the due weight. As to the rest, however, what
most advantageous places and obtain

110 about the necessity of putting the substitutive observations in
p. 108
the right place, holds good also, without any alteration, of the graphical normal places.
The method of the graphical adjustment enables us to execute the drawing with
put
in the
has been said
absolute correctness, and
it
leaves us full liberty to put the normal places where
consequently also in the places required for absolute correctness

it
we
like,
but in both these respects
and gives no formal help to it.

the graphical adjustment gives no information about the mean
leaves everything to our tact and practice,
As
to the criticism,
errors of its results.
nevertheless,
to
But,
if
we can
state the
mean
subject the graphical adjustments
the rule
V-'(o_m)2
I'And
with respect to the more special criticism
method even takes a very high rank.

remaining deviations,
same drawing, we
m,
particularly
error of each observation,
to
summary
criticism,
we
are able,
according to
m.
on systematical deviations, the graphical
Through graphical representations of the finally

if we can also lay down the mean errors on the
get the sharpest check on the objective correctness of any adjustment.

15*
116
From
and owing to tlie proportionally slight difficulties attached to it,

becomes
graphical adjustment
particularly suitable where we are to lay down new
the
this reason,
empirical laws.
such
In
of hypotheses
as
We
circumstances.
save
we have
cases
the
to
functional
much
to
work through,
and to
to check,
and
interdependency of observations
labour,
and
illustrate our results,
if
reject series
their
essential
we work by
graphical
adjustment.
Of
we
course,
are not obliged to subject
liminary stages, or as long as

stances
ought not
it is
be taken
to
observations to adjustment.
In the pre-
doubtful whether a greater number of essential circum-
into consideration,
may even
it
be the best thing to give
the observations just as they are.
But
drawing of a
widely
point.
even
in
order to illustrate such statements by the
which connects the several observed
line
the form
spread,
of a
Discontinuity in the curve
rectilinear
We
give his opinion on
to
have already, in
number
The theory
9,
limitation.
each of these events we

event
all
the
other
its
it
ought,
occurrence in reality.
treats
especially
of
an
observation
may
possible
of an event approaches,
number
such
of trials.
observations
But there
is
whose events
no compulsion in
can result in different numerical values, then for
very well speak of

ones.
the
or in other words, as the limit
of favourable events to the total
of probabilities
When
observed
every
defined ''probability' as the limit to which
cannot be naturally or immediately expressed in numbers.

this
in
be reserved for those eases in
to
of repetitions is increasing indefinitely;
of the ratio of the
broken
is
such a marked geometrical peculiarity that
is
law of the large numbers taken for granted
when the number
to give this
THE THEORY OF PROBABILITY.
XIV.
65.
we ought
fashion which unfortunately
to
polygon which
more than cusps, double-points, and asymptotes,
which the author expressly wants
then
points,
form of a continuous curve and not, according
line the
is
we use the graphical form
if
In this
way
its
probability, imagining as the opposite
the theory of probabilities has served as
the constant foundation of the theory of observation as a whole.
But,
on
the
other
hand, it is important to notice that the determination of the

law of errors by symmetrical functions may also be employed in the non-numerical cases
without the intervention of the notion of probability. For as we can always indicate the
mutually complementary opposite events as the "fortunate" or "unfortunate" one, or as
"Yes" and "No", we may also use the numbers
and
as such a formal
indication.
If
117
then we identify
sums
of
events.
with
the
with the unfavourable "No", the
"Yes"-event,
numbers got in a series of repetitions will give the frequency of affirmative

This relation, which has been used already in some of the foregoing examples, we
the
must here consider more

If
explicitly.
same observation, which admits
repetitions of the
the result "Yes"
times,
for the favourable event is

for the
favourable
Sg
=
=
=
tn
and
m-\-n
,u
0,
then
two alternatives, give
if
Sj
....
In order to determine the half-invariants by means of
m
m
of only
we employ the form of the symmetrical functions

then the sums of the powers are
But
errors,
m times "No"
against
m+M
same law of actual
3m
jo
this,
3m
-f
Sr
/ig
m.
we
solve the equations
+ (w
-j-
(121)
) /4
find then
m
Ml
mn
fit
{m -{- nf
mn {n tn)
Fa
fî
Compare
equal to
interchanging
half-invariants
In
The
fi^.
and
of odd degree (from
order
event becomes
(m-f-w)*
are integral
functions
of the relative frequency, which
qj- = fXi;
1
w, none of the half-invariants of even degree are changed,

fi.^
to
represent
q.
^
is
by
and those
upwards) only change their signs.
= ^-
= ^p =
^-^
==
relative frequency of the opposite result is
need only assume, in (122), that
m-{-n
{m-\-nY
mn {n^ 4mn + *)
23, example 2, and 24, example 3.
All the
itself
(122)
The
the connection between the laws of presumptive errors,
and n increase
indefinitely, while the probability of the
and the probability of the opposite event

half invariants are then
h =
we
is
represented by
= miip)
K = P<l{<f '^M+ V\
(123)
-^3
Our mean values are
therefore, respectively, the relative frequency
and the probability
itself.
118
We
must now
first
notice here
function of the probability
simple
that
complete
When
(the frequency).
stated in the form of one single probability,

a
half-invariant
every
is
own
its
and
fixed
a result of observation can be
we have
properly so called,
thereby given as
determination of the law of errors as by the whole series of half-invariants.
In such cases
it
simpler to employ the theory of probability instead of the symmetrical
is
functions and the method of the least squares.
The theory
of probability
its
thereby gets
much more
province determined in a
way than that employed in the beginning of this paragraph.

But at the same time we see that the form of the half-invariants is not only the
which must be employed where the conditions for the use of the probability
means
general
are not fulfilled, but also that, within the theory of probability itself, we shall require,
natural and suitable
of the
particularly, the notion
mean
error.
Even where the probability can replace all the half-invariants, we shall require all
the various sides of the notions which are distinctly expressed in the half-invariants. Now
we have
is
to
particularly to consider the probability as the definite
elicit
which
is
the
definite
particularly emphasised
to
what
justly due to them.
rely
mean
the
in
error.
is
value,
now
Finally,
we
shall see
the point
implied in the probability, and
Otherwise,
we should
constantly be
on the predictions of the theory of probability to an extent
tempted
is
which
of uncertainty
degree
mean
far
beyond
immediately that the laws of error of
the probabilities are far from typical, but that they have rather a type of their own, which
must sometimes be
All this
as a
means
we
especially emphasised.
shall be able to do here,
where we have the half-invariants in reserve
of representing the theory of probability.
66.
In particular,
we can now, though only
in the form of the half-invariants,
solve one of the principal problems of the theory of probability,
presumptive
errors
the frequency
for
only two events and which
is
repeated
of one
trial,
which can have
upon the supposition that the
times,
the law of the large numbers, and that the probability
for a single trial is
trial follows
known.
(123) give us already the corresponding law of error for each
The equations
and as the
and determine the law of
of the events of a
total absolute frequency
is
the
sum
of the partial ones,
trial,
we need only use
the
equations (35) to find:
>l,
(m)
Aîm)
A,{m)
Xi(m)
=
^
=
=
Np
Npq
Npq(q-p)
Npq{q^
=
=
Np{l-p)
Np{l-p){l-2p)
ipq->rp^}
Np{l-p)(l-{3 + VS)p){l-{S-VS)p).
(124)
119
The
mean frequency to
small the mean error
ratio of the
When ^
is
frequency; and
VNq. When
if
is
nearly
1,
the probability, p,
The law
of error
is
the
number
differs
the
of trials
from
little
mean
the
^Np
square root
the
of
mean
error of the opposite event is nearly equal to
nearly equal to
is
therefore the probability itself.
is
not strictly typical,
mean
the
i,
error will be about \ j/iV.
although the rational function of the r*
degree in kr{in) vanishes for r different values of
p between
and
1,
the limits included,
so that the deviation from the typical form must, on the whole, be small.
however, we
If,
consider the relative magnitude of the higher half-invariants as compared with the powers
of the
mean
error
>l3(m).(>(,(m))-l
= -^^
VNpq
and
(125)
the occurence of
denominators of the abridged fractions shows, not "only that
in the
Npq
great numbers of
here as always,
repetitions,
cause an approximation to the typical form,
but also that, in contrast to this, the law of error in the cases of certainty and impossibility,
when
high degree,
and
while
remarkable property
jp
at
the same time the square of the
is
still
traceable in the cases in
small or very nearly equal to

