IRLF 
 
THE THEORY OF MEASUREMENTS 
 
McGraw-Hill BookCompaiiy 
 
 Electrical World 
 Engineering Record 
 Railway Age Gazette 
 Signal Engineer 
 Electric Railway Journal 
 
 Engineering News 
 
 American Machinist 
 
 AraericanEngjneer 
 
 Coal Age 
 
 Metallurgical and Chemical Engineering 
 
 Power 
 

 THE THEORY OF 
 
 MEASUREMENTS 
 
 BY 
 
 A. DE FOREST PALMER, PH.D. 
 
 Associate Professor of Physics in Brown University. 
 
 McGRAW-HILL BOOK COMPANY 
 
 239 WEST 39TH STREET, NEW YORK 
 6 BOUVERIE STREET, LONDON, E.G. 
 
 1912 
 
COPYRIGHT, 1912, 
 
 BY THE 
 
 McGRAW-HILL BOOK COMPANY 
 
 Stanbopc jjbress 
 
 H.GILSON COMPANY 
 BOSTON, U.S.A. 
 
PREFACE. 
 
 THE function of laboratory instruction in physics is twofold. 
 Elementary courses are intended to develop the power of discrimi- 
 nating observation and to put the student in personal contact with 
 the phenomena and general principles discussed in textbooks and 
 lecture demonstrations. The apparatus provided should be of the 
 simplest possible nature, the experiments assigned should be for 
 the most part qualitative or only roughly quantitative, and emphasis 
 should be placed on the principles illustrated rather than on the 
 accuracy of the necessary measurements. On the other hand, 
 laboratory courses designed for more mature students, who are 
 supposed to have a working knowledge of fundamental principles, 
 are intended to give instruction in the theory and practice of the 
 methods of precise measurement that underlie all effective research 
 and supply the data on which practical engineering enterprises are 
 based. They should also develop the power of logical argument 
 and expression, and lead the student to draw rational conclusions 
 from his observations. The instruments provided should be of 
 standard design and efficiency in order that the student may gain 
 practice in making adjustments and observations under as nearly 
 as may be the same conditions that prevail in original investigation. 
 
 Measurements are of little value in either research or engineering 
 applications unless the precision with which they represent the 
 measured magnitude is definitely known. Consequently, the stu- 
 dent should be taught to plan and execute proposed measurements 
 within definitely prescribed limits and to determine the accuracy 
 of the results actually attained. Since the treatment of these 
 matters in available laboratory manuals is fragmentary and often 
 very inadequate if not misleading, the author some years ago under- 
 took to impart the necessary instruction, in the form of lectures, 
 to a class of junior engineering students. Subsequently, textbooks 
 on the Theory of Errors and the Method of Least Squares were 
 adopted but most of the applications to actual practice were still 
 given by lecture. The present treatise is the result of the experi- 
 
 257860 
 
VI PREFACE 
 
 ence gained with a number of succeeding classes. It has been 
 prepared primarily to meet the needs of students in engineering 
 and advanced physics who have a working knowledge of the differ- 
 ential and integral calculus. It is not intended to supersede but 
 to supplement the manuals and instruction sheets usually employed 
 in physical laboratories, Consequently, particular instruments and 
 methods of measurement have been described only in so far as they 
 serve to illustrate the principles under discussion. 
 
 The usefulness of such a treatise was suggested by the marked 
 tendency of laboratory students to carry out prescribed work in a 
 purely automatic manner with slight regard for the significance or 
 the precision of their measurements. Consequently, an endeavor 
 has been made to develop the general theory of measurements and 
 the errors to which they are subject in a form so clear and concise 
 that it can be comprehended and applied by the average student 
 with the prescribed previous training. To this end, numerical ex- 
 amples have been introduced and completely worked out whenever 
 this course seemed likely to aid the student in obtaining a thorough 
 grasp of the principles they illustrate. On the other hand, inherent 
 difficulties have not been evaded and it is not expected, or even 
 desired, that the student will be able to master the subject without 
 vigorous mental effort. 
 
 The first seven chapters deal with the general principles that 
 underlie all measurements, with the nature and distribution of the 
 errors to which they are subject, and with the methods by which 
 the most probable result is derived from a series of discordant 
 measurements. The various types of measurement met with in 
 practice are classified, and general methods of dealing with each 
 of them are briefly discussed. Constant errors and mistakes are 
 treated at some length, and then the unavoidable accidental errors 
 of observation are explicitly defined. The residuals corresponding 
 to actual measurements are shown to approach the true accidental 
 errors as limits when the number of observations is indefinitely 
 increased and their normal distribution in regard to sign and mag- 
 nitude is explained and illustrated. After a preliminary notion of 
 its significance has been thus imparted, the law of accidental errors 
 is stated empirically in a form that gives explicit representation to 
 all of the factors involved. It is then proved to be in conformity 
 with the axioms of accidental errors, the principle of the arithmetical 
 ij and the results of experience. The various characteristic 
 
PREFACE vii 
 
 errors that are commonly used as a measure of the accidental errors 
 of given series of measurements are clearly denned and their signifi- 
 cance is very carefully explained in order that they may be used 
 intelligently. Practical methods for computing them are developed 
 and illustrated by numerical examples. 
 
 Chapters eight to twelve inclusive are devoted to a general dis- 
 cussion of the precision of measurements based on the principles 
 established in the preceding chapters. The criteria of accidental 
 errors and suitable methods for dealing with constant and systematic 
 errors are developed in detail. The precision measure, of the result 
 computed from given observations, is defined and its significance is 
 explained with the aid of numerical illustrations. The proper basis 
 for the criticism of reported measurements and the selection of 
 suitable numerical values from tables of physical constants or other 
 published data is outlined ; and the importance of a careful estimate 
 of the precision of the data adopted in engineering and scientific 
 practice is emphasized. The applications of the theory of errors to 
 the determination of suitable methods for the execution of proposed 
 measurements are discussed at some length and illustrated. 
 
 In chapter thirteen, the relation between measurement and re- 
 search is pointed out and the general methods of physical research 
 are outlined. Graphical methods of reduction and representation 
 are explained and some applications of the method of least squares 
 are developed. The importance of timely and adequate publication, 
 or other report, of completed investigations is emphasized and some 
 suggestions relative to the form of such reports are given 
 
 Throughout the book, particular attention is paid to methods of 
 computation and to the proper use of significant figures. For the 
 convenience of the student, a number of useful tables are brought 
 together at the end of the volume. 
 
 A. DE FOREST PALMER. 
 
 BROWN UNIVERSITY, 
 July, 1912. 
 
CONTENTS. 
 
 PAGE 
 
 PREFACE v 
 
 CHAPTER I. 
 
 GENERAL PRINCIPLES 1 
 
 Introduction Measurement and Units Fundamental and 
 Derived Units Dimensions of Units Systems of Units in Gen- 
 eral Use Transformation of Units. 
 
 CHAPTER II. 
 
 MEASUREMENTS 11 
 
 Direct Measurements Indirect Measurements Classification of 
 Indirect Measurements Determination of Functional Relations 
 Adjustment, Setting, and Observation of Instruments Record 
 of Observations Independent, Dependent, and Conditioned 
 Measurements Errors and the Precision of Measurements Use 
 of Significant Figures Adjustment of Measurements Discus- 
 sion of Instruments and Methods. 
 
 CHAPTER III. 
 
 CLASSIFICATION OF ERRORS 23 
 
 Constant Errors Personal Errors Mistakes Accidental 
 Errors Residuals Principles of Probability. 
 
 CHAPTER IV. 
 
 THE LAW OF ACCIDENTAL ERRORS 29 
 
 Fundamental Propositions Distribution of Residuals Proba- 
 bility of Residuals The Unit Error The Probability Curve 
 Systems of Errors The Probability Function The Precision 
 Constant Discussion of the Probability Function The Proba- 
 bility Integral Comparison of Theory and Experience The 
 Arithmetical Mean. 
 
 CHAPTER V. 
 
 CHARACTERISTIC ERRORS 44 
 
 The Average Error The Mean Error The Probable Error 
 Relations between the Characteristic Errors Characteristic 
 Errors of the Arithmetical Mean Practical Computation of 
 Characteristic Errors Numerical Example Rules for the Use 
 of Significant Figures. 
 
 CHAPTER VI. 
 
 MEASUREMENTS OF UNEQUAL PRECISION 61 
 
 Weights of Measurements The General Mean Probable Error 
 of the General Mean Numerical Example. 
 
 ix 
 
x CONTENTS 
 
 CHAPTER VII. 
 
 PAGE 
 
 THE METHOD OF LEAST SQUARES 72 
 
 Fundamental Principles Observation Equations Normal Equa- 
 tions Solution with Two Independent Variables Adjustment of 
 the Angles about a Point Computation Checks Gauss's Method 
 of Solution Numerical Illustration of Gauss's Method Con- 
 ditioned Quantities. 
 
 CHAPTER VIII. 
 
 PROPAGATION OP ERRORS 95 
 
 Derived Quantities Errors of the Function Xi X z X 3 
 . . . X q Errors of the Function ai-Xi 0:2^2 013X3 =h . . . 
 aqXq Errors of the Function F (Xi, X ? , . . . , Xq) Example 
 Introducing the Fractional Error Fractional Error of the Func- 
 tion aX! n > X X 2 n ' X ... X X q n *. 
 
 CHAPTER IX. 
 
 ERRORS OF ADJUSTED MEASUREMENTS 105 
 
 Weights of Adjusted Measurements Probable Error of a Single 
 Observation Application to Problems Involving Two Unknowns 
 
 Application to Problems Involving Three Unknowns. 
 
 CHAPTER X. 
 
 DISCUSSION OF COMPLETED OBSERVATIONS 117 
 
 Removal of Constant Errors Criteria of Accidental Errors 
 Probability of Large Residuals Chauvenet's Criterion Preci- 
 sion of Direct Measurements Precision of Derived Measurements 
 
 Numerical Example. 
 
 CHAPTER XI. 
 
 DISCUSSION OF PROPOSED MEASUREMENTS 144 
 
 Preliminary Considerations The General Problem The Pri- 
 mary Condition The Principle of Equal Effects Adjusted Effects 
 
 Negligible Effects Treatment of Special Functions Numerical 
 Example. 
 
 CHAPTER XII. 
 
 BEST MAGNITUDES FOR COMPONENTS 165 
 
 Statement of the Problem General Solutions Special Cases 
 Practical Examples Sensitiveness of Methods and Instruments. 
 
 CHAPTER XIII. 
 
 RESEARCH 192 
 
 Fundamental Principles General Methods of Physical Research 
 
 Graphical Methods of Reduction Application of the Method 
 of Least Squares Publication. 
 
 TABLES 212 
 
 INDEX.. 245 
 
LIST OF TABLES. 
 
 PAGE 
 
 I. DIMENSIONS OF UNITS 212 
 
 II. CONVERSION FACTORS 213 
 
 III. TRIGONOMETRICAL RELATIONS 215 
 
 IV. SERIES 217 
 
 V. DERIVATIVES 219 
 
 VI. SOLUTION OF EQUATIONS 220 
 
 VII. APPROXIMATE FORMULA 221 
 
 VIII. NUMERICAL CONSTANTS 222 
 
 IX. EXPONENTIAL FUNCTIONS e x AND e~ x 223 
 
 X. EXPONENTIAL FUNCTIONS e* 2 AND e~ xZ 224 
 
 XI. THE PROBABILITY INTEGRAL P A 225 
 
 XII. THE PROBABILITY INTEGRAL P s 226 
 
 XIII. CHAUVENET'S CRITERION 226 
 
 XIV. FOR COMPUTING PROBABLE ERRORS BY FORMULAE (31) AND (32). 227 
 XV. FOR COMPUTING PROBABLE ERRORS BY FORMULA (34) 228 
 
 XVI. SQUARES OF NUMBERS 229 
 
 XVII. LOGARITHMS; 1000 TO 1409 231 
 
 XVIII. LOGARITHMS 232 
 
 XIX. NATURAL SINES 234 
 
 XX. NATURAL COSINES 236 
 
 XXI. NATURAL TANGENTS 238 
 
 XXII. NATURAL COTANGENTS 240 
 
 XXIII. RADIAN MEASURE. . 242 
 
THE 
 THEOEY OF MEASUREMENTS 
 
 CHAPTER I. 
 GENERAL PRINCIPLES. 
 
 i. Introduction. Direct observation of the relative position 
 and motion of surrounding objects and of their similarities and 
 differences is the first step in the acquisition of knowledge. 
 Such observations are possible only through the sensations pro- 
 duced by our environment, and the value of the knowledge thus 
 acquired is dependent on the exactness with which we corre- 
 late these sensations. Such correlation involves a quantitative 
 estimate of the relative intensity of different sensations and of 
 their time and space relations. As our estimates become more 
 and more exact through experience, our ideas regarding the 
 objective world are , gradually modified until they represent 
 the actual condition of things with a considerable degree of 
 precision. 
 
 The growth of science is analogous to the growth of ideas. 
 Its function is to arrange a mass of apparently isolated and un- 
 related phenomena in systematic order and to determine the in- 
 terrelations between them. For this purpose, each quantity that 
 enters into the several phenomena must be quantitatively deter- 
 mined, while all other quantities are kept constant or allowed 
 to vary by a measured amount. The exactness of the relations 
 thus determined increases with' the precision of the measure- 
 ments and with the success attained in isolating the particular 
 phenomena investigated. 
 
 A general statement, or a mathematical formula, that ex- 
 presses the observed quantitative relation between the different 
 magnitudes involved in any phenomenon is called the law of 
 that phenomenon. As here used, the word law does not mean 
 
 1 
 
2 THE THEORY OF MEASUREMENTS [ART. 2 
 
 that the phenomenon must follow the prescribed course, but 
 that, under the given conditions and within the limits of error 
 and the range of our measurements, it has never been found to 
 deviate from that course. In other words, the laws of science 
 are concise statements of our present knowledge regarding 
 phenomena and their relations. As we increase the range and 
 accuracy of our measurements and learn to control the condi- 
 tions of experiment more definitely, the laws that express our 
 results become more exact and cover a wider range of phenomena. 
 Ultimately we arrive at broad generalizations from which the 
 laws of individual phenomena are deducible as special cases. 
 
 The two greatest factors in the progress of science are the 
 trained imagination of the investigator and the genius of 
 measurement. To the former we owe the rational hypotheses 
 that have pointed the way of advance and to the latter the 
 methods of observation and measurement by which the laws of 
 science have been developed. 
 
 2. Measurement and Units. To measure a quantity is to 
 determine the ratio of its magnitude to that of another quan- 
 tity, of the same kind, taken as a unit. The number that 
 expresses this ratio may be either integral or fractional and is 
 called the numeric of the given quantity in terms of the chosen 
 unit. In general, if Q represents the magnitude of a quantity, 
 U the magnitude of the chosen unit, and N the corresponding 
 numeric we have 
 
 Q = NU, (I) 
 
 which is the fundamental equation of measurement. The two 
 factors N and U are both essential for the exact specification of 
 the magnitude Q. For example: the length of a certain line 
 is five inches, i.e., the line is five times as long as one inch. It 
 is not sufficient to say that the length of the line is five; for in 
 that case we are uncertain whether its length is five inches, five 
 feet, or five times some other unit. 
 
 Obviously, the absolute magnitude of a quantity is independent 
 of the units with which we choose to measure it. Hence, if we 
 adopt a different unit U', we shall find a different numeric N' 
 
 such that 
 
 Q = N'U', (II) 
 
 and consequently 
 
 NU = N'U', 
 
ART. 2] ' GENERAL PRINCIPLES 3 
 
 or $-^- (HI) 
 
 Equation (III) expresses the general principle involved in the 
 transformation of units and shows that the numeric varies in- 
 versely as the magnitude of the unit; i.e., if U is twice as large 
 as U', N will be only one-half as large as N'. To take a con- 
 crete example: a length equal to ten inches is also equal to 
 25.4 centimeters approximately. In this case N equals ten, 
 N' equals 25.4, U equals one inch, and U r equals one centi- 
 
 N f 
 meter. The ratio of the numerics -^ is 2.54 and hence the 
 
 inverse ratio of the units -, is also 2.54, i.e., one inch is equal to 
 
 2.54 centimeters. 
 
 Equation (III) may also be written in the form 
 
 (IV) 
 
 which shows that the numeric of a given quantity relative to the 
 unit U is equal to its numeric relative to the unit U' multiplied 
 
 w 
 
 by the ratio of the unit U f to the unit U. The ratio -jj is called 
 
 the conversion factor for the unit U f in terms of the unit U. 
 It is equal to the number of units U in one unit U', and when 
 multiplied by the numeric of a quantity in terms of U' gives 
 the numeric of the same quantity in terms of U. The con- 
 version factor for transformation in the opposite direction, i.e., 
 
 from U to U', is obviously the inverse of the above, or -== In 
 
 general, the numerator of the conversion factor is the unit in 
 which the magnitude is already expressed and the denominator 
 is the unit to which it is to be transformed. For example: 
 one inch is approximately equal to 2.54 centimeters, hence the 
 numeric of a length in centimeters is about 2.54 times its numeric 
 in inches. Conversely, the numeric in inches is equal to the 
 numeric in centimeters divided by 2.54 or multiplied by the 
 reciprocal of this number. 
 
 In so far as the theory of mensuration and the attainable 
 accuracy of the result are concerned, measurements may be made 
 in terms of any arbitrary unite and, in fact, the adoption oisuch 
 
4 THE THEORY OF MEASUREMENTS [ART. 3 
 
 units is frequently convenient when we are concerned only with 
 relative determinations. In general, however, measurements are 
 of little value unless they are expressed in terms of generally 
 accepted units whose magnitude is accurately known. Some 
 such units have come into use through common consent but most 
 of them have been fixed by government enactment and their per- 
 manence is assured by legal standards whose relative magnitudes 
 have been accurately determined. Such primary standards, pre- 
 served by various governments, have, in many cases, been very 
 carefully intercompared and their conversion factors are accu- 
 rately known. Copies of the more important primary standards 
 may be found in all well-equipped laboratories where they are 
 preserved as the secondary standards to which all exact measure- 
 ments are referred. Carefully made copies are, usually, sufficiently 
 accurate for ordinary purposes, but, when the greatest precision 
 is sought, their exact magnitude must be determined by direct 
 comparison with the primary standards. The National Bureau 
 of Standards at Washington makes such comparisons and issues 
 certificates showing the errors of the standards submitted for 
 test. 
 
 3. Fundamental and Derived Units. Since the unit is, neces- 
 sarily, a quantity of the same kind as the quantity measured, we 
 must have as many different units as there are different kinds of 
 quantities to be measured. Each of these units might be fixed 
 by an independent arbitrary standard, but, since most measur- 
 able quantities are connected by definite physical relations, it is 
 more convenient to define our units in accordance with these 
 relations. Thus, measured in terms of any arbitrary unit, a 
 uniform velocity is proportional to the distance described in 
 unit time; but, if we adopt as our unit such a velocity that the 
 unit of length is traversed in the unit of time, the factor of pro- 
 portionality is unity and the velocity is equal to the ratio of the 
 space traveled to the elapsed time. 
 
 Three independently defined units are sufficient, in connection 
 with known physical relations, to fix the value of most of the 
 other units used in physical measurements. We are thus led to 
 distinguish two classes of units; the three fundamental units, 
 defined by independent arbitrary standards, and the derived 
 units, fixed by definite relations between the fundamental units. 
 The .magnitude, and to some extent the choice, of the fundamental 
 
ART. 4] GENERAL PRINCIPLES 5 
 
 units is arbitrary, but when definite standards for each of these 
 units have been adopted the magnitude of all of the derived units 
 is fixed. 
 
 For convenience in practice, legal standards have been adopted 
 to represent some of the derived units. The precision of these 
 standards is determined by indirect comparison with the standards 
 representing the three fundamental units. Such comparisons are 
 based on the known relations between the fundamental and de- 
 rived units and are called absolute measurements. The practical 
 advantage gained by the use of derived standards lies in the fact 
 that absolute measurements are generally very difficult and require 
 great skill and experience in order to secure a reasonable degree 
 of accuracy. On the other hand, direct comparison of derived 
 quantities of the same kind is often a comparatively simple 
 matter and can be carried out with great precision. 
 
 4. Dimensions of Units. The dimensions of a unit is a 
 mathematical formula that shows how its magnitude is related 
 to that of the three fundamental units. In writing such formulae, 
 the variables are usually represented by capital letters inclosed 
 in square brackets. Thus, [M], [L] and [T]- represent the dimen- 
 sions of the units of mass, length and time respectively. 
 
 Dimensional formulae and ordinary algebraic equations are 
 essentially different in significance. The former shows the rela- 
 tive variation of units, while the latter expresses a definite mathe- 
 matical relation between the numerics of measurable quantities. 
 Thus if a point in uniform motion describes the distance L in the 
 time T its velocity V is defined by the relation 
 
 V = Y (V) 
 
 Since L and T are concrete quantities of different kind, the right- 
 hand member of this equation is not a ratio in the strict arithmet- 
 ical sense; i.e., it cannot be represented by a simple abstract num- 
 ber. Hence, in virtue of the definite physical relation expressed 
 by equation (V), we are led to extend our idea of ratio to include 
 the case of concrete quantities. From this point of view, the ratio 
 of two quantities expresses the rate of change of the first quantity 
 with respect to the second. It is a concrete quantity of the same 
 kind as the quantity it serves, to define. As an illustration, con- 
 sider the meaning of equation (V). Expressed in words, it is " the 
 
6 THE THEORY OF MEASUREMENTS [ART. 4 
 
 velocity of a point, in uniform motion, is equal to the time rate at 
 which it moves through space." 
 
 If we represent the units of velocity, length, and time by [7], 
 [L], and [T\, respectively, and the corresponding numerics by v, 
 I, and t, we have by equation (I), article two, 
 
 F = v(V], L = l(L], T = t[T], 
 and equation (V) becomes 
 
 w-m-i' 
 
 or 
 
 [V][T] t 
 
 Since, by definition, [V] and |~l are quantities of the same kind, 
 
 their ratio can be expressed by an abstract number k and equation 
 (VI) may be written in the form 
 
 v = kl, (VII) 
 
 which is an exact numerical equation containing no concrete 
 quantities. 
 
 The numerical value of the constant k obviously depends on 
 the units with which L, T, and V are measured. If we define the 
 unit of velocity by the relation 
 
 ryi-M 
 [TV 
 or, as it is more often written, 
 
 [F] = [L!T-'] f (VIII) 
 
 k becomes equal to unity and the relation (VII) between the 
 numerics of velocity, length, and time reduces to the simple form 
 
 The foregoing argument illustrates the advantage to be gained 
 by defining derived units in accordance with the physical rela- 
 tions on which they depend. By this means we eliminate the 
 often incommensurable constants of proportionality such as k 
 would be if the unit of velocity were defined in any other way 
 than by equation (VIII). 
 
ART. 5] GENERAL PRINCIPLES 7 
 
 The expression on the right-hand side of equation (VIII) is the 
 dimensions of the unit of velocity when the units of length, mass, 
 and time are chosen as fundamental. The dimensions of any 
 other units may be obtained by the method outlined above when 
 we know the physical relations on which they depend. The form 
 of the dimensional formula depends on the units we choose as 
 fundamental, but the general method of derivation is the same in 
 all cases. As an exercise to fix these ideas the student should 
 verify the following dimensional formulae: choosing [M], [L], and 
 [T] as fundamental units, the dimensions of the units of area, 
 acceleration, and force are [L 2 ], [LT~ 2 ], and [MLT~ 2 ] respectively. 
 As an illustration of the effect of a different choice of fundamental 
 units, it may be shown that the dimensions of the unit of mass is 
 [FL^T 2 ] when the units of length [L], force [F], and time [T] are 
 chosen as fundamental. The dimensions of some important 
 derived units are given in Table I at the end of this volume. 
 
 5. Systems of Units in General Use. Consistent systems 
 of units may differ from one another by a difference in the choice 
 of fundamental units or by a difference in the magnitude of the 
 particular fundamental units adopted. The systems in common 
 use illustrate both types of difference. 
 
 Among scientific men, the so-called c.g.s. system is almost 
 universally adopted, and the results of scientific investigations 
 are seldom expressed in any other units. The advantage of such 
 uniformity of choice is obvious. It greatly facilitates the com- 
 parison of the results of different observers and leads to general 
 advance in our knowledge of the phenomena studied. The units 
 of length, mass, and time are chosen as fundamental in this 
 system and the particular values assigned to them are the centi- 
 meter for the unit of length, the gram for the unit of mass, and 
 the mean solar second for the unit of time. 
 
 The units used commercially in England and the United States 
 of America are far from systematic, as most of the derived units 
 are arbitrarily defined. So far as they follow any order, they 
 form a length-mass-time system in which the unit of length is the 
 foot, the unit of mass is the mass of a pound, and the unit of time 
 is the second. This system was formerly used quite extensively 
 by English scientists and the results of some classic investigations 
 are expressed in such units. 
 
 English and American engineers find it more convenient to use 
 
8 THE THEORY OF MEASUREMENTS [ART. 6 
 
 a system in which the fundamental units are those of length, 
 force, and time. The particular units chosen are the foot as the 
 unit of length, the pound's weight at London as the unit of force, 
 and the mean solar second as the unit of time. We shall see that 
 this is equivalent to a length-mass-time system in which the units 
 of length and time are the same as above and the unit of mass is 
 the mass of 32.191 pounds. 
 
 6. Transformation of Units. When the relative magnitude 
 of corresponding fundamental units in two systems is known, a 
 result expressed in one system can be reduced to the other with 
 the aid of the dimensions of the derived units involved. Thus: 
 let A c represent the magnitude of a square centimeter, A t the 
 magnitude of a square inch, N c the numeric of a given area when 
 measured in square centimeters, and Ni the numeric of the same 
 area when measured in square inches; then, from equation (IV), 
 article two, we have 
 
 But if L c is the magnitude of a centimeter and LI that of an inch, 
 Ai is equal to Lf, and therefore 
 
 Hence, the conversion factor -p for reducing square centimeters 
 
 A-i 
 
 to square inches is equal to the square of the conversion factor 
 
 for reducing from centimeters to inches. Now the dimensions 
 Li 
 
 of the unit of area is [L 2 ], and we see that the conversion factor 
 for area may be obtained by substituting the corresponding con- 
 version factor for lengths in this dimensional formula. This is a 
 simple illustration of the general method of transformation of 
 units. When the fundamental units in the two systems differ in 
 magnitude, but not in kind, the conversion factor for correspond- 
 ing derived units in the two systems is obtained by replacing the 
 fundamental units by their respective conversion factors in the 
 dimensions of the derived units considered. 
 
 It should be noticed that the fundamental units in the c.g.s. 
 system are those of length, mass, and time, while on the engineer's 
 system they are length, force, and time. In the latter system, 
 
ART. 6] GENERAL PRINCIPLES 9 
 
 force is supposed to be directly measured and expressed by the 
 dimensions [F]. Consequently the dimensions of the unit of 
 mass are [FL~ 1 T 2 ], and the unit of mass is a mass that will acquire 
 . a velocity of one foot per second in one second when acted upon 
 by a force of one pound's weight. For the sake of definiteness, 
 the unit of force is taken as the pound's weight at London, where 
 the acceleration due to gravity (g) is equal to 32.191 feet per 
 second per second. Otherwise the unit of force would be variable, 
 depending on the place at which the pound is weighed. 
 
 From Newton's second law of motion we know that the relation 
 between acceleration, mass, and force is given by the expression 
 
 / = ma. 
 
 For a constant force the acceleration produced is inversely pro- 
 portional to the mass moved. Now the mass of a pound at London 
 is acted upon by gravity with a force of one pound's weight, and, if 
 free, it moves with an acceleration of 32.191 feet per second per 
 second. Hence a mass equal to that of 32.191 pounds acted 
 upon by a force of one pound's weight would move with an acceler- 
 ation of one foot per second per second, i.e., it would acquire a 
 velocity of one foot per second in one second. Hence the unit of 
 mass in the engineer's system is 32.191 pounds mass. This unit 
 is sometimes called a slugg, but the name is seldom met with since 
 engineers deal primarily with forces rather than masses, and are 
 
 W 
 content to write for mass without giving the unit a definite 
 
 7 
 
 name. This is equivalent to saying that the mass of a body, 
 expressed in sluggs, is equal to its weight, at London, expressed in 
 pounds, divided by 32.191. 
 
 After careful consideration of the foregoing discussion, it will 
 be evident that the engineer's length-force-time system is exactly 
 equivalent to a length-mass-time system in which the unit of 
 length is the foot, the unit of mass is the slugg or 32.191 pounds' 
 mass, and the unit of time is the mean solar second. In the latter 
 system the fundamental units are of the same kind as those of 
 the c.g.s. system. Hence, if the conversion factor for the unit 
 of mass is taken as the ratio of the magnitude of the slugg to that 
 of the gram, quantities expressed in the units of the engineer's 
 system may be reduced to the equivalent values in the c.g.s. 
 system by the method described at the beginning of this article. 
 
10 THE THEORY OF MEASUREMENTS [ART. 6 
 
 When, as is frequently the case, the engineer's results are expressed 
 in terms of the local weight of a pound as a unit of force in place 
 of the pound's weight at London, the result of a transformation 
 of units, carried out as above, will be in error by a factor equal to 
 the ratio of the acceleration due to gravity at London and at the 
 location of the measurements. Unless the local gravitational 
 acceleration is definitely stated by the observer and unless he 
 has used his length-force-time units in a consistent manner, it is 
 impossible to derive the exact equivalent of his results on the 
 c.g.s. system. 
 
CHAPTER II. 
 MEASUREMENTS. 
 
 IN article two of the last chapter we defined the term " measure- 
 ment " and showed that any magnitude may be represented by 
 the product of two factors, the numeric and the unit. The object 
 of all measurements is the determination of the numeric that ex- 
 presses the magnitude of the observed quantity in terms of the 
 chosen unit. For convenience of treatment, they may be classified 
 according to the nature of the measured quantity and the methods 
 of observation and reduction. 
 
 7. Direct Measurements. The determination of a desired 
 numeric by direct observation of the measured quantity, with the 
 aid of a divided scale or other indicating device graduated in 
 terms of the chosen unit, is called a direct measurement. 
 
 Such measurements are possible when the chosen unit, together 
 with its multiples and submultiples, can be represented by a 
 material standard, so constructed that it can be directly applied 
 to the measured quantity for the purpose of comparison, or when 
 the unit and the measured magnitudes produce proportional 
 effects on a suitable indicating device. 
 
 Lengths may be directly measured with a graduated scale, 
 masses by comparison with a set of standard masses on an equal 
 arm balance, time intervals by the use of a clock regulated to 
 give mean solar time, and forces with the aid of a spring balance. 
 Hence magnitudes expressible in terms of the fundamental units 
 of either the c.g.s. or the engineer's system may be directly 
 measured. 
 
 Many quantities expressible in terms of derived units, that can 
 be represented by material standards, are commonly determined 
 by direct measurement. As illustrations, we may cite the deter- 
 mination of the volume of a liquid with a graduated flask and the 
 measurement of the electrical resistance of a wire by comparison 
 with a set of standard resistances. 
 
 8. Indirect Measurements. The determination of a desired 
 numeric by computation from the numerics of one or more 
 
 11 
 
12 THE THEORY OF MEASUREMENTS [ART. 9 
 
 directly measured magnitudes, that bear a known relation to the 
 desired quantity, is called an indirect measurement. 
 
 The relation between the observed and computed magnitudes 
 may be expressed in the general form 
 
 y = Ffa, Xz, x 3 , . . . a, b, c . . . ), 
 
 where y, x t , x 2 , etc., represent measured or computed magnitudes, 
 or the numerics corresponding to them, a, b, c, etc., represent 
 constants, and F indicates that there is a functional relation 
 between the other quantities. This expression is read, y equals 
 some function of xi, x*, etc., and a, b, c, etc. In any particular 
 case, the form of the function F and the number and nature of the 
 related quantities must be known before the computation of the 
 unknown quantities is undertaken. 
 
 Most of the indirect measurements made by physicists and 
 engineers fall into one or another of three general classes, char- 
 acterized by the nature of the unknown and measured magnitudes 
 and the form of the function F. 
 
 9. Classification of Indirect Measurements. 
 I. 
 
 In the first class, y represents the desired numeric of a magni- 
 tude that is not directly measured, either because it is impossible 
 or inconvenient to do so, or because greater precision can be at- 
 tained by indirect methods. The form of the function F and the 
 numerical values of all of the constants a, 6, c, etc., appearing in 
 it, are given by theory. The quantities xi, Xz, etc., represent 
 the numerics of directly measured magnitudes. In the following 
 pages indirect measurements belonging to this class will sometimes 
 be referred to as derived measurements. 
 
 As an illustration we may cite the determination of the density 
 s of a solid sphere from direct measurements of its mass M and 
 its diameter D with the aid of the relation 
 
 M 
 
 = F^' 
 
 Comparing this expression with the general formula given above, 
 we note that s corresponds to y, M to xi, D to x a , J to a, TT to 6, 
 
 and that F represents the function y^^. The form of the func- 
 
ART. 9] MEASUREMENTS 13 
 
 tion is given by the definition of density as the ratio of the mass 
 to the volume of a body and the numerical constants and w are 
 given by the known relation between the volume and diameter of 
 a sphere. 
 
 II. 
 
 In the second class of indirect measurements, the numerical 
 constants a, b, c, etc., are the unknown quantities to be computed, 
 the form of the function F is known, and all of the quantities y, 
 Xi, x z , etc., are obtained by direct measurements or given by 
 theory. The functions met with in this class of measurements 
 usually represent a continuous variation of the quantity y with 
 respect to the quantities x\, x 2 , etc., as independent variables. 
 Hence the result of a direct measurement of y will depend on the 
 particular values of Xi, x 2 , etc., that obtain at the time of the 
 measurement. Consequently, in computing the constants a, b, c, 
 etc., we must be careful to use only corresponding values of the 
 measured quantities, i.e., values that are, or would be, obtained 
 by coincident observations on the several magnitudes. 
 
 Every set of corresponding values of the variables y, Xi, x 2 , etc., 
 when used in connection with the given function, gives an algebraic 
 relation between the unknown quantities a, b, c, etc., involving 
 only numerical coefficients and absolute terms. When we have 
 obtained as many independent equations as there are unknown 
 quantities, the latter may be determined by the usual algebraic 
 methods. We shall see, however, that more precise results can 
 be obtained when the number of independent measurements far 
 exceeds the minimum limit thus set and the computation is made 
 by special methods to be described hereafter. 
 
 The determination of the initial length L and the coefficient of 
 linear expansion a of a metallic bar from a series of measurements 
 of the lengths L t corresponding to different temperatures t with the 
 aid of the functional relation 
 
 L t = Lo (1 + at) 
 
 is an example of the class of measurements here considered. Such 
 measurements are sometimes called determinations of empirical 
 constants. 
 
14 
 
 THE THEORY OF MEASUREMENTS [ART. 9 
 
 III. 
 
 The third class of indirect measurements includes all cases in 
 which each of a number of directly measured quantities yi, y*, y s , 
 etc., is a given function of the unknown quantities Xi, x 2 , X B , etc., 
 and certain known numerical constants a, 6, c, etc. In such cases 
 we have as many equations of the form 
 
 y 1 = FI (xi, x 2 , 3 , . . . a, 6, c, . . . ), 
 2/2 = F 2 (xi, z 2 , $t, . . . a, M, . . . )> 
 
 as there are measured quantities yi, y 2 , etc. This number must 
 be at least as great as the number of unknowns Xi, x 2 , etc., and 
 
 may be much greater. 
 The functions F lt F 2 , 
 etc., are frequently dif- 
 ferent in form and some 
 of them may not con- 
 tain all of the un- 
 knowns. The numeri- 
 cal constants, appearing 
 in different functions, 
 are generally different. 
 But the form of each 
 of the functions and 
 the values of all of the 
 constants must be 
 known before a solu- 
 tion of the problem is 
 possible. 
 
 Problems of this type 
 are frequently met with 
 in astronomy and geod- 
 esy. One of the simplest is known as the adjustment of the 
 angles about a point. Thus, let it be required to find the most 
 probable values of the angles Xi, x 2 , and x 3 , Fig. 1, from direct 
 measurements of yi, y 2 , y 3) . . . y & . In this case the general 
 equations take the form 
 
 FIG. 
 
ART. 11] MEASUREMENTS 15 
 
 2/i = xi, 
 
 2/2 = xi + x 2 , 
 
 2/4 = X 2 , 
 2/5 = 2 
 2/6 = , 
 
 and all of the numerical constants are either unity or zero. The 
 solution of such problems will be discussed in the chapter on the 
 method of least squares. 
 