the
mean
will
error
becomes skew and deviates from the typical in an
0,
1.
be about
rare
in
more frequent than
be
the
case
under
V\-
Errors
degree.
ct.
mean
becomes
which the probability
beyond
error.
infinitely
This
0.
either very
is
though
The law
of
mean frequency
the
The
99iV
hey
error
is
must be
fully
In applications of adjustment to problems of probability,
it
must
will not be
counter-balanced
consequently skew
is,
in
ct.
cannot
great diminishing errors
and frequencies of 97 or 96
in typical cases,
consideration,
numerous cases of 100 per
errors
Jn a hundred trials with the probability == 99^ per
exceed |, and are therefore less than the

therefore
mean
by
perceptible
from this reason,
frequently necessary to reject extreme probabilities.
XV.
67.
THE FORMAL THEORY OF PROBABILITY.
The formal theory
of probability teaches us
that depend upon other probabilities,
which are supposed
how
to
to
determine probabilities
be given.
Of
course, there
are no mathematical rules specially applicable to computations that deal with probabilities,
and there are many computations with
probabilities
probability, for instance, adjustments of probabilities.
which do not
But
fall
under the theory of
in view of the direct application
120
not only
of probabilities,
life,
to
games, insurances, and
that a computation will
lead
us
a probability as
to
whole a determination of a law of
in the
but to
statistics,
understood that special importance attaches
be
will
it
the
to
marks which show
result, as this implies in part or
its
The formal theory
errors.
conditions of
all
of probabilities rests
on two theorems, one concerning the addition of probabilities, the other concerning their
multiplication.
The theorem concerning
I.
the addition
of probabilities can,
as
all
probabilities
are positive numbers, be deduced from the usual definition of addition as a putting together:
if
sum
of probabilities
is
be a probability
to
each of the probabilities that we are
a quality in
the sura
common,
add together as corresponding
to
mark
trial.
An "either or"
the result E^, whose probability
if either
whose probability
is
is,
therefore, the simple
The event E,, whose
of the addition of probabilities.
to the
same
of the
trial
trial,
their
equally
is
are
probability of each
we
If
require no other resem-
repeated
certainty, because then all events
1,
= p.
If
= p,
being
n events
then
the
of the
aggregate
the
multiplication of probabilities can,
is
of unity.
trial
be
probability
of
as
all
proba-
in
JO
the same proportional of the multiplicand as the
Only as probabilities presuppose
numbers of
infinite
i?,
is
a relative frequency,
has given favourable events in
it
must
relate to a trial
Np^ cases; and
times, must have given (Npi)Pi favourable events.

p must be the relative frequency of the compound
repetitions have given Np^p^ favourable events.
succeeded as conditional
please, the
trials,
we
shall
the corresponding proposition for relative frequencies.
= P1P2'
N times,
is
if
for the final event.
As
same proposition must hold good
Now
trials
The
the
if
repeated
in the multiplication
which out of the

trials
number
T, and
for probabilities.
Tj which,
p^, being also a relative
frequency, takes the place of multiplier, then the corresponding trial Tj,
we
same
proper fractions, be deduced from the definition of the multiplication of frac-
commence by proving
If,
than that they belong
np.
according to which the product
multiplier
Pi-\-Pi,
be the probability for a certain event, q the probability
The theorem concerning
II.
bilities
the
probable,
If
be the probability
we have p-\-q^\,
against the same, then
these events
sum must
gramma-
î, or the quite different event iJj,
is
p^, occurs, and not in any other case.
are favourable.
is
probability
between the events whose probabilities are added together,
blance
If
with this quality, the same quality must
to be the correct probability of events
is
must occur,
to its particular
which, after the addition, our whole attention must be given.
t(f
be found in no other event of the
tions,
be allowed to look upon
These events must mutually exclude one another, but must at the same time have
events.
tical
we must
itself,
T,^
total
Np^
P'=PiPi,
number
of
must both have
can be taken as large as
121
The probability p ^PiP-i-, as the product of the probabilities p^ and p^, relates to
compound trial, which is favourable only if both conditional trials, 1\ and
the event of a
Tj, have given favourable events;

bability
on
probability,
the factors
condition
be
may
we
proposition
the
compound
shall
in
of success
it
is
To
arranged in a definite order.
trials
Let both T^ and 1\ have succeeded in
without
trials
4-^4-^-1-/?
-^2
any
the
to
regard
other,
*^^ frequencies or
the multiplication for computation of the

as
only
or
by
as
multiplicands
= Pj
|)
= P1P2,
"~T~
a cases, while only T, has
mark
the
is
a:h
unless
probability of the trial
T^
if
we
trial is
if
i.
e.
compound
and
Pj
or
there
if
impossible,
P2
both p^
and p^
Example
1.
-r
P, and
But
favourable events.
P, and P^
is
found by
are
in
applicable
P^
but not
this proportion expresses that the frequency or

trial
This proportionality
T^.
^^
two observed values whose laws of errors are given.
any of the conditional
= PiPz-
for
we
0,
trials that
The condition
that
see
the
compound
cannot possibly succeed,
of certainty (probability
the favourable events of
must be proper
probability,
indicated by probability
is
trial is certainty
2^2 as probabilities
"J^
consider the multiplication of probabilities as the determinalion
is
impossibility
But
c:d.
of the law of errors for a function of
Since
compound
get
of their
not affected by the event of the
is
of freedom,
therefore
Considering each of the
the order in which the trials are executed,
according to
we
probabilities
the correct result
events, necessary
prove this very important
succeeded in b cases, only T^ in c cases, and neither in d cases.
two
is
probability
conditional trials are carried out in every case of
suppose that both
trial.
if
nevertheless,
product of the probabilities of conditional
the
as
imagine the conditional
to
Tj must have succeeded in the event, whose

However indifferent the order of
Tj, is p^.
trial
the numerical computation
in
correctly to be found
the trial T, must have had the event whose pro-
first
p,, and then the other
is
all
fractions, Pip.^ =p =
1) in
conditional trials; for as p^
be possible only
1 will
when
1.
When
the
favourable
events
of
the
all
conditional
trials,
in
number, have the same probability p, the compound event, which depends on the success
of
J
all
for
these, has the probability p".
If
by every single drawing there
is
the probability of
"red" and \ for "black", the probability of 10 drawings all giving red will be 7^24.
lities of
Example 2. Suppose a pack of 52 cards to be so well shuffled that the probabired and black may constantly be proportional to the remainder in the stock, then
the probability of the 10 uppermost cards being red will be
_2625242322212019
"~
18
17_
~
52'5r50*49"48'47"46*45'44'43
the
/9n(-c)
!gg Ijg
|^
|16 |10
^^^(K))
/352(10)
_
~
_19_ _
~ J__
56588
2978'
being binomial functions.