 10. Determination of Functional Relations. When the form 
 of the functional relation between the observed and unknown 
 magnitudes is not known, the solution of the problem requires 
 something more than measurement and computation. In some 
 cases a study of the theory of the observed phenomena, in con- 
 nection with that of allied phenomena, will suggest the form of the 
 required function. Otherwise, a tentative form must be assumed 
 after a careful study of the observations themselves, generally by 
 graphical methods. In either case the constants of the assumed 
 function must be determined by indirect measurements and the 
 results tested by a comparison of the observed and the computed 
 values of the related quantities. If these values agree within the 
 accidental errors of observation, the assumed function may be 
 adopted as an empirical representation of the phenomena. If 
 the agreement is not sufficiently close, the form of the function 
 is modified, in a manner suggested by the observations, and the 
 process of computation and comparison is repeated until a satis- 
 factory agreement is obtained. A more detailed treatment of 
 such processes will be found in Chapter XIII. 
 
 11. Adjustment, Setting, and Observation of Instruments. 
 Most of the magnitudes dealt with in physics and engineering 
 are determined by indirect measurements. But we have seen 
 that all such quantities are dependent upon and computed from 
 directly measured quantities. Consequently, a study of the 
 methods and precision of direct measurement is of fundamental 
 importance. 
 
 In general, every direct measurement involves three distinct 
 operations. First: the instrument adopted is so placed that its 
 
16 THE THEORY OF MEASUREMENTS [ART. 12 
 
 scale is in the proper position relative to the magnitude to be 
 measured and all of its parts operate smoothly in the manner and 
 direction prescribed by theory. Operations of this nature are 
 called adjustments. Second: the reference line of the instru- 
 ment is moved, or allowed to move, in the manner demanded by 
 theory, until it coincides with a mark chosen as a point of reference 
 on the measured magnitude. We shall refer to this operation as a 
 setting of the instrument. Third: the position of the index of 
 the instrument, with respect to its graduated scale, is read. This 
 is an observation. 
 
 As an illustration, consider the measurement of the normal 
 distance between two parallel lines with a micrometer microscope. 
 The instrument must be so mounted that it can be rigidly clamped 
 in any desired position or moved freely in the direction of its 
 optical axis without disturbing the direction of the micrometer 
 screw. The following adjustments are necessary: the axis of the 
 micrometer screw must be made parallel to the plane of the two 
 lines and perpendicular to a normal plane through one of them; 
 the eyepiece must be so placed that the cross-hairs are sharply 
 defined; the microscope must be moved, in the direction of its 
 optical axis, until the image of the two lines, or one of them if the 
 normal distance between them is greater than the field of view 
 of the microscope, is in the same plane with the cross-hairs. The 
 latter adjustment is correct when there is no parallax between the 
 image of the lines and the cross-hairs. The setting is made by 
 turning the micrometer head until the intersection of the cross- 
 hairs bisects the image of one of the lines. Finally the reading 
 of the micrometer scale is observed. A similar setting and ob- 
 servation are made on the other line and the difference between 
 the two observations gives the normal distance between the two 
 lines in terms of the scale of the micrometer. 
 
 12. Record of Observations. In the preceding article, the 
 word "observation" is used in a very much restricted sense to 
 indicate merely the scale reading of a measuring instrument. 
 This restriction is convenient in dealing with the technique of 
 measurement, but many other circumstances, affecting the accu- 
 racy of the result, must be observed and taken into account in a 
 complete study of the phenomena considered. There is, however^ 
 little danger of confusion in using the word in the two different 
 senses since the more restricted meaning is in reality only a 
 
ART. 13] MEASUREMENTS 17 
 
 special case of the general. The particular significance intended 
 in any special case is generally clear from the context. 
 
 The first essential for accurate measurements is a clear and 
 orderly record of all of the observations. The record should begin 
 with a concise description of the magnitude to be measured, and 
 the instruments and methods adopted for the purpose. Instru- 
 ments may frequently be described, with sufficient precision, by 
 stating their name and number or other distinguishing mark. 
 Methods are generally specified by reference to theoretical treatises 
 or notes. The adjustment and graduation of the instruments 
 should be clearly stated. The date on which the work is carried 
 out and the location of the apparatus should be noted. 
 
 Observations, in the restricted sense, should be neatly arranged 
 in tabular form. The columns of the table should be so headed, 
 and referred to by subsidiary notes, that the exact significance of 
 all of the recorded figures will be clearly understood at any future 
 time. All circumstances likely to affect the accuracy of the 
 measurements should be carefully observed and recorded in the 
 table or in suitably placed explanatory notes. 
 
 Observations should be recorded exactly as taken from the 
 instruments with which they are made, without mental computa- 
 tion or reduction of any kind even the simplest. For example: 
 when a micrometer head is divided into any number of parts 
 other than ten or one hundred, it is better to use two columns in 
 the table and record the reading of the main scale in one and 
 that of the micrometer head in the other than to reduce the head 
 reading to a decimal mentally and enter it in the same column 
 with the main scale reading. This is because mistakes are likely 
 to be made in such mental calculations, even by the most expe- 
 rienced observers, and, when the final reduction of the observations 
 is undertaken at a future time, it is frequently difficult or impos- 
 sible to decide whether a large deviation of a single observation 
 from the mean of the others is due to an accidental error of obser- 
 vation or to a mistake in such a mental calculation. 
 
 13. Independent, Dependent, and Conditioned Measure- 
 ments. Measurements on the same or different magnitudes are 
 said to be independent when both of the following specifications 
 are fulfilled: first, the measured magnitudes are not required to 
 satisfy a rigorous mathematical relation among themselves; 
 second, the same observation is not used in the computation of 
 
18 THE THEORY OF MEASUREMENTS [ART. 14 
 
 any two of the measurements and the different observations are 
 entirely unbiased by one another. 
 
 When the first of these specifications is fulfilled and the second 
 is not, the measurements are said to be dependent. Thus, when 
 several measurements of the length of a line are all computed 
 from the same zero reading of the scale used, they are all dependent 
 on that observation and any error in the position of the zero mark 
 affects all of them by exactly the same amount. When the position 
 of the index relative to the scale of the measuring instrument is 
 visible while the settings are being made, there is a marked tendency 
 to set the instrument so that successive observations will be exactly 
 alike rather than to make an independent judgment of the bisection 
 of the chosen mark in each case. The observations, corresponding 
 to settings made in this manner, are biased by a preconceived 
 notion regarding the correct position of the index and the measure- 
 ments computed from them are not independent. The impor- 
 tance of avoiding faulty observations of this type cannot be too 
 strongly emphasized. They not only vitiate the results of our 
 measurements, but also render a determination of their precision 
 impossible. 
 
 Measurements that do not satisfy the first of the above speci- 
 fications are called conditioned measurements. The different 
 determinations of each of the related quantities may or may not 
 be independent, according as they do or do not satisfy the second 
 specification, but the adjusted results of all of the measurements 
 must satisfy the given mathematical relation. Thus, we may 
 make a number of independent measurements of each of the 
 angles of a plane triangle, but the mean results must be so adjusted 
 that the sum of the accepted values is equal to one hundred and 
 eighty degrees. 
 
 14. Errors and the Precision of Measurements. Owing to 
 unavoidable imperfections and lack of constant sensitiveness in 
 our instruments, and to the natural limit to the keenness of our 
 senses, the results of our observations and measurements differ 
 somewhat from the true numeric of the observed magnitude. 
 Such differences are called errors of observation or measurement. 
 Some of them are due to known causes and can be eliminated, 
 with sufficient accuracy, by suitable computations. Others are 
 apparently accidental in nature and arbitrary in magnitude. 
 Their probable distribution, in regard to magnitude and frequency 
 
ART. 15] MEASUREMENTS 19 
 
 of occurrence, can be determined by statistical methods when a 
 sufficient number of independent measurements is available. 
 
 The precision of a measurement is the degree of approximation 
 with which it represents the true numeric of the observed magni- 
 tude. Usually our measurements serve only to determine the 
 probable limits within which the desired numeric lies. Looked 
 at from this point of view, the precision of a measurement may be 
 considered to be inversely proportional to the difference between 
 the limits thus determined. It increases with the accuracy, 
 adaptability, and sensitiveness of the instruments used, and with 
 the skill and care of the observer. But, after a very moderate 
 precision has been attained, the labor and expense necessary for 
 further increase is very great in proportion to the result obtained. 
 
 A measurement is of little practical value unless we know the 
 precision with which it represents the observed magnitude. 
 Hence the importance of a thorough study of the nature and dis- 
 tribution of errors in general and of the particular errors that 
 characterize an adopted method of measurement. At first sight 
 it might seem incredible that such errors should follow a definite 
 mathematical law. But, when the number of observations is 
 sufficiently great, we shall see that the theory of probability leads 
 to a definite and easily calculated measure of the precision of a 
 single observation and of the result computed from a number 
 of observations. 
 
 15. Use of Significant Figures. When recording the nu- 
 merical results of observations or measurements, and during all 
 of the necessary computations, the number of significant figures 
 employed should be sufficient to express the attained precision 
 and no more. By significant figures we mean the nine digits and 
 zeros when not used merely to locate the decimal point. 
 
 In the case of the direct observation of the indications of instru- 
 ments, the above specification is usually sufficiently fulfilled by 
 allowing the last recorded significant figure to represent the 
 estimated tenth of the smallest division of the graduated scale. 
 For example: in measuring the length of a line, with a scale 
 divided in millimeters, the position of the ends of the line would 
 be recorded to the nearest estimated tenth of a millimeter. 
 
 Generally, computed results should be so recorded that the 
 limiting values, used to express the attained precision, differ by 
 only a few units in the last one or two significant figures. Thus: 
 
20 THE THEORY OF MEASUREMENTS [ART. 15 
 
 if the length of a line is found to lie between 15.65 millimeters and 
 15.72 millimeters, we should write 15.68 millimeters as the result 
 of our measurement. The use of a larger number of significant 
 figures would be not only a waste of space and labor, but also a 
 false representation of the precision of the result. Most of the 
 magnitudes we are called upon to measure are incommensurable 
 with the chosen unit, and hence there is no limit to the number 
 of significant figures that might be used if we chose to do so; but 
 experienced observers are always careful to express all observa- 
 tions and results and carry out all computations with a number 
 just sufficient to represent the attained precision. The use of 
 too many or too few significant figures is strong evidence of inex- 
 perience or carelessness in making observations and computations. 
 More specific rules for determining the number of significant 
 figures to be used in special cases will be developed in connection 
 with the methods for determining the precision of measurements. 
 
 The number of significant figure^ in any numerical expression 
 is entirely independent of the position of the decimal point. 
 Thus: each of the numbers 5,769,600, 5769, 57.69, and 0.0005769 
 is expressed by four significant figures and represents the corre- 
 sponding magnitude within one-tenth of one per cent, notwith- 
 standing the fact that the different numbers correspond to differ- 
 ent magnitudes. In general, the location of the decimal point 
 shows the order of magnitude of the quantity represented and 
 the number of significant figures indicates the precision with which 
 the actual numeric of the quantity is known. 
 
 In writing very large or very small numbers, it is convenient 
 to indicate the position of the decimal point by means of a positive 
 or negative power of ten. Thus: the number 56,400,000 may 
 be written 564 X 10 5 or, better, 5.64 X 10 7 , and 0.000075 may 
 be written 75 X W~ or 7.5 X 10~ 5 . When a large number of 
 numerical observations or results are to be tabulated or used in 
 computation, a considerable amount of time and space is saved 
 by adopting this method of representation. The second of the 
 two forms, illustrated above, is very convenient in making com- 
 putations by means of logarithms, as in this case the power of 
 ten always represents the characteristic of the logarithm of the 
 corresponding number. 
 
 In rounding numbers to the required number of significant 
 figures, the digit in the last place held should be increased by one 
 
ART. 17] MEASUREMENTS 21 
 
 unit when the digit in the next lower place is greater than five, 
 and left unchanged when the neglected part is less than five- 
 tenths of a unit. When the neglected part is exactly five-tenths 
 of a unit the last digit held is increased by one if odd, and left 
 unchanged if even. Thus: 5687.5 would be rounded to 5688 and 
 5686.5 to 5686. 
 
 1 6. Adjustment of Measurements. The results of inde- 
 pendent measurements of the same magnitude by the same or 
 different methods seldom agree with one another. This is due to 
 the fact that the probability for the occurrence of errors of exactly 
 the same character and magnitude in the different cases is very 
 small indeed. Hence we are led to the problem of determining 
 the best or most probable value of the numeric of the observed 
 magnitude from a series of discordant measurements. The given 
 data may be all of the same precision or it may be necessary to 
 assign a different degree of accuracy to the different measure- 
 ments. In either case the solution of the problem is called the 
 adjustment of the measurements. 
 
 The principle of least squares, developed in the theory of errors 
 that leads to the measure of precision cited above, is the basis 
 of all such adjustments. But the particular method of solution 
 adopted in any given case depends on the nature of the measure- 
 ments considered. In the case of a series of direct, equally pre- 
 cise, measurements of a single quantity, the principle of least 
 squares leads to the arithmetical mean as the most probable, and 
 therefore the best, value to assign to the measured quantity. 
 This is also the value that has been universally adopted on a priori 
 grounds. In fact many authors assume the maximum probability 
 of the arithmetical mean as the axiomatic basis for the develop- 
 ment of the law of errors. 
 
 The determination of empirical relations between measured 
 quantities and the constants that enter into them is also based 
 on the principle of least squares. For this reason, such deter- 
 minations are treated in connection with the discussion of the 
 methods for the adjustment of measurements. 
 
 17. Discussion of Instruments and Methods. The theory 
 of errors finds another very important application in the discussion 
 of the relative availableness and accuracy of different instruments 
 and methods of measurement. Used in connection with a few 
 preliminary measurements and a thorough knowledge of the 
 
22 THE THEORY OF MEASUREMENTS [ART. 17 
 
 theory of the proposed instruments and methods, it is sufficient 
 for the determination of the probable precision of an extended 
 series of careful observations. By such means we are able to 
 select the instruments and methods best adapted to the particular 
 purpose in view. We also become acquainted with the parts of 
 the investigation that require the greatest skill and care in order 
 to give a result with the desired precision. 
 
 The cost of instruments and the time and skill required in 
 carrying out the measurements increase much more rapidly than 
 the corresponding precision of the results. Hence these factors 
 must be taken into account in determining the availableness of a 
 proposed method. It is by no means always necessary to strive 
 for the greatest attainable precision. In fact, it would be a 
 waste of time and money to carry out a given measurement with 
 greater precision than is required for the use to which it is to be 
 put. For many practical purposes, a result correct within one- 
 tenth of one per cent, or even one per cent, is amply sufficient. 
 In such cases it is essential to adopt apparatus and methods that 
 will give results definitely within these limits without incurring 
 the greater cost and labor necessary for more precise deter- 
 minations. 
 
CHAPTER III. 
 CLASSIFICATION OF ERRORS. 
 
 ALL measurements, of whatever nature, are subject to three 
 distinct classes of errors, namely, constant errors, mistakes, and 
 accidental errors. 
 
 18. Constant Errors. Errors that can be determined in 
 sign and magnitude by computations based on a theoretical 
 consideration of the method of measurement used or on a pre- 
 liminary study and calibration of the instruments adopted are 
 called constant errors. They are sometimes due to inadequacy of 
 an adopted method of measurement, but more frequently to 
 inaccurate graduation and imperfect adjustment of instruments. 
 
 As a simple illustration, consider the measurement of the 
 length of a straight line with a graduated scale. If the scale is 
 not held exactly parallel to the line, the result will be too great 
 or too small according as the line of sight in reading the scale is 
 normal to the line or to the scale. The magnitude of the error 
 thus introduced depends on the angle between the line and the 
 scale and can be exactly computed when we know this angle and 
 the circumstances of the observations. If the scale is not straight, 
 if its divisions are irregular, or if they are not of standard length, 
 the result of the measurement will be in error by an amount 
 depending on the magnitude and distribution of these inaccuracies 
 of construction. The sign and magnitude of such errors can 
 gener o1 ly be determined by a careful study and calibration of the 
 scai 
 
 If M represents the actual numeric of the measured magnitude, 
 M Q the observed numeric, and Ci, C 2 , C 3 , etc., the constant errors 
 inherent in the method of measurement and the instruments used, 
 
 M = Mo + Ci + C 2 + C 3 + - . (1) 
 
 The necessary number of correction terms Ci, G' 2 , C z , etc., is 
 determined by a careful study of the theory and practical appli- 
 cation of the apparatus and method used in finding M Q . The 
 magnitude and sign of each term are determined by subsidiary 
 
 23 
 
24 THE THEORY OF MEASUREMENTS [ART. 18 
 
 measurements or calculated, on theoretical grounds, from known 
 data. Thus, in the above illustration, suppose that the scale is 
 straight and uniformly graduated, that each of its divisions is 
 1.01 times as long as the unit in which it is supposed to be gradu- 
 ated, and that the line of the graduations makes an angle a with 
 the line to be measured. Under these conditions, the number of 
 correction terms reduces to two: the first, Ci, due to the false 
 length of the scale divisions, and the second, C 2 , due to the lack 
 of parallelism between the scale and the line. 
 
 Since the actual length of each division is 1.01, the .length of 
 Mo divisions, i.e., the length that would have been observed on 
 an accurate scale, is 
 
 M l = Mo X 1.01 = Mo + 0.01 Mo = Mo + Ci, 
 ... Ci = + 0.01 Mo. 
 
 If the line of sight is normal to the line in making the observa- 
 tions, the length M 2 that would have been obtained if the scale 
 had been properly placed is 
 
 M 2 = MO cos a = MO + Czj 
 /. C 2 =-M (l-cosa)=-2M sin 2 ^ 
 and (1) takes the form 
 
 M= Mo + 0.01 Mo - 2M sin 2 |> 
 
 = M (l+0.01-2sin 2 ^Y 
 
 The precision with which it is necessary to determine the cor- 
 rection terms Ci, C 2 , etc., and frequently the number of these 
 terms that should be employed depends on the precision with 
 which the observed numeric M is determined. If M is measured 
 within one-tenth of one per cent of its magnitude, the several 
 correction terms should be determined within one one-hundredth 
 of one per cent of M , in order that the neglected part of the sum 
 of the corrections may be less than one-tenth of one per cent of 
 M . If any correction term is found to be less than the. above 
 limit, it may be neglected entirely since it is obviously useless 
 to apply a correction that is less than one-tenth of the uncer- 
 tainty of M . 
 
 In our illustration, suppose that the precision is such that we 
 are sure that M is less than 1.57 millimeters and greater than 
 
ART. 19] CLASSIFICATION OF ERRORS 25 
 
 1.55 millimeters, but is not sufficient to give the fourth significant 
 figure within several units. Obviously, it would be useless to 
 determine Ci and C% closer than 0.001 millimeter, and if the mag- 
 nitude of either of these quantities is less than 0.001 millimeter 
 our knowledge of the true value of M is not increased by making 
 the corresponding correction. In fact, it is usually impossible 
 to determine the C's with greater accuracy than the above limit, 
 since, as in our illustration, M Q is usually a factor in the correction 
 terms. Hence the writing down of more than the required num- 
 ber of significant figures is mere waste of labor. 
 
 When considering the availableness of proposed methods and 
 apparatus, it is important to investigate the nature and magni- 
 tude of the constant errors inherent in their use. It sometimes 
 happens that the sources of such errors can be sufficiently elimi- 
 nated by suitable adjustment of the instruments or modification 
 of the method of observation. When this is not possible the 
 conditions should be so chosen that the correction terms can be 
 computed with the required precision. Even when all possible 
 precautions have been taken, it very seldom happens that the 
 sum of the constant errors reduces to zero or that the magni- 
 tude of the necessary corrections can be exactly determined. 
 Moreover, such errors are never rigorously constant, but present 
 small fortuitous variations, which, to some extent, are indistinguish- 
 able from the accidental errors to be described later. 
 
 A more detailed discussion of constant errors and the limits 
 within which they should be determined will be given after we 
 have developed the methods for estimating the precision of the 
 observed numeric M. 
 
 19. Personal Errors. When setting cross-hairs, or any other 
 indicating device, to bisect a chosen mark, some observers will 
 invariably set too far to one side of the center, while others will 
 as consistently set on the other side. Again, in timing a transit, 
 some persons will signal too soon and others too late. With 
 experienced and careful observers, the errors introduced in this 
 manner are small and nearly constant in magnitude and sign, 
 but they are seldom entirely negligible when the highest possible 
 precision is sought. 
 
 Errors of this nature will be called personal errors, since their 
 magnitude and sign depend on personal peculiarities of the 
 observer. Their elimination may sometimes be effected by a 
 
26 THE THEORY OF MEASUREMENTS [ART. 20 
 
 careful study of the nature of such peculiarities and the magnitude 
 of the effects produced by them under the conditions imposed 
 by the particular problem considered. Suitable methods for this 
 purpose are available in connection with most of the investiga- 
 tions in which an exact knowledge of the personal error is essential. 
 Such a study is .frequently referred to as a determination of the 
 "Personal Equation" of the observer. 
 
 20. Mistakes. Mistakes are errors due to reading the indi- 
 cations of an instrument carelessly or to a faulty record of the 
 observations. The most frequent of these are the following : 
 the wrong integer is placed before an accurate fractional reading, 
 e.g., 9.68 for 19.68; the reading is made in the wrong direction of 
 the scale, e.g., 6.3 for 5.7; the significant figures of a number are 
 transposed, e.g., 56 is written for 65. Care and strict attention 
 to the work in hand are the only safeguards against such mistakes. 
 
 When a large number of observations have been systematically 
 taken and recorded, it is sometimes possible to rectify an obvious 
 mistake, but unless this can be done with certainty the offending 
 observation should be dropped from the series. This statement 
 does not apply to an observation showing a large deviation from 
 the mean but only to obvious mistakes. 
 
 21. Accidental Errors. When a series of independent meas- 
 urements of the same magnitude have been made, by the same 
 method and apparatus and with equal care, the results generally 
 differ among themselves by several units in the last one or two 
 significant figures. If in any case they are found to be identical, 
 it is probable that the observations were not independent, the 
 instruments adopted were not sufficiently sensitive, the maximum 
 precision attainable was not utilized, or the observations were 
 carelessly made. Exactly concordant measurements are quite as 
 strong evidence of inaccurate observation as widely divergent 
 ones. 
 
 As the accuracy of method and the sensitiveness of instruments 
 is increased, the number of concordant figures in the result in- 
 creases but differences always occur in the last attainable figures. 
 Since there is, generally, no reason to suppose that any one of the 
 measurements is more accurate than any other, we are led to 
 believe that they are all affected by small unavoidable errors. 
 
 After all constant errors and mistakes have been corrected, the re- 
 maining differences between the individual measurements and the true 
 
ART. 22] CLASSIFICATION OF ERRORS 27 
 
 numeric of the measured magnitude are called accidental errors. 
 They are due to the combined action of a large number of inde- 
 pendent causes each of which is equally likely to produce a posi- 
 tive or a negative effect. Probably most of them have their 
 origin in small fortuitous variations in the sensitiveness and 
 adjustment of our instruments and in the keenness of our senses 
 of sight, hearing, and touch. It is also possible that the correla- 
 tion of our sense perceptions and the judgments that we draw 
 from them are not always rigorously the same under the same 
 set of stimuli. 
 
 Suppose that N measurements of the same quantity have been 
 made by the same method and with equal care. Let ai, a^, 3, 
 . . . a N represent the several results of the independent meas- 
 urements, after all constant errors and mistakes have been elim- 
 inated, and let X represent the true numeric of the measured 
 magnitude. Then the accidental errors of the individual measure- 
 ments are given by the differences, 
 
 Ai - ai - X, A 2 = a 2 - X, A 3 = a 3 - X } . . . A^ = a N -X. (2) 
 
 The accidental errors AI, A 2 , . . . A# thus denned are sometimes 
 called the true errors of the observations ai, a 2 , . . . a N . 
 
 22. Residuals. Since the individual measurements a\ t a?, 
 . . . a N differ among themselves, and since there is no reason to 
 suppose that any one of them is more accurate than any other, it 
 is never possible to determine the exact magnitude of the numeric 
 X. Hence the magnitude of the accidental errors A i, A 2 , . . . A# 
 can never be exactly determined. But, if x is the most probable 
 value that we can assign to the numeric X on the basis of our 
 measurements, we can determine the differences 
 
 ri = di x, r z = a 2 x, . . . r N = a N x. (3) 
 
 These differences are called the residuals of the individual measure- 
 ments dij 02, . . . a N . They represent the most probable values 
 that we can assign to the accidental errors AI, A 2 , . . . A# on the 
 basis of the given measurements. 
 
 It should be continually borne in mind that the residuals thus 
 determined are never identical with the accidental errors. How- 
 ever precise our measurements may be, the probability that x is 
 exactly equal to X is always less than unity. As the number 
 and precision of measurements increase, the difference between 
 
28 THE THEORY OF MEASUREMENTS [ART. 23 
 
 the magnitudes x and X decreases, and the residuals continually 
 approach the accidental errors, but exact equality is never attain- 
 able with a finite number of observations. 
 
 23. Principles of Probability. The theory of errors is an 
 application of the principles of probability to the discussion of 
 series of discordant measurements for the purpose of determining 
 the most probable numeric that can be assigned to the measured 
 quantity and making an estimate of the precision of the result 
 thus obtained. A discussion of the fundamental principles of 
 the theory of probability, sufficient for this purpose, is given in 
 most textbooks on advanced algebra, and the student should 
 master them before undertaking the study of the 1 theory of errors. 
 
 For the sake of convenience in reference, the three most useful 
 propositions are stated below without proof. 
 
 PROPOSITION 1. If an event can happen in n independent 
 ways and either happen or fail in N independent ways, the prob- 
 ability p that it will occur in a single trial at random is given by 
 
 the relation 
 
 n , A . 
 
 p - r w 
 
 Also if p' is the probability that it will fail in a single trial at 
 random, 
 
 p = l_p = !_.. ( 5 ) 
 
 PROPOSITION 2. If the probabilities for the separate occurrence 
 of n independent events are respectively pi, p%, . . . p n , the prob- 
 ability PS that some one of these events will occur in a single trial 
 at random is given by the relation 
 
 PS = Pi + Pz + Pz + ' ' ' + P^ (6) 
 
 PROPOSITION 3. If the probabilities for the separate occurrence 
 of n independent events are respectively pi, p 2 , . . . p n , the 
 probability P that all of the events will occur at the same time is 
 given by the relation 
 
 P = Pi X P2 X X Pn. (7) 
 
CHAPTER IV. 
 THE LAW OF ACCIDENTAL ERRORS. 
 
 24. Fundamental Propositions. The theory of accidental 
 errors is based on the principle of the arithmetical mean and the 
 three axioms of accidental errors. When the word " error " is used 
 without qualification, in the statement of these propositions and 
 in the following pages, accidental errors are to be understood. 
 
 Principle of the Arithmetical Mean. The most probable value 
 that can be assigned to the numeric of a measured magnitude, on 
 the basis of a number of equally trustworthy direct measurements, 
 is the arithmetical mean of the given 'measurements. 
 
 This proposition is self-evident in the case of two independent 
 measurements, made by the same method with equal care, since 
 one of them is as likely to be exact as the other, and hence it is 
 more probable that the true numeric lies halfway between them 
 than in any other location. Its extension to more than two 
 measurements is the only rational assumption that we can make 
 and is sanctioned by universal usage. 
 
 First Axiom. In any large number of measurements, positive 
 and negative errors of the same magnitude are equally likely to 
 occur. The number of negative errors is equal to the number 
 of positive errors. 
 
 Second Axiom. Small errors are much more likely to occur 
 than large ones. 
 
 Third Axiom. All of the errors of the measurements in a 
 given series lie between equal positive and negative limits. Very 
 large errors do not occur. 
 
 The foundation of these propositions is the same as that of the 
 axioms of geometry. Namely: they are general statements that 
 are admitted as self-evident or accepted as a basis of argument by 
 all competent persons. Their justification lies in the fact that 
 the results derived from them are found to be in agreement with 
 experience. 
 
 25. Distribution of Residuals. It was pointed out in article 
 twenty-two that the true accidental errors, represented by A's, 
 
 29 
 
30 
 
 THE THEORY OF MEASUREMENTS [ART. 26 
 
 cannot be determined in practice, but the residuals, represented 
 by r's, can be computed from the given observations by equation 
 (3). The A's may be considered as the limiting values toward 
 which the r's approach as the number of observations is indefinitely 
 increased. If the residuals corresponding to a very large num- 
 ber of observations are arranged in groups according to sign and 
 magnitude, the groups containing very small positive or negative 
 residuals will be found to be the largest, and, in general, the magni- 
 tude of the groups will decrease nearly uniformly as the magnitude 
 of the contained residuals increases either positively or negatively. 
 Let n represent the number of residuals in any group, and r their 
 common magnitude, then the distribution of the residuals, in 
 regard to sign and magnitude, may be represented graphically 
 by laying off ordinates proportional to the numbers n against 
 
 abscissae proportional to the corresponding magnitudes r. The 
 points, thus located, will be approximately uniformly distributed 
 about a curve of the general form illustrated in Fig. 2. 
 
 The number of residuals in each group will increase with the 
 total number of measurements from which the r's are computed. 
 Consequently the ordinates of the curve in Fig. 2 will depend on 
 the number of observations considered as well as on their accuracy. 
 Hence, if we wish to compare different series of measurements with 
 regard to accuracy, we must in some way eliminate the effect of 
 differences in the number of observations. Moreover, we are not 
 so much concerned with the total number of residuals of any given 
 magnitude as with the relative number of residuals of different 
 magnitudes. For, as we shall see, the acuracy of a series of 
 observations depends on the ratio of the number of small errors 
 to the number of large ones. 
 
 26. Probability of Residuals. Suppose that a very large 
 number N of independent measurements have been made and that 
 
AKF.27J THE LAW OF ACCIDENTAL ERRORS 31 
 
 the corresponding residuals have been computed by equation (3). 
 By arranging the results in groups according to sign and magni- 
 tude, suppose we find HI residuals of magnitude n, n 2 of magni- 
 tude r 2 , etc., and n\ of magnitude n, n/ of magnitude r 2 , etc. 
 If we choose one of the measurements at random, the probability 
 
 that the corresponding residual is equal to r\ is -^ , since there 
 
 are N residuals and n\ of them are equal to r\. In general, if y\, y 2 , 
 Hi, 2/2', represent the probabilities for the occurrence 
 of residuals equal to n, r 2 , . . . n, r 2 , . . . respectively, 
 
 When N is increased by increasing the number of measurements, 
 each of the n's is increased in nearly the same ratio since the 
 residuals of the new measurements are distributed in essentially 
 the same manner as the old ones, provided all of the measure- 
 ments considered are made by the same method and with equal 
 care. Consequently, the y's corresponding to a definite method 
 of observation are nearly independent of the number of measure- 
 ments. As N increases they oscillate, with continually decreas- 
 ing amplitude, about the limiting values that would be obtained 
 with an infinite number of observations. Hence the form of a 
 curve, having y's for ordinates and corresponding r's for abscissae, 
 depends on the accuracy of the measurements considered and is 
 sensibly independent of N, provided it is a large number. 
 
 27. The Unit Error. The relative accuracy of different 
 series of measurements might be studied with the aid of the corre- 
 sponding y : r curves, but since the y's are abstract numbers, and 
 the r's are concrete, being of the same kind as the measurements, 
 it is better to adopt a slightly different mode of representation. 
 For this purpose, each of the r's is divided by an arbitrary con- 
 stant k, of the same kind as the measurements, and the abstract 
 
 numbers y^> -^> etc., are used as abscissae in place of the r's. In 
 
 A/ K 
 
 the following pages, k will be called the unit error. Its magnitude 
 may be arbitrarily chosen in particular cases, but, when not 
 definitely specified to the contrary, it will be taken equal to the 
 least magnitude that can be directly observed with the instru- 
 ments and methods used in making the measurements. To 
 
32 
 
 THE THEORY OF MEASUREMENTS I ART. 28 
 
 illustrate: suppose we are measuring a given length with a scale 
 divided in millimeters. By estimation, the separate observations 
 can be made to one-tenth of a millimeter. Hence, in this case 
 we should take k equal to one-tenth of a millimeter. 
 
 If the residuals are arranged in the order of increasing magni- 
 tude, it is obvious that the successive differences TI r , r? TI 
 etc., are all equal to k. Hence, if the most probable value of the 
 measured quantity, x in equation (3), is taken to the same num- 
 ber of significant figures as the individual measurements, all of 
 the residuals are integral multiples of k and we have 
 
 k 
 
 k 
 
 28. The Probability Curve. The result of a study of the 
 distribution of the residuals may be arranged as illustrated in the 
 following table, where n is the number of residuals of magnitude 
 r; y is the probability that a single residual, chosen at random, is 
 of magnitude r; N is the total number of measurements, and k is 
 the unit error. 
 
 r 
 
 n 
 
 V 
 
 r 
 ~k 
 
 -r p 
 
 n' p 
 
 ~N~ 
 
 -P 
 
 -n 
 
 * 
 
 w 
 
 -1 
 
 "0 
 
 no 
 
 N 
 
 
 
 ri 
 
 ni 
 
 N 
 
 +1 
 
 rp 
 
 n p 
 
 w 
 
 +P 
 
 M 
 
 Plotting y against ^ we obtain 2 p discrete points as in Fig. 3. 
 
 When N is large, these points, are somewhat symmetrically dis- 
 tributed about a curve of the general form illustrated by the 
 dotted line. If a larger number of observations is considered, 
 
ART. 29] THE LAW OF ACCIDENTAL ERRORS 
 
 33 
 
 some of the points will be shifted upward while others will be 
 shifted downward, but the distribution will remain approxi- 
 mately symmetrical with respect to the same curve. In general, 
 successive equal increments to N cause shifts of continually de- 
 creasing magnitude; and in the limit, when TV becomes equal to 
 infinity, and the residuals are equal to the accidental errors, the 
 points would be on a uniform curve symmetrical to the y Q ordi- 
 nate. The curve thus determined represents the relation between 
 the magnitude of an error and the probability of its occurrence 
 in a given series of measurements. For this reason it is called 
 the probability curve. 
 
 29. Systems of Errors. The coordinates of the probability 
 curve are y and-r-, since it represents the distribution of the true 
 
 accidental errors AI, A 2 , etc., in regard to relative frequency and 
 magnitude. Since the curve is uniform, it represents not only 
 the errors of the actual observations, but also the distribution of 
 all of the accidental errors that would be found if the sensitive- 
 ness of our instruments were infinitely increased and an infinite 
 number of observations were made, provided only that all of the 
 observations were made with the same degree of precision and 
 entirely independently. 
 
 All of the errors represented by a curve of this type belong to a 
 definite system, characterized by the magnitude of the maximum 
 ordinate yo and the slope of the curve. Hence, every probability 
 curve represents a definite system of errors. It also represents 
 the accidental errors of a series of measurements of definite pre- 
 cision. Hence, the accidental errors of series of measurements of 
 different precision belong to different systems, and each series 
 is characterized by a definite system of errors. 
 
 The probability curves A and B in Fig. 4 represent the systems 
 
34 
 
 THE THEORY OF MEASUREMENTS [ART. 30 
 
 of errors that characterize two series of measurements of different 
 precision. As the precision of measurement is increased it is 
 obvious that the number of small errors will increase relatively 
 to the number of large ones. Consequently the probability of 
 small errors will be greater and that of large ones will be less in 
 the more precise series A than in the less precise series B. Hence, 
 the curve A has a greater maximum ordinate and slopes more 
 rapidly toward the horizontal axis than the curve B. 
 
 30. The Probability Function. The maximum ordinate and 
 the slope of the probability curve depend on the constants that 
 appear in the equation of the curve. When we know the form 
 of the equation and have a method of determining the numerical 
 value of the constants, we are able to determine the relative pre- 
 cision of different series of measurements. Since the curve repre- 
 sents the distribution of the true accidental errors, we are also able 
 to compare the distribution of these errors with that of the resid- 
 uals and thus develop workable methods for finding the most 
 probable numeric of the measured magnitude. 
 
 It is obvious, from an inspection of Figs. 3 and 4, that y is a 
 continuous function of A, decreasing very rapidly as the magni- 
 tude of A increases either positively or negatively and symmetrical 
 with respect to the y axis. Hence, the probability curve sug- 
 gests an equation in the form 
 
 (9) 
 
ART. 31] THE LAW OF ACCIDENTAL ERRORS 35 
 
 where e is the base of the Napierian system of logarithms, o> is a 
 constant depending on the precision of the series of measurements 
 considered, and the other variables have been defined above. 
 This equation can be derived analytically from the three axioms 
 of accidental errors, with the aid of several plausible assumptions 
 regarding the constitution of such errors, or from the principle 
 of the arithmetical mean. However, the strongest evidence of 
 its exactness lies in the fact that it gives results in substantial 
 agreement with experience. Consequently, we will adopt it as an 
 empirical relation, and proceed to show that it is in conformity 
 with the three axioms and leads to the arithmetical mean as the 
 most probable numeric derivable from a series of equally good 
 independent measurements of the same magnitude. 
 
 Equation (9) is the mathematical expression of the law of 
 accidental errors and is often referred to simply as the law of 
 errors. Its right-hand member is called the probability function 
 and, for the sake of convenience, is represented by (A), giving 
 the relations 
 
 2/ = 0(A); ^(A)^' 2 ^. (10) 
 
 31. The Precision Constant. The curves in Fig. 4 were 
 plotted, to the same scale, from data computed by equation (9). 
 The constant w was taken twice as great for the curve A as for 
 the curve B, and in both cases values of y were computed for suc- 
 cessive integral values of the ratio r-- The maximum ordinate of 
 
 each of these curves corresponds to the zero value of A and is 
 equal to the value of co used in computing the y's. The curve 
 A, corresponding to the larger value of o>, approaches the hori- 
 zontal axis much more rapidly than the curve B. 
 
 Obviously, the constant co determines both the maximum 
 ordinate and the slope of the probability curve. But we have 
 seen that these characteristics are proportional to the precision 
 of the measurements that determine the system of errors repre- 
 sented. Hence co characterizes the system of errors consid- 
 ered and is proportional to the precision of the corresponding 
 measurements. Some writers have called it the precision measure, 
 but, as it depends only on the accidental errors and takes no 
 account of the accuracy with which constant errors are avoided 
 or corrected, it does not give a complete statement of the pre- 
 
36 THE THEORY OF MEASUREMENTS [ART. 32 
 
 cision. Consequently the term " precision measure " will be re- 
 served for a function to be discussed later, and a; will be called the 
 precision constant in the following pages. 
 
 When A is taken equal to zero in equation (9), y is equal to co. 
 Hence the precision of measurements, so far as it depends upon 
 accidental errors, is proportional to the probability for the occur- 
 rence of zero error in the corresponding system of errors. In 
 this connection, it should be borne in mind that the system of 
 errors includes all of the errors that would have been found 
 with an infinite number of observations, and that it cannot be 
 restricted to the errors of the actual measurements for the pur- 
 pose of computing o> directly. Indirect methods for computing 
 a> from given observations will be discussed later. 
 
 32. Discussion of the Probability Function. Inspection of 
 the curves in Fig. 4, in connection with equation (9), is sufficient to 
 show that the probability function is in agreement with the first 
 two axioms. Since y is an even function of A, positive and nega- 
 tive errors of the same magnitude are equally probable, and conse- 
 quently equally numerous in an extended series of measurements. 
 Hence the first axiom is fulfilled. Since A enters the function 
 only in the negative exponent, the probability for the occurrence 
 of an error decreases very rapidly as its magnitude increases 
 either positively or negatively. Hence small errors are much more 
 likely to occur than large ones and the second axiom is fulfilled. 
 
 Since the function </> (A) is continuous for values of A ranging 
 from minus infinity to plus infinity, it is apparently at variance 
 with the third axiom. For, if all of the errors lie between definite 
 finite limits L and + L, (A) should be continuous while A 
 lies between these limits and equal to zero for all values of A 
 outside of them. But we have no means of fixing the limits 
 -f- L and L, in any given case; and we note that 0(A) becomes 
 very small for moderately large values of A. Hence, whatever the 
 true value of L may be, the error involved in extending the limits 
 to oo and +00 is infinitesimal. Consequently, </>(A) is in sub- 
 stantial agreement with the third axiom provided it leads to the 
 conclusion that all possible errors lie between the limits oo and 
 + oo . This will be the case if it gives unity for the probability 
 that a single error, chosen at random, lies between oo and -f oo . 
 For, if all of the errors lie between these limits, the probability 
 considered is a certainty and hence is represented by unity. 
 
ART. 33] THE LAW OF ACCIDENTAL ERRORS 
 
 37 
 
 33. The Probability Integral. The accidental errors, corre- 
 sponding to actual measurements, may be arranged in groups ac- 
 cording to their magnitude in the same manner that the residuals 
 were arranged in article twenty-eight. When this is done the 
 errors in succeeding groups differ in magnitude by an amount 
 equal to the unit error k t since k is the least difference that can 
 be determined with the instruments used in making the obser- 
 vations. Hence, if A p is the common magnitude of the errors 
 in the pth group, 
 
 -A = A 
 
 (P+2) 
 
 -A 
 
 (p+i) 
 
 or, expressing the same relation in different form, 
 
 
 where a- is an indeterminate quantity that enters each of the 
 equations because we do not know the actual magnitude of the 
 
 A's. 
 
 FIG. 5. 
 
 Let the probability curve in Fig. 5 represent the system of 
 errors to which the errors of the actual measurements belong. 
 Then the ordinates y p , 2/( p +i), 2/( P +2), 2/(p+a) represent the 
 probabilities of the errors A p , A( p +i>, . . . A( p + e ) respectively. 
 Since the errors of the actual measurements satisfy the relation 
 (i), none of them correspond to points of the curve lying between 
 the ordinates y p , 2/( P + i), . . . 2/( P +). Hence, in virtue of equa- 
 tion (6), article twenty-three, if we choose one of the measure- 
 ments at random the probability that the magnitude of its error 
 lies between A p and A( P + Q ) is 
 
 2/CP+8)- 
 
38 THE THEORY OF MEASUREMENTS [ART. 33 
 
 Multiplying and dividing the second member by q, 
 
 where y pq is written for the mean of the ordinates between y p 
 and 2/(p+ fl ). From equation (i) 
 
 & 
 Hence, 
 
 In the limit, when we consider the errors of an infinite number 
 of measurements made with infinitely sensitive instruments, every 
 point of the curve represents the probability of one of the errors 
 of the system. Consequently, for any finite value of q, Ihe inter- 
 val between the ordinates y p and y( P +q> is infinitesimal, and all 
 of the ordinates between these limits may be considered equal. 
 Hence, in the limit, 
 
 p = , y pq = 2/ A = 
 and (iii) reduces to 
 
 =* (A) , (11) 
 
 where y% +d * represents the probability that the magnitude of a 
 single error, chosen at random, is between A and A + dA. 
 
 By applying the usual reasoning of the integral calculus, it is 
 evident that the expression 
 
 rf = I /% (A) JA, (12) 
 
 /t i/ a 
 
 represents the probability that the magnitude of an error, chosen 
 at random, lies between the limits a and b. The integral in this 
 expression also represents the area under the probability curve 
 
 between the ordinates at T and T. Consequently the probability 
 
 in question is represented graphically by the shaded area in Fig. 6. 
 
 The probability that an error, chosen at random, is numerically 
 
 less than a given error A is equal to the probability that it lies 
 
ART. 33] THE LAW OF ACCIDENTAL ERRORS 39 
 
 between the limits A and -J-A. Hence, if we designate this 
 probability by PA, 
 
 A 
 
 A 
 
 since (A) is an even function of A. Introducing the complete 
 expression for (A) from equation (10) we obtain 
 
 A 2 
 
 k jo 
 For the sake of simplification, put 
 
 2 A 2 
 
 then 
 
 /Y'ett, 
 
 Jo 
 
 (13) 
 
 which is an entirely general expression for the probability PA, 
 applicable to any system of errors when we know the correspond- 
 ing values of the constants o> and k. A series of numerical values 
 of the right-hand member of (13), corresponding to successive 
 values of the argument t, is given in Table XI, at the end of 
 this volume. Obviously, this table may be used in computing 
 the probability PA corresponding to any system of errors, since 
 the characteristic constants o> and k appear only in the limit of the 
 integral. 
 
 Whatever the values of the constants w and k, the limit vVw T 
 
40 THE THEORY OF MEASUREMENTS [ART. 34 
 
 becomes infinite when A is equal to infinity. Hence, in every 
 system of errors, 
 
 * dt = l ) (13a) 
 
 where the numerical value is that given in Table XI, for the limit 
 t equals infinity. Consequently the probability function (A) 
 leads to the conclusion that all of the errors in any system lie 
 between the limits <x> and +00, and, therefore, it fulfills the 
 condition imposed by the third axiom as explained in the last 
 paragraph of article thirty-two. 
 
 34. Comparison of Theory and Experience. Equation (13) 
 may be used to compare the distribution of the residuals actually 
 found in any series of measurements with the theoretical distri- 
 bution of the accidental errors. If N equally trustworthy meas- 
 urements of the same magnitude have been made, all of the N 
 corresponding accidental errors belong to the same system, and 
 the probability that the error of a single measurement is numer- 
 ically less than A is given by PA in equation (13). Consequently, 
 if N is sufficiently large, we should expect to find 
 
 # A = NP* (iv) 
 
 errors less than A. For, if we consider only the errors of the 
 actual measurements, the probability that one of them is less 
 than A is equal to the ratio of the number less than A to the total 
 number. In the same manner, the number less than A 7 should 
 be 
 
 Hence, the number lying between the limits A and A' should be 
 
 N* = N* - N*. (v) 
 
 These numbers may be computed by equation (13) with the aid 
 of Table XI, when we know N and the value of the expression 
 
 V^co 
 
 corresponding to the given measurements. The number, 
 
 N r r , of residuals lying between the limits r equals A and r' equals 
 A' may be found by inspecting the series of residuals computed 
 from the given measurements by equation (3), article twenty-two. 
 If N is large and the errors of the given measurements satisfy 
 the theory we have developed, the numbers N% and N r r ' should 
 
ART. 34] THE LAW OF ACCIDENTAL ERRORS 
 
 41 
 
 be very nearly equal, since in an extended series of measurements 
 the residuals are very nearly equal to the accidental errors. 
 
 The following illustration, taken from Chauvenet's "Manual 
 of Spherical and Practical Astronomy," is based on 470 obser- 
 vations of the right ascension of Sirius and Altair, by Bradley. 
 The errors of these measurements belong to a system character- 
 ized by a particular value of the ratio T that has been computed, 
 
 by a method to be described later (articles thirty-eight and forty- 
 two), and gives the relation 
 
 VTTCO 
 
 k 
 
 = 1.8086. 
 
 Consequently, to find the theoretical value of PA, corresponding 
 to any limit A, we take t equal to 1.8086 A in equation (13) and 
 find the corresponding value of the integral by interpolation from 
 Table XL 
 
 The third column of the following table gives the values of 
 PA corresponding to the chosen values of A in the first column 
 and the computed values of t in the second column. The fourth 
 column gives the corresponding values of N&. computed by equa- 
 tion (iv), taking N equal to 470. The sixth column, computed 
 by equation (v), gives the number, Nj[, of errors that should 
 lie between the limits A and A' given in the fifth. The seventh 
 column gives the number of residuals actually found between the 
 same limits. 
 
 A 
 
 t 
 
 ^A 
 
 ^A 
 
 Limits 
 
 A A' 
 
 < 
 
 N r 
 
 // 
 
 0.1 
 
 0.1809 
 
 0.2019 
 
 95 
 
 0.0-0.1 
 
 95 
 
 94 
 
 0.2 
 
 0.3617 
 
 0.3910 
 
 184 
 
 0.1-0.2 
 
 89 
 
 88 
 
 0.3 
 
 0.5426 
 
 0.5571 
 
 262 
 
 0.2-0.3 
 
 78 
 
 78 
 
 0.4 
 
 0.7234 
 
 0.6937 
 
 326 
 
 0.3-0.4 
 
 64 
 
 58 
 
 0.5 
 
 0.9043 
 
 0.7990 
 
 376 
 
 0.4-0.5 
 
 50 
 
 51 
 
 0.6 
 
 1.0852 
 
 0.8751 
 
 411 
 
 0.5-0.6 
 
 35 
 
 36 
 
 0.7 
 
 1.2660 
 
 0.9266 
 
 436 
 
 0.6-0.7 
 
 " 25 
 
 26 
 
 0.8 
 
 1.4469 
 
 0.9593 
 
 451 
 
 0.7-0.8 
 
 15 
 
 14 
 
 0.9 
 
 1.6277 
 
 0.9787 
 
 460 
 
 0.8-0.9 
 
 9 
 
 10 
 
 1.0 
 
 1.8086 
 
 0.9895 
 
 465 
 
 0.9-1.0 
 
 5 
 
 7 
 
 00 
 
 GO 
 
 1.0000 
 
 470 
 
 l.O-oo 
 
 5 
 
 8 
 
 Comparison of the numbers in the last two columns shows very 
 good agreement between theory, represented by N%, and expe- 
 
42 THE THEORY OF MEASUREMENTS [ART. 35 
 
 rience, represented by N r r f , when we remember that the theory 
 assumes an infinite number of observations and that the series 
 considered is finite. Numerous comparisons of this nature have 
 been made, and substantial agreement has been found in all 
 cases in which a sufficient number of independent observations 
 have been considered. In general, the differences between N% 
 and N^' decrease in relative magnitude as the number of obser- 
 vations is increased. 
 
 35. The Arithmetical Mean. In article twenty-four it was 
 pointed out, as one of the fundamental principles of the theory 
 of errors, that the arithmetical mean of a number of equally trust- 
 wor^hy direct measurements on the same magnitude is the most 
 probable value that we can assign to the numeric of the measured 
 magnitude. In order to show that the probability function (A) 
 leads to the same conclusion, let eft, a 2 , . AT represent the 
 given measurements, and let x represent the unknown numeric 
 of the measured magnitude. If the actual value of this numeric 
 is X, the true accidental errors of the given measurements are 
 
 Ai = ai X, A 2 = 02 X, . . . AAT = ax X, (2) 
 
 and all of them belong to the same system, characterized by a 
 particular value .of the precision constant co. The probability 
 that one of the errors of this system, chosen at random, is equal 
 to an arbitrary magnitude A p is given by the relation 
 
 Since we cannot determine the true value X, the most probable 
 value that we can assign to x is that which gives a maximum 
 probability that N errors of the system are equal to the N resid- 
 uals 
 
 TI = ai x, r z = a 2 x, . . . r N = a N x. (3) 
 
 This is equivalent to determining x, so that the residuals are as 
 nearly as possible equal to the accidental errors. 
 
 If 2/1, 2/2, ... VN represent the probabilities that a single error 
 of the system, chosen at random, is equal to r\, r 2 , . . . r N respec- 
 tively, 
 
 2/i = (n), 2/2 = (r 2 ), . . . y N = 
 
 Hence, if P is the probability that N of the errors chosen together 
 
ART. 35] THE LAW OF ACCIDENTAL ERRORS 43 
 
 are equal to n, r 2 , . . . r N respectively, we have, by equation (7), 
 article twenty-three, 
 
 P = 2/1 X 2/2 X ... X y N 
 
 Since the exponent in this expression is negative and -^ is con- 
 
 K 
 
 stant, the maximum value of P will correspond to the minimum 
 value of (ri 2 + r 2 2 + . . . -f ?W 2 ). Hence the most probable 
 value of x is that which renders the sum of the squares of the 
 residuals a minimum. 
 
 In the present case, the r's are functions of a single independent 
 variable x. Consequently the sum of the squares of the r's will 
 be a minimum when x satisfies the condition 
 
 -f-(ri 2 + r 2 2 + . ... +/VO =0. 
 
 (JJU 
 
 Substituting the expression for the r's in terms of x from equation 
 (3) this becomes 
 
 (a, - xY + (a 2 - xY + . . . + (a* - z) 2 = 0. 
 dx( ) 
 
 Hence, (i - x) + (a 2 - x) + . . . + (a N - x) = 0, (14) 
 
 ai -f 2 + + AT 
 and x = jy- 
 
 Consequently, if we take x equal to the arithmetical mean of the 
 a's in (3), the sum of the squares of the computed r's is less than 
 for any other value of x. Hence the probability P that N errors 
 of the system are equal to the N residuals is a maximum, and the 
 arithmetical mean is the most probable value that we can assign 
 to the numeric X on the basis of the given measurements. 
 
 Equation (14) shows that the sum of the residuals, obtained 
 by subtracting the arithmetical mean from each of the given 
 measurements, is equal to zero. This is a characteristic property 
 of the arithmetical mean and serves as a useful check on the 
 computation of the residuals. 
 
 The argument of the present article should be regarded as a 
 justification of the probability function 0(A) rather than as a 
 proof of the principle of the arithmetical mean. As pointed out 
 above, this principle is sufficiently established on a priori grounds 
 and by common consent. 
 
CHAPTER V. 
 CHARACTERISTIC ERRORS. 
 
 SEVERAL different derived errors have been used as a measure 
 of the relative accuracy of different series of measurements. Such 
 errors are called characteristic errors of the system, and they de- 
 crease in magnitude as the accuracy of the measurements, on which 
 they depend, increases. Those most commonly employed are the 
 average error A , the mean error M, and the probable error E, any 
 one of which may be used as a measure on the relative accuracy 
 of a single observation. 
 
 36. The Average Error. The average error A of a single 
 observation is the arithmetical mean of all of the individual errors 
 of the system taken without regard to sign. That is, all of the 
 errors are taken as positive in forming the average. Hence, if 
 N is the total number of errors, 
 
 ! _ 
 
 ~N~ "W 
 
 where the square bracket [ ] is used as a sign of summation, and 
 the ~~ over the A indicates that, in taking the sum, all of the A's 
 are to be considered positive. 
 
 In accordance with the usual practice of writers on the theory 
 of errors, the square bracket [ ] will be used as a sign of summa- 
 tion, in the following pages, in place of the customary sign S. 
 This notation is adopted because it saves space and renders com- 
 plicated expressions more explicit. 
 
 In equation (15) all of the errors of the system are supposed 
 to be included in the summation. Hence, both [A] and N are 
 infinite and the equation cannot be applied to find A directly 
 from the errors of a limited number of measurements. Conse- 
 quently we will proceed to show how the average error can be 
 derived from the probability function, and to find its relation 
 to the precision constant co. A little later we shall see how A 
 can be computed directly from the residuals corresponding to a 
 limited number of measurements. 
 
 44 
 
ART. 36] CHARACTERISTIC ERRORS 45 
 
 If yd is the probability that the magnitude of a single error, 
 chosen at random, lies between A and A + dA, and rid is the num- 
 ber of errors between these limits, 
 
 and consequently 
 
 n d = Ny d 
 
 = N4> (A) ^ (16) 
 
 in virtue of equation (11), article thirty-three, where A represents 
 the mean magnitude of the errors lying between A and A + dA. 
 Hence, the sum of the errors between these limits is 
 
 and the sum of the errors between A = a and A = b is 
 
 N 
 
 Substituting the complete expression for </>(A) from equation (10) 
 this becomes 
 
 Hence, the sum of the positive errors of the system is 
 
 Nu / -*, 
 -; I Ae kz dA, 
 k Jo 
 
 and the sum of the negative errors is 
 
 Nu r 
 
 k J -<* 
 
 These two integrals are obviously equal in magnitude and opposite 
 in sign. Consequently the sum of all of the errors of the system 
 taken without regard to sign is 
 
 Ae -^A (17) 
 
 7TCO 
 
46 THE THEORY OF MEASUREMENTS [ART. 37 
 
 Hence from equation (15), 
 
 ~ N 
 and introducing the numerical value of IT, 
 
 A =0.3183-- (19) 
 
 CO 
 
 37. The Mean Error. The mean error M of a single meas- 
 urement in a given series is the square root of the mean of the 
 squares of the errors in the system determined by the given 
 measurements. Expressed mathematically 
 
 A^ + A^-f-.* + A^_[A1 
 N ' N 
 
 This equation includes all of the errors that belong to the given 
 system. Hence, as pointed out in article thirty-six, in regard to 
 equation (15), it cannot be applied directly to a limited series of 
 measurements. 
 By equation (16) the number of errors with magnitudes between 
 
 the limits A and A + dA is equal to , . Consequently 
 
 /c 
 
 the sum of the squares of the errors between these limits is equal 
 #A 2 4>(A)dA 
 
 k 
 in the last article, 
 
 to - .; . Hence, by reasoning similar to that employed 
 
 (21) 
 
 / 
 
 / 
 
 2N r A% -, * 
 
 since the integrand is an even function of A. Integrating by 
 parts, 
 
 7TCO 
 
 The first term of the second member of this equation reduces to 
 
AKT.38] CHARACTERISTIC ERRORS 47 
 
 zero when the limits are applied. Putting t 2 for in the 
 
 K 
 
 second term, 
 
 [Al-^P^a- (22) 
 
 TT^CO 2 Jo 2 7TC0 2 
 
 in virtue of equation (13a). Hence, 
 
 N 2* 
 and 
 
 M = 
 
 = 0.3989-- 
 
 CO 
 
 (23) 
 
 38. The Probable Error. The probable error E of a single 
 measurement is a magnitude such that a single error, chosen at 
 random from the given system, is as likely to be numerically 
 greater than E as less than E. In other words, the probability 
 that the error of a single measurement is greater than E is equal 
 to the probability that it is less than E. Hence, in any extended 
 series of measurements, one-half of the errors are less than E and 
 one-half of them are greater than E. 
 
 The name " probable error," though sanctioned by universal 
 usage, is unfortunate; and the student cannot be too strongly 
 cautioned against a common misinterpretation of its meaning. 
 The probable error is NOT the most probable magnitude of the 
 error of a single measurement and it DOES NOT determine the 
 limits within which the true numeric of the measured magnitude 
 may be expected to lie. Thus, if x represents the measured 
 numeric of a given magnitude Q and E is the probable error of x, 
 it is customary to express the result of the measurement in the 
 form 
 
 Q = x E. 
 
 This does not signify that the true numeric of Q lies between the 
 limits x E and x + E, neither does it imply that x is probably 
 in error by the amount E. It means that the numeric of Q is as 
 likely to lie between the above limits as outside of them. If a 
 new measurement is made "by the same method and with equal 
 care, the probability that it will differ from x by less than E is 
 equal to the probability that it will differ by more than E. 
 
48 
 
 THE THEORY OF MEASUREMENTS [ART. 38 
 
 In article thirty-three it was pointed out that the probability 
 that an error, chosen at random from a given system, lies between 
 the limits A = a and A = b is represented by the area under the 
 probability curve between the ordinates corresponding to the 
 limiting values of A. Hence, the probability that the error of a 
 single measurement is numerically less than E may be represented 
 by the area under the probability curve between the ordinates y- E 
 and y+ E , in Fig. 7, and the probability that it is greater than E by 
 the sum of the areas outside of these ordinates. Since these two 
 
 FIG. 7. 
 
 probabilities are equal, by definition, the ordinates correspond- 
 ing to the probable error bisect the areas under the two branches 
 of the probability curve. 
 
 Since the probability that the error of a single measurement is 
 less than E is equal to the probability that it is greater than E 
 and the probability that it is less than infinity is unity, the 
 probability that it is less than E is one-half. Consequently, 
 putting A equal to E in equation (13), article thirty-three, 
 
 Pw = ~ 
 
 rw T" 1 
 
 e-dt - 2- 
 
 \J 
 
 From Table XI, 
 
 PA = 0.49375 for the limit t = 0.47, 
 PA = 0.50275 for the limit t = 0.48, 
 
 and by interpolation, 
 
 P E = 0.50000 for the limit t = 0.47694. 
 Hence, equation (24) is satisfied when 
 
 (24) 
 
 = 0.47694, 
 
ART. 39] 
 and we have 
 
 CHARACTERISTIC ERRORS 
 
 E 
 
 0.47694 k 
 
 VTT w 
 
 = 0.2691 - 
 
 CO 
 
 49 
 
 (25) 
 
 39. Relations between the Characteristic Errors. Elimina- 
 
 k 
 ting- from equations (18), (23), and (25), taken two at a time, we 
 
 obtain the relations 
 
 (26") 
 E = 0.4769 VTT -A = 0.8453 -A, 
 
 E = 0.4769 V2 M = 0.6745 M,. 
 
 which express the relative magnitudes of the average, mean, and 
 probable errors. These relations are universally adopted in com- 
 
 MAE 
 k k k 
 
 FIG. 8. 
 
 puting the precision of given series of measurements, and they 
 should be firmly fixed in mind. 
 
 The three equations from which the relations (26) are derived 
 may be put in the form 
 
 A = 0.3183 
 
 k co 
 M _ 0.3989 
 k co 
 E = 0.2691 
 
 k co 
 
 The probability curve in Fig. 8 represents the distribution of 
 the errors in a system characterized by a particular value of co, 
 
 (27) 
 
50 THE THEORY OF MEASUREMENTS [ART. 39 
 
 determined by a given series of measurements. The ordinates 
 
 AM A E 
 VA> VM> an d Us correspond to the abscissae -^> -jp and -"& > com " 
 
 puted by the above equations. Consequently, y A represents the 
 probability that the error of a single measurement is equal to 
 +A, y M the probability that it is equal to +M, and y E the prob- 
 ability that it is equal to +E. In like manner y- A , y- M , and 
 y~ E represent the respective probabilities for the occurrence of 
 errors equal to A, M, and E. 
 
 A curve of this type can be constructed to correspond to any 
 given series of measurements, and in all cases the relative loca- 
 tion of the ordinates y A , y M) and y E will be the same. It was 
 pointed out in the last article that the ordinates y E and y- E bisect 
 the areas under the two branches of the curve. Consequently, 
 in an extended series of measurements, somewhat more than one- 
 half of the errors will be less than either the average or the mean 
 error. Moreover, it is obvious from Fig. 8 that an error equal to 
 E is somewhat more likely to occur than one equal to either A or M. 
 
 Since each of the characteristic errors A, M, and E, bears a 
 constant relation to the precision constant co, any one of them 
 might be used as a measure of the precision of a single measure- 
 ment in a given series, so far as this depends on accidental errors. 
 The probable error is more commonly employed for this purpose 
 on account of its median position in the system of errors deter- 
 mined by the given measurements. 
 
 It is interesting to observe that the ordinate y M corresponds to 
 a point of inflection in the probability curve. By the ordinary 
 method of the calculus we know that this curve has a point of 
 inflection corresponding to the abscissa that satisfies the relation 
 
 Substituting the complete expression for y 
 
 Hence, 
 
ART. 40] CHARACTERISTIC ERRORS 51 
 
 is the abscissa of the point of inflection. Comparing this with 
 equation (23) we see that 
 
 and consequently that the ordinates y M and y- M meet the prob- 
 ability curve at points of inflection. 
 
 40. Characteristic Errors of the Arithmetical Mean. Equa- 
 tion (23) may be put in the form 
 
 CO 2 1 
 
 where M is the mean error of a single measurement in a series 
 corresponding to the unit error k and the precision constant w. 
 Consequently the probability function, 
 
 "***& 
 
 y = we k y 
 corresponding to the same series may be put in the form 
 
 y = ae 2M *. (i) 
 
 If A i, A 2 , . . . AJV are the accidental errors of N direct measure- 
 ments in the same series, the probability P that they all occur in 
 a system characterized by the mean error M is equal to the product 
 of the probabilities for the occurrence of the individual errors in 
 that system. Hence, 
 
 If the individual measurements are represented by a\ t 0,2, 
 . . . a N , and the true numeric of the measured quantity is X, 
 
 Ai = ai - X; A 2 = a z - X\ . . . A# = a N - X, 
 
 and, if x is the arithmetical mean of the measurements, the corre- 
 sponding residuals are 
 
 n = ai x', r z = 2 x; . . . r N = a N x. 
 Consequently, if the error of the arithmetical mean is 5, 
 
 X - x = 5, 
 and 
 
 Ai = n - 5; A 2 = r 2 - 5; . . . A# = r N 8. 
 
 Squaring and adding, 
 
 [A 2 ] = [r 2 ]-25M+ATS 2 ; 
 
 (28) 
 
52 THE THEORY OF MEASUREMENTS [ART. 40 
 
 since [r] Is equal to zero in virtue of equation (14), article thirty- 
 five. When this value of [A 2 ] is substituted in (ii), the resulting 
 value of P is the probability that the arithmetical mean is in 
 error by an amount 6. For, as we have seen in article thirty-five, 
 the minimum value of [r 2 ] occurs when x is taken equal to the 
 arithmetical mean. Consequently, P is a maximum when <5 is 
 equal to zero and decreases in accordance with the probability 
 function as 5 increases either positively or negatively. 
 
 We do not know the exact value of either X or 5; but, if y a is 
 the probability that the error of the arithmetical mean is equal 
 to an arbitrary magnitude 5, the foregoing reasoning leads to the 
 relation 
 
 2M2 
 
 But the arithmetical mean is equivalent to a single measurement 
 in a series of much greater precision than that of the given meas- 
 urements. Hence, if o> a is the precision constant correspond- 
 ing to this hypothetical series and M a is the mean error of the 
 arithmetical mean, we have by analogy with (i) 
 
 a* 
 
 y a = w a e 2 M 2 . (iv) 
 
 Equations (iii) and (iv) are two expressions for the same prob- 
 ability and should give equal values to y a whatever the assumed 
 value of 5. This is possible only when 
 
 2M , 
 
 and 
 
 1 N 
 
 ~ 2M 2 
 Hence, 
 
 M M 
 
 M a = = 
 
 VN 
 
 Consequently, the mean % error of the arithmetical mean is equal 
 to the mean error of a single measurement divided by the square 
 root of the number of measurements. 
 
 Since the average, mean, and probable errors of a single meas- 
 urement are connected by the relations (26), the corresponding 
 
Art. 41] CHARACTERISTIC ERRORS 53 
 
 errors of the arithmetical mean, distinguished by th.e subscript 
 a, are given by the relations 
 
 4 = -4=; M a = -^=; E a = -?j=. (29) 
 
 VN VN VN 
 
 41. Practical Computation of Characteristic* Errors. As 
 
 pointed out in article thirty-seven, the square of the mean error 
 
 [A 2 1 
 M is the limiting value of the ratio ^rp when both members 
 
 become infinite, i.e., when all of the errors of the given system 
 are considered. But the errors of the actual measurements fall 
 into groups, as explained in article thirty-three, and the errors in 
 succeeding groups differ in magnitude by a constant amount k, 
 depending on the nature of the instruments used in making the 
 observations. Consequently, the ordinates, of the probability 
 curve, corresponding to these errors are uniformly distributed 
 along the horizontal axis. Hence, if we include in [A 2 ] only the 
 errors of the actual measurements, the limiting value of the ratio 
 
 fA 2 l 
 
 L -^- when N is indefinitely increased will be nearly the same as if 
 
 all of the errors of the system were included. Since the ratio 
 approaches its limit very rapidly as N increases, the value of M 
 can be determined, with sufficient precision for most practical 
 purposes, from a somewhat limited series of measurements. 
 
 If we knew the true accidental errors, the mean error could be 
 computed at once from the relation 
 
 (v) 
 
 and, since the residuals are nearly equal to the accidental errors 
 when N is very large, an approximate value can be obtained by 
 using the r's in place of the A's. A better approximation can be 
 obtained if we take account of the difference between the A's 
 and the r's. From equation (28) 
 
 [A 2 ] = [r 2 ] + AT5 2 , (vi) 
 
 where 6 is the unknown error of the arithmetical mean. Probably 
 the best approximation we can make to the true value of 8 is to 
 set it equal to the mean error of the arithmetical mean. Hence, 
 from the second of equations (29) 
 
54 THE THEORY OF MEASUREMENTS [ART. 41 
 
 Consequently, (vi) becomes 
 
 NM 2 = [r 2 ] + 
 and we have 
 
 (30) 
 
 Thus the square of the mean error of a single measurement is 
 equal to the sum of the squares of the residuals divided by the 
 number of measurements less one. 
 
 Combining (30) with the third of equations (26), article thirty- 
 nine, we obtain the expression 
 
 E = 0.6745 V^rj < 31 ) 
 
 for the probable error of a single measurement. Hence, by equa- 
 tions (29), the mean error M a and the probable error E a of the 
 arithmetical mean are given by the relations 
 
 and * = - (32) 
 
 When the number of measurements is large, the computation 
 of the probable errors E and E a by the above formulae is some- 
 what tedious, owing to the necessity of finding the" square of 
 each of the residuals. In such cases a sufficiently close approx- 
 imation for practical purposes can be derived from the average 
 error A with the aid of equations (26). The first of these equa- 
 tions may be written in the form 
 
 [A3 = T [A] 2 
 
 N 2 N 2 ' 
 
 If we assume that the distribution of the residuals is the same as 
 that of the true accidental errors, a condition that is accurately 
 fulfilled when N is very large, we can put 
 
 N 
 Consequently, 
 
 
ART. 41] CHARACTERISTIC ERRORS 55 
 
 When the mean error M is expressed in terms of the A's, equation 
 (30) becomes 
 
 [A 2 ]_ M 
 N ' N-l' 
 or 
 
 [Ag = N [Sp. 
 
 [r 2 ] tf- 1 [r]2 ' 
 Consequently 
 
 [A? [r? 
 
 and, since this ratio is equal to A 2 , we have 
 
 == and A = - X (33) 
 
 -1) NVN-1 
 
 Combining this result with the second of equations (26) and the 
 third of (29), we obtain 
 
 E = 0.8453 . ^ ; E a = 0.8453 - ^ . (34) 
 
 VN(N-1)' NVN-1 
 
 The above formulae for computing the characteristic errors from 
 the residuals have been derived on the assumption that the true 
 accidental errors and the residuals follow the same law of dis- 
 tribution. This is strictly true only when the number of measure- 
 ments considered is very large. Yet, for lack of a better method, 
 it is customary to apply the foregoing formulas to the discussion 
 of the errors of limited series of measurements and the results 
 thus obtained are sufficiently accurate for most practical purposes. 
 When the highest attainable precision is sought, the number of 
 observations must be increased to such an extent that the theo- 
 retical conditions are fulfilled. 
 
 The choice between the formulae involving the average error 
 A and those depending on the mean error M is determined largely 
 by the number of measurements available and the amount of 
 time that it is worth while to devote to the computations. When 
 the number of measurements is very large, both sets of formulae 
 lead to the same values for the probable errors E and E a , and 
 much time is saved by employing those depending on A. For 
 limited series of observations a better approximation to the true 
 values of these errors is obtained by employing the formulae in- 
 volving the mean error. In either case the computation may be 
 
56 
 
 THE THEORY OF MEASUREMENTS [ART. 42 
 
 facilitated by the use of Tables XIV and XV at the end of this 
 volume. These tables give the values of the functions 
 
 0.6745 0.8453 0.8453 
 
 0.6745 
 
 VN(N-1)' 
 
 and 
 
 NVN-l' 
 
 corresponding to all integral values of N between two and one 
 hundred. 
 
 42. Numerical Example. The following example, represent- 
 ing a series of observations taken for the purpose of calibrating 
 the screw of a micrometer microscope, will serve to illustrate the 
 practical application of the foregoing methods. Twenty inde- 
 pendent measurements of the normal -distance between two 
 parallel lines, expressed in terms of the divisions of the micrometer 
 head, are given in the first and fourth columns of the following 
 table under a. 
 
 a 
 
 r 
 
 r i 
 
 a 
 
 r 
 
 r 2 
 
 194.7 
 
 +0.53 
 
 0.2809 
 
 194.3 
 
 +0.13 
 
 0.0169 
 
 194.1 
 
 -0.07 
 
 0.0049 
 
 194.3 
 
 +0.13 
 
 0.0169 
 
 194.3 
 
 +0.13 
 
 0.0169 
 
 194.0 
 
 -0.17 
 
 0.0289 
 
 194.0 
 
 -0.17 
 
 0.0289 
 
 194.4 
 
 +0.23 
 
 0.0529 
 
 193.7 
 
 -0.47 
 
 0.2209 
 
 194.5 
 
 +0.33 
 
 0.1089 
 
 194.1 -0.07 
 
 0.0049 
 
 193.8 
 
 -0.37 
 
 0.1369 
 
 193.9 -0.27 
 
 0.0729 
 
 193.9 
 
 -0.27 
 
 0.0729 
 
 194.3 +0.13 
 
 0.0169 
 
 193.9 
 
 -0.27 
 
 0.0729 
 
 194.3 +0.13 
 
 0.0169 
 
 194.8 
 
 +0.63 
 
 0.3969 
 
 194.4 +0.23 
 
 0.0529 
 
 193.7 
 
 -0.47 
 
 0.2209 
 
 
 
 194.17 
 
 5.20 
 
 1.8420 
 
 
 
 .r 
 
 
 
 [r 2 ] 
 
 Since the observations are independent and equally trust- 
 worthy, the most probable value that we can assign to the numeric 
 of the measured magnitude is the arithmetical mean x; and we 
 find that x is equal to 194.17 micrometer divisions. Subtracting 
 194.17 from each of the given observations we obtain the residuals 
 in the columns under r. The algebraic sum of these residuals is 
 equal to zero as it should be, owing to the properties of the arith- 
 metical mean. The sum without regard to sign, [r], is equal to 
 5.20. Squaring each of the residuals gives the numbers in the 
 columns under r 2 and adding these figures gives 1.8920 for the 
 sum of the squares of the residual [r 2 ]. 
 