16
122
after
Example 3. Compute the probability that a man whose age is a

n years, and that he will die in one of the succeeding m years.
If
we suppose
n years
will be
-Pr.
The
that the
(1 5a)(l
die before
(I q+-i).
2a+l)
years will be
Qm
â+n+i) [qa+n+2 + + qa+n+m-2) qa+n+mi]}'t

+ (1 qa+n) {qa+n+i
I Qm = (I qa+n) (1 2a++i)
{I qa+m+n~\)-
qa+n
-f" (1
or
The required
probability of death
The most convenient form

a table of the probabilities
the
men from
After this
answer
a large
2,
player,
in
We
4.
Example
certain
-t^ )
is
will
not, as
we
+ n
years, is
here supposed,
but of the absolute frequencies
i,
population who
gets
be a
only
reach the
will
age
case of the
special
Z,
of
of
i.
general
of the 5
then
who
It
directly.
We
2,
3, 4,
5,
6, 7,
9,
8,
getting 11 and not
there that the
k^ player
in succession will be
9,
that none of the players will get 11?
be found perhaps that
it
is
probability,
11?
is
it
is
first
then
we
require.
not quite easy to compute these probabilities
a good plan to reconnoitre the problem by first bringing out
present themselves
as
quite
easily
and simply, without considering whether
In this case, for instance,
we take the
probabilities, p^, that
players will get 11.
attack
the problem
more
seriously,
and examine
simple functions of the probabilities we have found, p., which

babilities of the
For 5 players we
the probability,
they are just those
each of the
1,
loses
and 10 of the same colour.
is
In such cases
such results
two cards of the well-shuffled pack, and wins or
r*,
what
will
such a way that each
the probability of h players (named beforehand)
is
what
gets
Lastly,
of cards arranged in
game
others?
Secondly,
first
'a+fi+m
of the points on his two cards is eleven or not.
for instance, only the cards
What
'a+n
/^
imagine
order,
sum
according as the
any
(^
use,
statements of mortality
for
for all integral ages
q^
but before the elapse of w
years,
(properly infinitely large)
= -j^
q.
after
Pn Pn+m-
PQm
consequently
the
is i will
be alive at the end
of his then dying in either one or the other of the succeeding
Qm
probability
man whose age

man whose age is a will
the probability that a
q. is
the probability
his next birthday,
of
that
will be still alive
same
or similar sort as those inquired after.
may
if
there are not any
be interpreted as pro-
123
^îP2 Ps)
+ ^(lU Pi)
PO Ih-
-i-
Eepetitions of the same trial occur very frequently in problems solvable by
68.
the
theory of probabilities,
and should always by treated by means of a very simple and
important law, the polynomial formula.

Let us suppose that the various events of the single
and
may
trial
be indicated by colours,
that, in the single trial, the probability of white is tv, of black b,
The
probability that
we
x white, y
shall get in x-\-i/-\-z trials
and of red
black,
r.
and z red
results, in a given order, is then
w" -P
The number
r'.
of the events of this kind that differ only in order,
is
the trinomial
coefficient
^
nx,y,z)
which
And
is
this
1-2-3...
+ y + g)
i.2...x-l-2...y\-2...z'
the coefficient of the term w'' -hy-r'
in
the
development of
(m;
+ 6 + r)
(^+J'+^).
same term
r {x, y, z)
is
(a;
the required probability of getting
white
iV
-hy -r'
times,
(126)
and red z times by
black y times,
{x-{'y-\-z) repetitions.
When
the
that w-\-h-\-r-\-
probabilities of all possible single results are

.
1,
and when the number of repetitions
is
known and employed, so

w, we must consequently
imagine
(w-\-h-{-r-\- ...)" developed by the polynomial theorem, and the single terms of
the development will then give us the probabilities of the different possible events of the
repetitions without regard to the order of succession.
Example
1.
If the question is of the probability of getting, in
10
there are the three possible events of white, black, and red, even
numbers
each colour, and
and
we must
if
the probabilities of the single events are w,
retain the terms of {w-\-b-\-rY'^
which have even
b,
indices,
trials in
which
and z of
x, y,
r, respectively,
and we thus
then
find:
V+r^) + 210z<;^ (66+156 V2+156V<+r6) +

+45p2 (68-f 286r'''+706V+286V6+r8) + 610+456 V+2106V^+ 2106 V+456V8+ ro
= i{(M'+6+r)i + (t<;+6+r)"' + (M; 6+r)<' + (w+6 r)i} =
= |{l+(l 2w) + (l 26)' + (l 2^)"}.
,io_|_45m;8 (^,2.^,.2) _|_ 210m;6 (^,4+ 66
The
probability,
consequently,
is
always greater than \, but only a
the probability of getting some of the events in a single
trial,
Example 2. Peter and Paul play at heads-or-tails

But Peter throws with 3 coins, Paul only with
against).
(i.
2,
little greater,
is
very small.
e.
probability
and the one wins

16*
unless
a for and
who
gets
124
the greatest
of "heads".
both get the same number of heads they throw again,
If
may be necessary. What is the

we write for Peter's probability
as often as
If
=
qi
number
win?
probability that Peter will
and against throwing heads p, ==
for
and
i^
p^ =1^ and g'2

5' ^h^n we should develop {Pi-\- qi)^ -{Pi-}- q^)^, and
the terms in which the index oi p^ is greater than that of p^, are in favour of Peter;
those in which the indices are equal, give a drawn game; and those in which the index
of
|, for Paul's
greater than that of pi, are in favour of Paul.
is
^2
For the single game there
is
the
probability
for Peter of
Paul of
for
drawn game of
for a
^^,
^.
o
1
As
same way, when they play the games over

-~ *s
again, we need not consider the possibilities of drawn games at all, and we find
the
are
probabilities
distributed
the
in
Peter's final probability.
What
3.
Example
game which
won once
is
out of four
is
times,
repeated 10 times.
the probability of winning at most 2 of these?
is
551124
1048576
69.
which
we
It often occurs that
inquire in a general
a function of one or more numbers.
is
Often
it
is
way concerning a
problem into such a one of a more general character, where the unknown
Pn of
Pi, p^r P31
the
probabilities,
suffixes
being
the
hold good for an
arbitrary value
this
theory of probabilities,
method almost
p)
0,
table.
And
particularly such
with dilference-equations.
e.
i.
a whole table
Integration of
indeed of so great importance in the art of solving problems
finite difference-equations is
of the
of n,
is
arguments of the
then we must generally work with implicit equations, f{pi,

as
probability,
also easier to transform a special
we can almost understand
that
that
Laplace has treated
as the one to be used in all cases, in fact as the scientific quintessence
of the theory of probabilities.
Since finite difference-equations like differential equations cannot as a rule be integrated
by
known
cases, especially
equations
that,
with
functions,
we can
in
an elementary treatise deal only with the simplest
such as can be solved by iexponential functions, namely the linear differenceconstant
coefficients.
As
to these,
it
is
only
necessary to mention here
when
CmPn+m
the solution
is
given by
Pn
+ c^Pn =
=
k,r",
-^
(w being arbitrary),
-\-
KK
(127)
125
where
ri, ... are
the roots in the equation
Cmr'"-\- ...
=0,
Co
the corresponding roots occur singly
while k,, ... km are integration-constants whenever

but
rational
functions
integral
corresponding root occurs
the integration
leads to,
of
I,
if
the
it
can really lead
difference-equations which the theory of probabilities

which the exponential functions occur in connection with
of the
many
those in
particularly
times.
mention one other means, however, not only because
shall
to
with arbitrary constants, and of the degree
binomial functions and factorials, but also because
it
has played an important part in the
conception of this book.
The
late Professor L.
of transformation, which
Oppermann,
finite or infinite series
communicated
to
me
method
with an unessential alteration.
shall here state
in April 1871,
of numbers
Uq^
...
Ufi
can univocally be expressed by another:
*^0
"o
**!
^2
^3
~\~
2u 2
Wo
w,
=
=
where the sum
2'
may
In order, vice versa,
-f-
(128)
{iY2:jip{z)up,
be taken from
compute the
m's
to
M4
4m,
^"'i
6M4 +
4M4
U3
u, +
3M3
-f-
UK
W,
-\-
3o
"""3
oo
to
+00,
provided that Up
when
p^n.
by means of the w's, we have equations of just
the same form:
Wq ==
u^
M3
=
=
M,
U-r
expressed
in
as
in (17)
single
~\~
~{'
4<>4
=
=
M2
Here,
+ '^ ^3 ^^4 +
Wi 2^2 3MJ3
+ 3^3 6^4
tv^ 4to^
^Ô "1~ '^1
-|-
*<'2
-j-
(129)
w. 4ilYl"f3r(x)Wr.
and
(18), the
equation,
general dependency between the m, and
be means of an independent variable
z.
From
w,-
can be
(129)
we
get identically
If
we here put
the same form.
=
e'
e",
then
e'
1-
=--
e'
will reduce (128) to
an equation of
126
If M, is the
frequency or probability of
'
the relations of the values in Oppermann's transformation to the half-invariants
illustrate
In particular we have
and sums of powers.
!^\
^ î(ô+î)
Ws
'iPo(M'o
'^
now Mo,Mi,
If
then
w^, w^, ... w as function
probabilities,
^''o'
and
equation
of their
pretty
often
we know
if
Oppermann's transformation
this
Wi( Wo4-Oi)(m?o4-2m> J
But
turns out that,
it
a probability m, as the functional law of errors for
the same law of errors
we have
the
reverse
process to find
w,-
If
(for
we can look upon
expresses
and frequently we want nothing more.
itself,
the
series
are
pretty simple
if
is
but they are often less favourable for numerical computation, as they frequently
simple;
give
to
problems of
as the observed value, then
by symmetrical functions,
in
with than the original one
to deal
easier
through a difference-
only to a difference-equation
course leads
of
suffix.
is
this law only
instance the more difficult ones in Laplace's collection of problems).
If
which depend
... M are a series of probabilities or other quantities
their suffix according to a fixed law,
equation,
+ M',)
Wo'
M'o
for
1-2
'
on
an observed value, then also
taiten as
unknown
the
as a difference between
of remedying this, but
it
much
would carry us too
There
larger quantities.
far to enter into a closer
a means
exists
examination of the
question here.
Example
1.
throw a
twice, or lose by throwing
what
wth throw,
what
is
is
my
die,
and go on throwing
"two" or "three".
probability of
If the
winning?
If
till
either
game is to
number
the
win by getting "one"
be over at latest by the

of throws
is
unlimited,
the probability of another "one" appearing before any "two" or "three"?
Four results are to be distinguished from one another. At any throw, say the ^*,
Let the
the game can in general be won, lost, half won (by only one "one"), or drawn.
probability of the i^^ throw resulting in a win be p., of the
in half
win
Thus the
s,,
and in a drawn game be
probability of a second throw
s^,_j -j-r,._j.
It is easy
to
express p.,
is
g.,
r,-,
then pi
^0,
same resulting
in a loss be q.,
gi
g,
^,
Si
and r, == a.
* throw
|, and, generally, the probability of an

r.,
and
s,.
in terms
of r._,
and
s._^,
and
also
127
Pi-fi
!?,_i'
^^^
''i-i'
and
^" terms of r._2
*,-i
By elimination then the
etc.
s^_^,
diifereuce-
equations can be found.
When we
the
or
replace
g'
or s or r by
common form
a;,
which
x,_,
integrated as
is
Xi
for
ri
+ 6J)2-';
-ir,_,.
we have determined the
probabilities of the first throws,
When, by the
we
(a
difference-equation can be written in
=0,
-\-{Xi^2
r we have the simpler form
x the
constants,
get
Pi-
3.
'
.= 42-,
and
r,
We
then have the formulae P
making the winning
probabilities of
In
repeated a great many, w, times.
If
successive
game
games
are
where
of dice,
won,
ra
the
1. (2-l)_n
w
3(2-l)
game
2.
=î -j-
probability
it
what
is
the
5in for
and we get
Poc___l
+ (?oo
9;
^^^
of
= 9't+
P<
winning
is
The same game
is
then happens at least once in this series that
you get a prize.

^
and ^n
-|-p
or losing throw,
P+<?
Example
2-'.
the
What
is
the probability p of this?
In
probability of getting a series of 5 "sixes" in
10000 throws?
It will
got in the
first
be simplest to find the probability,
r repetitions.
q^=
The difference-equation
^>^,
that the prize will not be
for this is
or
^'C,
where
How?).
where
is
(b)
Ci,
2,^
- (1- ) {q^^^_^ + ra?,.+_2 +.+ "'-=' ?.+! + '3'"^^ ?.} -
the
first
integral
of
(a).
(As well as
(a)
we can
directly demonstrate
(b)
(b).
Hence
Cm are constants, which as well as Cq
must be determined by means
of
128
==
""'
^' Vm
1m-i
the
n*
of
equation
degree, which
Q'o
?!
only a
than 1;
less
little
The
vi.
small
done, and
be
this
use the difference-equation

of those
which are known
unknown
values of the
enables us to get
(b)
plying (b) to 5,,
of these
q^_^_^
ffi""
(c)
(for small
roots
or large m's)
ra's
when
in
is
even;
will
be
the others are
highly desirable to avoid the complete solution of
it is
problem
most important
illustrate a
will
(c).
We may
artifice.
compute a single value of the unknown function by means

us from the conditions of the problem, and then successive
to
to
means
in terms of
q^^J^^
negative root occurs
function by
"+'
roots of an irreducible
/)m are the
*'0
Pi
also small.
In the actual computation
This can
/>"
largest
always imaginary, and they are
aiid
got from
is
/>"+'
by dividing out
"''
of those
Then we
q.
q^,
already obtained; here, for instance,

get
q^_^_^,
either
by again ap-
or by applying (a) to q^ and

best in both ways for the
q^^^ (or
sake of the check), etc.

It is
evident that the table of the numerical values of the function which
we can
form in this way, cannot easily become of any great extent or give us exact information
as
to
the form
of the
form of the function

In
our
example
is
But we
function.
known
are able to interpolate,
we may be
(as here),
we need only continue
the
and,
when the
general
justified in using extrapolation also.
computations
above
described
the
until
to the greatest root

term in q^
dominates the others to
c^p'^-\- ..., corresponding
p^,
such a degree that the first difference of Log q^ becomes constant, and the computation of
indices can then

q^ for higher
numerical case
q^
1-004078
made
be
as
by a simple geometrical progression.
(0-9998928)'-
qmoo = 0-6577.
white and a n black
Example 3. A bag contains n balls, a

drawn out of the bag and a black ball then placed
After the y* operation the white and black
times.
the
probability
ones
m^(/)
that
the
numbers
white
and
it,
this process
balls in the
balls
will
have
"^(i')
^^"^(2/-l)+^-'*'+i(y-l)
and
Ux(0)
By Oppermann's transformation we
-co
to
a:
except Ma(0)
1.
find
.(/)
taken from
0,
-{-cc
=
,
( l)''2'jff^(2;)-<T(2/),
or
is
be
x and
ball
is
repeated y
bag are counted.
then
n x.
We
I'
of
in
ones.
In the
the
Find
black
129
The
x under
limits of
infinite,
being
This difference-equation, in which
^'
is
-\-l can be replaced by x, consequently
the variable,
As we have,
easily be integrated.
further,
set
By Oppermann's
inverse transformation
we
find
= (-i)"^^/?.!^)
a;= oo toa; = -f-<- This
My) =
My)
2"
may
taken from
now:
(-1)'
â{z)
(-^)'',
expression
M^)i{-iY+'^-â-r(z-x).(^-^y
has the above mentioned practical short-comings, which are sensible particularly
if ,
x,
y are large numbers; in these cases an artifice like that used by Laplace (problem 17)
becomes necessary. But our exact solution has a simple interpretation. The sum that mulor
tiplies /?
(x)mux{y),
of the values
all
is
the
(^)', ("^^^^f,
these consecutive values.
easy,
solve
to
analogous
If
this
to that
difference of the function
(axy^
problem
We
f^=^)', (")',
and
that
it
expressed by Ux{y), then
is
and
square of the
is
formed by
possible,
if
not
without the integration of any difference-equation, in a way

4.
of wîy) to give us the half-invariants /i,, fi^
of errors as
found by a table
is
as the final difference
learn from this interpretation
used in 07, example
we make use
("-)
we
find for the
mean value
of
for
the
after
same law
y drawings
XAy)-a{^'^J
for the
-"
XVI.
mean
error
(("')'-(~T)+K("^)'-("i-')")-
THE DETERMINATION OF PROBABILITIES A PRIORI