 Taking N equal to twenty, in formulae (33) and (34), we find 
 the average and probable errors 
 
ART. 42] CHARACTERISTIC ERRORS 57 
 
 = =b 0.267; A a = Ar ^ = 0.0596, 
 
 NVN-l 
 
 E = 0.8453 7== = 0.226; # = 0.8453 ^-^ = = 0.0504, 
 
 where the numerical results are written with the indefinite sign 
 since the corresponding errors are as likely to be positive as nega- 
 tive. 
 
 When formulae (30), (31), and (32) are employed we obtain the 
 mean errors, 
 
 and the probable errors 
 
 E = 0.6745 
 
 The values of the probable errors E and jEk, computed by the 
 two methods, agree as closely as could be expected with so small 
 a number of observations. Probably the values d= 0.210 and 
 0.047, computed from the mean errors M and M a , are the more 
 accurate, but those derived from the average errors A and A a are 
 sufficiently exact for most practical purposes. An inspection of 
 the column of residuals is sufficient to show that eleven of them 
 are numerically greater, and nine are numerically less than either 
 of the computed values of E. Consequently, both of these values 
 fulfill the fundamental definition of the probable error of a single 
 measurement as nearly as we ought to expect when only twenty 
 observations are considered. 
 
 If we use D to represent the measured distance between the 
 parallel lines, in terms of micrometer divisions, we may write 
 the final result of the measurements in the form 
 
 D = 194.170 =t 0.047 mic. div. 
 
 This does not mean that the true value of D lies between the 
 specified limits, but that it is equally likely to lie between these 
 limits or outside of them. Thus, if another and independent 
 series of twenty measurements of the same distance were made 
 
58 THE THEORY OF MEASUREMENTS [ART. 43 
 
 with the same instrument, and with equal care, the chance that 
 the final result would lie between 194.123 and 194.217 is equal to 
 the chance that it would lie outside of these limits. 
 
 Equation (25), article thirty-eight, may be written in the form 
 
 -co 0.4769 
 
 Taking E equal to 0.210, we find that 
 
 v = 2.271 
 k 
 
 for the particular system of errors determined by the above meas- 
 urements. Consequently, the probability for the occurrence of an 
 error less than A in this system is, by equation (13), article thirty- 
 three, 
 
 2.271.A 
 
 and, since there are twenty measurements, we should expect to 
 find 20 PA errors numerically less than any assigned value of A. 
 
 The values of PA, corresponding to various assigned values of 
 A, can be easily computed with the aid of Table XI and applied, 
 as explained in article thirty-four, to compare the theoretical 
 distribution of the accidental errors with that of the residuals 
 given under r in the above table. Such a comparison would have 
 very little significance in the present case, however it resulted, 
 since the number of observations considered is far too small to 
 fulfill the theoretical requirements. But it would show that, 
 even in such extreme cases, the deviations from the law of errors 
 are not greater than might be expected. The actual comparison 
 is left as an exercise for the student. 
 
 43. Rules for the Use of Significant Figures. The funda- 
 mental principles underlying the use of significant figures were 
 explained in article fifteen. General rules for their practical ap- 
 plication may be stated in terms of the probable error as follows: 
 
 All measured quantities should be so expressed that the last 
 recorded significant figure occupies the place corresponding to the 
 second significant figure in the probable error of the quantity 
 considered. 
 
 The number of significant figures carried through the compu- 
 
ART. 43] CHARACTERISTIC ERRORS 59 
 
 tations should be sufficient to give the final result within one unit 
 in the last place retained and no more. 
 
 For practical purposes probable errors should be computed to 
 two significant figures. 
 
 The example given in the preceding article will serve to illus- 
 trate the application of these rules. The second significant figure 
 in the probable error of the arithmetical mean occupies the third 
 decimal place. Consequently, the final result is carried to three 
 decimal places, notwithstanding the fact that the last place is 
 occupied by a zero. It would obviously be useless to carry out 
 the result farther than this, since the probable error shows that 
 the digit in the second decimal place is equally likely to be in 
 error by more or less than .five units. If less significant figures 
 were used, the fifth figure in computed results might be vitiated 
 by more than one unit. 
 
 In order to apply the rules to the individual measurements, it 
 is necessary to make a preliminary series of observations, under 
 as nearly as possible the same conditions that will prevail during 
 the final measurements, and compute the probable error of a 
 single observation from the data thus obtained. Then, if possible, 
 all final measurements should be recorded to the second significant 
 figure in this probable error and no farther. It sometimes happens, 
 as in the above example, that the graduation of the measuring 
 instruments used is not sufficiently fine to permit the attainment 
 of the number of significant figures required by the rule. In such 
 cases the observations are recorded to the last attainable figure, 
 .or, if possible, the instruments are so modified that they give 
 the required number of figures. Thus, in the example cited, the 
 second significant figure in the probable error of a single measure- 
 ment is in the second decimal place, but the micrometer can 
 be read only to one-tenth of a division. Hence the individual 
 measurements are recorded to the first instead of the second 
 decimal place. In this case the accuracy attained in making the 
 settings of the instrument was greater than that attained in 
 making the readings, and an observer, with sufficient experience, 
 would be justified in estimating the fractional parts to the nearest 
 hundredth of a division. A better plan would be to provide the 
 micrometer head with a vernier reading to tenths or hundredths of 
 a division. In the opposite case, when the accuracy of setting is 
 less than the attainable accuracy of reading, it is useless to record 
 
60 THE THEORY OF MEASUREMENTS [ART. 43 
 
 the readings beyond the second significant figure in the probable 
 error of a single observation. 
 
 For the purpose of computing the residuals, the arithmetical 
 mean should be rounded to such an extent that the majority of 
 the residuals will come out with two significant figures. This 
 greatly reduces the labor of the computations and gives the calcu- 
 lated characteristic errors within one unit in the second significant 
 figure. 
 
CHAPTER VI. 
 MEASUREMENTS OF UNEQUAL PRECISION. 
 
 44. Weights of Measurements. In the preceding chapter 
 we have been dealing with measurements of equal precision, and 
 the results obtained have been derived on the supposition that 
 there was no reason to assume that any one of the observations 
 was better than any other. Under these conditions we have 
 seen that the most probable value that we can assign to the 
 numeric of the measured magnitude is the arithmetical mean of 
 the individual observations. Also, if M and E are the mean and 
 probable errors of a single observation, M a and E a the mean and 
 probable errors of the arithmetical mean, and A/" the number of 
 observations, we have the relations 
 
 # = 0.6745 M; ' E a = 0.6745 M n , 
 M E 
 
 v 
 
 (35) 
 
 The true numeric X of the measured magnitude cannot be 
 exactly determined from the given observations, but the final 
 result of the measurements may be expressed in the form 
 
 X = x E a , 
 
 which signifies that X is as likely to lie between the specified 
 limits as outside of them. 
 
 Now suppose that the results of m independent series of meas- 
 urements of the same magnitude, made by the same or different 
 methods, are given in the form 
 
 X = xi E lt 
 X = x% it EZ, 
 
 X = x m d= E m . 
 61 
 
62 THE THEORY OF MEASUREMENTS [ART. 44 
 
 What is the most probable value that can be assigned to X on 
 the basis of these results? Obviously, the arithmetical mean of the 
 x's will not do in this case, unless the E's are all equal, since the 
 x's violate the condition on which the principle of the arithmetical 
 mean is founded. If we knew the individual observations from 
 which each of the x's were derived, and if the probable error of 
 a single observation was the same in each of the series, the most 
 probable value of X would be given by the arithmetical mean of 
 all of the individual observations. Generally we do not have the 
 original observations, and, when we do, it frequently happens that 
 the probable error of a single observation is different in the differ- 
 ent series. Consequently the direct method is seldom applicable. 
 
 The E's may differ on account of differences in the number of 
 observations in the several series, or from the fact that the prob- 
 able error of a single observation is not the same in all of them, or 
 from both of these causes. Whatever the cause of the difference, 
 it is generally necessary to reduce the given results to a series of 
 equivalent observations having the same probable error before 
 taking the mean. For it is obvious that a result showing a small 
 probable error should count for more, or have greater weight, 
 in determining the value of X than one- that corresponds to a 
 large probable error, since the former result has cost more in time 
 and labor than the latter. 
 
 The reduction to equivalent observations having the same 
 probable error is accomplished as follows: m numerical quanti- 
 ties wi, w 2 , . . . w m , called the weights of the quantities Xi, x 2 , 
 . . . x m , are determined by the relations 
 
 E* E a 2 E* 
 
 W ^E?> W *=Ef'> ' ' Wm =E^' (36) 
 
 where E a is an arbitrary quantity, generally so chosen that all 
 of the w's are integers, or may be placed equal to the nearest 
 integer without involving an error of more than one or two units 
 in the second significant figure of any of the E's. In the following 
 pages E 8 will be called the probable error of a standard observa- 
 tion. Obviously, the weight of a standard observation is unity 
 on the arbitrary scale adopted in determining, the w's; for, by 
 equations (36), 
 
ART. 45] MEASUREMENTS OF UNEQUAL PRECISION 63 
 
 Such an observation is not assumed to have occurred in any of 
 the series on which the x's depend, but is arbitrarily chosen as a 
 basis for the computation of the weights of the given results. 
 
 By comparing equations (35) and (36), we see that E\ is equal 
 to the probable error of the arithmetical mean of w\ standard 
 observations. But it is also the probable error of the given 
 result XL Consequently x\ is equivalent to the arithmetical 
 mean of wi standard observations. Similar reasoning can be 
 applied to the other E's and in general we have 
 
 Xi = mean of w\ standard observations, 
 x 2 = mean of w 2 standard observations, 
 
 x m = mean of w m standard observations. 
 
 (i) 
 
 The weights Wi, w 2} . . . w m are numbers that express the rela- 
 tive importance of the given measurements for the determination 
 of the most probable value of the numeric of the measured mag- 
 nitude. Each weight represents the number of hypothetical 
 standard observations that must be combined to give an arith- 
 metical mean with a probable error equal to that of the given 
 measurement. 
 
 45. The General Mean. From equations (i) it is obvious 
 that 
 
 = the sum of Wi standard observations, 
 = the sum of w z standard observations, 
 
 w m x m = the sum of w m standard observations, 
 
 and, consequently, 
 
 -f + w m x m 
 
 is equal to the sum of w\ + ^2 + . . + W TO standard observa- 
 tions. Since the probable error E 8 is common to all of the 
 standard observations, they are equally trustworthy and their 
 arithmetical mean is the most probable value that we can assign 
 to the numeric X on the basis of the given data. Representing 
 this value of X Q we have 
 
 _ WiXi + W 2 X 2 + * + W m X m X Q( _V 
 
 Wl+W2 + . . . + Wm 
 The products W&1, etc., are called weighted observations or meas- 
 
64 THE THEORY OF MEASUREMENTS [ART. 45 
 
 urements, and x is called the general or weighted mean. The 
 weight W Q of X Q is obviously given by the relation 
 
 wo = wi + w 2 + - + w m , (38) 
 
 since X Q is the mean of w standard observations. 
 
 Equation (37) for the general mean can be established inde- 
 pendently from the law of accidental errors in the following manner: 
 Let coi, o> 2 , . . . w m represent the precision constants correspond- 
 ing to the probable errors EI, E z , E m , and let w s be an 
 arbitrary quantity connected with the arbitrary quantity E 8 by 
 the relation 
 
 # 8 = 0.2691 - 
 fc> 
 
 Then, by equations (25) and (36), 
 
 i 2 C0 2 2 CO TO 2 
 
 Wl = ~^> W2 = l^> *- IF- (39) 
 
 If XQ is the most probable value of the numeric X, the residuals 
 corresponding to the given aj's are 
 
 ri = xi XQ', r 2 = x z XQ', . . . r m = x m x . 
 The probability that the true accidental error of x\ is equal to r\ 
 
 s 
 
 in virtue of equations (39). Similarly, if 2/1, 2/2, Vm are the 
 probabilities that r\, r 2 , . . . r m are the true accidental errors of 
 
 x m} 
 
 OJ.2 
 T-TT 
 
 2/2 = co 2 e 
 
 Hence, if P is the probability that all of the r's are simultaneously 
 equal to true accidental errors, we have 
 
 w z 
 
 -Tr-- 
 
 P = (wio> 2 . . . ov)e 
 
 and the most probable value of X is that which renders P a 
 maximum. Obviously, the maximum value of P occurs when 
 
ART. 45] MEASUREMENTS OF UNEQUAL PRECISION 65 
 
 (wirf + w 2 r 2 2 + . . . + w m r m 2 ) is a minimum. Consequently the 
 most probable value X Q is given by the relation 
 
 ^T (wiri 2 + w 2 r 2 2 + + w m r m 2 ) = 0. 
 Substituting the values of the r's and differentiating this becomes 
 
 Wi (Xi XQ) + W 2 (X 2 XQ) + W m (x m XQ) = 0. 
 
 Hence, 
 
 WiXi + W 2 X 2 + + W m X m 
 
 XQ ; : : j 
 
 as given above. 
 
 If we multiply or divide the numerator and denominator of 
 equation (37) by any integral or fractional constant, the value 
 of #o is unaltered. Hence, from (36), it is obvious that we are at 
 liberty to choose any convenient value for E a) whether or not it 
 gives integral values to the w's. Equations (36) also show that 
 the weights of measurements are inversely proportional to the 
 squares of their probable errors and consequently we may take 
 
 #! 2 E? EJ 
 
 w 2 = wi-^-', w 3 = w 1 ^-; . . . w m = wi-^-- (40) 
 
 Etf 1 &m 
 
 Hence, if we choose, we can assign any arbitrary weight to one of 
 the given measurements and compute the weights of the others 
 by equation (40). 
 
 The foregoing methods for computing the weights w\, w 2 , etc., 
 are applicable only when the given measurements x\, x 2 , etc., are 
 entirely free from constant errors and mistakes. When this 
 condition is not fulfilled the method breaks down because the 
 errors of the x's do not follow the law of accidental errors. In 
 such cases it is sometimes possible to assign weights to the given 
 measurements by combining the given probable errors with an 
 estimate of the probable value of the constant errors, based on a 
 thorough study of the methods by which the x's were obtained. 
 Such a procedure is always more or less arbitrary, and requires 
 great care and experience, but when properly applied it leads to a 
 closer approximation to the true numeric of the measured magni- 
 tude than would be obtained by taking the simple arithmetical 
 mean of the x's. Since it involves a knowledge of the laws of 
 propagation of errors and of the methods for estimating the pre- 
 
66 THE THEORY OF MEASUREMENTS [ART. 46 
 
 cision attained in removing constant errors and mistakes, it can- 
 not be fully developed until we take up the study of the under- 
 lying principles. 
 
 46. Probable Error of the General Mean. When the given 
 x's are free from constant errors and the E's are known, the weights 
 of the individual measurements are given by (36), and the weight 
 W of the general mean is given by (38). Consequently, if E is 
 the probable error of the general mean, we have by analogy with 
 equations (36) 
 
 1*0=14 and #0=-- (41) 
 
 If we choose, E may be expressed in terms of any one of the E's 
 in place of E 8 . Thus, let E n and w n be the probable error and 
 the weight of any one of the x's, then by (36) 
 
 E > 
 
 W 
 
 and eliminating E a between this equation and (41) we have 
 
 (42) 
 
 When the weights are assigned by the method outlined in the 
 last paragraph of the preceding article, or when, for any reason, 
 the w's are given but not the E's, (41) and (42) cannot be applied 
 until E a or E n has been derived from the given x's and w's. If 
 the number of given measurements is large, the value of E 8 corre- 
 sponding to the given weights can be computed with sufficient 
 precision by the application of the law of errors as outlined below. 
 If the number of given measurements is small, or if constant 
 errors and mistakes have not been considered in assigning the 
 weights, the following method gives only a rough approximation 
 to the true value of E s , and consequently of E Q) since the condi- 
 tions underlying the law of errors are not strictly fulfilled. It will 
 be readily seen that while E 8 may be arbitrarily assigned for the 
 purpose of computing the weights, when the E's are given, its 
 value is fixed when the weights are given. 
 
 Let xi, z 2 , . . . x m represent the given measurements and 
 Wi, ^ 2 , ... w m , the corresponding weights. Then, if o? 8 repre- 
 
ART. 46] MEASUREMENTS OF UNEQUAL PRECISION 67 
 
 sents the precision constant of a standard observation, and wi 
 that of an observation of weight w\, we have by (39) 
 
 Consequently, if 2/ A is the probability that the error of x i is equal 
 to A, 
 
 and, by equation (11), article thirty-three, the probability that 
 the error of x\ lies between the limits A and A + dA is 
 
 
 Now, WiA 2 is the weigh ted square of the error A, and in the follow- 
 ing pages the product VwA will be called a weighted error. Hence, 
 if we put d = VwjA, and dd = Vw { dA, we have for the probability 
 that the weighted error of Xi lies between the limits 5 and d -\- dd 
 
 Since the same result would have been obtained if we had started 
 with any other one of the x's and w's, it is obvious that this equa- 
 tion expresses the probability that any one of the x's, chosen at 
 random, is affected by a weighted error lying between the limits 
 5 and d + dd. But, if rid is the number of #'s affected by weighted 
 errors lying between these limits, and m is the total number of 
 as's, we have also 
 
 or 
 
 Hence, the sum of the squares of the weighted errors lying between 
 5 and 5 -f- dd is given by the relation 
 
 S2 u s - TO- ,* , 
 = m8 2 -re dS, 
 
 = 
 " m 
 
68 THE THEORY OF MEASUREMENTS [A RT . 46 
 
 and, by the method adopted in articles thirty-six and thirty-seven, 
 we have 
 
 [g] = 2 a), r 
 
 m A: Jo 
 
 where [5 2 ] is supposed to include all possible weighted errors 
 between the limits plus and minus infinity. Introducing the 
 values of the S's in terms of the w's and A's this becomes 
 
 m m 
 
 which is an exact equation only when the number of measure- 
 ments considered is practically infinite. 
 
 If M 8 is the mean error of a standard observation, we have from 
 equation (23) 
 
 Hence, from equation (26) 
 
 . = 0.6745 
 
 Now, we do not know the true value of the A's and the number of 
 given measurements is seldom sufficiently large to fulfill the con- 
 ditions underlying this equation. But we can compute the gen- 
 eral mean X Q and the residuals 
 
 Ti = Xi XQ] r 2 = X 2 XQ] . . . T m = X m X , 
 
 and, by a method exactly analogous to that of article forty-one, 
 it can be shown that the best approximation that we can make is 
 given by the relation 
 
 [wr 2 ] 
 
 m m 1 
 Hence, as a practicable formula for computing E 8 , we have 
 
 E a = 0.6745 V-T' (43) 
 
 ~ m 1 
 
 and consequently E is given by the relation 
 
 Eo = 0.6745V... r ,,' 
 in virtue of equation (41). 
 
ART. 47] MEASUREMENTS OF UNEQUAL PRECISION 69 
 
 When the probable errors of the given measurements are 
 known, and the weights are computed by equation (36), the value 
 of E 8 computed by equation (43) will agree with the value arbi- 
 trarily assigned, for the purpose of determining the w's, provided 
 the x's are sufficiently numerous and free from constant errors 
 and mistakes. The number of measurements considered is 
 seldom sufficient to give exact agreement, but a large difference 
 between the assigned and computed values of E 8 is strong evidence 
 that constant errors have not been removed with sufficient pre- 
 cision. On the other hand, satisfactory agreement may occur 
 when all of the x's are affected by the same constant error. Con- 
 sequently such agreement is not a criterion for the absence of 
 constant errors, but only for their equality in the different meas- 
 urements. 
 
 47. Numerical Example. As an illustration of the applica- 
 tion of the foregoing principles, consider the micrometer measure- 
 ments given under x in the following table. They represent the 
 results of six series of measurements similar to that discussed in 
 article forty-two, the last one being taken directly from that 
 article. The probable errors, computed as in article forty-two, 
 are given under E. They differ partly on account of differences 
 in the number of observations in the several series, and partly 
 from the fact that the individual observations were not of the 
 same precision in all of the series. The squares of the probable 
 errors multiplied by 10 4 are given under E 2 X 10 4 to the nearest 
 digit in the last place retained. It would be useless to carry them 
 out further as the weights are to be computed to only two signifi- 
 cant figures. 
 
 X 
 
 E 
 
 E* X 10* 
 
 w 
 
 ^5? 
 
 w 
 
 194.03 
 
 0.066 
 
 44 
 
 11 
 
 0.066 
 
 193.79 
 
 0.12 
 
 144 
 
 3 
 
 0.127 
 
 194.15 
 
 0.091 
 
 83 
 
 6 
 
 0.090 
 
 193.85 
 
 0.11 
 
 121 
 
 4 
 
 0.110 
 
 194.22 
 
 0.099 
 
 98 
 
 5 
 
 0.098 
 
 194.17 
 
 0.047 
 
 22 
 
 22 
 
 0.047 
 
 Taking E a equal to 0.22 gives E 8 * X 10 4 equal to 484, and by 
 applying equation (36), we obtain the weights given under w to 
 the nearest integer. Inverting the process and computing the 
 
70 
 
 THE THEORY OF MEASUREMENTS [ART. 47 
 
 E's from the assigned w's and E 8 gives the numbers in the last 
 column of the table. Since these numbers agree with the given 
 E's within less than two units in the second significant figure, we 
 may assume that the approximation adopted in computing the 
 w's is justified. If the agreement was less exact and any of the 
 differences exceeded two units in the second significant figure, it 
 would be necessary to compute the w's further, or, better, to adopt 
 a different value for E 8 , such that the agreement would be suffi- 
 cient with integral values of the w's. 
 
 For the purpose of computation, equation (37) may be written 
 in the form 
 
 X Q = C + 
 
 - C) + w, (x 2 - C) + 
 
 W m (X m C) 
 
 where C is any convenient number. In the present case 193 is 
 chosen, and the products w (x 193) are given in the first column 
 of the following table. 
 
 w (x - 193) 
 
 T 
 
 r2 X 10< 
 
 wr* X 10< 
 
 11.33 
 
 -0.065 
 
 42 
 
 462 
 
 2.37 
 
 -0.305 
 
 930 
 
 2790 
 
 6.90 
 
 +0.055 
 
 30 
 
 180 
 
 3.40 
 
 -0.245 
 
 600 
 
 2400 
 
 6.10 
 
 +0.125 
 
 156 
 
 780 
 
 25.74 
 
 +0.075 
 
 56 
 
 1232 
 
 55.84 
 
 
 
 7844 
 
 Substitution in the above equation for the general mean gives 
 
 and this is the most probable value that we can assign to the 
 numeric of the measured magnitude on the basis of the given 
 measurements. 
 
 By equation (38) the weight, w , of the general mean is 51. 
 Hence equation (41) gives 
 
 0.22 
 
 /= 
 
 V51 
 
 0.031 
 
 for the probable error of x . Selecting the first measurement 
 
ART. 47] MEASUREMENTS OF UNEQUAL PRECISION 71 
 
 since its weight corresponds exactly to its probable error, equa- 
 tion (42) gives 
 
 Eo = 0.066 i/ = 0.031. 
 51 
 
 If the second, third, or fifth measurement had been chosen, the 
 results derived by the two formulae would not have been exactly 
 alike; but the differences would amount to only a few units in the 
 second significant figure, and consequently would be of no prac- 
 tical importance. However, it is better to proceed as above and 
 select a measurement whose weight corresponds exactly with its 
 probable error as shown by the fifth column of the first table 
 above. 
 
 The residuals, computed by subtracting x from each of the 
 given measurements, are given under r in the second table; and 
 their squares multiplied by 10 4 are given, to the nearest digit in 
 the last place retained, under r 2 X 10 4 . The last column of the 
 table gives the weighted squares of the residuals multiplied by 
 10 4 . The sum, [wr 2 ], is equal to 0.784. Hence by equation (43) 
 
 E 8 = 0.6745 1/ ' 784 = =t 0.27, 
 o 
 
 and by equation (44) 
 
 JB, = 0.6745 J^- = 0.037. 
 51 X o 
 
 These results agree with the assumed value of E 8 and the pre- 
 viously computed value of E as well as could be expected when 
 so small a number of measurements are considered. Conse- 
 quently we are justified in assuming that the given measurements 
 are either free from constant errors or all affected by the same 
 constant error. 
 
 In practice the second method of computing E Q is seldom used 
 when the probable errors of the given measurements are known, 
 since its value as an indication of the absence of constant errors 
 is not sufficient to warrant the labor involved. When the prob- 
 able errors of the given measurements are not known it is the 
 only available method for computing EQ and it is carried out here 
 for the sake of illustration. 
 
CHAPTER VII. 
 THE METHOD OF LEAST SQUARES. 
 
 48. Fundamental Principles. Let Xi, X 2 , . . . X g , and FI, 
 Y 2 , . . . Y n represent the true numerics of a number of quan- 
 tities expressed in terms of a chosen system of units. Suppose 
 that the quantities represented by the Y's have been directly 
 measured and that we wish to determine the remaining quantities 
 indirectly with the aid of the given relations 
 
 YZ = FZ (Xl, Xz, . . . X q ), 
 
 Y n = F n (Xi,Xz, . . X q ). 
 
 (45) 
 
 The functions FI, F 2 , . . . F n may be alike or different in form 
 and any one of them may or may not contain all of the X's, but 
 the exact form of each of them is supposed to be known. 
 
 If the F's were known and the number of equations were equal 
 to the number of unknowns, the X's could be derived at once 
 by ordinary algebraic methods. The first condition is never ful- 
 filled since direct measurements never give the true value of the 
 numeric of the measured quantity. Let s i; s 2 , . . . s n represent 
 the most probable values that can be assigned to the F's on the 
 basis of the given measurements. If these values are substituted 
 for the F's in (45), the equations will not be exactly fulfilled and 
 consequently the true value of the X's cannot be determined. The 
 differences 
 
 Fi(Xi,X Z) . . . X q )-si = k 
 
 Fz(Xi,Xz, . . . Xq)-s 2 = k 
 
 *, . . . X q )-s n = A n 
 
 (46) 
 
 represent the true accidental errors of the s's. 
 
 Let Xi, Xz, . . . x q represent the most probable values that we 
 can assign to the X's on the basis of the given data. Then, since 
 
 72 
 
ART. 48] THE METHOD OF LEAST SQUARES 73 
 
 the s's bear a similar relation to the Y's } equations (45) may be 
 written in the form 
 
 Fi (Xi, X 2) . . . X q ) = S b 
 
 F 2 (xi, x 2) . . . x q ) = s 2} 
 F n (xi, x 2} . . . x q ) = s n , 
 
 (47) 
 
 where the functions F i} F 2 , etc., have exactly the same form as 
 before. When the number of s's is equal to the number of x's, 
 these equations give an immediate solution of our problem by 
 ordinary algebraic methods; but in such cases we have no data 
 for determining the precision with which the computed results 
 represent the true numerics Xi, X 2) etc. 
 
 Generally the number of s's is far in excess of the number of 
 unknowns and no system of values can be assigned to the x's 
 that will exactly satisfy all of the equations (47). If any assumed 
 values of the x's are substituted in (47), the differences 
 
 ^1 (Xi, X 2) . . . X q ) Si = 7*1, 
 
 F 2 (xi, x 2) . . . x q ) - s 2 = r 2 , 
 
 F n (Xi, X 2 , . . . X q ) - S- n = T n 
 
 represent the residuals corresponding to the given s's. ^Obviously, f 
 the most probable values that we can assign to the x's will be 
 those that give a maximum probability that these residuals are 
 equal to the true accidental errors AI, A 2 , etc. 
 
 If the s's are all of the same weight, the A's all correspond to 
 the same precision constant co. Consequently, as in article thirty- 
 five, the probability that the A's are equal to the r's is 
 
 
 and this is a maximum when 
 
 ri 2 + r 2 2 + . . . + r n 2 = [r 2 ] = a minimum. (49) 
 
 Hence, as in direct measurements, the most probable values that 
 we can assign to the desired numerics are those that render the 
 sum of the squares of the residuals a minimum. For this reason 
 the process of solution is called the method of least squares. 
 
74 THE THEORY OF MEASUREMENTS [ART. 49 
 
 Since the r's are functions of the q unknown quantities x i} x 2) 
 etc., the conditions for a minimum in (49) are 
 
 provided the x's are entirely independent in the mathematical 
 sense, i.e., they are not required to fulfill any rigorous mathe- 
 matical relation such as that which connects the three angles of 
 a triangle. The equations (47) are not such conditions since the 
 functions F i} F 2 , etc., represent measured magnitudes and may 
 take any value depending on the particular values of the x's that 
 obtain at the time of the measurements. When the r's are re- 
 placed by the equivalent expressions in terms of the x's and s's as 
 given in (48), the conditions (50) give q, and only g, equations 
 from which the x's may be uniquely determined. 
 
 If the weights of the s's are different, the A's correspond to 
 different precision constants coi, 0)2, . . . , co n given by the rela- 
 tions 
 
 where w a is the precision constant corresponding to a standard 
 measurement, i.e., a measurement of weight unity; and wi, w 2 , 
 . . . , w n are the weights of the s's. Under these conditions, as 
 in article forty-five, the most probable values of the re's are those 
 that render the sum of the weighted squares of the residuals a 
 minimum. Thus, in the case of measurements of unequal weight, 
 the condition (49) becomes 
 
 wiri 2 f w 2 2 + + MV 2 = [wr 2 ] = a minimum, (51) 
 and conditions (50) become 
 
 A M = ; ^M = 0; ... A M = . (52) 
 
 49. Observation Equations. The equations (50) or (52) can 
 always be solved when all of the functions FI, F 2) . . . F n are 
 linear in form. Many problems arise in practice which do not 
 satisfy this condition and frequently it is impossible or incon- 
 venient to solve the equations in their original form. In such 
 cases, approximate values are assigned to the unknown quantities 
 and then the most probable corrections for the assumed values 
 are computed by the method of least squares. Whatever the form 
 
ART. 50] THE METHOD OF LEAST SQUARES 
 
 75 
 
 of the original functions, the relations between the corrections can 
 always be put in the linear form by a method to be described in a 
 later chapter. 
 
 When the given functions are linear in form, or have been 
 reduced to the linear form by the device mentioned above, equa- 
 tions (47) may be written in the form 
 
 + to + 
 + to + 
 
 + piX q = si, 
 = s 2 , 
 
 p n x q = 
 
 (53) 
 
 where the a's, 6's, etc., represent numerical constants given either 
 by theory or as the result of direct measurements. These equa- 
 tions are sometimes called equations of condition; but in order 
 to distinguish them from the rigorous mathematical conditions, 
 to be treated later, it is better to follow the German practice and 
 call them observation equations, "Beobachtungsgleichungen." 
 
 By comparing equations (47), (48), and (53), it is obvious that 
 the expressions 
 
 + to + CiX 3 + 
 -f to + c 2 x 3 + 
 
 b n x 
 
 c n x 3 
 
 s 2 = r 2 , 
 p n x q - s n = r n 
 
 (54) 
 
 give the resi'duals in terms of the unknown quantities x\, x z , etc., 
 and the measured quantities si, s 2 , etc. 
 
 50. Normal Equations. In the case of measurements of 
 equal weight, we have seen that the most probable values of the 
 unknowns x\, x 2 , etc., are given by the solution of equations (50) 
 provided the x's are independent. Assuming the latter condition 
 and performing the differentiations we obtain the equations 
 
 dr, dr. 
 
 dr 3 
 
 dx t 
 
 (0 
 
76 
 
 THE THEORY OF MEASUREMENTS [ART. 50 
 
 Differentiating equations (54) with respect to the x's gives 
 
 dri _ dr 2 _ 
 
 ~dx\ ~ ai ' dxi ~~ 
 
 
 dx c 
 
 = a n , 
 = b n , 
 
 dr 2 
 
 and hence equations (i) become 
 r 2 a 2 + 
 i + r 2 6 2 + . 
 
 . drn 
 ' dx q 
 
 + r n a n = 0, 
 + r n b n = 0, 
 
 (ii) 
 
 (iii) 
 
 - . . . + r n p n = 0. 
 
 Introducing the expressions for the r's in terms of the x's from 
 equations (54) and putting 
 
 [aa] = didi -{- a 2 a 2 -|- a 3 a 3 ~h ~h d n d n} 
 
 w> 
 
 [as] = diSi + a 2 s 2 + a 3 s 3 + 
 
 [bd] = bidi + 6 2 a 2 + b s d s + 
 [66] = &!&! + 6 2 6 2 + 6 3 6 3 + 
 
 [be] = 6iCi + 6 2 c 2 + 6 3 c 3 + 
 
 a n s n , 
 
 6 n a n = [ab]j 
 
 b n b n , 
 
 6 n c n 
 
 (55) 
 
 equations (iii) reduce to 
 
 [aa] x-i + [ab] x z + [ac] x 3 
 
 [ac] 
 
 [be] x 2 + [cc] x 3 
 
 [bp]x q =[bs], 
 
 [CP] X* = N, 
 
 (56) 
 
 giving us q, so-called, normal equations from which to determine 
 the q unknown x's. 
 
 Since the normal equations are linear in form and contain only 
 numerical coefficients and absolute terms, they can always be 
 solved, by any convenient algebraic method, provided they are 
 entirely independent, i.e., provided no one of them can be ob- 
 tained by multiplying any other one by a constant numerical 
 
ART. 50] THE METHOD OF LEAST SQUARES 77 
 
 factor. This condition, when strictly applied, is seldom violated 
 in practice; but it occasionally happens that one of the equations 
 is so nearly a multiple or submultiple of another that an exact 
 solution becomes difficult if not impossible. In such cases the 
 number of observation equations may be increased by making 
 additional measurements on quantities that can be represented 
 by known functions of the desired unknowns. The conditions 
 under which these measurements are made can generally be so 
 chosen that the new set of normal equations, derived from all of 
 the observation equations now available, will be so distinctly 
 independent that the solution can be carried out without difficulty 
 to the required degree of precision. 
 
 By comparing equations (53) and (56), it is obvious that the 
 normal equations may be derived in the following simple manner. 
 Multiply each of the observation equations (53) by the coefficient 
 of xi in that equation and add the products. The result is the 
 first normal equation. In general, q being any integer, multiply 
 each of the observation equations by the coefficient of x q in that 
 equation and add the products. The result is the gth normal 
 equation. The form of equations (56) may be easily fixed in 
 mind by noting the peculiar symmetry of the coefficients. Those 
 in the principal diagonal from left to right are [aa], [66], [cc], etc., 
 and coefficients situated symmetrically above and below this 
 diagonal are equal. 
 
 When the given measurements are not of equal weight, the 
 observation equations (53), and the residual equations (54) remain 
 unaltered, but the normal equations must be derived from (52) 
 in place of (50). Since the weights Wi, w 2 , etc., are independent 
 of the x's, if we treat equations (52) in the same manner that we 
 have treated (50), we shall obtain the equations 
 
 * + w n r n a n = 0, 
 '. .4 Wn&n = 0, 
 
 (iv) 
 
 + Wtfzpz + ' ' ' + W n r n p n = 0, 
 
 in place of equations (iii). Hence, if we put 
 
 [iWia] = Wididi -f~ WzClzCLz ~\~ ' ' ' ~\~ W n d n CLnj 
 
 (57) 
 
 [was] = WidiSi + w&zSz + + w n a n s nj 
 
 ' -\-WnpnPn, 
 
78 
 
 THE THEORY OF MEASUREMENTS [ART. 51 
 
 the normal equations become 
 [waa] xi + [wab] x 2 + [wac] z 3 
 [wab] Xi + [wbb] x z + [wbc] x z 
 [wac] X! + [wbc] x 2 + [wcc] x z 
 
 + [wap] x q = [was], 
 + [wbp] x q = [wbs], 
 + [wcp] x q = [wcs], 
 
 (58) 
 
 [wap]xi + [wbp]x 2 + [wcp]x$ + + [wpp]x q = [wps]. 
 
 These equations are identical in form with equations (56), and 
 they may be solved under the same conditions and by the same 
 methods as those equations. Consequently, in treating methods 
 of solution, we shall consider the measurements to be of equal 
 weight and utilize equations (56). All of these methods may be 
 readily adapted to measurements of unequal weight by substitut- 
 ing the coefficients as given in (57) for those given in (55). 
 
 51. Solution with Two Independent Variables. When only 
 two independent quantities are to be determined the observation 
 equations (53) become 
 
 " 
 
 = s, 
 and the normal equations (56) reduce to 
 
 [aa] Xi + [ab] x 2 = [as], 
 
 [ab] X! + [bb] x 2 = [bs]. 
 Solving these equations we obtain 
 
 [bb] [as] - [ab] [bs] 
 
 [aa] [bb] - [ab] 2 
 _ [aa] [bs] [ab] [as] 
 
 [aa] [bb] - [ab] 2 
 
 As an illustration, consider the determination of the length Z/ 
 at C., and the coefficient of linear expansion a of a metallic 
 bar from the following measurements of its length L t at temper- 
 ature t C. 
 
 (56a) 
 
 (59) 
 
 t 
 
 L t 
 
 C. 
 
 20 
 
 mm. 
 1000.36 
 
 30 
 
 1000.53 
 
 40 
 
 1000.74 
 
 50 
 
 1000.91 
 
 60 
 
 1001.06 
 
Ara.51] THE METHOD OF LEAST SQUARES 
 
 79 
 
 or 
 
 Within the temperature range considered, L t and t are connected 
 with LO and a by the relation 
 
 L t = Lo (1 + at), 
 
 L t = Lo + L at, (v) 
 
 and a set of observation equations might be written out at once 
 by substituting the observed values of L t and t in this equation. 
 But the formation of the normal equations and the final solution 
 is much simplified when the coefficients and absolute terms in the 
 observation equations are small numbers of nearly the same order 
 of magnitude. To accomplish this simplification, the above func- 
 tional relation may be written in the equivalent form 
 
 and if we put 
 
 it becomes 
 
 L t - 1000 = Lo - 1000 + WL<xx 
 
 L t - 1000 = s; JQ = 6, 
 
 LO 1000 = Xi] 10 LOCK = Xz, 
 Xi -J- 6^2 = s. 
 
 (vi) 
 
 Using this function, all of the a's in equation (53a) become equal 
 to unity and the 6's and s's may be computed from the given 
 
 observations by equations (vi). 
 
 the observation equations are 
 
 xi + 2 z 2 = 
 xi + 3 x 2 = 
 
 Hence, in the present case, 
 
 .36, 
 .53, 
 
 x l +x 2 = .74, 
 
 zi + 5z 2 = .91, 
 
 Xl + 6x 2 = 1.06. 
 
 For the purpose of forming the normal equations, the squares 
 and products of the coefficients and absolute terms are tabulated 
 as follows : 
 
 Obs. 
 
 aa 
 
 ab 
 
 as 
 
 bb 
 
 bs 
 
 1 
 
 
 2 
 
 0.36 
 
 4 
 
 0.72 
 
 2 
 
 
 3 
 
 0.53 
 
 9 
 
 1.59 
 
 3 
 
 
 4 
 
 0.74 
 
 16 
 
 2.96 
 
 4 
 
 
 5 
 
 0.91 
 
 25 
 
 4.55 
 
 5 
 
 
 6 
 
 1.06 
 
 36 
 
 6.36 
 
 
 5 
 
 20 
 
 3.60 
 
 90 
 
 16.18 
 
 
 [aa] 
 
 W 
 
 [as] 
 
 [bb] 
 
 [bs] 
 
 Substituting these values of the coefficients in (56a) gives the 
 normal equations 
 
80 
 
 THE THEORY OF MEASUREMENTS [ART. 51 
 
 = 3.60, 
 = 16.18, 
 
 and by (59) we have 
 
 _ 90 X 3.60 - 20 X 16.18 
 
 5 X 90 - 400 
 5 X 16.18 - 20 X 3.60 
 
 = 0.008, 
 = 0.178. 
 
 5 X 90 - 400 
 
 From these results, with the aid of relations (vi), we find 
 Lo = xi + 1000 = 1000.008, 
 
 L a = ^ = 0.0178, 
 
 0.0178 
 
 = 0.0000178, 
 
 and finally 
 
 L t = 1000.008 (1 + 0.0000178 1) millimeters. (vii) 
 
 The differences between the values of L t computed by equation 
 (vii), and the observed values give the residuals. But they can 
 be more simply determined by using the above values of x\ 
 and Xz in the observation equations and taking the difference 
 between the computed and observed values of s. Thus, if s' 
 represents the computed value and r the corresponding residual 
 
 s' = 0.008 + 0.178 6, 
 and r = s f s. 
 
 With the values of s and 6 used hi the observation equations we 
 obtain the residuals as tabulated below: 
 
 s' 
 
 8 
 
 r 
 
 7-2 X 10* 
 
 0.364 
 0.542 
 0.720 
 0.898 
 1.076 
 
 0.36 
 0.53 
 0.74 
 0.91 
 1.06 
 
 +0.004 
 +0.012 
 -0.020 
 -0.012 
 +0.016 
 
 0.16 
 1.44 
 4.00 
 1.44 
 2.56 
 
 
 [r 2 ] = 9.60XlO~ 4 
 
 Since the above values of x\ and x 2 were computed by the method 
 of least squares, the resulting value of [r 2 ], i.e., .000960, should be 
 less than that obtainable with any other values of x\ and x%. 
 That this is actually the case may be verified by carrying out the 
 computation with any other values of x\ and x z . 
 
ART. 52] THE METHOD OF LEAST SQUARES 
 
 81 
 
 52. Adjustment of the Angles About a Point. As an illus- 
 tration of the application of the method of least squares to the 
 solution of a problem involving more than two unknown quanti- 
 ties, suppose that we wish to determine the most probable value 
 of the angles AI, AZ, and A 3 , Fig. 9, from a series of independent 
 measurements of equal weight on the angles Mi, M 2 , . . . M 6 . 
 If the given measurements were all exact, the equations 
 
 AI = Mi; AZ = M 2 ; A 3 = M 3 ; 
 
 AI-\- AZ = M 4 ; AI + AZ -{-As = MS; and Az -\- As = Me, 
 
 would all be fulfilled identically. In practice this is never the 
 case and it becomes 
 necessary to adjust the 
 values of the A's so that 
 the sum of the squares 
 of the discrepancies will 
 be a minimum. The 
 adjustment may be ef- 
 fected by adopting the 
 above equations as ob- 
 servation equations and 
 proceeding at once to 
 the solution for the A's 
 by the method of least 
 squares. But the ob- 
 served values of the M's 
 usually involve so many 
 significant figures that 
 the computation would 
 be tedious. It is better 
 to adopt approximate 
 values for the A's and then compute the necessary corrections by 
 the method of least squares. 
 
 For this purpose, suppose we adopt MI, M 2 , and M 3 as approxi- 
 mate values of A\, A 2 , and A s respectively and let xi, Xz, and x 3 
 represent the corrections that must be applied to the M's in order 
 to give the most probable values of the A's. Then, putting 
 
 AI = MI + xi, AZ = MZ + Xz, and A 3 = M 3 + # 3 , (viii) 
 the above equations become 
 
 FIG. 9. 
 
82 
 
 THE THEORY OF MEASUREMENTS [ART. 52 
 
 + x 2 
 
 = 0, 
 = 0, 
 = 0, 
 = M 4 - (Af ! + M 2 ), 
 
 To render the problem definite, suppose that the following 
 values of the M's have been determined with an instrument read- 
 ing to minutes of arc by verniers: 
 
 Mi = 10 49'.5, M 4 = 45 24'.0, 
 
 M 2 = 34 36'.0, M 6 = 60 53'.5, 
 
 M 3 = 15 25'.5, M 6 = 50 O'.O. 
 
 Substituting these values in the above equations we obtain 
 xi = 0, 
 
 x 2 = 0, 
 
 2'.5, 
 
 Adopting these as our observation equations and comparing with 
 (53) we obtain the coefficients and absolute terms tabulated below: 
 
 Oba. 
 
 a 
 
 b 
 
 c 
 
 s 
 
 1 
 
 1 
 
 
 
 
 
 
 
 2 
 
 
 
 1 
 
 
 
 
 
 3 
 
 
 
 
 
 1 
 
 
 
 4 
 
 1 
 
 1 
 
 
 
 -1.5 
 
 5 
 
 1 
 
 1 
 
 1 
 
 2.5 
 
 6 
 
 
 
 1 
 
 1 
 
 -1.5 
 
 The squares and products of the coefficients and absolute terms 
 may be tabulated, for the purpose of forming the normal equations, 
 as follows : 
 
 M 
 
 ab 
 
 ac 
 
 as 
 
 66 
 
 be 
 
 bs 
 
 cc 
 
 cs 
 
 1 
 
 
 
 
 1 
 1 
 
 
 
 
 
 
 
 1 
 1 
 
 
 
 
 
 
 
 
 1 
 
 
 
 
 
 
 -1.5 
 2.5 
 
 
 
 
 
 1 
 1 
 1 
 
 
 
 
 
 
 1 
 1 
 
 2 
 
 [be] 
 
 
 
 
 -1.5 
 2.5 
 -1.5 
 
 
 
 
 1 
 
 
 
 1 
 
 1 
 
 
 
 
 
 2.5 
 -1.5 
 
 3 
 
 [aa] 
 
 [ab] 
 
 1 
 
 [ac] 
 
 1 
 
 fas] 
 
 4 
 [66] 
 
 -0.5 
 
 [6s] 
 
 3 
 
 [cc] 
 
 1 
 
 [cs] 
 
ART. 53] THE METHOD OF LEAST SQUARES 83 
 
 Substituting these values in (56) the three normal equations 
 become 
 
 -0.5, 
 1 xi -f 2 x 2 + 3 z 3 = 1, 
 
 and solution by any method gives 
 
 xi = 0.625; x 2 = - 0.75; x 3 = 0.625. 
 
 With these results together with the given values of MI, Mz, 
 and M 3 we obtain from equations (viii) 
 
 A! = 10 50M25, 
 A 2 = 34 35'.25, 
 A 3 = 15 26M25. 
 
 In a problem so simple as the present the normal equations are 
 generally written out at once from the observation equations by 
 the rule stated in article fifty, without taking the space and time 
 to tabulate the coefficients, etc. But, until the student is thor- 
 oughly familiar with the process, it is well to form the tables as 
 a check on the computations and to make sure that none of the 
 coefficients or absolute terms have been omitted. For this reason 
 the tabulation has been given in full above and the student is 
 advised to carry out the formation of the normal equations by 
 the shorter method as an exercise. 
 
 53. Computation Checks. When the number of unknowns 
 is greater than two and a large number of observation equations 
 are given with coefficients and absolute terms involving more than 
 two significant figures, the formation of the normal equations is 
 the most tedious and laborious part of the computations. It is, 
 therefore, advantageous to devise a means of checking the com- 
 puted coefficients and absolute terms in the normal equations 
 before we proceed to the final solution. 
 
 For this purpose compute the n quantities t\ t ^2, ... t n by the 
 equations 
 
 ai + &i -f ci + - + pi = ti,~ 
 
 02 + &2 4- c 2 -f + pz = h, 
 
 On + & + C n -f - + p n = 
 
 (60) 
 
84 THE THEORY OF MEASUREMENTS [ART. 54 
 
 where the a's, b's, etc., are the coefficients in the given observa- 
 tion equations. Multiply the first of equations (60) by Si, the 
 second by s 2 , etc., and add the products. The result is 
 
 [as] + [bs] + [cs] + + \ps] = [ts]. (61) 
 
 In the same way, multiplying by the a's in order and adding, then 
 by the b's in order and adding, etc., we obtain the following rela- 
 tions 
 
 [aa] + [db] + [ac] + .-. + [ap] = [at], 
 
 [ab] + [bb] + [be] + + [bp] = [bt], 
 
 [ac] + [be] + [cc] + + [cp] = [ct], (62) 
 
 [ap] + \bp] + [ep] + . . . + \pp] = \pt]. 
 
 If the absolute terms in the normal equations have been accu- 
 rately computed, equation (61) reduces to an identity. If the 
 coefficients have been accurately computed equations (62) all 
 become identities. Hence (61) is a check on the computation of 
 the absolute terms and equations (62) bear the same relation to 
 the coefficients. The extra labor involved in computing the quan- 
 tities [ts] t [at], . . . , [pt] is more than repaid by the added confi- 
 dence in the accuracy of the normal equations. 
 
 When all attainable significant figures are retained throughout 
 the computations, the checks (61) and (62) should be identities. 
 In practice the accuracy of the measurements is seldom sufficient 
 to warrant so extensive a use of figures, and, consequently, the 
 squares and products, aa, ab, . . . as, at, etc., are rounded to such 
 an extent that the computed values of the x's will come out with 
 about the same number of significant figures as the given data. 
 Judgment and experience are necessary in determining the number 
 of significant figures that should be retained in any particular 
 problem and it would be difficult to state a general rule that 
 would not meet with many exceptions. When the computed 
 coefficients and absolute terms are rounded, as above, the checks 
 may not come out absolute identities, but they should not be 
 accepted as satisfactory when the discrepancy is more than two 
 units in the last place retained. 
 
 54. Gauss's Method of Solution. When the normal equa- 
 tions (56) are entirely independent, they may be solved by any 
 of the well-known methods for the solution of simultaneous 
 linear equations and lead to unique values of the unknown quan- 
 
ART. 54] THE METHOD OF LEAST SQUARES 85 
 
 titles xi, x 2) etc. Gauss's method of substitution is frequently 
 adopted for this purpose since it permits the computation to be 
 carried out in symmetrical form and provides numerous checks 
 on the accuracy of the numerical work. The general principles 
 of the method will be illustrated and explained by completely 
 working out a case in which there are only three unknowns. 
 Since the process of solution is entirely symmetrical, it can be 
 easily extended for the determination of a larger number of 
 unknowns, but too much space would be required to carry through 
 the more general case here. 
 
 When only three unknowns are involved, the normal equations 
 (56) and the check equations (60) and (61) may be completely 
 written out in the following form, the computed quantities and 
 equations being placed at the left, and the checks at the right. 
 
 [aa] xi + [ab] x 2 + [ac] x 3 = [as]. [aa] + [ab] + [ac] = [at]. 
 
 [ab] xi + [bb] x 2 + [be] x 3 = [bs]. [ab] + [bb] + [be] = [bt]. 
 
 [ac] xi + [be] x 2 + [cc] x 3 = [cs]. [ac] + [be] + [cc] = [ct]. 
 
 [as] + [bs]+[cs] =[st].\ 
 
 Solve the first equation on the left for xi y giving 
 [as] [ab] [ac] 
 
 Xi = 7 7 f 1 X 2 f 1 X$. 
 
 [aa] [aa] [aa\ 
 Compute the following auxiliary quantities: 
 
 (63) 
 
 [56] _ P4 [ &] = [bb 1], [bt] - pi M = [^ ' 1L 
 
 L aa] 
 
 [a61 
 
 [6c]- 
 
 M L " M 
 
 ~ M = [6s ' 1] ' M ~ N1 = [st 
 
 As a check on these computations we notice that 
 
 [bb 1] + [be - 1] = [bb] + [be] - |^| ([ab] + [c]), 
 
 [aaj 
 
 = [bt] - lab] - ([at] - [aa]), 
 
86 THE THEORY OF MEASUREMENTS [ART. 54 
 
 In a similar way we may show that we should have 
 
 [6c-l] + [cc-l] = [cM] and [6s- 1] + [cs- 1] = [st- 1]. 
 
 Substituting (64) in the last two of (63) and placing the above 
 checks to the right, we have the equations 
 [bb -I]x 2 + [be 1] x s = [bs 1], [bb 1] + [be- 1] = [fa 1], 
 [be -I]x 2 + [cc 1] z 3 = [cs 1], [be 1] + [cc- 1] = [ct 1], (65) 
 
 [6s-l] + [cs.l] = [s*. 1],. 
 
 which show the same type of symmetry as (63), but contain only 
 two unknown quantities. Solve the first of (65) for x 2 giving 
 
 __ 
 * 2 ~[6&.l] [bb-lf 3 ' 
 
 and compute the following auxiliaries: 
 
 [<*!] - [l^jlfc-1] = [-2], [cM] - l ~^ } [bt. 1} = let- 2], 
 
 (cs 1} - |^jj [bs 1} - [cs 2], [st 1] - |^|j (bt 1] = [* 2}. 
 
 By a method similar to that used above we can show that we 
 should have 
 
 [cc 2] = [ct 2] and [cs 2] = [st 2]. 
 
 Hence, substituting (66) in the last of (65), we have 
 [cc 2] x 3 = [cs 2], [cc 2] = [ct 2], 
 [cs.2] = N-2], 
 and consequently 
 
 [cs 2] _. 
 
 *"fc^t' (67) 
 
 Having determined the value of x 3 from (67), x% may be cal- 
 culated from (66), and then Xi from (64). 
 
 A very rigorous check on the entire computation is obtained as 
 follows: using the computed values of Xi, x z , and z 3 in equations 
 (54), derive the residuals 
 
 (68) 
 
 - s 2 , 
 
 T n = d n Xi ~|- O n X 2 ~\- C n Xs S n , 
 
 and then form the sums 
 
 [rr] = n 2 + r 2 2 + r 3 2 + - - - + r n 2 , 
 
 [SS] = Si 2 + S 2 2 + S 3 2 + + n 2 . 
 
ART. 55] THE METHOD OF LEAST SQUARES 87 
 
 If the computations are all correct, the computed quantities will 
 satisfy the relation 
 
 W = M-[a S ]-M[6 S .l]- [cs . 2] . (69) 
 
 To prove this, multiply the first of (68) by ri, the second by r%, 
 etc., and add the products. The result is 
 
 [rr] = [ar] Xi + [br] x 2 -f [cr] 3 - [sr]. 
 But from equations (iii), article fifty, 
 
 [ ar ] = [br] = [cr] = 0, 
 consequently 
 
 [rr]=- []. (70) 
 
 Multiply each of equations (68) by its s; add, taking account of 
 (70), and we obtain 
 
 [rr] = [ss] - [as] Xi - [6s] x z - [cs] x z . 
 
 Eliminating x\, X 2 , and z 3 , in succession with the aid of (64), (66), 
 and (67) we find 
 
 [rr] = [ss] - [as] - [6s 1] x 2 - [cs 1] x 9 , 
 
 and finally 
 
 r i r i l as ] r i [& s * 1] n n t cs ' 2] r Ol 
 
 [rr] = M ~ y M - I667i] [6s ' 1] - RT2] [cs ' 2] ' 
 
 which is identical with (69). 
 
 55. Numerical Illustration of Gauss's Method. The fore- 
 going methods are most frequently used for the adjustment of 
 astronomical and geodetic observations, and their application to 
 particular problems is fully discussed in practical treatises on 
 such observations. The physical problems, to which they are 
 applicable, usually involve the determination of an empirical 
 relation between mutually varying quantities. Such problems 
 will be discussed at some length in Chapter XIII, and the corre- 
 sponding observation equations will be developed. 
 
 It would require too much space to carry out the complete dis- 
 cussion of such a problem, in this place, with all of the observa- 
 tions made in any actual investigation. But, for the purpose of 
 illustration, the most probable values of xi, Xz, and x 3 will be 
 
88 
 
 THE THEORY OF MEASUREMENTS [ART. 55 
 
 derived, from the following typical observation equations, by 
 Gauss's method of solution: 
 
 + 2x 2 + 0.4z 3 = 
 + 4x 2 + 1.6x3 = 
 + 6 x 2 + 3.6 z 8 = 
 + 8x 2 + 6.4x 3 = 
 +10x 2 +10.0^3 = 
 
 0.24, 
 
 - 1.18, 
 
 - 1.53, 
 
 - 0.69, 
 1.20, 
 4.27. 
 
 Since the coefficient of xi is unity in each of these equations, 
 the products aa, ab, aCj as, and at are equal to a, 6, c, s, and t, 
 respectively. Consequently the first five columns of the follow- 
 ing table show the coefficients, absolute terms, and check terms 
 (t = a + b + c) of the observation equations as well as the 
 squares and products indicated at the head of the columns. The 
 sums [aa], [ab], etc., are given at the foot of the columns and the 
 checks, by equations (61) and (62), are given below the tables. 
 In the present case, the coefficients are expressed by so few signifi- 
 cant figures that it is not necessary to round the computed products 
 and consequently the checks come out identities. 
 
 aa 
 
 ab 
 
 ac 
 
 as 
 
 at 
 
 bb 
 
 be 
 
 
 
 2 
 4 
 6 
 8 
 10 
 
 0.0 
 0.4 
 1.6 
 3.6 
 6.4 
 10.0 
 
 0.24 
 -1.18 
 -1.53 
 -0.69 
 1.20 
 4.27 
 
 1.0 
 3.4 
 6.6 
 10.6 
 15.4 
 21.0 
 
 
 4 
 16 
 36 
 64 
 100 
 
 0.0 
 0.8 
 6.4 
 21.6 
 51.2 
 100.0 
 
 6 
 
 M 
 
 30 
 [ab] 
 
 22.0 
 
 M 
 
 2.31 
 [as] 
 
 58.0 
 
 M 
 
 220 
 
 m 
 
 180.0 
 [be] 
 
 
 
 Check: [ 
 
 aa] + [ab] + [c 
 
 ic] = 58.0. 
 
 
 
 bs 
 
 cc 
 
 cs 
 
 bl 
 
 ct 
 
 st 
 
 0.00 
 -2.36 
 -6.12 
 -4.14 
 9.60 
 42.70 
 
 0.00 
 0.16 
 2.56 
 12.96 
 40.96 
 100.00 
 
 0.00 
 -0.472 
 -2.448 
 -2.484 
 7.680 
 42.700 
 
 0.0 
 6.8 
 26.4 
 63.6 
 123.2 
 210.0 
 
 0.00 
 1.36 
 10.56 
 38.16 
 98.56 
 210.00 
 
 0.24 
 - 4.012 
 -10.098 
 - 7.314 
 
 18.480 
 89.670 
 
 39.68 
 [bs] 
 
 156.64 
 [cc] 
 
 44.976 
 [cs] 
 
 430.0 
 
 M 
 
 358.64 
 [ct] 
 
 86.966 
 [st] 
 
 Checks: [ab] + [66] + [be] = 430.0 
 [ac] + [be] + [cc] = 358.64 
 M + [6s] + [cs]= 86.966 
 
ART. 55] THE METHOD OF LEAST SQUARES 
 
 89 
 
 The normal equations and their checks might now be written 
 out in the form of equations (63), but, since the coefficients and 
 other data necessary for their solution are all tabulated above, it 
 is scarcely worth while to repeat the same data in the form of 
 equations. The computation of the auxiliaries [bb 1], [be 1], 
 etc., and the final solution for x i} x 2) and # 3 by logarithms is best 
 carried out in tabular form as illustrated on pages 90 and 91. 
 The meaning of the various quantities appearing in these tables, and 
 the methods by which they are computed, will be readily under- 
 stood by comparing the numerical process with the literal equa- 
 tions of the preceding article. When the letter n appears after a 
 logarithm it indicates that the corresponding number is to be taken 
 negative in all computations. 
 
 The computation of the residuals by equations (68) and the 
 final check by (69) is carried out in the following table, where 
 Scale, is written for the value of the expression axi + bx 2 + cxs, 
 when the computed values of x\, x 2 , and x 3 are used and s bs. is 
 the corresponding value of s in the observation equations. Thus 
 
 +- 
 
 Si = Si calc. ~ Si obs.- 
 
 I* 
 
 S ObB. 
 
 r 
 
 *Xio. 
 
 ss 
 
 0.245 
 -1.195 
 -1.512 
 -0.709 
 1.215 
 4.264 
 
 0.24 
 -1.18 
 -1.53 
 -0.69 
 1.20 
 4.27 
 
 +0.005 
 -0.015 
 +0.018 
 -0.019 
 +0.015 
 -0.006 
 
 25 
 225 
 324 
 361 
 225 
 36 
 
 0.0576 
 1.3924 
 2.3409 
 0.4761 
 1.4400 
 18.2329 
 
 
 
 
 .001196 
 [rr] 
 
 23.9399 
 
 [as] r , [6s 1] , .,, [cs 2] 
 
 :s-2] 
 
 52 = 23.9387 
 0.0012 
 
 [aa\ [oo i\ [cc A\ 
 0.8893 + 11.3042 + 11.74 
 Final check by (69): [rr] 
 
 Since the checks are all satisfactory, we are justified in assum- 
 ing that the computations are correct. Hence the most probable 
 values of the unknowns, derivable from the given observation 
 equations, are 
 
 xi = 0.245; x 2 = - 1.0003; z 3 = 1.4022, 
 
90 
 
 THE THEORY OF MEASUREMENTS [ART. 55 
 
 jfl 
 
 3 
 
 r [ 
 
 is 
 
 co ^^ 
 
 O 
 
 , ^ 
 
 _|- Oi 
 
 1 ^ 
 
 g 
 
 rH 
 
 - , IQ 
 
 r^^ 
 
 
 4 
 
 1 
 
 i 
 
 CO 
 
 00 ^ 
 
 1-H 
 
 o c^ o o 
 
 HII II II 
 
 t^ CO 
 O CD CO 
 ^^ t^ CD 1>- 
 
 GO CO <M O 
 HII II II 
 
 H" 
 
 "e "e 
 
 II II II 
 
 ^r " ^^ 
 
 tj ~Q O r ~ l 
 
 " "e o ^ 
 
 I 
 
 '|^ 
 
 be * 
 
 bfi ^ 
 
 bC ^ 
 
 c\j d II-H 
 
 OO 
 
 OO <N >O CO 
 CD CD O i i 
 
 Oi i i 00 
 CO rH rH (M 
 
 II II II II 
 
 05 ""ST 5P ^H 1 
 o o -2 . 
 
 CD GO O CD 
 
 
 b bfi 
 
 -2 -2 
 
 
 -STe-i 3 
 1111 i - 
 
 I 
 
 
 CO 1C 
 <M i i O 
 
 0^00 
 
 dodo 
 S <N ^^ 
 
 CD CD CD t^* 
 
 CD Oi 
 O O >O 
 
 r-< i 1 OO !> 
 
 
 II II II 
 
 II H II II 
 
 II II II II 
 
 
 o W 53 
 
 5" o^ 
 
 0^ 
 
 
 bfi M 
 
 mi^ 
 
 ffi 1 
 
 
 
 M 
 
 bD * 
 
 
 00 
 
 s^ 2 
 
 a! se 
 
 
 
 H H H 
 
 ii H H n 
 
 
 
 11111 
 
 1 1 1? 
 
 
 
 -2 J2 
 
 ^l^ 
 
 
 
 1 
 
 
 
 
 co d 
 
 
 
 
 II II 
 
 
 
 
 11 
 
 
 
 
 bO 
 
 
 
 
ART. 55] THE METHOD OF LEAST SQUARES 91 
 
 03 
 
 
 
 
 M 
 
 <N CY^ 
 
 <M CO 
 
 
 (0 
 
 O Oi 
 
 Ji 
 
 3 ^ 
 
 
 6 
 
 10 
 
 " 
 
 
 CO 
 
 CO 
 
 co 
 
 CO i ( CO 
 CO CO CO 
 
 CO 1 CO 
 
 
 
 OS T-( OS 
 
 CO ^ ^ CO 
 
 
 o c^i 
 
 o d o 10 
 
 ^ T-H CO OO 
 
 CO 10 
 
 
 II II 
 
 II II II II 
 
 II II II II 
 
 8 
 
 1 1 T 1 
 
 
 
 | i fffl 
 
 ifL rSl 
 
 "o S -^ "S 
 
 "S S - "5 
 
 ^ ^ 8 8 
 
 
 ' ' '~~ ' c 
 
 
 d d H r-i d 
 
 bfi 
 
 ^ ^r 03 
 
 ' 1- _ ( ' ^ ( Q^ 
 
 1 
 
 j3 
 
 
 B . 
 
 
 
 ^ s 
 
 CO rO 
 
 II II II II || 
 
 
 I 
 
 M 
 
 r-O -O *T? O 
 
 b- b- 
 
 l>. 
 
 ^ (N 
 
 1 " ' co Uo C3 ^ 
 
 CO 
 
 co OS CO 
 
 tO "^t 1 T 1 CO 
 
 ill 
 
 jjO o3 
 
 bfi 
 ^O 
 
 ?7 7 
 
 CO T-I OO 00 
 CO <M 
 
 II II II II 
 
 d d i-i 
 II II II 
 
 
 T- ' 1 ' T-H T-l 
 
 TH TH ^ (M 
 
 <M H H 
 
 
 =0 =0 oo -O 
 *O "O ^j ^^ 
 
 n i 2 
 
 
 
 bfi bfi 
 
 ' T ' c3 
 
 bfi 
 
 
 -2 
 
 TH II 
 
 O 
 
 
 
 -is 
 
 
 
 
 bfi 
 
 
 
 O Q 
 
 g 1 g 
 
 1 
 
 
 00 8 
 
 OS OO O OS 
 
 : 
 
 
 d- 
 
 iO ii O IO 
 
 d 
 
 
 II II II 
 
 II II II II 
 
 II 
 
 
 T i T i ^n" 1 i i 
 
 ^ ^ ^o ^ 
 
 cT 
 
 e 
 
 o co ^ *o 
 
 O t ^ -O -O 
 
 ^J o 1 o 
 
 O "O *^ O 
 
 8 
 
 | 2 8 jo^g 
 
 bfi bfi 
 
 _0 jD 
 
 ^T^T 
 
 bO 
 ^O 
 
 CO CO C^ !> ^H C*l CO 
 
 
 * ^L 
 
 
 ^^ ^^ *-O ^^ O ^D I^H 
 
 
 jO^ -O^ 
 
 
 1 
 
 
 
 
 
 1 II II II II 
 
 
 w 
 
 
 "17 ^ ^ ^HH 
 
 O 
 
 TH' 
 
 
 
 W) o3 bfi o3 
 
 II II 
 
 Tf 
 
 
 
 . 
 
92 THE THEORY OF MEASUREMENTS [ART. 56 
 
 and the corresponding empirical relation becomes 
 s = 0.245 a - 1.0003 6 + 1.4022 c. 
 
 A small number of observation equations with simple coefficients 
 have been chosen, in the above illustration, partly to save space 
 and partly in order that the computations may be more readily 
 followed. In practice it would seldom be worth while to apply 
 the method of least squares to so small a number of observations 
 or to adopt Gauss's method of solution with logarithms when the 
 normal equations are so simple. When the number of observa- 
 tions is large and the coefficients involve more than three or four 
 significant figures, the method given above will be found very 
 convenient on account of the numerous checks and the symmetry 
 of the computations. In order to furnish a model for more 
 complicated problems, the process has been carried out completely 
 even in the parts where the results might have been foreseen 
 without the use of logarithms. 
 
 56. Conditioned Quantities. When the unknown quantities, 
 Xi, Xz, etc., are not independent in the mathematical sense, the 
 foregoing method breaks down since the equations (50) no longer 
 express the condition for a minimum of [rr]. In such cases the 
 number of unknowns may be reduced by eliminating as many of 
 them as there are rigorous mathematical relations to be fulfilled. 
 The remaining unknowns are independent and may be deter- 
 mined as above. The eliminated quantities are then determined 
 with the aid of the given mathematical conditions. 
 
 For the purpose of illustration, consider the case of a single 
 rigorous relation between the unknowns, and let the correspond- 
 ing mathematical condition be represented by the equation 
 
 0(x lt x,, . . . , x q ) =0. (71) 
 
 As in the case of unconditioned quantities, the observation equa- 
 tions (53) are 
 
 + C&s + + piX q = Si, 
 
 c n x 3 p n x q = 
 
 The solution of (71) for x\, in terms of Xz, x s , . . ., x q , may be 
 written in the form 
 
 xi=f(xz,x a , ..*,*) (72) 
 
ART. 56] THE METHOD OF LEAST SQUARES 93 
 
 Introducing this value of xi, equations (53) become 
 
 + ClX* + * + PlX q = Si, 
 + C 2 X 3 + + P2Z a = S 2 , 
 
 4- c n z 3 + + p n x q = s. 
 
 Since the form of 6 is known, that of / is also known. Hence, by 
 collecting the terms in x%, x S} etc., and reducing to linear form, 
 if necessary, we have 
 
 bixz + ci'x s + + p\x q = s/, 
 
 The x's in these equations are independent, and, consequently, 
 they may be determined by the methods of the preceding articles. 
 Using the values thus obtained in (71) or (72) gives the remaining 
 unknown x\. The #'s, thus determined, obviously satisfy the 
 mathematical condition (71) exactly, -and give the least magnitude 
 to the quantity [rr] that is consistent with that condition. They 
 are, consequently, the most probable values that can be assigned 
 on the basis of the given data. 
 
 As a very simple example, consider the adjustment of the 
 angles of a plane triangle. Suppose that the observed values of 
 the angles are 
 
 si = 60 1'; s 2 = 59 58'; s 3 = 59 59'. 
 
 The adjusted values must satisfy the condition 
 
 xi + x 2 + x* = 180, 
 or 
 
 xi = 180 - x 2 - x 3 . 
 
 Eliminating Xi from the observation equations, 
 xi = Si' t Xz = s 2 ; and x s s 3 ; 
 and substituting numerical values we have 
 x z +x 3 = 119 59', 
 x 2 = 59 58', 
 
 x 3 = 59 59'. 
 
 The corresponding normal equations are 
 2z 2 + z 3 = 179 57', 
 = 179 58', 
 
94 THE THEORY OF MEASUREMENTS [ART. 56 
 
 from which we find 
 
 x 2 = 59 58'.7 and x s = 59 59'.7. 
 
 Then, from the equation of condition, 
 
 xi = 60 1'.6. 
 
 When there are two relations between the unknowns, expressed 
 by the equations 
 
 01 (xi, x t , . . . , x q ) = 0, 
 
 02 (xi, x 2 , . . . , x q ) = 0, 
 
 they may be solved simultaneously for xi and x 2 , in terms of the 
 other x's, in the form 
 
 xi = fi(x 3 , xt, . . . , x q ), 
 
 x z = /2(z3, $,..., x q ). 
 
 Using these in the observation equations (53) we obtain a new set 
 of equations, independent of x\ and x* t that may be solved as 
 above. It will be readily seen that this process can be extended 
 to include any number of equations of condition. 
 
 When the number of conditions is greater than two, the compu- 
 tation by the above method becomes too complicated for practical 
 application and special methods have been devised for dealing 
 with such cases. The development of these methods is beyond 
 the scope of the present work, but they may be found in treatises 
 on geodesy and practical astronomy in connection with the prob- 
 lems to which they apply. 
 
CHAPTER VIII. 
 PROPAGATION OF ERRORS. 
 
 57. Derived Quantities. In one class of indirect measure- 
 ments, the desired numeric -X" is obtained by computation from 
 the numerics Xi, X z , etc., of a number of directly measured mag- 
 nitudes, with the aid of the known functional relation 
 
 X = F(X 1 ,X i , . . . ,X q ). 
 
 We have seen that the most probable value that we can assign to 
 the numeric of a directly measured quantity is either the arith- 
 metical mean of a series of observations of equal weight or the 
 general mean of a number of measurements of different weight. 
 Consequently, if x\, Xz, . . , x q represent the proper means of 
 the observations on Xi, X 2 , . . . , X q the most probable value 
 x that we can assign to X is given by the relation 
 x = F (xi, x z , . . . , x q ) 
 
 where F has the same form as in the preceding equation. 
 
 Obviously, the characteristic errors of x cannot be easily deter- 
 mined by a direct application of the methods discussed in Chapters 
 V and VI, as this would require a separate computation of x from 
 each of the individual observations on which Xi, Xz, etc., depend. 
 Furthermore, it frequently happens that we do not know the 
 original observations and are thus obliged to base our computa- 
 tions on the given mean values, x\, Xz, etc., together with their 
 characteristic errors. 
 
 Hence it becomes desirable to develop a process for computing 
 the characteristic errors of x from the corresponding errors of 
 Xij xz, etc. For this purpose we will first discuss several simple 
 forms of the function F and from the results thus obtained we 
 will derive a general process applicable to any form of function. 
 
 58. Errors of the Function Xi Xz X 3 =t . . . X q . 
 Suppose that the given function is in the form 
 
 X = Xi + X 2 , or X = Xi - X 2 . 
 
 These two cases can be treated together by writing the function in 
 the form 
 
 X = X\ db Xz, 
 95 
 
96 THE THEORY OF MEASUREMENTS [ART. 58 
 
 and remembering that the sign indicates two separate problems 
 rather than, as usual, an indefinite relation in a single problem. 
 If the individual observations on Xi are represented by ai, a 2 , 
 . . . , a n , and those on X 2 by 61, 6 2 , . . . , b n , we have 
 
 n n 
 
 and the most probable value of X is given by the relation 
 
 x = Xi xz. 
 
 From the given observations we can calculate n independent 
 values of X as follows : 
 
 Ai = ai &i, A 2 = az d= 6 2 , . . . , A n = a w db 6 n , 
 
 and it is obvious that the mean of these is equal to x. The true 
 accidental errors of the a's are 
 
 Aai = oi Xi, Aa 2 = a z Xi, . . . , Aa n = a n Zi; 
 those of the 6's are 
 
 Ah = 61 - Z 2 , A6 2 = 6 2 - Z 2 , . . . , A6 n = b n - X 2 ; 
 and those of the A's are 
 
 ^A l =A 1 -X ) &A 2 =A 2 -X, . . . , &A n =A n -X. 
 