AND A POSTERIORI.
70.
The computations
of probabilities with
babilities to be given,
common
which we have been dealing
in
the
we always assume one or several proand then deduce from them the required ones. If now we ask, how
foregoing chapters have this point in
that
17
130
we
those "given" probabilities,
obtain
it
Without experience
it
is
theory in these as well as in other determinations
agreement with
is
we cannot
must be
first
and without
reality,
distinguish
sight
between two
to be purely theoretical,
of probabilities is based on estimate of equality,
inequality, or ratio of the probabilities of the several events, and in this process
assume the operative causes,
or
at
mode
rate their
any
clear
as purely empirical.
is
The a priori determination
71.
it
get any firmness or exactness.
reason to
special
methods, one of which, the a priori method seems at

while the other, the a posteriori method,
are necessary than
in these original determinations of proba-
impossible to insure
in determining probabilities, however, there
means
other
tliat
mention, and provisionally
to
and experience must cooperate
that both theory

bilities.
evident
is
which we have hitherto been able
those
of operation,
to be
we always
more
or less
known.
On
the one hand
we have the
respect to the events but that each of

for preferring
in
typical cases
them
is
possible,
which we know nothing else with

and in the absence of any reason
any one of them to any other, we estimate them
to be equally probable
For instance
What is the probability of
though certainly with the upmost uncertainty.
seeing, in the course'ôr|time, the back of the moon? Shall we say | or .? 9?t: j/.
:
On
in
the other hand
we have the
cases
which, by virtue of a good theory, we
causes
at
work
for
that,
equally typical,
know
so
much
theory must occur as frequently.
and
-g
of the causes or combinations of
each of those which will produce one event, we can point out
another (or n others) which will produce the opposite event,
result at
but far more important
In this case
and
pr respectively,
we must
and which according
to
the
estimate the probability of the
the conditions stated be strictly fulfilled, such a
if
determination of probability will be exact.
But even
if
such a theory
way obtain
probabilities,
errors, that
we may
may
be used
as
which are
very well
is
not absolutely unimpeachable, we can often in this
so
exact and have such
nearly
make use
of them,
We
our theoretically given probabilities.
kinds of computations.
are
In astronomical adjustment, for instance,
not more strict in

it
practice to consider all times of observation as theoretically given.

ever, will often give occasion to sensible
infinitely small
mean
and compute from them values which
is
other
almost an established
Their real errors, how-
bonds between the observed co-ordinates; but the
would require great labour to avoid the drawback.

Such an a priori determination of probabilities is particularly applicable in games.
fact is .that it
For
it
way
that,
is
essential to the idea of a
game
on the one hand they exclude
that
babilities possible.
The procedure employed
must be
laid
down
in
such a
computation beforehand of the result in a
all
particular case, while on the other hand they
the rules
make a
in
pretty exact computation of the pro-
a game,
e.
g.
throwing of dice or shuffling
^^'^''^
131
of cards, ought therefore to exclude all circumstances that
causes
But
in
which could bring about
train,
of the probabilities very insecure, but, above
judgment
become a
When
The
struggle.
so-called stock-jobbing is rather a
we speak
of a subjective probability.
which ignorant people have of
understand
that
there
who know
those
make
depend on the
it
game and
fair
war than a game.
the estimate of the probabilities depends essentially upon personal knowledge,
The
fear
all,
Otherwise the game will cease to be a
theoretical insight of the parties.

will
to set
(corriger la fortune).
ought to be eliminated, which not only by their incalculability
also those circumstances
make
might permit the players
or further a certain event
is
but a
little,
This too plays a great part, especially in daily

that
all
than for
estimate, and that
the
uncertainty in
great
new and unknown, proves
is
it
is
life.
that they
greater for
those who know more and are therefore better able
to judge.
Koulette
be taken as an example of the objective probability which arises
may
a well arranged game.

scale
of a
and the
circle
whose center
in
is
suppose a red one
The pointer
the pivot.
game depends on where
result of the
it
stops.
There we have as essential circumstances:
3) the
manner
in
which the
The length of the arc

small velocities, and when the
only
be
may
is
made
to revolve quickly,
If the pointer stops opposite a space
previously selected as favourable, the
is
game
won.
the length of the arc which
1)
and the
versed, this being determined by the initial velocity
and
in
pointer turns on an almost frictionless pivot and points to the
is
tra-
friction, 2) the initial position,
circle is divided.
unknown,
is
friction,
regarded as given,
as
especially
when we take
already mentioned,
is
care to exclude very
very slight.
So
much
that the frequency of a given length of the arc must, as
function of this length, be expressed by a functional law of errors of a nearly typical form.
For the frequency must go down, asymptotically, as

of the
least
arc which will be separated
maximum
one
full
many
by
between these limits.
If
far as 0,
revolutions
now,
both below and above limits

of the pointer,
for instance, it
and with
at
depended on, whether
the arc traversed was greater or smaller than a certain value, the apparatus would be in-
expedient,
it
winning space
repeated
of the
would not allow any tolerably trustworthy a

(or spaces)
regularly
priori
estimate.
will
be
even
exact.
very
For an area ABab, bounded by any
continuous curve whatever (in the present case the curve of

errors of the different possible events), by the axis of abscissae,
and two ordinates, can always as a

expressed as the
pP, qQ'
....
sum
with
But
if
the
small in proportion to the total circumference and, moreover,
each of the numerous revolutions, then the a priori determination
for
probabilities
is
of
first
approximation be
numerous equidistant small areas

constant base,
multiplied
by
the
17*
finite,
132
interval
^ qr =^
the total
imation,
...
even very convergent, in such a degree as 6 (x)
is
small x, and the said approximation
for
and divided by the base pp'

...
And
q(f
area of a curve of errors, then the series of which the first term
pq
That the
initial
the result of the
is
x -\- x* -^
1 -\-
we speak
is
this
x^
approx-
x^^
-\-
of
-\-
sufficient for all practical purposes.
of the roulette is
position
if
unknown, does not
foregoing, viz. that the probability of
is
winning
essentially
~.
can only cause an improvement of the accuracy of this approximation.
change
This uncertainty
If
we may assume
that the pointer will as probably start from any point in the circle as from any other, this
determination
known
The
the
will even
third
essential
The a
priori
circle
whose mean
error
different values,
whole circumference pq,
to the
essential to
is
it
determination of probability
most
^;^'
cannot be determined wholly
circumstance,
or a counting
give results of the

to
winning space on the
of the
ratio
measurement
up
any regard to the special kind of the un-
be exact, without
function of frequency.
can thus,
but demands a
a priori,
know.
according to
circumstances,
from the very poorest through gradual transition
such exact probabilities as agree with the suppositions in 65 seqq., and permit the
probability to replace
method cannot
give,
is
quantitative
numerical value of the probability

the roulette,
that
itself.
statement
Only when
computations that are to be relied on.
we cannot
entirely avoid building on altogeth

is
necessary, and in
measure, we must
order
the
of
this uncertainty is infinitely small,
bilities in
great caution
But what the a
the whole law of errors for our predictions.
it
is
uncertainty which
evident,
as
in
priori
aiTects
the
the example of
can we make use of a priori proba-
If in the
work and struggles of our
life,
uncertain and subjective a priori estimates,
not to overdo this caution for want of a proper
try, by tact or experience, without any real method, to get
an estimate
of the uncertainty.
PJven
the dice
may
by the best a priori determinations of probability caution

be
72.
large numbers,
false,
By
the pivot of the roulette
inferring
and count the numbers
Owing
for every
we
build on the law of the
from a law of actual errors in the form of frequency to the law

that
of
the probability.
for the favourable
We
and n
event,
vourable event must be transferred
repeat the trial or the observation,
for
to the signification of a probability as
unfavourable
not superfluous;
be worn out or bent, and so on.
may
the a posteriori determination of probabitity
of presumptive errors in
is
the unfavourable events.
mean
for every favourable event,
value, the single values being
the probability
for the fa-
unchanged from the law of actual errors to that of
presumptive errors; consequently
= -^.
(130)
133
same consideration, the square
Since, according to the
'^
a single
trial is
'
the square of the
mean
rs
is,
we
are to
If
m A^
we have made,
mean
we then ask
the
for
The
mean
whether this
is
one of
still
make, and
for
which
to
mean
=-
^^ 7.
'
,t
6*
*'^^
^''O'"
i> (1
(i32\'
1)
error.
mn
(m -{- n)^ (m
or
Pi
shows that the mean error at a single
m -j-
repetitions, can be
sources of errors, one of which
error of the probability
identity
nm
The a
approximation
is
is
-\-
the square of the
probability.
trial,
-\-
n)
(m
-f~
1)
(133)
^2
when the
probability
determined a poste-
is
in
two mutually
to the probability, for
error.
therefore
never gives an exact result, but only an

of repetitions
be regarded as insignificant,
But even by the very smallest number
it
is
may
we employ
is
so
we can immediately
the law
for
of
we not only obtain

the mean error, which may
of repetitions of the trial,
probability, but also a determination of
be useful in predictions, and
free
the inaccuracy of the a posteriori determination,
is
mean
may
found by means of them as complete expressions
employ the probabilities
must be admitted that
{m
1)
Only when the number
large that their reduction by a unit
some knowledge of the
nm
+ ^2 iP) =
posteriori determination
to the
computed by (34), as originating

the normal uncertainty belonging
while the other
p>q (123),
which a^ (p)
errors.
trial,
- P)
_
m^n\
(w 4- n)2 (m +
,,,x
{m-\-ti)^
which Xo
m~\-n,
\)
which we are
'<^
^ "'^
as the square of this
for
is
compute the uncertainty.
by
error for a single
or is a repetition
n repetitions, we have
riori
toi
errors must, according to (47), be
therefore, the square of the
those which
deviation
and the number s^ of the repetitions
(m-\-n){m-\-n
which
mean
of the
serve as a measure of the caution that
is
necessary.
It
not such a simple thing to employ these mean errors as those
in the ideal theory of probability, but it is not at all difficult.
As above mentioned, the a

empiric; theory, however,
trials
we make use
of
posteriori determination of probability
takes part in
must be
it,
repetitions,
but
in
is
the
seems
concealed in the demand,
same way
as
the
future
to be purely
that
trials
all
the
whose
and uncertainty are predicted by the a posteriori probabilities. Transgressions of

rule, which reveal themselves by unsuccessful predictions, are by no means rare, and
results
this
compel
statistics
and the other sciences which work with
probabilities, to
many
alterations
134
and hypotheses, and
of their theories
to the division
of the materials obtained by trial into
more and more homogeneous subdivisions.

Example. A die is inaccurate and suspected of being false.
we have on throwing it 126 times got "six" exactly 21 times, and
The probability of "six"

mean error is k^îp)
11
is
found,
15
= ~q--q'i^
=
~^.
consequently
now we
If
still
mean
Ideally,
and
mean
all
far,
i;
"
that
we
mean
by the
shall
not get "six" in
right.
errors
are
6 throws,
...,
(1
the
mean
is
the square of the
^)6 == j| = _-_j- but what

an accurate die

= +
square should be
can have a small error dp,
the
limits indicated
(1
its
then the square of the
:^
'^^"
however,
trial,
]-.
the probability
as by
probability is
error?
or
seek
^^^
900'
be
to
consequently,
On
so
error oi
I's,
(1
i*) (1
consequent error in
(1
p)^
/')*)
p) -
p{l
But
---
-5-
will be
the
now
is
the
if
6(1 pf dp;
^
if
the total square of
error of the probability of not getting "six" in 6 throws will be
h = (l-pf{l-{l-pf) + ^&{\-pY''-p(l-p)-^
In every single
an accurate
game
die,
but
mean
of this sort the

its
error is therefore only slightly larger than with
actual value is so large (nearly \)
on the part both of the player and of his opponent,

their laying a wager.
tions of the game.
This
may
6 throws without "six" 72 times.

expectation
mean
of
winning 72
error of this result,
-^
the
there
is
With
the
== 24 games.
first
terra
in
if
we
In
the
the
number
are to play this
above approximate
much
caution
much chance
not
be remedied by stipulating for a large
Let us examine the conditions
that
as to call for so
game
fractions
of
of repetiof
making
there will
be
computation of the square of the
above
X^
must be multiplied by
72, but
the second by 72^; hence

A,
The mean
=
=
+ 3^-^.5184
16 + 33 = 49,
72
|-.
error will be about 7, while it
would only have been
4,
if
the
die
had been
quite trustworthy.
73.
which
is
We
have mentioned already, in 66, the skewness of the laws of errors
peculiar to all probability.
law of actual errors
to that of
It
does not disappear, of course, in passing from the
presumptive
errors,
and in the a posteriori determination of
produces what we may call the paradox of unanimity: if all the repetitions
probability
we have made agree in giving the same event, the probability deduced from this, a posteit
riori,
must not only be
or 0, but the square of the
mean
error
X (p) of these determi-
135
nations (as well as the higher half-invariants) becomes
to
certainty or to impossibility,
= 0.
Must we
because a smaller or
only
infer then, respectively,
number
greater
of
repetitions
mutually agree? must we consider a unanimous agreement as a proof of the absolute

correctness of that which is thus agreed upon?
Of course not; nor can this inference be
maintained,
if
Such a law of
n
fin-fjtl
CO.
we
look
more
law of errors
law of errors which
skew
is
^=
(i^
not a strict consequence in practical calculations
must be
mean
is
error,
so small that
it
may
The paradox being thus
the
0,
that,
number
Add
must be
it
explained,
follows
that
it.
//^
must indicate
that
of a
is
must warn us
it
it
is
number,
but only that
0,
it
indeterminate.
no objections against the use
But
0.
the ratio
to this
the square
because
itself
as here,
be treated as a differential, which otherwise
posteriori probabilities in general can be based on

in
0,
an infinitely high degree,
in
something peculiar, even though the mean error be ever so small.
here that of the
Oj /13
but not when,
be sure, matj signify certainty,
errors, to
closely at the
of
to be cautious
computations with such probabilities as observed values, where the computation, as the
method
For this reason, we must
of the least squares, presupposes typical laws of errors.
for
such computations reject
as
unsuitable
material
of
hypothesis or theory of the
all
unanimously or nearly unanimously determined probabilities
Another thing
observation.
computation,
example we may take an examination

a posteriori statistics before the
if
it
is
does not
we must
that
explain
the
also
reject
unanimity.
As an
of the probability of marriage at dift'erent ages.
year and after the
c. 20'''
c.
60"^
must not be used
the
The
in the
computation of the sought constants of the formula, but the formula can be employed only
when
it
towards
has the quality of a functional law of errors so

0,
it
approaches asymptotically
both for low and high ages.
The paradox
of unanimity has played rather a considerable part in the history of
the theory of probabilities.

riori
that
It
has even been thought that we ought to compute a poste-
probabilities by another formula
-J^Jl1_
(134)
(Bayes's Rule)
and not, as above, by the formula of the mean number
P
The
proofs
some authors have
In the "Tidsskrift
for
tried
m
m -\- n
to give of Bayes's rule are
open to serious objections.
Mathematik" (Copenhagen, 1879), Mr. Bing has given a crushing

and their traditional basis, to which I shall refer those of my
criticism of these proofs
readers
who take an
interest in the attempts
that
have been
of probabilities mathematically from certain definitions.
made
to
deduce the theory
136
Bayes's rule has not been employed in practice to any
not in
statistics,
though
makes the paradox of unanimity disappear
it
a convenient way,
in
can neither prove nor disprove the exact validity
formula
of a
an a posteriori probability, any more than we can do so

the law of actual errors to that of presumptive errors,
we
tested by its consequences in practice before
never deviate more than at most the amount
rule
particularly
as
we
as,
the
determination of
after all,
any transition whatever from

deserves
certainly
the
give
mean
of the
But
and
up altogether.
will
m -j- n
events out of
for
for
the rule
it
give
a test will be that the hypothesis that Bayes's
series of repetition, viz. that
greater extent,
works entirely with a posteriori probability.
this science
error
The
true
from
be
to
result of such
probability,
the
can
of the
result
In order to
have proved favourable.
demonstrate this proposition we shall consider a somewhat more general problem.

If
V
we assume
made which have given
that trials have been previously
we have now
unfavourable events, and that
in continuing the trials found
and H unfavourable events, then the probability, being looked upon as the
//
favourable,
favourable
mean
value,
is
determined by
m -j- w
of which Bayes's formula
the special case corresponding
is
would therefore agree with the general rule,

so
much
-|- /i -|-
of the probability of both cases,
if
we knew
to
= =
v
1.
Bayes's rule
before the a posteriori determination
a report of one earlier favourable event and
as
one unfavourable event.