 We cannot determine these errors in practice, since we do not 
 know the true value of the X's, but we can assume them in literal 
 form as above for the purpose of finding the relation between the 
 characteristic errors of the x's. 
 
 Combining the equations of the preceding paragraph with the 
 given functional relation, we have 
 
 AA X = (ai 60 - (Zi Z 2 ) 
 
 = (a! - ZO (61 - Xz) 
 = Aai A&i, 
 
 and similar expressions for the other A A's. Consequently 
 
 (AAO 2 = (AaO 2 d= 2 AaiA&i + (A6i) 2 , 
 (AA 2 ) 2 = (Aa 2 ) 2 d= 2 Aa 2 A6 2 
 
 (AA n ) 2 = (Aa n ) 2 2 ka n tU) n 
 Adding these equations, we find 
 
 [(AA) 2 ] = [(Aa) 2 ] 2 [AaA6] + [(A6) 2 ]. 
 
ART. 58] PROPAGATION OF ERRORS 97 
 
 Since A a and A b are true accidental errors, they are distributed 
 in conformity with the three axioms stated in article twenty-four. 
 Consequently equal positive and negative values of Aa and A6 
 are equally probable and the term [AaA6] would vanish if an 
 infinite number of observations were considered. In any case it 
 is negligible in comparison with the other terms in the above 
 equation. Hence, on dividing through by n, we have 
 
 [(AA)1 = [(Aa)l [(A6)*]_ 
 n n n 
 
 and by equation (20), article thirty-seven, this becomes 
 
 M A 2 = M a 2 + M b 2 , (73) 
 
 where M A is the mean error of a single A, M a that of a single a, 
 and M b that of a single b. Since x, xi, and z 2 are the arithmetical 
 means of the A's, a's, and 6's, respectively, their respective mean 
 errors, M , MI, and M 2 , are given by the relations 
 
 M 2 M 2 Tlf i 2 
 
 M* = ^, itf-=, and M, = ^- 
 
 n n n 
 
 in virtue of equations (29), article forty. Consequently, by (73) 
 M 2 = Mi 2 + M 2 2 , 
 
 or M = VMi 2 + M 2 2 . (74) 
 
 Since the mean and probable errors, corresponding to the same 
 series of observations, are connected by the constant relation (26), 
 article thirty-nine, we have also 
 
 + Ef, (75) 
 
 where E, EI, and E z are the probable errors of x, x\, and #2, 
 respectively. 
 
 It should be noticed that the ambiguous sign does not appear 
 in the expressions for the characteristic errors. The square of 
 the error of the computed quantity is equal to the sum of the 
 squares of the corresponding errors of the directly measured quan- 
 tities; whether the sign in the functional relation is positive or 
 negative. Thus the error of the sum of two quantities is equal 
 to the corresponding error of the difference of the same two quan- 
 tities. 
 
 Now suppose that the given functional relation is in the form 
 X = Xi d= X 2 X t . 
 
98 THE THEORY OF MEASUREMENTS [ART. 59 
 
 The most probable value of X is given by the relation 
 
 x = xi x z x 3y 
 
 where the notation has the same meaning as in the preceding 
 case. Represent x\ x z by x p , then 
 
 a; = x p =t z 3 , 
 
 and, by an obvious extension of the notation used above, we have 
 M P 2 = Mi 2 + M 2 2 , 
 M z = M P 2 + M 3 2 
 
 = Mi 2 + M 2 2 + M 3 2 . 
 
 Passing to the more general relation 
 
 X = Xi X 2 X 3 - - - X,, 
 
 we have a; = 1 db # 2 x 3 z fl , 
 
 and, by repeated application of the above process, 
 
 M 2 = M M 2 MJ + - - + M 3 2 , ) 
 
 + -E- 
 
 Thus the square of the error of the algebraic sum of a series of 
 terms is equal to the sum of the squares of the corresponding 
 errors of the separate terms whatever the signs of the given terms 
 may ba 
 
 59. Errors of the Function a\Xi =t 0:2^2 db a s X 3 =b - a q X q . 
 
 Let the given functional relation be in the form 
 
 X = 
 
 where a\ is any positive or negative, integral or fractional, con- 
 stant. The most probable value that we can assign to X on the 
 basis of n equally good independent measurements of X is 
 
 x = aiXi, 
 
 where Xi is the arithmetical mean of the n direct observations 
 ai, a 2 , a s , . . . , a n . 
 
 The n independent values of X obtainable from the given obser- 
 vations are 
 
 AI ami, Az aids, . . . , A n = a\a n . 
 The accidental errors of the a's and A's are 
 
 Aai = a\ Xij Aa 2 = a 2 X\ t . . . , Aa n = a n X\, 
 and 
 
 A4i = Ai - X, A^ 2 = A t -X, . . . , AA n = A n -X. 
 
ART. 60] PROPAGATION OF ERRORS 99 
 
 Combining these equations we find 
 
 and similar expressions for the other AA's. Consequently 
 
 (AAO 2 = ai 2 (Aax) 2 , 
 and [(AA) 2 ] = ai [(Aa) 2 ]. 
 
 If M and Af i are the mean errors of x and xi t respectively, 
 
 and Jf,..I3. 
 
 Hence M 2 = onWi 2 , (77) 
 
 and, since the probable error bears a constant relation to the 
 mean error, 
 
 E 2 = a^! 2 . (78) 
 
 When the given functional relation is in the more general form 
 
 X = aiXi =b 0:2^2 =b 0.3X3 =b otqXqj 
 
 we have 
 
 x = 
 
 where the a;'s are the most probable values that can be assigned 
 to the X's on the basis of the given measurements. Applying 
 (77) and (78) to each term of this equation separately and then 
 applying (76) we have 
 
 t 
 E 2 = 
 
 where the ATs and E's represent respectively the mean and prob- 
 able errors of the x's with corresponding subscripts. 
 
 60. Errors of the Function F (X l} X 2 , . . . , X q ). 
 
 We are now in a position to consider the general functional 
 relation 
 
 X = F (Xi, Xz, . . . , X q ), 
 
 where F represents any function of the independently measured 
 quantities Xi, X 2 , etc. Introducing the most probable values of 
 the observed numerics, the most probable value of the computed 
 numeric is given by the relation 
 
 x = F fa, x 2) . . . , Xq). (80) 
 
 This expression may be written in the form 
 
 & l ), (Z 2 -f-5 2 .. . . . , (* + ,)!, 0) 
 
100 THE THEORY OF MEASUREMENTS [ART. 60 
 
 where the I's represent arbitrary constants and the.S's are small 
 corrections given by relations in the form 
 
 Obviously, the errors of the 5's are equal to the errors of the corre- 
 sponding x's. For, if Mi, Ms, and MI are the errors of Xi, 5i, and 
 Zi, respectively, we have by equation (74) 
 
 M s * = Mi 2 + Mf. 
 
 But MI is equal to zero, because I is an arbitrary quantity and any 
 value assigned to it may be considered exact. Consequently 
 
 Mi 2 . (ii) 
 
 Since the I's are arbitrary, they may be so chosen that the 
 squares and higher powers of the-5's will be negligible in compari- 
 son with the 8's themselves. Hence, if the x's are independent, 
 (i) may be expanded by Taylor's Theorem in the form 
 
 dF d , \ ** 
 
 where = F (z, z, . . . , x) = > 
 
 and the other differential coefficients have a similar significance. 
 When the observed values of the x's are substituted in these 
 coefficients, they become known numerical constants. 
 
 The mean error of F (li, Z 2 , . . . , l q ) is equal to zero, since it 
 is a function of arbitrary constants; and the mean errors of the 
 5's are equal to the mean errors of the corresponding x's by (ii). 
 Consequently, if M, Mi, M 2 , . . . , M q represent the mean errors 
 of x, Xi, x z , . . . , x q , respectively, we have by equation (79) 
 
 /dF - . V , fdF , , V , 
 
 = F~ MI ) + brr^ 2 ) + 
 
 \dxi I \dx 2 I N ~~ , , . 
 
 (OL) 
 
 where the E's represent the probable errors of the x's with corre- 
 sponding subscripts. 
 
 Equations (81) are general expressions for the mean and prob- 
 able errors of derived quantities in terms of the corresponding 
 errors of the independent components. Generally x\ t x 2 , etc., 
 
ART. 61] PROPAGATION OF ERRORS 101 
 
 represent either the arithmetical or the general means of series of 
 direct observations on the corresponding components, and EI, E z , 
 etc., can be computed by equations (32) or (41). In some cases, 
 the original observations are not available but the mean values 
 together with their probable errors are given. 
 For the purpose of computing the numerical value of the differ- 
 
 r\Tj1 r\Tj1 
 
 ential coefficients -r ; > etc., the given or observed values of 
 oXi 0X2 
 
 the components x i} x 2) etc., may generally be rounded to three 
 significant figures. This greatly reduces the labor of computa- 
 tion and does not reduce the precision of the result, since the E's 
 and M's are seldom given or desired to more than two significant 
 figures. 
 
 61. Example Introducing the Fractional Error. The prac- 
 tical application of the foregoing process is illustrated in the follow- 
 ing simple example: the volume V of a right circular cylinder is 
 computed from measurements of the diameter D and the length L, 
 and we wish to determine the probable error of the result. In 
 this case, V corresponds to x, D to xi, L to x 2) and the functional 
 relation (80) becomes 
 
 Also, if EV, E D , and EL are the probable errors of V, D, and L, 
 respectively, the second of equations (81) becomes 
 
 where 
 
 sv 
 
 and 
 
 dV d /I \ 1 n2 
 -r^F- = ^F \ -7 TTL) L ] = -TrD*. 
 dL dL\4 / 4 
 
 Hence 
 
 The computation can be simplified by introducing the frac- 
 
 TTT 
 
 tional error -^~- Thus, dividing the above equation by 
 
 we have 
 
 ^ =4 ^! + ^ 
 
 T7"O 7~^9 I T O 
 
102 THE THEORY OF MEASUREMENTS [ART. 62 
 
 or, writing PV, PD, and PL for the fractional errors, 
 Py 2 = 4 Pz> 2 + P L \ 
 P V 
 and finally 
 
 E v = FP F = V 
 
 A similar simplification can be effected, in dealing with many 
 other practical problems, by the introduction of the fractional 
 errors. Consequently it is generally worth while to try this ex- 
 pedient before attempting the direct reduction of the general 
 equation (81).- 
 
 In order to render the problem specific, suppose that 
 D = 15.67 0.13 mm., 
 L = 56.25 d= 0.65 mm., 
 then V = 10848 
 
 PD = = = 
 
 P L = ^ = ^ = .0116; Pz, 2 = 135 X 10- 6 , 
 
 = 0.020, 
 
 E v = VTV = 220 mm 
 Hence 
 
 7= 10.85 0.22 cln. 3 
 62. Fractional Error of the Function aX^ 1 X Z 2 U2 X 
 
 X a n5 .- 
 
 Suppose the given relation is in the form 
 X = F(X l ) =aXi 
 
 where a and n are constants and the =fc sign of the exponent n is 
 used for the purpose of including the two functions aXi +n and 
 aX-r^ in the same discussion. In this case equation (80) becomes 
 
 x = axi n , 
 
 and the second of (81) reduces to 
 But 
 
 _=_ 
 
 Consequently 
 
ART. 62] PROPAGATION OF ERRORS 103 
 
 If P and PI are the fractional errors of x and xi, respectively, we 
 have 
 
 E* 
 
 - 
 
 Hence 
 
 i P = nP,. (82) 
 
 If we replace n by in the above argument, (80) becomes 
 
 _ 
 
 x = aXi m , 
 
 and we find 
 
 m 
 
 Hence the fractional error of any integral or fractional power of 
 a measured numeric is equal to the fractional error of the given 
 numeric multiplied by the exponent of the power. 
 
 If the given function is in the form of a continuous product 
 
 X = aX l X X, X X X qt 
 (80) becomes x = axi X x 2 X X x q . 
 
 dF 
 
 Hence = ax z X x 3 X X x g , 
 
 ox\ 
 
 I dF 1 
 
 and - = 
 
 Hence, by (81), 
 
 JP _ Ei 2 EJ Eg 2 
 
 r z ~ 7~2 ~f~ ~~2 ~r T > 
 
 Js JL>1 JU2 Lq 
 
 and, if P, PI, P 2 , . . . , P q represent the fractional errors of the 
 #'s with corresponding subscripts, 
 
 Combining the above cases we obtain the more general rela- 
 tion 
 
 X = aXi 1 X Xz 2 X * * X X q , 
 and the corresponding expression for (80) is 
 
 Applying (82) to each factor separately and then applying (83) to 
 the product, we find 
 
 f - - - +nfPf. (84) 
 
104 THE THEORY OF MEASUREMENTS [ART. 62 
 
 For the sake of illustration and to fix the ideas this result may 
 be compared with the example of the preceding article. If we 
 
 put x = V, Xi = D, HI = 2, x 2 = L, n 2 = 1, a = -7 , P = Py, 
 PI = PD, and PZ = PL the above expression for x becomes 
 
 V = %TrD 2 L, 
 and (84) becomes 
 
 Occasionally it is convenient to express the probable error in 
 the form of a percentage of the measured magnitude. If E and 
 p are respectively the probable and percentage errors of x, 
 
 p= 100 - = 100 P. (85) 
 
 x 
 
 Consequently (84) may be written in the form 
 
 P 2 = niW + n 2 2 p 2 2 + + nfp*, (84a) 
 
 where pi, p 2 , . . . , p q are the percentage errors of Xi, x 2 , . . . , x q , 
 respectively 
 
CHAPTER IX. 
 ERRORS OF ADJUSTED MEASUREMENTS. 
 
 WHEN the most probable values of a number of numerics 
 Xi, X 2 ,etc., are determined by the method of least squares, the 
 results Xi, x 2 ,etc., are called adjusted measurements of the quan- 
 tities represented by the X's. In Chapter VII we have seen how 
 the x's come out by the solution of the normal equations (56) or 
 (58), and how these equations are derived from the given obser- 
 vations through the equations (53). In the present chapter we 
 will determine the characteristic errors of the computed x's in 
 terms of the corresponding errors of the direct measurements on 
 which they depend. 
 
 63. Weights of Adjusted Measurements. When there are q 
 unknowns and the given observations are all of the same weight, 
 the normal equations, derived in article fifty, are 
 
 [aa] Xi + [ab] x 2 + [ac] x 3 + - + [ap] x q = [as], 
 
 [db] x, + [66] x 2 + [6c] *,+ + [bp] x q = [bs], (56) 
 
 [ap] xi + [bp] x z + [cp] x 3 + + [pp] x q = [ps]. 
 
 Since these equations are independent, the resulting values of the 
 x's will be the same whatever method of solution is adopted. In 
 Chapter VII Gauss's method of substitution was used on account 
 of the numerous checks it provides. For our present purpose 
 the method of indeterminate multipliers is more convenient as it 
 gives us a direct expression for the x's in terms of the measured 
 s's. Obviously this change of method cannot affect the errors of 
 the computed quantities. 
 
 Multiply each of equations (56) in order by one of the arbitrary 
 quantities AI, A 2 , . . . , A q and add the products. The result- 
 ing equation is 
 
 (86) 
 
 + ([db] A 1 + [bb] A, + + [bp] A q ) x 2 
 
 + > 
 
 = [as] A l + [6s] A 2 + + [ps] A q . 
 105 
 
[ob] A, + [66] A* + + [6p] A q = 0, 
 
 106 THE THEORY OF MEASUREMENTS [ART. 63 
 
 Since the A's are arbitrary and q in number, they can be made to 
 satisfy any q relations we choose without affecting the validity 
 of equation (86). Hence, if we determine the A's in terms of the 
 coefficients in (56) by the relations 
 
 (g7) 
 
 equation (86) gives an expression for x\ in the form 
 
 xi = [as] Ai + [&*] 4i +!-'+ \ps]A t . (88) 
 
 If we repeat this process q times, using a different set of multipliers 
 each time, we obtain q different equations in the form of (86). 
 In each of these equations we may place the coefficient of one of 
 the x's equal to unity and the other coefficients equal to zero, giv- 
 ing q sets of equations in the form of (87) for determining the q sets 
 of multipliers. Representing the successive sets of multipliers by 
 A's, B's, C"s, etc., we obtain (88), and the following expressions 
 for the other x's : 
 
 x 2 = [as] Bi + [bs] ft +...;+ \p 8 ] B q , 
 x 3 = [as] Ci + [6s] C 2 + + \ps] C q , 
 
 x q = [as] P! + [6s] P 2 + + \ps] P q . 
 
 From equations (87), it is obvious that the A's do not involve 
 the observations Si, s 2 , etc. Consequently (88) may be expanded 
 in terms of the observations as follows: 
 
 Xi = ctiSi + a z s 2 -f + ctgS q , (89) 
 
 where the a's depend only on the coefficients in the observation 
 
 equations (53) and are independent of the s'a. Since we are con- 
 
 sidering the case of observations of equal weight, each of the s's 
 
 in (89) is subject to the same mean error M 8 . Her e, if MI is 
 
 the mean error of Xi, we have by equations (79), article fifty-nine, 
 
 Mx 2 = ai 2 M s 2 + 2 2 M a 2 + - + a n 2 M, 2 
 
 = M M, 2 . 
 
 But, if Wi is the weight of x\ in comparison with that of a single s, 
 we have by (36), article forty-four, 
 
 Wl w i (90) 
 
 Mi 2 [act] 
 
ART. 63] ERRORS OF ADJUSTED MEASUREMENTS 107 
 
 since the ratio of the mean errors of two quantities is equal to the 
 ratio of their probable errors. 
 
 Comparing equations (88) and (89), with the aid of equations 
 (55), article fifty, we see that 
 
 biA 2 + +piA q , 
 
 (i) 
 
 a n = 
 
 p n A q . 
 
 Multiply each of these equations by its a and add the products, 
 then multiply each by its b and add, and so on until all of the 
 coefficients have been used as multipliers. We thus obtain the 
 q sums [aa], [ba], . . . , [pa], and by taking account of equations 
 (87) we have 
 
 [aa] = 1, > 
 
 [ba] = [ca] = . = [pa] = 0. ) 
 
 Hence, if we multiply each of equations (i) by its a and add the 
 
 products, we have 
 
 [aa] = A i. 
 
 Consequently equation (90) becomes 
 
 A l 
 
 (91) 
 
 The weights of the other x's may be obtained, by an exactly 
 similar process, from equations (88a). The results of such an 
 analysis are as follows: 
 
 M 
 
 M a 2 P t 
 
 (91a) 
 
 Obviously the coefficients of the sums [as], [bs], etc., in equa- 
 tions (88) and (88a) do not depend upon the particular method by 
 which the normal equations are solved, since the resulting values 
 of the x's must be the same whatever method is used. Conse- 
 quently, if the absolute terms [as], [bs], . . . , [ps] are kept in literal 
 form during the solution of the normal equations by any method 
 whatever, the results may be written in the form of equations 
 
108 THE THEORY OF MEASUREMENTS [ART. 64 
 
 (88) and (88a); and the quantities AI, B 2) etc., will be numerical 
 if the coefficients [aa], [ab], . . . , [bb], . . . , [pp] are expressed 
 numerically. 
 
 Hence, in virtue of (91) and (91 a), we have the following rule 
 for computing the weights of the z's. 
 
 Retain the absolute terms of the normal equations in literal 
 form, solve by any convenient method, and write out the solution 
 in the form 
 
 a?i = [as] A! + [bs] A 2 + [cs] A 3 + - + \ps] A qt 
 x 2 = [as] B l + [bs] B 2 + [cs] B 3 + - - - + \ps] B q , 
 
 x q = [as] P 1 + [bs] P 2 + [cs] P, + - - + [ps] P q . 
 
 Then the weight of x\ is the reciprocal of the coefficient of [as] in 
 the equation for x\, the weight of x 2 is the reciprocal of the co- 
 efficient of [bs] in the equation for x%, and in general the weight of 
 x q is the reciprocal of the coefficient of [ps] in the equation for x q . 
 
 As an aid to the memory, it may be noticed that the coefficients 
 AI, B 2 , Cs, . . . , P q , that determine the weights, all lie in the 
 main diagonal of the second members of the above equations. 
 When the number of unknowns is greater than two, the labor of 
 computing all of the A's, B's, etc., would be excessive, and conse- 
 quently it is better to determine the x's by the methods of Chap- 
 ter VII. The essential coefficients AI, B 2 , C 3 , . . . , P q can be 
 determined independently of the others by the method of deter- 
 minants as will be explained later. 
 
 If the given observations are not of equal weight, the weights 
 of the x's may be determined by a process similar to the above, 
 starting with normal equations in the form of (58), article fifty. 
 The result of such an analysis can be expressed by the rule stated 
 above if we replace the sums [as], [bs], . . . , [ps] by the weighted 
 sums [was], [wbs], . . . , [wps], the notation being the same as in 
 article fifty. 
 
 64. Probable Error of a Single Observation. By definition, 
 article thirty-seven, the mean error M 8 of a single observation is 
 given by the expression 
 
 _ Af + A^+.-.+A.' _ [AA] , (iii) 
 
 n n 
 
 where the A's represent the true accidental errors of the s's. 
 When the number of observations is very great, the residuals given 
 
ART. 64] ERRORS OF ADJUSTED MEASUREMENTS 109 
 
 by equations (54) may be used in place of the A's without causing 
 appreciable error in the computed value of M 8 . But, in most 
 practical cases, n is so small that this simplification is not admis- 
 sible and it becomes necessary to take account of the difference 
 between the residuals and the accidental errors. 
 
 Let Ui, u 2 , . . . , u q represent the true errors of the x's ob- 
 tained by solution of the normal equations (56). Then the true 
 accidental error of the first observation is given by the relation 
 
 Ol (Xi + Ui) + 61 (X 2 + U 2 ) + + Pl (X q + U q ) - Si = Ai. 
 
 But, by the first of equations (54), 
 
 aiXi + 6ix 2 -f cix s + + pix q si = ri, 
 
 where r\ is the residual corresponding to the first observation. 
 Combining these equations and applying them in succession to 
 the several observations, we obtain the following expressions for 
 the A's in terms of the r's: 
 
 ri + aiui + biu 2 + CiU 3 + - + piUq = Ai, 
 
 A 2 , 
 
 ,.* 
 
 + b n u 2 + c n u 3 + + p n u q = A n . 
 Multiply each of these equations by its r and add; the result is 
 
 [rr] + [ar] HI + [br] u 2 + [cr] u 3 + + [pr] u q = [Ar]. 
 But by equations (iii), article fifty, 
 
 [ar] = [br] = [cr] = = for] = 0, (v) 
 
 and, consequently, 
 
 [rr] = [Ar]. (vi) 
 
 Multiply each of equations (iv) by its A and add. Then, taking 
 account of (vi), we have 
 
 [rr] + [aA] Ul + [6A] u 2 + + [pA] u q = [AA]. (vii) 
 
 In order to obtain an expression for the u's in terms of the A's, 
 multiply each of equations (iv) by its a and add, then multiply 
 by the b's in order and add, and so on with the other coefficients. 
 The first term in each of these sums vanishes in virtue of (v), and 
 we have 
 
 [aa] ui + [ab] w 2 + + [ap] u q = [aA], 
 
 [db] Ul + [bb] u, + + \bp] u q = [6A], 
 
 lap] ui + [bp] u 2 + - - - + [pp] u q = 
 
 (viii) 
 
110 THE THEORY OF MEASUREMENTS [Am. 64 
 
 These equations are in the same form as the normal equations (56) 
 with the z's replaced by u's and the s's by A's. Hence any solu- 
 tion of (56) for the x's may be transposed into a solution of (viii) 
 for the u's by replacing the s's by A's without changing the coeffi- 
 cients of the s's. Consequently, by (89), we have 
 
 and similar expressions for the other u's. 
 The coefficients of the u's in (vii) expand in the form 
 
 [aA] = aiAi + a 2 A 2 + + a n A n . 
 Hence 
 
 [aA] ui = aiaiAi 2 + a 2 2 A 2 2 +.+ a n a n A n 2 , 
 
 Since positive and negative A's are equally likely to occur, the 
 sum of the terms involving products of A's with different subscripts 
 will be negligible in comparison with the other terms. The sum 
 of the remaining terms cannot be exactly evaluated, but a suffi- 
 ciently close approximation is obtained by placing each of the A 2 's 
 
 equal to the mean square of all of them, - - -* Consequently, as 
 the best approximation that we can make, we may put 
 
 n 
 But, by equations (ii), [aa] is equal to unity. Hence 
 
 [aA] - M. 
 
 iv 
 
 Since there is nothing in the foregoing argument that depends on 
 the particular u chosen, the same result would have been obtained 
 with any other u. .Consequently, in equation (vii), each term that 
 
 involves one of the u's must be equal to - - !i and, since there 
 
 tv 
 
 are q such terms, the equation becomes 
 
 Hence, by equation (iii), 
 and 
 
ART. 64] ERRORS OF ADJUSTED MEASUREMENTS 111 
 
 where the r's represent the residuals, computed by equations (54) ; 
 n is the number of observations ; and q is the number of unknowns 
 involved in the observation equations (53). In the case of direct 
 measurements, the number of unknowns is one, and (92) reduces 
 to the form already found in article forty-one, equation (30), for 
 the mean error of a single observation. 
 
 When the observations are not of equal weight, the mean error 
 M 8 of a standard observation, i.e. an observation of weight 
 unity, is given by the expression 
 
 2 = 
 
 n 
 
 where the w's are the weights of the individual observations. 
 Starting with this relation in place of (iii) and making correspond- 
 ing changes in other equations, an analysis essentially like the 
 preceding leads to the result 
 
 Ma = ^'^-, (93) 
 
 T n q 
 
 which reduces to the same form as (92) when the weights are all 
 unity. 
 
 Introducing the constant relation between the mean and probable 
 errors, we have the expressions 
 
 E 8 = 0.6741/-M- , (94) 
 
 V n q 
 
 for the probable error of a single observation in the case of equal 
 weights, and 
 
 E 8 = 0.674\/-^i, (95) 
 
 V n q 
 
 for the probable error of a standard observation in the case of 
 different weights. 
 
 Finally, if M k , E k , and w k represent the mean error, the probable 
 error, and the weight of x k , any one of the unknown quantities, 
 we may derive the following relations from the above equations 
 by applying equations (36), article forty-four: 
 
 M s 
 
 - = 7= V ' 
 
 A/in. T n o 
 
 (96) 
 
112 
 
 THE THEORY OF MEASUREMENTS [ART. 65 
 
 when the weights of the given observations are equal, and 
 
 M k = -^= = L Y/-^-> 
 
 v Wk vWk n ~ Q 
 
 E, 0.674 
 
 Ek = / - = 
 
 (97) 
 
 ~ 2 
 when the weights of the given observations are not equal. 
 
 65 . Application to Problems Involving Two Unknowns . When 
 the observation equations involve only two unknown quantities, 
 the solution of the normal equations is given by (59), article 
 fifty-one, in the form 
 
 _ [66] [as] - [ab] [bs] 
 [aa] [bb] - [ab] 2 ' 
 _ [aa] [bs] [ab] [as] 
 
 [aa] [bb] - [ab] 2 
 
 By the rule of article sixty-three, the weight of Xi is equal to the 
 reciprocal of the coefficient of [as] in the equation for Xi, and the 
 weight of #2 is equal to the reciprocal of the coefficient of [bs] in 
 the equation for x 2 . Hence, by inspection of the above equations, 
 we have 
 
 [aa] [bb] - [ab] 2 
 
 
 _ 
 
 W 2 = 
 
 [bb] 
 
 [aa] [bb] - [ab] 2 
 [aa] 
 
 (98) 
 
 Since there are only two unknown quantities, and the observa- 
 tions are of equal weight, equation (92) gives the mean error of a 
 single observation when q is taken equal to two. Hence 
 
 (99) 
 
 where n is the number of observation equations and [rr] is the 
 sum of the squares of the residuals that are obtained when the 
 computed values of Xi and Xz are substituted in equations (53a), 
 article fifty-one. 
 
 Combining equations (98) and (99) with (96), we obtain the 
 following expressions for the probable errors of Xi and x 2 : 
 
 0.674 
 
 E 2 = 0.674 
 
 v/ 
 v/ 
 
 [66] 
 
 
 [aa][bb] - [ab] 2 n-2 
 
 [aa] 
 
 [rr> 
 
 [aa] [bb] - [ab] 2 n-2 
 
 (100) 
 
ART. 65] ERRORS OF ADJUSTED MEASUREMENTS 113 
 
 For the purpose of illustration, we will compute the probable 
 errors of the values of x\ and x 2 obtained in the numerical prob- 
 lem worked out in article fifty-one. Referring to the numerical 
 tables in that article, we find 
 
 [aa] = 5; [ab] = 20; [bb] = 90; n = 5; 
 [rr] = 9.60 X 1Q- 4 . 
 
 Hence, by equations (100), 
 
 *' 
 
 V / 
 
 5X90-400 
 
 By equations (vi), article fifty-one, the length L of the bar at 
 C., and the coefficient of linear expansion a are given by the 
 relations 
 
 L = iooo + si; a = -L.*. 
 
 10 -L70 
 
 Since L is equal to #1 plus a constant, its probable error is equal 
 to that of Xi by the argument underlying equation (ii), article 
 sixty. Hence 
 
 EL. = E! = =fc 0.016. 
 
 To find the probable error of a, we have by equations (81), article 
 sixty, 
 
 But, since L is very large in comparison with x 2 , the second term 
 on the right-hand side is negligible in comparison with the first. 
 Consequently, without affecting the second significant figure of 
 the result, we may put 
 
 = Ei X 10- 4 = =fc 0.038 X 10- 5 . 
 
 Hence the final results of the computations in article fifty-one may 
 be more comprehensively expressed in the form 
 
 L Q = 1000.008 db 0.016 millimeters, 
 a = (1.780 db 0.038) X 10~ 5 , 
 
114 
 
 THE THEORY OF MEASUREMENTS [AET. 66 
 
 when we wish to indicate the precision of the observations on 
 which they depend. 
 66. Application to Problems Involving Three Unknowns. The 
 
 normal equations, for the determination of three unknowns, take 
 the form 
 
 [aa] Xi + [ah] x 2 + [ac] x 3 = [as], 
 
 [ac] xi + [be] x 2 + [cc] x 3 = [cs]. 
 Solving by the method of determinants and putting 
 
 we have 
 
 [as] 
 
 x 2 = [as 
 
 [as] 
 
 Hence, by the rule of article sixty-three, 
 
 D 
 
 Wl [bb][cc] -[be] 2 ' 
 
 = D 
 
 2 ~~ [aa] [cc] [ac] 2 ' 
 
 D 
 [aa][bb]-[ab]*' 
 
 [aa] 
 [ab] 
 [ac] 
 
 [ab] [ac 
 [66] [be 
 [be] [cc 
 
 ] 
 
 = A 
 
 [bb] [be] 
 [be] [cc] 
 
 1 J 
 
 [be] [cc] 
 [06] [ac] 
 
 4 
 
 -[cs] 
 
 [06] [ac] 
 [bb] [be] 
 
 t 
 
 D 
 
 D 
 
 D 
 
 [ac] [cc] 
 [06] [6c] 
 
 - + [6s] 
 
 [aa] [ac] 
 [ac] [cc] 
 
 - 
 
 -[cs] 
 
 [ab] [be] 
 [aa] [ac] 
 
 , 
 
 D 
 
 D 
 
 D 
 
 [ah] [66] 
 [ac] [6c] 
 
 + N- 
 
 [ac] [be] 
 [aa] [ab] 
 
 + [cs] 
 
 [aa] [ab] 
 [ab] [bb] 
 
 
 D 
 
 D 
 
 D 
 
 w s = 
 
 (ix) 
 
 (x) 
 
 The determinant D can be eliminated from equations (x), if 
 we can obtain an independent expression for any one of the w's. 
 The solution of the normal equations by Gauss's Method in 
 article fifty-four led to the result 
 
 - 
 X3 ~ 
 
 [cc'2] 
 
ART. 66] ERRORS OF ADJUSTED MEASUREMENTS 115 
 
 The auxiliary [cc 2] is independent of the absolute terms [as], 
 [6s], and [cs]. The auxiliary [cs 2] may be expanded as follows: 
 
 [oc] r , [6cl] ( , [ab] 
 
 [6cl] ( , 
 ~ PTTJ \ M - 
 
 Hence the coefficient of [cs] in the above expression for x$ is 
 r - ~y, and, consequently, the weight of x$ is equal to [cc2]. 
 
 [CC ZJ 
 
 Substituting this value for w s in the third of equations (x) and 
 eliminating D from the other two we have 
 
 [aa] [bb 1] 
 
 [66 
 
 (101) 
 
 w 3 = [cc 2], 
 
 where the auxiliary quantities [66 1], [cc 1], and [cc 2] have the 
 same significance as in article fifty-four. 
 
 The weights of the x's having been determined by equations 
 (101), their probable errors may be computed by equations (96). 
 In the present case q is taken equal to three, since there are three 
 unknowns, and the r's are given by equations (68). 
 
 In the numerical illustration of Gauss's Method, worked out in 
 article fifty-five, we found the following values of the quantities 
 appearing in equations (96) and (101): 
 
 [aa] = 6; [66] = 220; [6c] = 180; [cc] = 157; 
 [66 1] = 70; [cc 1] = 76.0; [cc 2] = 5.97; 
 [rr] = 0.00120; n = 6; q = 3. 
 
 These values have been rounded to three significant figures, when 
 necessary, since the probable errors of the #'s are desired to only 
 two significant figures. Substituting in equations (101) we have 
 
 Wl = 6X7 _ 2 5.97 -1.17, 
 
 220 X 157 - 180 
 
 70 
 ^2 = y^5.97 = 5.50, 
 
 w 3 = 5.97, 
 
116 THE THEORY OF MEASUREMENTS [ART. 66 
 
 From equation (94) 
 
 \E. 
 
 and, by equations (96), 
 
 a = 0.674 1/ ' 0012 = 0.0135, 
 
 0.0135 
 . O.UUoo. 
 
 Consequently the precision of the measurements, so far as it 
 depends on accidental errors, may be expressed by writing the 
 computed values of the x's in the form 
 
 xi = 0.245 0.012, 
 X2 =- 1.0003 0.0057, 
 z 3 = 1.4022 0.0055. 
 
 Since the last significant figure in each of the x's occupies the same 
 place as the second significant figure in the corresponding prob- 
 able error, it is evident that the proper number of figures were 
 retained throughout the computations in article fifty-five. 
 
CHAPTER X. 
 DISCUSSION OF COMPLETED OBSERVATIONS. 
 
 67. Removal of Constant Errors. The discussion of acci- 
 dental errors and the determination of their effect on the result 
 computed from a given series of observations, as carried out in the 
 preceding chapters, are based on the assumption that the meas- 
 urements are entirely free from constant errors and mistakes. 
 Hence the first matter of importance, in undertaking the reduction 
 of observations, is the determination and removal of all constant 
 errors and mistakes. Also, in criticizing published or reported 
 results, judgment is based very largely on the skill and care with 
 which such errors have been treated. In the former case, if suit- 
 able methods and apparatus have been chosen and the adjust- 
 ments of instruments have been properly made, sufficient data is 
 usually at hand for determining the necessary corrections within 
 the accidental errors. In the latter case we must rely on the dis- 
 cussion of methods, apparatus, and adjustments given by the 
 author and very little weight should be given to the reported 
 measurements if this discussion is not clear and 'adequate. 
 
 No evidence can be obtained from the observations themselves 
 regarding the presence or absence of strictly constant errors. 
 The majority of them are due to inexact graduation of scales, 
 imperfect adjustment of instruments, personal peculiarities of the 
 observer, and faulty methods of manipulation. They affect all 
 of the observations by the same relative amount. Their detec- 
 tion and correction or elimination depend entirely on the judg- 
 ment, experience, and care of the observer and the computer. 
 When the same magnitude has been measured by a number of 
 different observers, using different methods and apparatus, the 
 probability that the constant errors have been the same in all of 
 the measurements is very small. Consequently if the corrected 
 results agree, within the accidental errors of observation, it is 
 highly probable that they are free from constant errors. This is 
 the only criterion we have for the absence of such errors and it 
 
 117 
 
118 THE THEORY OF MEASUREMENTS [ART. 67 
 
 breaks down in some cases when the measured magnitude is not 
 strictly constant. 
 