In the more general
(w
^^
and
for the
If
in-\-n
It
-\-
^2 {on
-j-
-\-
computed
this
value, that
(fin
{m
(w -f w)
-\- fi) [n
w)
(m
(n
is,
mean
-\-
(m -^n)
the square
between
m)^
I/)
fi)
-\- v)
trials is
(w -\-n)pY
^2
{m
we now compare with
servation and its
(m
the square of the
case
v)
-\- [I -\-
'
1)
deviation
between the ncAV ob-
and {m-\-n)p, we
m -\- n -{- -\- v

m ^- n -\- -\- v
fi
(m -|- w)
now
k^.
of the
error at the single trial is
find
[i
appears at once from the latter formula that the greatest imaginable value of the ratio
the greatest of the two numbers

the absolute
With
maximum
of
it
will
v.
In Bayes's rule
/^
the ratio of the square of deviation
respect to Bayes's rule the
the same time
and
{i
= =
postulated
be seen that
we
proposition
is
1.
Here, therefore,
to that of the
mean
hereby demonstrated.
can replace Bayes's rule by a better one,
if
is
1 is
error.
But
at
there
is
137
only an a priori determination, however
we take the a
priori probabilities
ra
of the
uncertain,
for,
and
(1
we
probability
s) against, instead of
are
seeking.
and
fi
m^
(137)
then we are certain to avoid the paradox of unanimity where
much
so
deviating
the
as
mean
from
error
If
so that
v,
the
might do harm, without
it
observation
the
in
posteriori
determination.
we can always
Neither Bayes's rule nor this latter one can be of any great use; but
employ them, when the found probabilities can be looked upon as

the other hand, the formula of the
mean
Where
the paradox of unanimity correctly.
adjustment, the latter formula, as
value
have
may
the
said,
be used in
definitive
all
cases,
results.
if
we
On
interpret
found probabilities are to be subjected to
must be employed; nor can the other rules
be of any help in the cases where observed probabilities
have to be rejected on
account
of the skewness of the law of errors.
MATHEMATICAL EXPECTATION AND
XVII.
Whether the theory
74.
of probability
elsewhere, in all cases nearly in which

of the practical
result
The gain which a favourable event

it must as a rule be
bought by a
the value
game
it
is
of the
entails
stake.
each of them
is
V,
mathematical expectation.
of the
has a value
The question
and
is
the
How
chance
of
winning
we
to
compare
are
Imagine the
and the number of repetitions iV to become indefinitely
according to the definition of
clear,
of a favourable event, the prediction
a computation
with that of which the game gives us expectation?
latter
to be repeated,
employed in games, in insurances, or
we can speak
won through
is
is
MEAN ERROR.
ITS
probability,
that the
must be pNV, when p indicates the
sum
of
probability.
large, then
the prizes won,
The gain
if
to "be
expected from every single game is consequently pV, and this product of the probability
and the value of the prize is what we call mathematical expectation.
The
adjective "mathematical" warns us not to consider
the possible gain has for a single player.
pV
as the real value
which
This value, certainly, depends, not only objectively
on the quantity of good things which form the prize, but also on purely subjective circumstances,
sort of
among
others on
good things.
called the
An
how much
the player previously possesses and requires of the same
attempt which has been made to determine by means of what
"moral expectation", whether a game
regarded as a failure.
For
it
is
is
advantageous or not, must certainly be
takes into account the probable change in the logarithm of

18
138
the
but
property,
player's
it
We
subordinate circumstances.
at the
of the
of that
shall not here try to solve this difficulty.
with respect to the mathematical expectation, that
It is evident,
unbound games
does not take into consideration his requirements and other
same time, the
several
games.
mathematical expectation
total
The same
the
is
we
if
case,
play a
event entitles the player to a special (positive or negative) prize.
speak of the total mathematical expectation as
Example
a
nothing;
expectation
We
1.
play several
sum
equal to the
is
in
game
this
which each
we
case
latter
of partial ones.
and one of 6 wins 8
s.,
ixO + Jx2-|--Jx8^2s. A
then
we
play with a die in such a way that a throw of 1 or 2 or 3 wins
throw of 4 or 5 wins 2
is
made up
In
if
The
s.
stake
of 2
total
mathematical
will
consequently
s.
We
might also deduce the 2 s. throughout, so that a throw

of 1, or 2, or 3, causes a loss of 2 s. and a throw of 6 a gain of 6 s.
the total mathematical expectation then becomes
0.
correspond to an even game.
1)
In computations
2.
Example
number
the table of the
of persons
such a person living x years

of
his
the
at
dying
of the various
exact
is
living
1(a)
ii-.
may
at a given
The
a.
age
of his dying within
age of a-\-x years
other necessary probabilities
of life-insurances the basis is
kinds
dx,
and
The value
now a
an endowment of
of
years old
is
then
see, is
most
Such
is
now a
of an
years old
the formula
a:
where /(()
But
it
from
these
all
V, payable in
(1
(1-j-yo)-')
who
years, if the person
^'
^^^^^^
J)(a)
computed by means of a table of the function
easily
l{z){l+p)-'.
of great use for other purposes also.
The value
person
capital,
l{a)(\+p)-
D{z)
a table
j,
thus equal to the mathematical expectation
is
^^+Pf
l{a)
which, as we
alive,
'
be found; 2) the rate of interest p, which serves for the
valuation of future payments of capital, {l-\-p)-''V, or annuities certain
is
years
of
probability
annuity,
shall
live,
v,
due
at
the end
can be computed as a
x
00
and Z)(qo)
of
every
sum
year
of such payments,
or
by
00
0.
deserves to be mentioned
that this same mathematical expectation
safely looked upon as a total mathematical expectation in a
game whose
possible years of death
first
through which
the probability of death
in
the
is
most
events are the various

^^
year being
'
T/I""^^^
'
'"
139
the second
,
I
and
so
on
while the corresponding values are annuities certain
(a)
of V for varying duration.
we
In this way
expression
X
find
for
the value of the life-annuity
00
x
-^S^{lia + x)-lia + + l)){l-il+p)--).
l(a-\-x-\-l)) =
00
sum ^(l(a-\-x)
Since the
thesis that the expression
may be
7- J
and
this
we
1(a),
(140)
find by solution of the last paren-
written
^)2î^+-)-Hu+x+i))ii+pr
X
shows that the value of the life-annuity
which the yearly interest
of
the
is
the difference between the capital
is
of
and the value of a life-insurance
sum
payable at the
beginning of the year of death.

In life-insurance computations integrals are often employed with great advantage,
instead of the
sums we have used
here; periodical payments (yearly, half-yearly, or quarterly)
being reduced to continuous payments, and vice versa.

75.
is
expressed in
mean
it
error
That mathematical expectation is not a solid value, but an uncertain claim,

the law of errors for the mathematical expectation, and particularly in its
for,
owing
does not matter
often
skew.
If
to the frequent repetitions
much
the
value
V is given
=fV,
mathematical expectation, tl
free
is,
XîH')
total expectation of
By
free
games we may pretty
safely
events of the same trial or game.
The mean
error
upon a proposed game,
is
or
=^
computed by
(141)
p{l-p)NV^;
(142)
H" = EpiNiVt,
we have
Ipi{].-pi)NiVi\
(143)
understand such as are not settled by the various
(As to these, see 76.)
highly
we
are
to
computing whether we ought

value
uncertain claims
or
to
enter
outstanding
balance of accounts.
the person
the
= pNV
excellently adapted for
how
error of
get
mutually free games,
X^(H")
mean
=^(l-p)F.
E'
for the
games and insurances,
of error, the square of the
same game we
iV repetitions of the
and
and
in
according to the general rule, to be
/i,(/^)
If there are
and combinations
that the isolated laws of errors, here as for the probabilities, are
Such things of course are regulated by the boldness or caution of

concerned; but even the most cautious man may under fairly typical circum18*
140
stances be contented with
diminishing the value of his mathematical expectation
mean
times the amount of the
and
error,
would be sheer foolhardiness,
it
by 4
a passionate
if
player would venture a stake which exceeded the mathematical expectation by the quadruple
mean
of its
On
error.
mean
the other hand, a simple subtraction or addition of the
error
cannot be counted a very strong proof of caution or boldness respectively.
A game
1.
Example
arranged in such a way that the probability of winning
is
from the person who keeps the bank