 Sometimes constant errors are not strictly constant but vary 
 progressively from observation to observation owing to gradual 
 changes in surrounding conditions or in the adjustment of instru- 
 ments. The slow expansion of metallic scales due to the heat 
 radiated from the body of the observer is an illustration of a 
 progressive change. Such variations are usually called systematic 
 errors. They may be corrected or eliminated by the same methods 
 that apply to strictly constant errors when adequate means are 
 provided for detecting them and determining the magnitude of 
 the effects produced. When their range in magnitude is compara- 
 ble with that of the accidental errors, their presence can usually be 
 determined by a critical study of the given observations and their 
 residuals. But, if they have not been foreseen and provided for 
 in making the observations, their correction is generally difficult 
 if not impossible. In many cases our only recourse is a new series 
 of observations taken under more favorable conditions and accom- 
 panied by adequate means of evaluating the systematic errors. 
 
 A general discussion of the nature of constant errors and of the 
 methods by which they are eliminated from single direct observa- 
 tions was given in Chapter III. These processes will now be con- 
 sidered a little more in detail and extended to the arithmetical 
 mean of a number of direct observations. Let a\ t d 2 , a s , . . . , a n 
 represent a series of direct observations after each one of them 
 has been corrected for all constant errors. Then the most prob- 
 able value that can be assigned to the numeric of the measured 
 magnitude is the arithmetical mean 
 
 x = q i + fl2 + +a n /jx 
 
 IV 
 
 Now suppose, that the actual uncorrected observations are 01, o 2 , 
 
 o 3 , , o n , then 
 
 ai = 01 + cj + cj' + cj" + + ci<*> = 01 + [cj, 
 a 2 = o 2 + cj + c 2 " + cj" + + c 2 ("> = o 2 + 
 
 C*n = O n + C n ' + C" + C n '" + + cj* = O n + [c 
 
 where the c's represent the constant errors to be eliminated and 
 may be either positive or negative. There are as many c's in 
 each equation as there are sources of constant error to be consid- 
 
ART. 67] DISCUSSION OF COMPLETED OBSERVATIONS 119 
 
 ered. Usually, when all of the observations are made by the 
 same method and with equal care, the number of c's is the same 
 in all of the equations. Substituting (ii) in (i) 
 
 J . = 0l + 02+ +. [Cj + [cj+ - - +[ftj 
 
 n n ' 
 
 When there are no systematic errors 
 
 Cl = Cz = C 3 ' = 
 Cl " = C 2 " = C," = = Cn " = C ", 
 
 = C 3 ' = * = Cn 
 
 Consequently 
 
 [ci] = [c z ] = [c 3 ] = = [c n ] = [c], (iv) 
 
 and we have 
 
 x = + [c] 
 
 n 
 
 = Om + c' + c" + c"' + -f c<>, (102) 
 
 where o m is written for the mean of the actual observations. 
 Hence, when all of the observations are affected by the same con- 
 stant errors, the corrections may be applied to the arithmetical 
 mean of the actual observations and the resulting value of x will 
 be the same as if the observations were separately corrected before 
 taking the mean. 
 
 The residuals corresponding to the corrected observations ai, 
 a 2 , a 3 , . . . , a n are given by equations (3), article twenty-two. 
 Replacing x and the a's by their values in terms of o m and the 
 o's as given in (102) and (ii), and taking account of (iv), equations 
 (3) become 
 
 ri = di X = Oi+ [Ci] - Om- [C\ = 01 - O m , 
 
 r 2 = a 2 x = o 2 + [c 2 ] o m [c] = o 2 o m , (103) 
 
 r n = a n - X = O n + [C n ] -Om- [c] = O n - O m . 
 
 Consequently, when there are no systematic errors, the residuals 
 computed from the o's and o m will be identical with those com- 
 puted from the a's and x. Hence, if the uncorrected observations 
 are used in computing the probable error of x, by the formula 
 
 / W 
 
 E = 0.674\/ / J 1X > 
 V n (n 1) 
 
120 THE THEORY OF MEASUREMENTS [ART. 67 
 
 the result will be the same as if the corrected observations had 
 been used; and, as pointed out above, the observations and their 
 corresponding residuals give no evidence of the presence of strictly 
 constant errors. 
 
 When the constant errors affecting the different observations 
 are different or when any of them are systematic in character, 
 equation (iv) no longer holds, and, consequently, the simplifica- 
 tion expressed by (102) is no longer possible. In the former case 
 the observations should be individually corrected before the mean 
 is taken. The same result might be obtained from equation (iii), 
 but the computation would not be simplified by its use. In the 
 latter case the several observations are affected by errors due to 
 the same causes but varying progressively in magnitude in response 
 to more or less continuous variations in the conditions under 
 which they are made. 
 
 In equations (ii) the c's having the same index may be con- 
 sidered to be due to the same cause, but to vary in magnitude 
 from equation to equation as indicated by the subscripts. The 
 arithmetical means of the errors due to the same causes are 
 
 , _ Ci' + C 2 ' + + C n ' 
 
 Cm '~ ~ 
 
 _ 
 
 Cm - 
 
 n 
 and the mean of the observations is 
 
 01 + 02 + ' ' ' 
 
 O m = 
 
 n 
 
 Substituting (ii) in (i) and taking account of the above relations 
 we have 
 
 X = O m + C m ' + C m " + ' ' ' + C w <> . (104) 
 
 Hence, in the case of systematic errors, the most probable value 
 of the numeric of the measured magnitude may be obtained from 
 the mean of the uncorrected observations by applying mean cor- 
 rections for the systematic errors. When all of the errors are 
 strictly constant equation (104) becomes identical with (102) 
 because all of the errors having the same index are equal. Obvi- 
 
ART. 68] DISCUSSION OF COMPLETED OBSERVATIONS 121 
 
 ously it also holds when part of the c's are strictly constant and the 
 remainder are systematic. 
 
 If we use the value of x given by (104) in place of that given 
 by (102) in the residual equations (103), the c's will not cancel. 
 Hence, if any of the constant errors are systematic in nature,. the 
 residuals computed from the o's and o m will be different from 
 those computed from the a's and x; and, consequently, they will 
 not be distributed in accordance with the law of accidental errors. 
 
 In practice it is generally advisable to correct each of the ob- 
 servations separately before taking the mean rather than to use 
 equation (104), since the true residuals are required in computing 
 the probable error of x, and they cannot be derived from the un- 
 corrected observations. Whenever possible the conditions should 
 be so chosen that systematic errors are avoided and then the 
 necessary computation can be made by equations (102) and (103). 
 
 68. Criteria of Accidental Errors. We have seen that the 
 residuals computed from observations affected by systematic errors 
 do not follow the law of accidental errors. Hence, if it can be 
 shown that the residuals computed from any given series of obser- 
 vations are distributed in conformity with the law of errors, it is 
 probable that the given observations are free from systematic 
 errors or that such errors are negligible in comparison with the 
 accidental errors. Observations that satisfy this condition may 
 or may not be free from strictly constant errors, but necessary 
 corrections can be made by equation (102) and the probable error 
 of the mean may be computed from the residuals given by 
 equation (103). 
 
 Systematic errors should be very carefully guarded against in 
 making the observations, and the conditions that produce them 
 should be constantly watched and recorded during the progress 
 of the work. After the observations have been completed they 
 should be individually corrected for all known systematic errors 
 before taking the mean. The strictly constant errors may then 
 be removed from the mean, but before this is done it is well to 
 compute the residuals and see if they satisfy the law of accidental 
 errors. If they do not, search must be made for further causes 
 of systematic error in the conditions surrounding the measure- 
 ments and a new series of observations should be made, under 
 more favorable conditions, whenever sufficient data for this pur- 
 pose is not available. 
 
122 THE THEORY OF MEASUREMENTS [ART. 68 
 
 Residuals, when sufficiently numerous, follow the same law of 
 distribution as the true accidental errors. Consequently system- 
 atic errors and mistakes might be detected by a direct comparison 
 of the actual distribution with the theoretical, as carried out in 
 article thirty-four, provided the number of observations is very 
 large. However, in most practical measurements, the residuals 
 are not sufficiently numerous to fulfill the conditions underlying 
 the law of errors, and a considerable difference between their 
 actual and theoretical distribution is quite as likely to be due to^ 
 this fact as to the presence of systematic errors. Whatever the 
 number of observations, a close agreement between theory and 
 practice is strong evidence of the absence of such errors but it is 
 seldom worth while to carry out the comparison with less than 
 one hundred residuals. 
 
 When the residuals are numerous and distributed in the same 
 manner as the accidental errors, the average error of a single 
 observation, computed by the formula 
 
 Vn(n- 1)' 
 and the mean error, computed by the formula 
 
 satisfy the relation 
 
 M = 1.253 A. 
 Also the formulae 
 
 E = 0.8453 A and E = 0.6745 M 
 
 give the same value for the probable error of a single observation. 
 When the number of observations is limited, exact fulfillment of 
 these relations ought not to be expected, but a large deviation 
 from them is strong evidence of the presence of systematic errors 
 or mistakes. Unless the number of observations is very small, 
 ten or less, the relations should be fulfilled within a few units in 
 the second significant figure, as is the case in the numerical example 
 worked out in article forty-two. 
 
 Obviously the arithmetical mean is independent of the order 
 in which the observations are arranged in taking it, but the order 
 of the residuals in regard to sign and magnitude depends on the 
 order of the observations. When there are systematic errors and 
 the observations are arranged in the order of progression of their 
 
ART. 68] DISCUSSION OF COMPLETED OBSERVATIONS 123 
 
 cause, the residuals will gradually increase or decrease in absolute 
 magnitude in the same order; and, if the systematic errors are 
 large in comparison with the accidental errors, there will be but 
 one change of sign in the series. Thus, if the temperature is 
 gradually rising while a length is being measured with a metallic 
 scale and the observations are arranged in the order in which they 
 are taken, the first half of them will be larger than the mean and 
 the last half smaller, except for the variations caused by accidental 
 errors. For the purpose of illustration, suppose that the observa- 
 tions are 
 
 1001.0; 1000.9; 1000.8; 1000.7; 1000.6; 1000.5; 1000.4. 
 The mean is 1000.7 and the residuals 
 
 + .3; +.2; +.1; .0; -.1; -.2; -.3 
 
 decrease in absolute magnitude from left to right, i.e., in the order 
 in which the observations were made. There are five cases in 
 which the signs of succeeding residuals are alike and one in which 
 they are different; the former cases will be called sign-follows and 
 the^latter a sign-change. This order of the residuals in regard to 
 magnitude and sign is typical of observations affected by sys- 
 tematic errors when they are arranged in conformity with the 
 changes in surrounding conditions. Since such changes are usually 
 continuous functions of the time, the required arrangement is 
 generally the order in which the observations are taken. 
 
 Such extreme cases as that illustrated above are seldom met 
 with in practice owing to the impossibility of avoiding accidental 
 errors of observation and the complications they produce in the 
 sequence of residuals. Generally the systematic errors that are 
 not readily discovered and corrected before making further re- 
 ductions are comparable in magnitude with the accidental errors. 
 Consequently they cannot control the sequence in the signs of 
 the residuals but they do modify the sequence characteristic of 
 true accidental errors. 
 
 In any extended series of observations there should be as many 
 negative residuals as positive ones, since positive and negative 
 errors are equally likely to occur. After any number of observations 
 have been made, the probability that the residual of the next obser- 
 vation will be positive is equal to the probability that it will be nega- 
 tive, since the possible number of either positive or negative errors 
 is infinite. Consequently the chance that succeeding residuals 
 
124 THE THEORY OF MEASUREMENTS [ART. 69 
 
 will have the same sign is equal to the chance that they will have 
 different signs. Hence, if the residuals are arranged in the order 
 in which the corresponding observations were made, the number 
 of sign-follows should be equal to the number of sign-changes. 
 
 The residuals, computed from limited series of observations, 
 seldom exhibit the theoretical sequence of signs exactly because 
 they are not sufficiently numerous to fulfill the underlying condi- 
 tions. Nevertheless, a marked departure from that sequence 
 suggests the presence of systematic errors or mistakes and should 
 lead to a careful scrutiny of the observations and the conditions 
 under which they were made. If the disturbing causes cannot be 
 detected and their effects eliminated, it is generally advisable to 
 repeat the observations under more favorable conditions. The 
 numerical example, worked out in article forty-two, may be cited 
 as an illustration from practice. The observations were made in 
 the order in which they are tabulated, beginning at the top of the 
 first column and ending at the bottom of the fourth column. In 
 the second and fifth columns we find ten positive and ten negative 
 residuals. The number of sign-follows is ten and the number of 
 sign-changes is nine. This is rather better agreement with the 
 theoretical sequence of signs than is usually obtained with so few 
 residuals. It indicates that the observations were made under 
 favorable conditions and are sensibly free from systematic errors 
 but it gives no evidence whatever that strictly constant errors 
 are absent. 
 
 Although the foregoing criteria of accidental errors are only 
 approximately fulfilled when the number of observations is lim- 
 ited, their application frequently leads to the detection and elimi- 
 nation of unforeseen systematic errors. The first method is rather 
 tedious and of little value when less than one hundred obser- 
 vations are considered, but the last two methods may be easily 
 carried out and are generally exact enough for the detection of 
 systematic errors comparable in magnitude with the probable error 
 of a single observation. 
 
 69. Probability of Large Residuals. In discussing the dis- 
 tribution of residuals in regard to magnitude, the words large and 
 small are used in a comparative sense. A large residual is one that 
 is large in comparison with the majority of residuals in the series 
 considered. Thus, a residual that would be classed as large in a 
 series of very precise observations would be considered small in 
 
ART. 69] DISCUSSION OF COMPLETED OBSERVATIONS 125 
 
 dealing with less exact observations. Consequently, in expressing 
 the relative magnitudes of residuals, it is customary to adopt a 
 unit that depends on the precision of the measurements considered. 
 The probable error of a single observation is the best magnitude 
 to adopt for this purpose, since it is greater than one-half of the 
 errors and less than the other half. If we represent the relative 
 magnitude of a given error by S, the actual magnitude by A, and 
 the probable error of a single observation by E, 
 
 S = |- (105) 
 
 The relative magnitudes of the residuals may be represented in 
 the same way by replacing the error A by the residual r. It is 
 obvious that values of S less than unity correspond to small re- 
 siduals and values greater than unity to large residuals in any 
 series of observations. 
 
 In equation (13), article thirty-three, the probability that an 
 error chosen at random is less than a given error A is expressed 
 
 by the integral 
 
 */~ A 
 o / v j 
 
 PA = -^= e-*dt. (13) 
 
 V-n-Jo 
 
 Equation (25), article thirty-eight, may be put in the form 
 
 V ** k 
 
 & = 7= -> 
 
 VTT a? 
 
 where $ is written for the numerical constant 0.47694. Hence, 
 introducing (105), 
 
 and (13) becomes 
 
 P 8 = 'eft. (106) 
 
 Obviously this integral expresses the probability that an error 
 chosen at random is less than S times the probable error of a 
 single observation. It is independent of the particular series to 
 which the observations belong and its values, corresponding to 
 a series of values of the argument S, are given in Table XII. 
 
 Since all of the errors in any system are less than infinity, Poo 
 is equal to unity. Hence the probability that a single error, 
 
126 
 
 THE THEORY OF MEASUREMENTS [ART. 69 
 
 chosen at random, is greater than S times E is given by the rela- 
 tion 
 
 Qs = 1 - Pa- (V) 
 
 Now the residuals, when sufficiently numerous and free from 
 systematic errors and mistakes, should follow the same distri- 
 bution as the accidental errors. Hence, if n s is the number of 
 residuals numerically greater than SE and N is the total number 
 in any series of observations, we should have 
 
 Qs = T?" (vi) 
 
 Since the numerical value of P 8 , and consequently that of Q 8 
 depends only on the limit S and is independent of the precision 
 
 of the particular series of measurements considered, the ratio jj. > 
 
 corresponding to any given limit S, should be the same in all 
 cases. Consequently, if N observations have been made on any 
 magnitude and by any method whatever, n 8 of them should corre- 
 spond to residuals numerically greater than SE. Conversely, if 
 we assign any arbitrary number to n a , equation (vi) defines the 
 number of observations that we should expect to make without 
 exceeding the assigned number of residuals greater than SE. 
 Hence, if N a is the number of observations among which there 
 should be only one residual greater than S times the probable 
 error of a single observation, we have, by placing n s equal to 
 one in (vi), and substituting the value of Q 8 from (v), 
 
 *--r^>r (107) 
 
 The fourth column of the following table gives the values of N a , 
 to the nearest integer, corresponding to the integral values of the 
 limit S given in the first column. The values of P 8 in the second 
 column are taken from Table XII, and those of Q 8 in the third 
 column are computed by equation (v). 
 
 S 
 
 P. 
 
 e. 
 
 N s 
 
 1 
 
 0.50000 
 
 0.50000 
 
 2 
 
 2 
 
 0.82266 
 
 0.17734 
 
 6 
 
 3 
 
 0.95698 
 
 0.04302 
 
 23 
 
 4 
 
 0.99302 
 
 0.00698 
 
 143 
 
 5 
 
 0.99926 
 
 0.00074 
 
 1351 
 
ART. 70] DISCUSSION OF COMPLETED OBSERVATIONS 127 
 
 To illustrate the significance of this table, suppose that 143 
 direct observations have been made on any magnitude by any 
 method whatever. The probable error E of a single observation 
 in this series may be computed from the residuals by equation (31) 
 or (34). Then, if the residuals follow the law of errors, not more 
 than one of them should be greater than four times as large as E. 
 If the number of observations had been 1351, we should expect 
 to find one residual greater than five times E, and on the other 
 hand if the number had been only twenty-three, not more than 
 one residual should be greater than three times E. 
 
 Although the probability for the occurrence of large residuals 
 is small, and very few of them should occur in limited series 
 of observations, their distribution among the observations, in 
 respect to the order in which they occur, is entirely fortuitous. 
 A large residual is as likely to occur in the first, or any other, 
 observation of an extended series as in the last observation. Con- 
 sequently the limited series of observations, taken in practice, 
 frequently contain abnormally large residuals. This is not due 
 to a departure from the law of errors, but to a lack of sufficient 
 observations to fulfill the theoretical conditions. In such cases 
 there are not enough observations with normal residuals to balance 
 those with abnormally large ones. Consequently a closer approxi- 
 mation to the arithmetical mean that would have been obtained 
 with a more extended series of observations is obtained when the 
 abnormal observations are rejected from the series before taking 
 the mean. 
 
 Observations should not be rejected simply because they show 
 large residuals, unless it can be shown that the limit set by the 
 theory of errors, for the number of observations considered, is 
 exceeded. This can be judged approximately by comparing the 
 residuals of the given observations with the numbers given in the 
 first and last columns of the above table, but a more rigorous test 
 is obtained by applying Chauvenet's Criterion, as explained in the 
 following article. 
 
 70. Chauvenet's Criterion. The probability that the error 
 of a single observation, chosen at random, is less than SE is 
 expressed by P a in equation (106). Now, the taking of N inde- 
 pendent observations is equivalent to N selections at random from 
 the infinite number of possible accidental errors. Hence, by 
 equation (7), article twenty-three, the probability that each of 
 
128 THE THEORY OF MEASUREMENTS [ART. 70 
 
 the N observations in any series is affected by an error less than 
 SE is equal to P N . Since all of the N errors must be either greater 
 or less than SE } the probability that at least one of them is greater 
 than this limit is equal to 1 P 8 N . Placing this probability 
 equal to one-half, we have 
 
 i - P." = i, 
 
 or 
 
 P. - (1 - (vii) 
 
 If the limit S is determined by this equation, there is an even 
 chance that at least one of the N observations is affected by an 
 error greater than SE. 
 
 Expanding the second member of (vii) by the Binomial Theorem 
 
 11 N -I I (N- l)(2N-l) 1 
 
 N 2 1-2-N 2 4 1-2- 3- N* 8 
 
 1-2-3 . . . K-N K 
 
 The terms of this series decrease very rapidly and all but the first 
 are negative. Consequently the sum of the terms beyond the 
 second is small in comparison with the other two; and, whatever 
 
 the value of N, (1 %) N is nearly equal to, but always slightly 
 less than, - ^-^ - . Since P 8 and S increase together, the limit 
 T determined by the relation 
 
 2N-1 
 
 2N 
 
 (108) 
 
 is slightly greater than the limit S determined by (vii). Hence, 
 if N independent direct observations have been made, the prob- 
 ability against the occurrence of a single error greater than 
 
 A r = TE (109) 
 
 is greater than the probability for its occurrence. Consequently, 
 if the given series contains a residual greater than A r , the prob- 
 able precision of the arithmetical mean is increased by excluding 
 the corresponding observation. 
 
ART. 70] DISCUSSION OF COMPLETED OBSERVATIONS 129 
 
 Equations (108) and (109) express Chauvenet's Criterion for the 
 rejection of doubtful observations. In applying them, the prob- 
 able error E of a single observation is first computed from the 
 residuals of all of the observations by either equation (31) or the 
 first of equations (34) with the aid of Table XIV or XV. If any 
 of the residuals appear large in comparison with the computed 
 value of E, PT is determined from (108) by placing N equal to 
 the number of observations in the given series. T is then obtained 
 by interpolation from Table XII, and finally A r is computed by 
 (109). If one or more of the residuals are greater than the com- 
 puted A r , the observation corresponding to the largest of them is 
 excluded from the series and the process of applying the criterion is 
 repeated from the beginning. If one or more of the new residuals 
 are greater than the new value of A r , the observation correspond- 
 ing to the largest of them is rejected. This process is repeated 
 and observations rejected one at a time until a value of A r is ob- 
 tained that is greater than any of the residuals. 
 
 When more than one residual is greater than the computed 
 value of Ay, only the observation corresponding to the largest 
 of them should be rejected without further study. The rejection 
 of a single observation from the given series changes the arith- 
 metical mean, and hence all of the residuals and the value of E 
 computed from them. If r and r' are the residuals corresponding 
 to the same observation before and after the rejection of a more 
 faulty observation, and if A r and A r ' are the corresponding 
 limiting errors, it may happen that r' is less than A/, although r 
 is greater than Ay. Hence the second application of the criterion 
 may show that a given observation should be retained notwith- 
 standing the fact that its residual was greater than the limiting 
 error in the first application, provided an observation with a 
 larger residual was excluded on the first trial. 
 
 To facilitate the computation of Ay, the values of T corre- 
 sponding to a number of different values of N have been 
 interpolated from Table XII and entered in the second column 
 of Table XIII. 
 
 For the purpose of illustration, suppose that ten micrometer 
 settings have been made on the same mark and recorded, to the 
 nearest tenth of a division of the micrometer head, as in the first 
 column of the following table. 
 
130 
 
 THE THEORY OF MEASUREMENTS [ART. 71 
 
 Obs. 
 
 r 
 
 r' 
 
 2.567 
 
 +0.0118 
 
 
 2.559 
 
 +0.0038 
 
 +0.0051 
 
 2.556 
 
 +0.0008 
 
 +0.0021 
 
 2.552 
 
 -0.0032 
 
 -0.0019 
 
 2.551 
 
 -0.0042 
 
 -0.0029 
 
 2.553 
 
 -0.0022 
 
 -0.0009 
 
 2.555 
 
 -0.0002 
 
 +0.0011 
 
 2.548 
 
 -0.0072 
 
 -0.0059 
 
 2.554 
 
 -0.0012 
 
 +0.0001 
 
 2.557 
 
 +0.0018 
 
 +0.0031 
 
 x =2.5552 
 
 [r] = 0.0364 
 
 [r>] = 0.0231 
 
 z'=2.5539 
 
 # = 0.0032 
 
 #' = 0.0023 
 
 
 IF = 2. 91 
 
 T' = 2.84 
 
 
 Ar = 0.0093 
 
 A/ = 0.0065 
 
 The residuals, computed from the mean x, are given under r. 
 The probable error E } computed from [r] by the first of equations 
 (34), with the aid of Table XV, is 0.0032. The value of T corre- 
 sponding to ten observations is 2.91 from Table XIII, and the 
 limiting error Ay is equal to 0.0093. Since this is less than the 
 residual 0.0118, the corresponding observation (2.567) should be 
 rejected from the series. 
 
 The mean of the retained observations, xi, is 2.5539, and the 
 corresponding residuals are given under r' in the third column of 
 the above table. The new value of the limiting error (A/), com- 
 puted by the same method as above, is 0.0065. Since none of 
 the new residuals are larger than this, the nine observations left 
 by the first application of the criterion should all be retained. 
 
 71. Precision of Direct Measurements. The first step in 
 the reduction of a series of direct observations is the correction 
 of all known systematic errors and the test of the completeness of 
 this process by the criteria of article sixty-eight. In general, the 
 systematic errors represent small variations of otherwise constant 
 errors; and, in making the preliminary corrections, it is best to 
 consider only this variable part, i.e., the corrections are so applied 
 that all of the corrected observations are left with exactly the 
 same constant errors. Thus, suppose that the temperature of a 
 scale is varying slowly during a series of observations, and is 
 never very near to the temperature at which the scale is standard. 
 It is better to correct each observation to the mean temperature 
 of the scale and leave the larger correction, from mean to standard 
 
ART. 71] DISCUSSION OF COMPLETED OBSERVATIONS 131 
 
 temperature, until it can be applied to the arithmetical mean in 
 connection with the corrections for other strictly constant errors. 
 This is because the systematic variations in the length of the 
 scale are so small that the unavoidable errors in the observed 
 temperatures and the adopted coefficient of expansion of the scale 
 can produce no appreciable effect on the corrections to mean 
 temperature. The effect of these errors on the larger correction 
 from mean to standard temperature is more simply treated in 
 connection with the arithmetical mean than with the individual 
 observations. 
 
 Let 01, 02, . . . , o n represent a series of direct observations 
 corrected for all known systematic errors and satisfying the 
 criteria of accidental errors. We have seen that the most prob- 
 able value that we can assign to the numeric of the measured mag- 
 nitude, on the basis of such a series, is given by the relation 
 
 x = o m + c'+c"+ - +cfe>, (102) 
 
 where o m is the arithmetical mean of the o's, and the c's represent 
 corrections for strictly constant errors. If the c's could be deter- 
 mined with absolute accuracy, or even within limiting errors that 
 are negligible in comparison with the accidental errors of the o's, 
 the only uncertainty in the above expression for x would be that 
 due to the accidental error of o m . Hence, by equations (103), if 
 E x and E m are the probable errors of x and o m , respectively, we 
 should have 
 
 *. = *_ = 0.674 Vy '.' (HO) 
 
 . . 
 
 If we follow the usual practice and regard the probable error of a 
 quantity as a measure of the accidental errors of the observations 
 from which it is directly computed, equation (110) still holds 
 when the accidental errors of the c's are not negligible; but, as we 
 shall see, E x is no longer a complete measure of the precision of x 
 in such cases. 
 
 In practice each of the c's must be computed, on theoretical 
 grounds, from subsidiary observations with the aid of physical 
 constants that have been previously determined by direct or 
 indirect measurements. For the sake of brevity the quantities 
 on which the c's depend will be called correction factors. Since all 
 of them are subject to accidental errors, the computed c's are 
 affected by residual errors of indeterminate sign and magnitude. 
 
132 THE THEORY OF MEASUREMENTS [ART. 71 
 
 When the probable errors of the correction factors are known the 
 probable errors of the c's may be computed by the laws of propa- 
 gation of errors with the aid of the correction formulae by which 
 the c's are determined. 
 
 Equation (102) gives x as a continuous sum of o m and the c's. 
 Consequently, if we represent the probable errors of the c's by 
 Ei t E 2 , . . . , E q , respectively, we have by equation (76), article 
 fifty-eight, 
 
 R x 2 = E m * + Ei* + +E q *, (111) 
 
 wnere R x is the resultant probable error of x due to the correspond- 
 ing errors of o m and the c's. To distinguish R x from the probable 
 error E X) which depends only on the accidental error of o m , we 
 shall call it the precision measure of x. 
 
 Although equation (111) is simple in form, the separate compu- 
 tation of the E'SJ from the errors of the correction factors on which 
 they depend, is frequently a tedious process. Moreover several 
 of the c's may depend on the same determining quantities. Con- 
 sequently the computation of x and R x is frequently facilitated by 
 bringing the correction factors into the equation for x explicitly, 
 rather than allowing them to remain implicit in the c's. Thus, 
 if a, )8, . . . , p represent the correction factors on which the c's 
 depend, equation (102) may be put in the form 
 
 x = F(o m ,a,0, . . . , P). (112) 
 
 Hence, by equation (81), article sixty, 
 
 where E a , Ep, etc., are the probable errors of a, ft, etc. 
 
 For example, suppose that o m represents the mean of a num- 
 ber of observations of the distance between two parallel lines 
 expressed in terms of the divisions of the scale used in making 
 the measurements. Let t\ represent the mean temperature of the 
 scale during the observations; L the mean length of the scale 
 divisions at the standard temperature U, in terms of the chosen 
 unit; a the coefficient of expansion of the scale; and ft the angle 
 between the scale and the normal to the lines. Then, if the 
 individual observations have been corrected to mean temperature 
 ti before computing the mean observation o m , the best approxima- 
 
ART. 71] DISCUSSION OF COMPLETED OBSERVATIONS 133 
 
 tion that we can make to the true distance between the lines is 
 given by the expression 
 
 x = o m L\l]+a(ti - t ) I , 
 
 in which the correction factors L, a, /?, fa, and to appear explicitly , 
 as in the general equation (112). A more detailed discussion of 
 this example will be found in article seventy-three. 
 
 If we represent the separate effects of the errors E m , E a , . . . , 
 E p on the error R x by D m , D a , D$, . . . , D PJ respectively, we 
 have 
 
 *-*/ D - - S E *-> :.:i ' D > * T P E < m > 
 
 and (113) becomes 
 
 R* 2 = D m * + D a 2 + Df + - - - + D P 2 . (115) 
 
 In some cases the fractional effects 
 
 _Drn, _D. . _D, 
 
 m ~ x ' a ~ x ' ' ' ' p ~ x 
 
 can be more easily computed numerically than the corresponding 
 D's. When this occurs, the fractional precision measure 
 
 is first computed and then R x is determined by the relation 
 
 R x = x-P x . (117) 
 
 While equations (112) to (117) are apparently more complicated 
 than (102) and (111), they generally lead to more simple numerical 
 computations. Moreover the probable errors of some of the 
 correction factors are frequently so small that they produce no 
 appreciable effect on R x . When either equation (115) or (116) is 
 used, such cases are easily recognized because the corresponding 
 D's or P's are negligible in comparison with D m or P m . Obvi- 
 ously the same condition applies to the E's in equation (111), but 
 the numerical computation of either the D's or the P's is generally 
 more simple than that of the E's in (111) because approximate 
 values of o m and the correction factors may be used in evaluat- 
 ing the differential coefficients in (114). The allowable degree of 
 approximation, the limit of negligibility of the D's, and some other 
 
134 THE THEORY OF MEASUREMENTS [ART. 71 
 
 details of the computation will be discussed more extensively 
 in the next article. 
 
 If the true numeric of the measured magnitude is represented 
 by Xj the final result of a series of direct measurements may be 
 expressed in the form 
 
 X = xR x , (118) 
 
 where x is the most probable value that can be assigned to X on 
 the basis of the given observations, and R x is the precision measure 
 of x. In practice x may be computed by either equation (102) 
 or (112), or the arithmetical mean of the individually corrected 
 observations may be taken, and R x is given by equations (111), 
 (115), or (117), the choice of methods depending on the nature 
 of the given data and the preference of the computer. 
 
 The exact significance of equation (118) should be carefully 
 borne in mind, and it should be used only when the implied condi- 
 tions have been fulfilled. Briefly stated, these conditions are as 
 follows : 
 
 1st. The accidental errors of the observations on which x 
 depends follow the general law of such errors. 
 
 2nd. A careful study of the methods and apparatus used has 
 been made for the purpose of detecting all sources of constant 
 or systematic errors and applying the necessary corrections. 
 
 3rd. The given value of x is the most probable that can be 
 computed from the observations after all constant errors, system- 
 atic errors, and mistakes have been as completely removed as 
 possible. 
 
 4th. The resultant effect of all sources of error, whether acci- 
 dental errors of observation or residual errors left by the correc- 
 tions for constant errors, is as likely to be less than R x as greater 
 than R x . 
 
 The expressions in the form X = x E x , used in preceding 
 chapters, are not violations of the above principles because, in 
 those cases, we were discussing only the effects of accidental 
 errors and the observations were assumed to be free from all con- 
 stant errors and mistakes. Such ideal conditions never occur in 
 practice. Consequently R x should not be replaced by E x in 
 expressing the result of actual measurements in the form of equa- 
 tion (118), unless it can be shown by equation (115), and the given 
 data that the sum of the squares of the D's corresponding to all 
 of the correction factors is negligible in comparison with Z) m 2 . 
 
ART. 72] DISCUSSION OF COMPLETED OBSERVATIONS 135 
 
 In the latter case E x and R x are identical as may be easily seen 
 by comparing equations (110), (111), and (115). 
 
 72. Precision of Derived Measurements. When a desired 
 numeric Z is connected with the numerics Xi, X 2 , . . . , X q 
 of a number of directly measured magnitudes by the relation 
 
 XQ = F (Xi, X%, . . . , X q ), 
 
 the most probable value that we can assign to X Q is given by the 
 expression 
 
 x = F(x 1 ,xt, . . . , x q ), (119) 
 
 where the x's are the most probable values of the X's with corre- 
 sponding subscripts. Each of the component x's, together with 
 its precision measure, can be computed by the methods of the pre- 
 ceding article. The precision measure of X Q may be computed 
 with the aid of equation (81), article sixty, by replacing the E's in 
 that equation by the R's with corresponding subscripts. 
 
 Sometimes the numerical computations are simplified and the 
 discussion is clarified by bringing the direct observations and the 
 correction factors explicitly into the expression for XQ. If o a , 
 Ob, . . . , Op are the arithmetical means of the direct observa- 
 tions, after correction for systematic errors, on which Xi, x z , . . . , 
 x q respectively depend, and a, /?, . . . , p are the correction 
 factors involved in the constant errors of the observations, equa- 
 tion (119) may be put in the form 
 
 x = d (o a , o b , . . . , o p , a, j8, . . . , p). (120) 
 
 The function 6 is always determinable when the function F in 
 (119) is given and the correction formulae for the constant errors 
 are known. 
 
 Representing the precision measure of XQ by R , and adopting 
 an obvious extension of the notation of the preceding article, we 
 have, by equation (81), 
 
 Introducing the separate effects of the E's, 
 
 *-*' ' ' ' = *=l^' 
 
 (121) becomes 
 
 *' ' ' ' ; '-*- (122) 
 
 . (123) 
 
136 THE THEORY OF MEASUREMENTS [ART. 72 
 
 The fractional effects of the E's are 
 
 P _. . P =5*. P = ^. . P _A? 
 
 ^ " XQ ' ' p x ' a Z ' p " X Q ' 
 
 and the fractional precision measure of x is given by the relation 
 
 XQ 
 
 When the numerical computation of the P's is simpler than that 
 of the D's, PO is first computed by equation (124) and then RQ 
 is determined by the relation 
 
 #o = z Po. (125) 
 
 The expression of the final result of the observations and com- 
 putations in the form 
 
 XQ = XQ RQ 
 
 has exactly the same significance with respect to X Q , XQ, and R Q 
 that (118) has with respect to X, x, and R x . It should not be 
 used until all of the underlying conditions have been fulfilled as 
 pointed out in the preceding article. Confusion of the precision 
 measure R with the probable error E 0) and insufficient rigor in 
 eliminating constant errors have led many experimenters to an 
 entirely fictitious idea of the precision of their measurements. 
 
 When the correction factors are explicitly expressed in the 
 reduction formulae, as in equations (112) and (120), the only 
 difference between the expressions for direct and derived measure- 
 ments is seen to lie in the greater number of directly observed 
 quantities, o a , o&, etc., that appear in the latter equation. The 
 same methods of computation are available in both cases and the 
 following remarks apply equally well to either of them. 
 
 For practical purposes, the precision measure R is computed 
 to only two significant figures and the corresponding x is carried 
 out to the place occupied by the second significant figure in R. 
 The reasons underlying this rule have been fully discussed in 
 article forty-three, in connection with the probable error, and 
 need not be repeated here. In computing the numerical value 
 of the differential coefficients in equations (113), (114), (121), and 
 (122), the observed components, o m , o a , o&, etc., and the correc- 
 tion factors, a, , etc., are rounded to three significant figures, 
 and those that affect the result by less than one per cent are neg- 
 lected. This degree of approximation will always give R within 
 
ART. 72] DISCUSSION OF COMPLETED OBSERVATIONS 137 
 
 one unit in the second significant figure and usually decreases the 
 labor of computation. 
 