expectation with mean error
but
is
^, the
then (0-8
is
prize
+ 2'41/w)
imprudence estimate his negative mathematical hope,
He must
(J*.
much
has got so
2304 games,
460 g 80
enter
will
w games
fi.
game
only
or
capital
also so
become very
many
safe;
to
regard
part
of
game,
ij^
mean
error
game
is arranged in such an entertaining
of taking
in
part
tinue for 144 games,
(1-0
144(0-8
the
0-4)) ^.
who must
for
But
man owns
every
way
game, then even
mean
error
cts.
cts.
per
x 48.
which
after all,
game
is
But who
not worse
is
already greater than the mathematical expectation;
is
the
(a
40
will
however, he
If,
the average gain of 20
upon such a game against the banker
and
he can play about
customers, that
4 times as great as the mean error 2
or exactly
cts.,
as
credit,
his business will
prudent
if
x 144 + 2'4 x 12
at about one dollar,
will
only augment
No prudent man will enter upon such a game, unless he can thereby
in this way we insure our risks, because it is too dangerous to be
If the
the mathematical
banker has no property,
If the
his fear, lower than 0-8
the stake for each
fix
consequently
than so many others)? The very stake

every
In
Jf.
a chance of seeing the bank broken about once in six times.
stand
thus
expect 144 games to be played before the prizes are to be paid, he cannot without
may
= 144
is
that
we pay 40
rather a
{-^2-AViM
cautious
as above)
the disproportion.
avoid a greater risk:
own
"one's
insurer".
cts. for
the excitement
person
may
being only 28
if
also con-
80
cts.
or
For a poor fellow, who has only one dollar in his pocket, but
some reason
necessarily get 8
only 2304
and
fi,
such a game may also be the best resource.
Jf,
fails
if
he cannot get 8 times as much, then he
would be exceedingly foolhardy if he played 2304 times or more in that bank. If we must
run the risk, we can do no better than venturing everything on one card; if we distribute
our chances over n repetitions, then we must, beyond the mathematical expectation, hope
for
Vn
Example
total
insurance
2.
of
Two
mean
which might help by the one attempt.

fire-insurance companies have each msured 10,000 farms
times that part of the
f 10,000,000.
error
The
yearly
probability
both must every year spend f 5000 on management.
of
damage by
Both have
to rest satisfied
with one single mean error as security against a
How
high must
either fix its annual
pany
premium, when
has 10,000 risks of 1,000, while
there
has insured:
is
sufficient
deficit in
fire
is
for
^,
and
guaranty-fund
each
fiscal year.
the difference that the
com-
141
142
we may
their differences,
even
the computation
in
game by subtracting the mathematical

Thus we may write:
mean
of the squares of the
expectation from
errors reduce to
the gains, and adding
all
it
to the losses.
X^ (N,H(l,
This rule then
il-p,),
...
w))
N{p, {a,-
Ha,
can
now compute
mean
the
))+
+^(a_//(l,
(146)
n)y-).
errors
in
the
examples
and
2,
in
In
74.
we have
-i(0)2
In the life-annuity example
+ i(2) + ^(8)2_2^ =
we now
(140) for the value of the annuity, rather
sum
of a
number
and only the deaths
of
endowments;
see the advantage of using the longer formula
than the formula (139) which gives the value as

the partial expectations are here not unbound,
for
in the several years of age exclude
co-ordinate events in the same game.
we
from the rule of unbound games only in the absence of the factors
differs
i,(H)
the
{l-p).
We
No.
one another and can be considered
For the square of the mean error of the life-annuity
have, from (144):

a:
=^Qg
+ ^)-H(^ + ^+i))a-a+p)-')y
^îJJa)2{^(^(''
In the above studies on the
77.
mean
have supposed that the probabilities we use are
errors
free
of mathematical
from
(147)
a priori by good theory or found a posteriori from very large numbers of repetitions.
determination
is
observed
numbers of
trials,
computed a priori presuppose

The same warning must be taken with
or if probabilities
with sensibly large mean errors.
respect to other values which
may
value of the gains, for instance,
enter into the computed mathematical expectations; the
may depend on
the future rate of interest.
of the manifold sources of errors are to be omitted
must
or not,
total ^2-
of the
As
mean
for
each special case
errors,
computation of the mean
error,
smallness of their parts of the
considered in 75 and 76, are, as a rule, very important, while
problems are often insignificant.
computed by the method of the

for
depend on
in
the relative
Whether some
to the theory of probability it is characteristic only that the parts of the squares
analogous parts in other
research
This
not complete in the cases in which the probabilities determined a posteriori
are found only by small
values
we
expectations
being either determined
error,
its
next discovery,
least
we have
squares,
to
then,
compute
in
the
When
order
mean
the
the orbit of a planet

to
restrict
errors
of its
the limits
is
of
co-ordinates
143
at the next opposition. Ordinarily these
mean
errors are so large that the X^ for its future
may be wholly omitted, though this yij is analogous to those from 75 and
But when we have computed a table of mortality by the method of the least squares,
we can certainly find by that method the mean error Vki {P) of the probability of
observations
76.
computed from the table; but if we are to predict anything as to the uncertainty
with regard to n lives, and with regard to the corresponding mathematical expectation npa,
then we must not, unless n is very great, take the mean error as naVXîp), but we
life
must, as a rule,
first
a]/np{\p)-\-n^/ii{p).
take
^2
(^)
'i^
consideration,
(Comp. example, 72).
-"C>*xK(c>"
and consequently use the formula
Fig.l
Fig.2
Fi3.3.
'_^-"-^'^-'^Z'\,ISltiL
14 DAY USE
RETURN TO DESK FROM WHICH BORJROWED
LOAN
DEPT.
This book is due on the last date stamped below,

or on the date to which teoewed. Renewals only:
Tel. No. 642-3405
Renewals may be made 4 days ptior to date due.
Renewed books are subject to immediate recall.
HECDLD
MAY
U. C.
JUN
2 1999
BERKELEY
IJ)21A-20m-3,'73
(Q8677sl0)476-A-31
R)UC.-KYCH. LIBRAW
General Library
University of Califortiia
Berkeley
LD
21-100m-8,'34
s#'"
*^f#3ifflf#feAvA'
.'/^'^%^
^'J^-KSCSSriv,
I'^.TT^^*^ >V>'T
.V
;e^,:
l3S^-:?c^^

Theory of Observations

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Theory of Observations

Hochgeladen von

Copyright:

Verfügbare Formate

UC-NRLF

DIRECTOR OF THE COPENHAGEN OBSERVATOHY

CHARLES & EDWIN LAYTON

PRINTED BT BIANCO LUNO

The Law of Causality.

The Observations and

Theoretical and Empirical Science

Actual Errors and Laws of Presumptive Errors

Large Numbers of Repetitions

Frequency and Probability

Curves of Actual Errors of Observations in Discontinued Values

Their Determination by Interpolation

Actual Errors for Rounded Observations

Typical Curves of Errors

913. The Binomial Functions

of Errors expressed by Symmetrical Functions.

1516. Coefficients of the Equation of Errors. Suras

very general Series by Half-Invariants

3435. Linear Functions

of Errors of Functions of Observations.

More Observations. Bonds

One Single Observation

of Errors and Half-Invariants.

Mean Value. Approximation to the Typical Form

Presumptive Errors when the Presumptive

Probable Errors and other Dangerous Notions

Possibility of regarding certain

5961. Every Function

Two, which belong

Complete Sets of Free Functions

General Remarks about Computation with Observed or Inexactly Given Values

be Determined by Means of Observations whicli are not Repeti64

Mutually Free Systems of Functions

can be divided into a

Adjustment by Correlates and by Elements

General Solution of the Problem

Schedule for Adjustment by (Correlates

for each Observation.

Under- Adjustment and Over- Adjustment

Special Auxiliary Methods.

Lawful and Illegitimate Facilities

XIV. The Theory of Probability.

121123. Relation of Probability to the Laws of Errors by Half-Invariants

XV. The Formal Theory

Addition and Multiplication of Probabilities

Linear Equations of Differences.

The Determination of Probabilities a Priori and a Posteriori.

Theory and Experience

130-133. Determination a Posteriori and

Mathematical Expectation and

THE LAW OF CAUSALITY.

assumption that everything tJiuf exists, and everything

start with the

repeated in every detail,

difference in the other part of the causes.

between the results of causes that are

This assumption, which

believe the fundamental assumptions of religion,

and intimately connected.

but not in practice.

of causality forces itself upon

in that very question he bears witness

are consistently to deny the law of causality,

observation, and particularly