 Generally the components o m , o a , o b , etc., represent the arith- 
 metical means of series of direct observations that have been 
 corrected for systematic errors. In such cases the corresponding 
 probable errors E mt E a , Eb, etc., can be computed, by equations 
 in the form of (110), from the residuals determined by equations 
 in the form of (103), with the aid of the observations on which 
 the o's depend. If the observations are sufficiently numerous, 
 the computation of the .27's.may be simplified by using formulae 
 depending on the average error in the form 
 
 E = 0.845 fl=> (34) 
 
 n Vn 1 
 
 where [f] is the sum of the residuals without regard to sign and n 
 is the number of observations. If the observations on which any 
 of the o's depend are not of equal weight, the general mean should 
 be used in place of the arithmetical mean and the corresponding 
 probable errors should be computed by equations (41), (42), or 
 (44), depending on the circumstances of the observations. 
 
 The o's in equation (120) are supposed to represent simultane- 
 ous values of the directly observed magnitudes. When any of 
 these quantities are continuous functions of the time, or of any 
 other independent variables, it frequently happens that only a 
 single observation can be made on them that is simultaneous 
 with the other components. In such cases this single observation 
 must be used in place of the corresponding o in (120), and its 
 probable error must be determined for use in equation (122). 
 For the latter purpose, it is sometimes possible to make an auxil- 
 iary series of observations under the same conditions that pre- 
 vailed during the simultaneous measurements except that the 
 independent variables are controlled. The required E may be 
 assumed to be equal to the probable error of a single observation 
 in the auxiliary series. Consequently it may be computed by 
 formulae in the form, 
 
 E = 0.674* /W 
 E = 0.845 
 
 n- I 
 or 
 
 [r] 
 
138 THE THEORY OF MEASUREMENTS [ART. 72 
 
 where n is the number of auxiliary observations, and the r's are 
 the corresponding residuals. In some cases this simple expedient 
 is not available; and approximate values must be assigned to the 
 E's on theoretical grounds, depending on the nature of the meas- 
 urements; or more or less extensive experimental investigations 
 must be undertaken to determine their values more precisely. 
 
 Such investigations are so various in character and their utility 
 depends so much on the skill and ingenuity of the experimenter, 
 that a detailed general discussion of them would be impossible. 
 They may be illustrated by the following very common case. 
 Suppose that one of the components in equation (120) repre- 
 sents the gradually changing temperature of a bath. In com- 
 puting x Q we must use the thermometer reading o t taken at the 
 time the other components are observed. The errors of the fixed 
 points of the thermometer and its calibration errors enter the 
 equation among the correction factors a, /?, etc., and do not con- 
 cern us in the present discussion. In order to determine the 
 probable error of o t , the temperature of the bath may be caused 
 to rise uniformly, through a range that includes o t , by passing a 
 constant current through an electric heating coil, or the bath 
 may be allowed to cool off gradually by radiation. In either case 
 the rate of change of temperature should be nearly the same as 
 prevailed when o t was observed. A series of corresponding obser- 
 vations of the time T and the temperature t are made under 
 these conditions, and the empirical relation between T and t is 
 determined graphically or by the method of least squares. The 
 probable error of o t may be assumed to be equal to the probable 
 error of a single observation of t in this series, and may be com- 
 puted by equation (94), article sixty-four. 
 
 Some of the correction factors a, ft, etc., appearing as com- 
 ponents in equations (112) and (120), represent subsidiary obser- 
 vations, and some of them represent physical constants. The 
 subsidiary observations may be treated by the methods outlined 
 above. When the highest attainable precision is desired, the 
 physical constants, together with their probable errors, must be 
 determined by special investigation. In less exact work they 
 may be taken from tables of physical constants. Such tabular 
 values seldom correspond exactly to the conditions of the experi- 
 ments in hand and their probable errors are seldom given. 
 Generally a considerable range of values is given, and, unless 
 
ART. 72] DISCUSSION OF COMPLETED OBSERVATIONS 139 
 
 there is definite reason in the experimental conditions for the 
 selection of a particular value, the mean of all of them should be 
 adopted and its probable error placed equal to one-half the range 
 of the tabular values. The deviations of the tabular values from 
 the mean are due more to differences in experimental conditions 
 and in the material treated than to accidental errors. Conse- 
 quently a probable error calculated from the deviations would 
 have no significance unless these differences could be taken into 
 account. The selection of suitable values from tables of physical 
 constants requires judgment and experience, and the general 
 statements above should not be blindly followed. In many cases 
 the original sources of the data must be consulted in order to 
 determine the values that most nearly satisfy the conditions of 
 the experiments in hand. 
 
 In good practice the conditions of the experiment are usually 
 so arranged that the D's, in equation (123), corresponding to the 
 direct observations o a , o&, etc., are all equal. None of the D's 
 corresponding to correction factors should be greater than this 
 limit, but it sometimes happens that some of them are much 
 smaller. Since R is to be computed to only two significant 
 figures, any single D which is less than one-tenth of the average 
 of the other D's may be neglected in the computation. If the 
 sum of any number of D's is less than one-tenth of the average 
 of the remaining D's they may all be neglected. A somewhat 
 more rigorous limit of rejection can be developed for use in plan- 
 ning proposed measurements, but it is scarcely worth while in 
 the present connection since the correction factors and all other 
 quantities must be taken as they occurred in the actual measure- 
 ments, and negligible D's are very easily distinguished by inspec- 
 tion after a little experience. 
 
 After #o has been determined, x may be computed by either 
 equation (119) or (120). If (119) is used the x's must first be 
 determined by (102) or (112). Sometimes the computation may 
 be facilitated by using a modification of (120), in which some of 
 the correction factors appear explicitly while others are allowed 
 to remain implicit in the z's to which they apply. Such cases 
 cannot be treated generally, but must be left to the ingenuity of 
 the computer. Whatever formula is used, the observed quanti- 
 ties and the correction factors should be expressed by sufficient 
 significant figures to give the computed X Q within a few units in 
 
140 THE THEORY OF MEASUREMENTS [ART. 73 
 
 the place occupied by the second significant figure of R . Occa- 
 sionally the total effect of one or more of the correction factors is 
 less than this limit and may be neglected in the computation. For 
 
 f$ W 7? 
 
 a single factor, say a, this is the case when a is less than ~ 
 
 73. Numerical Example. The following illustration repre- 
 sents a series of measurements taken for the purpose of cali- 
 brating the interval between the twenty-fifth and seventy-fifth 
 graduations on a steel scale supposed to be divided in centimeters. 
 The observations were made with a cathetometer provided with 
 a brass scale and a vernier reading to one one-thousandth of a 
 division. One division of the level on this instrument corre- 
 sponds to an angular deviation of 3 X 10~ 4 radians, and the ad- 
 justments were all well within this limit. The steel scale was 
 placed in a vertical position with the aid of a plumb-line, and, 
 since a deviation of one-half, millimeter per meter could have 
 been easily detected, the error of this adjustment did not exceed 
 5 X 10~ 4 radians. Consequently the angle between the two 
 scales was not greater than 8 X 10~ 4 radians, and it may have 
 been much smaller than this. The temperature of the scales was 
 determined by mercury in glass thermometers hanging in loose 
 contact with them. The probable error of these determinations 
 was estimated at five-tenths of a degree centigrade, due partly 
 to looseness of contact and partly to an imperfect knowledge of 
 the calibration errors of the thermometers. 
 
 Twenty independent observations, when tested by the last 
 two criteria of article sixty-eight, showed no evidence of the pres- 
 ence of systematic errors or mistakes. Consequently the mean 
 o m , in terms of cathetometer scale divisions, and its probable 
 error E m were computed before the removal of constant errors. 
 The following numerical data represents the results of the obser- 
 vations and the known calibration constants of the cathetometer. 
 
 Mean temperature of the steel scale, T 20 0.5 C. 
 
 Mean temperature of the brass scale, ti 21.3 =t 0.5 C. 
 
 Mean of twenty observations on the measured 
 
 interval in terms of brass scale divisions, o m . . 50.0051 db 0.0015 scale div. 
 Mean length, at standard temperature, of the 
 
 brass scale divisions in the interval used, S. . 0.999853 d= 0.000024 cm. 
 
 Standard temperature of brass scale, t 15.0 C. 
 
 Coefficient of linear expansion of brass scale, a. (182 12) X 10~ 7 . 
 
 Angle between two scales, /3, less than 8 X 10- 4 rad. 
 
ART. 731 DISCUSSION OF COMPLETED OBSERVATIONS 141 
 
 The most probable value that can be assigned to the measured 
 interval is given by the expression 
 
 Since ft is a very small angle, -- - may be treated by the approxi- 
 
 COS p 
 
 mate formulae of Table VII, and the above expression becomes 
 
 where 
 
 t = fa-to. 
 
 The quantity S (1 -f- at) is very nearly equal to unity. Hence, 
 neglecting small quantities of the second and higher orders, the 
 correction due to the angle ft is 
 
 < 0.000016. 
 
 Since this is less than two per cent of the probable error of o m , it is 
 negligible in comparison with the accidental errors of observation. 
 Consequently the precision of x is not increased by retaining the 
 term involving ft, and we may put 
 
 x = OmS (1 + at). (a) 
 
 The probable error of t Q is zero, because the accidental errors of 
 the temperature observations, made during the calibration of the 
 brass scale, are included in the probable errors of S and a com- 
 puted by the method of article sixty-five. Consequently the 
 probable error of t is equal to that of fa, and we have 
 
 t = 6,3 0.5 C. 
 
 In the present case equation (115) is the most convenient for 
 computing the precision measure ,.R X of x. Only two significant 
 figures are to be retained in the separate effects computed by 
 equation (114). Consequently the factor (1 + at) may be taken 
 equal to unity, and the numerical values of o m and S may be 
 rounded to three significant figures for the purpose of this com- 
 putation. Thus, taking o m equal to 50.0, S equal to 1.00, and 
 the other data as given above, we have 
 
142 THE THEORY OF MEASUREMENTS [ART. 73 
 
 D m = -E m = S(l+ at) E m = 1 X E m = 0.0015. 
 
 oo m 
 
 D,= ~ Q E t =o m (l + at) E,= 50 X E a = 0.0012. 
 
 do 
 
 D a =~E a = OmStE a = 50 X 6.3 X E a = 0.00038. 
 
 da 
 
 m =50 X 182 X 10~ 7 X E t = 0.00046. 
 
 ot 
 
 D m 2 = 225.0 X 10~ 8 
 A, 2 = 144.0 X 10~ 8 
 Z> 2 = 14.4 X 10~ 8 
 A 2 = 21.2 X 10~ 8 
 [D 2 ] = 404.6 X 10~ 8 
 Hence, by equation (115), 
 
 R x *= [D 2 ] = 404.6 X 10- 8 , 
 
 JB X = V404.6 X 10- 8 = 0.0020. 
 
 For the purpose of computing x, it is convenient to put the 
 given data in the form 
 
 Om = 50 (1+0.000102), 
 S = 1- 0.000147, 
 at = 0.000115. 
 
 Then, by equation (a), 
 
 x = 50 (1 + 0.000102) (1 - 0.000147) (1 + 0.000115), 
 and by formula 7, Table VII, 
 
 x = 50 (1 + 0.000102 - 0.000147 + 0.000115) 
 = 50 (1 + 0.00007) 
 = 50.0035. 
 
 This method of computation, by the use of the approximate 
 formulae of Table VII, gives x within less than one unit in the last 
 place held, and is much less laborious than the use of logarithms. 
 Since the length S of the cathetometer scale divisions is given 
 in centimeters, the computed values of x and R x are also expressed 
 in centimeters and our uncertainty regarding the true distance L 
 between the twenty-fifth and the seventy-fifth graduations of the 
 steel scale is definitely stated by the expression 
 
 L = 50.0035 d= 0.0020 centimeters, 
 at the temperature 
 
 T r = 20.00.5C. 
 
ART. 73] DISCUSSION OF COMPLETED OBSERVATIONS 143 
 
 The above discussion shows that the precision of the result 
 would not have been materially increased by a more accurate 
 determination of T, fa, and a, since the effects of the errors of 
 these quantities are small in comparison with that of the errors 
 of o m and S. The probable error of o m might have been reduced 
 by making a larger number of observations and taking care to 
 keep the instrument in adjustment within one-tenth of a level 
 division or less. But the given value of E m is of the same order 
 of magnitude as the least count of the vernier used, and, since 
 each observation represents the difference of two scale readings, 
 it would not be decreased in proportion to the increased labor of 
 observation. Moreover, the terms D m and D 8 in the above value 
 of R x are nearly equal in magnitude, and it would not be worth 
 while to devote time and labor to the reduction of one of them 
 unless the other could be reduced in like proportion. 
 
CHAPTER XI. 
 DISCUSSION OF PROPOSED MEASUREMENTS. 
 
 74. Preliminary Considerations. The measurement of a 
 given quantity may generally be carried out by any one of several 
 different, and more or less independent, methods. The available 
 instruments usually differ in type and in functional efficiency. A 
 choice among methods and instruments should be determined by 
 the desired precision of the result and the time and labor that it is 
 worth while to devote to the observations and reductions. 
 
 Since the labor of observation and the cost of instruments in- 
 crease more rapidly than the inverse square of the precision 
 measure of the attained result, a considerable waste of time and 
 money is involved in any measurement that is executed with 
 greater precision than is demanded by the use to which the result 
 is to be put. On the other hand, if the precision attained is not 
 sufficient for the purpose in hand, the measurement must be 
 repeated by a more exact method. Consequently the labor and 
 expense of the first determination contributes very little to the 
 final result and the waste is quite as great as in the preceding 
 case. Sometimes the expense of a second determination is 
 avoided by using the inexact result of the first, but such a saving 
 is likely to prove disastrous unless the uncertainty of the adapted 
 data is duly considered. 
 
 In general the greatest economy is attained by so planning 
 and executing the measurement that the result is given with the 
 desired precision and neglecting all refinements of method and 
 apparatus that are not essential to this end. While these con- 
 siderations have greater weight in connection with measurements 
 carried out for practical purposes they should never be neglected 
 in planning investigations undertaken primarily for the advance- 
 ment of science. In the former case the cost of necessary measure- 
 ments may represent an appreciable fraction of the expense of 
 a proposed engineering enterprise and must be taken into account 
 in preparing estimates. In the latter case there is no excuse for 
 burdening the limited funds available for research with the expense 
 
 144 
 
ART. 75] DISCUSSION OF PROPOSED MEASUREMENTS 145 
 
 of ill-contrived and haphazard measurements. The precision 
 requirements may be, and indeed usually are, quite different in 
 the two cases, 'but the same process of arriving at suitable methods 
 applies to both. 
 
 75. The General Problem. In its most general form the 
 problem may be stated as follows : Required the magnitude of a 
 quantity X within the limits R, X being a function of several 
 directly measured quantities X\, X 2 , etc. ; within what limits must 
 we determine the value of each of the components X\, X z , etc.? 
 In discussing this problem, all sources of error both constant and 
 accidental must be taken into account. For this purpose the 
 various methods available for the measurement of the several 
 components are considered with regard to the labor of execution 
 and the magnitude of the errors involved as well as with regard to 
 the facility and accuracy with which constant errors can be removed. 
 
 After such a study, certain definite methods are adopted pro- 
 visionally, and examined to determine whether or not the re- 
 quired precision in the final result can be attained by their use. 
 As the first step in this process, the function that gives the rela- 
 tion between X and the components, Xi, X 2 , etc., is written out 
 in its most complete form with all correction factors explicitly 
 represented. Thus, as in article seventy-two, the most probable 
 value of the quantity X may be expressed in the form 
 
 X Q = 0(o a ,o bj . . . , p ,a,/3, . . . , p), (120) 
 
 where the o's represent observed values of X\ t X 2 , etc., and a, /3, 
 . . . , p, represent the factors on which the corrections for con- 
 stant errors depend as pointed out in connection with equation 
 (112), article seventy-one. 
 
 The form of the function 0, and the nature and magnitude of 
 the correction factors appearing in it, will depend on the nature 
 of the proposed methods of measurement. Since all detectable 
 constant errors are explicitly represented by suitable correction 
 factors, all of the quantities appearing in the function may be 
 treated as directly measured components subject to accidental 
 errors only. Hence the problem reduces to the determination 
 of the probable errors within which each of the components must 
 be determined in order that the computed value of XQ may come 
 out with a precision measure equal to the given magnitude R Q . 
 If all of the components can be determined within the limits set 
 
146 THE THEORY OF MEASUREMENTS [ART. 76 
 
 by the probable errors thus found, without exceeding the limits 
 of time and expense imposed by the preliminary considerations, 
 the provisionally adopted methods are adequate for the purpose 
 in hand and the measurements may be carried out with con- 
 fidence that the final result will be precise within the required 
 limits. When one or more of the components cannot be deter- 
 mined within the limits thus set without undue labor or expense, 
 the proposed methods must be modified in such a manner that the 
 necessary measurements will be feasible. 
 
 76. The Primary Condition. The present problem is, to 
 some extent, the inverse of that treated in articles seventy-one 
 and seventy-two. In the latter case the given data represented 
 the results of completed series of observations on the several 
 component quantities appearing in the function 0, together with 
 their respective probable errors. The purpose of the analysis was 
 the determination of the most probable value XQ that could be 
 assigned to the measured magnitude and the precision measure 
 of the result. In the present case approximate values of x and 
 the components in 6 are given, and the object of the analysis is 
 the determination of the probable errors within which each of the 
 components must be measured in order that the value of XQ, 
 computed from the completed observations, may come out with a 
 precision measure equal to a given magnitude R . 
 
 If D , Db, . . . , D p , D a) Dp, . . . , D p represent the separate 
 effects of the probable errors E a , Eb, . . . , E p , E a , Ep, . . . , 
 E p of the components o aj o b , . . . , o p , a, /3, . . . , p, respec- 
 tively, we have, as in article seventy-two, 
 
 and the primary condition imposed on these quantities is given by 
 the relation 
 #o 2 = Da 2 + ZV + - + ZV + ZV + iy + - - . +D P 2 . (123) 
 
 The precision measure R and approximate values of the com- 
 ponents are given by the conditions of the problem and the pro- 
 posed methods of measurement. The E's, and hence also the 
 D's, are the unknown quantities to be determined. Conse- 
 quently there are as many unknowns in equation (123) as there 
 are different components in the function 0. Obviously the problem 
 is indeterminate unless some further conditions can be imposed 
 
ART. 77] DISCUSSION OF PROPOSED MEASUREMENTS 147 
 
 on the D's; for otherwise it would be possible to assign an infinite 
 number of different values to each of the D's which, by proper 
 selection and combination, could be made to satisfy the primary 
 condition (123). 
 
 77. The Principle of Equal Effects. An ideal condition to 
 impose on the D's would specify that they should be so determined 
 that the required precision in the final result X Q would be attained 
 with the least possible expense for labor and apparatus. Un- 
 fortunately this condition cannot be put into exact mathematical 
 form since there is no exact general relation between the difficulty 
 and the precision of measurements. However, it is easy to see 
 that the condition is approximately fulfilled when the measure- 
 ments are so made that the D's are all equal to the same magnitude. 
 For, the probable error of any component is inversely proportional 
 to the square root of the number of observations on which it 
 depends and the expense of a measurement increases directly 
 with the number of observations. Consequently the expense 
 
 W a of the component o a is approximately proportional to 7^-5 or, 
 
 &a 
 n/j 1 
 
 since r is constant, to -^ 9 . Similar relations hold for the other 
 do a D a 2 
 
 components. Hence, as a first approximation, we may assume 
 that 
 
 A2 A2 A2 A2 
 
 where W is the total expense of the determination of x , and A is 
 a constant. By the usual method of finding the minimum value 
 of a function of conditioned quantities, the least value of W con- 
 sistent with equation (123) occurs when the D's satisfy (123) and 
 also fulfill the relations 
 
 _ 
 
 dD a "* ^ dD a = 
 
 ML + ***?- o 
 
 dD b ^ * dD b - 
 
 = 
 
 SD * ^ dD 
 
148 THE THEORY OF MEASUREMENTS [ART. 77 
 
 where K is a constant. Introducing the expressions for R<? and 
 W in terms of the D's, differentiating, and reducing, we have 
 
 and by equation (123) 
 
 where AT is the number of D's in (123) or the equal number of 
 components in the function 6. Consequently equation (123) is 
 fulfilled and the condition of minimum expense is approximately 
 satisfied when the components are so determined that the separate 
 effects of their probable errors satisfy the relation 
 
 D a = D b = - . - = D a = Dp = = -. (127) 
 
 Equation (127) is the mathematical expression of the principle 
 of equal effects. It does not always express an exact solution of 
 the problem, since A is seldom strictly constant; but it is the 
 best approximation that we can adopt for the preliminary com- 
 putation of the D's and E's. The results thus obtained will 
 usually require some adjustment among themselves before they 
 will satisfy both the preliminary considerations and the primary 
 condition (123). We shall see that the necessary adjustment is 
 never very great; and, in fact, that a marked departure from the 
 condition of equal effects is never possible when equation (123) is 
 satisfied. 
 
 Combining equations (122) and (127), we find 
 
 E ^ . ^ - E ^ . ^ 
 
 VAT " de ' a VN ' de ' 
 
 da 
 
 w Ro i . 
 
 * = ~~7= ' ~^7T > 
 
 VN <&' VN y, 
 
 do b 5/3 
 
 (128) 
 
 Hence, if the final measurements are so executed that the probable 
 errors of the several components are equal to the corresponding 
 values given by equations (128), the final result XQ, computed by 
 equation (120), will come out with a precision measure equal to 
 
ART. 78] DISCUSSION OF PROPOSED MEASUREMENTS 149 
 
 the specified R Q , and the condition of equal effects (127) will be 
 fulfilled. 
 
 In computing the E's by equation (128), R Q is taken equal to 
 the given precision measure of X Q and N is placed equal to the 
 
 J/3 
 
 number of components in the function 0. The derivatives T 
 
 do a 
 
 etc., are evaluated with the aid of approximate values of the 
 components obtained by a preliminary trial of the proposed 
 methods or by computation, on theoretical grounds, from an 
 approximate value of XQ and a knowledge of the conditions under 
 which the measurements are to be made. Since only two sig- 
 nificant figures are required in any of the E's, the adopted values 
 of the components may be in error by several per cent, without 
 affecting the significance of the results. Moreover, any number 
 of components, whose combined effect on any derivative is less 
 than five per cent, may be entirely neglected in computing that 
 derivative. Consequently the function frequently may be sim- 
 plified very much for the purpose of computing the derivatives and 
 this simplification may take different forms in the case of differ- 
 ent derivatives. No more than three significant figures should be 
 retained at any step of the process and sometimes the required pre- 
 cision can be attained with the approximate formulae of Table VII. 
 
 Since equation (127) is an approximation, the E's derived from 
 equations (128) are to be regarded as provisional limits for the 
 corresponding components. If all of them are attainable, i.e., if 
 all of the components can be determined within the provisional 
 limits, without exceeding the limit of expense set by the prelim- 
 inary considerations, the solution of the problem is complete and 
 the proposed methods are suitable for the work in hand. 
 
 78. Adjusted Effects. Generally some of the E's given by 
 (128) will be unattainable in practice while others will be larger 
 than a limit that can be easily reached. In other words, it will 
 be found that the labor involved in determining some of the 
 components within the provisional limit is prohibitive while 
 other components can be determined with more than the pro- 
 visional precision without undue labor. In such a case the pro- 
 visional limits are modified by increasing the E's corresponding 
 to the more difficult determinations and decreasing the E's that 
 correspond to the more easily determinable components in such a 
 way that the combined effects satisfy the condition (123). 
 
150 THE THEORY OF MEASUREMENTS [ART. 78 
 
 The maximum allowable increase in a single E is by the factor 
 . For, taking E a for illustration, 
 
 B0 a 
 
 and consequently 
 
 Hence (123) cannot be satisfied unless all of the rest of the D's 
 are negligibly small. For example, if there are nine components, 
 VN is equal to three. Consequently no one of the E's can be 
 increased to more than three times the value given by the condi- 
 tion of equal effects if (123) is to be satisfied. When, as is fre- 
 quently the case, the number of components is less than nine, or 
 when more than one of the E's is to be increased, the limit of 
 allowable adjustment is much less than the above. The extent 
 to which any number of E's may be increased is also limited 
 by the difficulty, or impossibility, of reducing the effects of the 
 remaining E's to the negligible limit. 
 
 If the probable errors given by equations (128) can be modified, 
 to such an extent that the corresponding measurements become 
 feasible, without violating the condition (123), the proposed 
 methods are suitable for the final determination of XQ. Other- 
 wise they must be so modified that they satisfy the conditions of 
 the problem or different methods may be adopted provisionally 
 and tested for availability as above. 
 
 Sometimes it will be found that the proposed methods are 
 capable of greater precision than is demanded by equations (128). 
 In such cases the expense of the measurements may be reduced 
 without exceeding the given precision measure of XQ by using less 
 precise methods. But such methods should never be finally 
 adopted until their feasibility has been tested by the process out- 
 lined above. 
 
 A discussion on the foregoing lines not only determines the 
 practicability of the proposed methods, but also serves as a guide 
 in determining the relative care with which the various parts of 
 the work should be carried out. For, if the final result is to come 
 out with a precision measure R Q , it is obvious that all adjustments 
 and measurements must be so executed that each of the com- 
 
ART. 79] DISCUSSION OF PROPOSED MEASUREMENTS 151 
 
 ponents is determined within the limits set by equations (128), 
 or by the adjusted E's that satisfy (123). 
 
 79. Negligible Effects. In the preceding article it was 
 pointed out that the availableness of proposed methods of meas- 
 urement frequently depends on the possibility of so adjusting the 
 E's given by equations (128) that they are all attainable and 
 at the same time satisfy the primary condition (123). Generally 
 this cannot be accomplished unless some of the E's can be reduced 
 in magnitude to such an extent that their effect on the precision 
 measure R is negligible. 
 
 On account of the meaning of the precision measure, and the 
 fact that it is expressed by only two significant figures, it is obvi- 
 ous that any D is negligible when its contribution to the value of 
 
 73 
 
 #0 is less than y^. Thus, if Ri is the value of the right-hand 
 
 member of equation (123), when D a is omitted, D a is negligible 
 provided 
 
 or 
 
 0. 
 Squaring gives 
 
 0.81 Bo 2 < #i 2 , 
 and by definition 
 
 R<? - RS = D*. 
 Consequently 
 
 0.81 #o 2 < #o 2 - D*, 
 and 
 
 Z> a 2 <0.19# 2 , 
 or 
 
 D a < 0.436 #o. 
 
 Hence, if D a is less than 0.436 # , it will contribute lees than ten 
 per cent of the value of R Q . Since the true error of x is as likely 
 to be greater than R as it is to be less than R Q , a change of ten 
 per cent in the value of R Q can have no practical importance. 
 Consequently D a is negligible when it satisfies the above condi- 
 tion. However, the constant 0.436 is somewhat awkward to 
 handle, and if D a is very nearly equal to the limit 0.436 RQ, the 
 propriety of omitting it is doubtful. These difficulties may be 
 avoided by adopting the smaller and more easily calculated limit 
 of rejection given by the condition 
 
 D = R Q . (129) 
 
152 THE THEORY OF MEASUREMENTS [ART. 79 
 
 This limit corresponds to a change of about six per cent in the 
 value of Ro given by equation (123), and is obviously safe for all 
 practical purposes. Since the above reasoning is independent of 
 the particular D chosen, the condition (129) is perfectly general 
 and applies to any one of the D's in equation (123). 
 
 When two or more of the D's satisfy (129) independently, any 
 one of them may be neglected, but all of them cannot be neg- 
 lected without further investigation for otherwise the change in 
 Ro might exceed ten per cent. This would always happen if all 
 
 T~) 
 
 of the D's considered were very nearly equal to the limit ~^- 
 
 o 
 
 However, by analogy with the above argument, it is obvious that 
 any q of the D's are simultaneously negligible when 
 
 + D 2 2 + . . . + D 3 2 == Jflo, (130) 
 
 where the numerical subscripts 1, 2, . . . , q are used in place 
 of the literal subscripts occurring in equation (123) in order to 
 render the condition (130) entirely general. Thus DI may corre- 
 spond to any one of the D's in (123), D 2 to any other one, etc. 
 By applying the principle of equal effects, the condition (130) 
 may be reduced to the simple form 
 
 D, = D 2 = ... = D q = - ^ (131) 
 
 3 Vg 
 
 If some of the D's in (131) can be easily reduced below the limit 
 
 p 
 
 j=. , the others may exceed that limit somewhat without violating 
 3 V q 
 
 the condition (130). However, equation (131) generally gives the 
 best practical limit for the simultaneous rejection of a number of 
 D's, and all departures from it should be carefully checked by (130). 
 To illustrate the practical application of the foregoing discussion, 
 suppose that the practicability of certain proposed methods of 
 measurement is to be tested by the principle of equal effects 
 developed in article seventy-seven. Let there be N components 
 in the function 0, and suppose that q of them, represented by 
 ai, 2, . . . , a q , can be easily determined with greater precision 
 than is demanded by equations (128), while the measurement 
 of the remaining N q components within the^limits thus set 
 would be very difficult. Obviously some adjustment of the E's 
 given by (128) is desirable in order that the labor involved in the 
 various parts of the measurement may be more evenly balanced. 
 
ART. 79] DISCUSSION OF PROPOSED MEASUREMENTS 153 
 
 The greatest possible increase in the E's corresponding to the 
 N q difficult components will be allowable when the E's of the 
 q easy components can be reduced to the negligible limit. To 
 determine the necessary limits, R is taken equal to the given 
 precision measure of XQ, and the negligible D's corresponding to 
 the q easy components are determined by equation (131). Then 
 by equations (122), the corresponding E's will be negligible when 
 
 E!=Z -^ 
 
 3 Vq 
 
 1 1 
 If 
 
 
 dai 
 
 E 2 = -^L< 
 
 1 
 
 w 
 
 (132) 
 
 A r 
 
 J_^ 
 
 6^ 
 da q 
 
 If these limits can be attained with as little difficulty as the pre- 
 viously determined E's of the N q remaining components, the 
 corresponding D's may be omitted from equation (123) during 
 the further discussion of precision limits. 
 
 Since q of the D's have disappeared, the others may be some- 
 what increased and still satisfy the primary condition (123). 
 The corresponding new limits for the E's of the difficult components 
 may be obtained from equations (127) and (128) by replacing 
 N by N q. If these new limits together with the negligible 
 limits given by equations (132) can all be attained, without 
 exceeding the expense set by the preliminary considerations, the 
 proposed methods may be considered suitable for the final deter- 
 mination of XQ with the desired precision. Otherwise new methods 
 must be devised and investigated as above. 
 
 Equations (132) may also be used to determine the extent to 
 which mathematical constants should be carried out during the 
 computations. For this purpose the components i, 0% , , 
 or part of them, represent the mathematical constants appearing 
 in the function 8. The corresponding E's, determined by equa- 
 tions (132), give the allowable limits of rejection in rounding the 
 numerical values of the constants for the purpose of simplifying 
 
154 
 
 THE THEORY OF MEASUREMENTS [ART. 79 
 
 the computations. Thus, suppose that the volume of a right 
 circular cylinder of length L and radius a is to be computed 
 within one-tenth of one per cent, how many figures should be 
 retained in the constant TT? In this case 
 
 n / \ 17 9 T 
 
 (Oa , , , ) = y = *&lt, 
 RQ = 0.001 V = 0.001 7ra 2 L, 
 60 6V 
 
 = 0.00105. 
 
 0.001 7T 
 
 If TT is taken equal to 3.142 the error due to rounding is 0.00041 . 
 Since this is less than the negligible limit E r , four significant 
 figures in TT are sufficient for the purpose in hand. 
 
 It sometimes happens that the total effect of one or more of the 
 components in the function 0, on the computed value of x , is 
 negligible in comparison with RQ. This will obviously be the case 
 when 
 
 60 RQ 
 
 a^ a ^ IF' 
 
 for a single component a or when 
 
 KM \ 2 -L-/ de 
 z~~ a i) + (^~~ 
 dai I \da2 
 
 da 
 
 for q components. Thus, on the principle of equal effects, the 
 components i, <* 2 , , <* 3 will be simultaneously negligible 
 when they satisfy the conditions 
 
 1 RQ 1 
 
 * 155 i 
 
 (133) 
 
 RQ 1 
 
 daz 
 
 7"> 1 
 
 \7^'~d0~ 
 
 Such cases frequently arise in connection with the components 
 that represent correction factors. 
 
ART. 80] DISCUSSION OF PROPOSED MEASUREMENTS 155 
 
 80. Treatment of Special Functions. During the foregoing 
 argument, it has been assumed that the function 6 in equation (120) 
 is expressed in the most general form consistent with the pro- 
 posed methods of measurement. Such an expression involves the 
 explicit representation of all directly measured quantities, and 
 all possible correction factors. Part of the latter class of com- 
 ponents represent departures of the proposed methods from the 
 theoretical conditions underlying them, and others depend upon 
 inaccuracies in the adjustment of instruments. In practice it 
 frequently happens that the general function is very compli- 
 cated, and consequently that the direct discussion of precision 
 as above is a very tedious process. Under these conditions it is 
 desirable to modify the form of the function in such a manner as 
 to facilitate the discussion. 
 
 Sometimes the general function 9 can be broken up into a series 
 of independent functions or expressed as a continuous product 
 of such functions. Thus, it may be possible to express 6 in the 
 form 
 XQ = 6 (o a , o b , . . ., a, |8, . . .) 
 
 = /i(ai,a 2 , . . . )/ 2 (&i,& 2 , . . . )/ 3 (ci,c 2 , . . . 
 
 or in the form 
 
 XQ = d (O a , O b) . 
 
 (134) 
 
 (135) 
 
 = /i(ai,a 2 , . . . ) X/2(&i,&2, . ) X/ 3 (ci,c 2 , . . . 
 
 X ... X / (mi, m 2) . . . ), 
 where the a's, &'s, . . . , and m's represent the same components, 
 o a , o b , . . . , a, 0, . . . , that appear in 6 by a new and more 
 general notation. The functions /i, / 2 , . . . , f n may take any 
 form consistent with the problem in hand, but the precision dis- 
 cussion will not be much facilitated unless they are independent 
 in the sense that no two of them contain the same or mutually 
 dependent variables. Sometimes the latter condition is imprac- 
 ticable and it becomes necessary to include the same component 
 in two or more of the functions. Under such conditions the expan- 
 sion has no advantage over the general expression for 0, unless 
 the effect of the errors of each of the common components can 
 be rendered negligible in all but one of the functions. It is 
 scarcely necessary to point out that equations (134) and (135) 
 represent different problems, and that if it were possible to expand 
 
156 THE THEORY OF MEASUREMENTS [ART. 80 
 
 the same function in both ways, the component functions /i, 
 /2, , fn would be different in the two cases. 
 For the sake of convenience let 
 
 /I (Oi, 2, ) = 2 
 /2 (6l, 6 2 , . . . ) = ^2 
 
 jfn (Wi,m 2 ,. . . ) = 2 
 
 Then equation (134) may be written in the form 
 
 X = Zi 2 2! 3 . . . d= 2, (137) 
 
 and (135) may be put in the form 
 
 x = z l Xz z Xz 3 X . . . Xz n . (138) 
 
 First consider the case in which the function representing the 
 proposed methods of measurement has been put in the form of 
 (137). Since the precision measure follows the same laws of 
 propagation as the probable error, the discussion given in article 
 fifty-eight leads to the relation 
 
 # 2 = 7^2 + # 2 2 + R f + _ m + Rn 2 } ( 139) 
 
 where RQ is the precision measure of x , and each of the other R's 
 represents the precision measure of the z with corresponding sub- 
 script. Hence, by the principle of equal effects, provisional 
 values of the R's may be obtained from the relation 
 
 R, = R 2 = R, = . . . = R n = A . ( 140 ) 
 
 The R's having been determined by (140), the corresponding 
 probable errors of the a's, 6's, etc., may be computed by the 
 methods of the preceding articles with the aid of equations (136). 
 If the provisional limits of precision thus found are not all attain- 
 able with approximately equal facility, the conditions of the 
 problem may be better satisfied by moderately adjusted relative 
 values of the probable errors as pointed out in article seventy- 
 eight. Obviously the adjusted values must satisfy equation (139) 
 if the value of x computed by (137) is to come out with a pre- 
 cision measure equal to the given R . 
 
 When the function representing the proposed methods can be 
 put in the form of (138) the computation is facilitated by intro- 
 ducing the fractional errors 
 
 P = ; Pl = ! ; P 2 = f2;...; P n = f" (141) 
 
 XQ Zi Zz Z n 
 
ART. 81] DISCUSSION OF PROPOSED MEASUREMENTS 157 
 
 For, by the argument underlying equation (83), article sixty-two, 
 Po 2 = Pi 2 + P 2 2 + Pa 2 + . . . + P 2 , (142) 
 
 and, by the principle of equal effects, provisional values of the 
 P's are given by the relation 
 
 Pi = P 2 = P 3 = . . . = P = *=. (143) 
 
 Vn 
 
 Since RQ and approximate values of the components are given, 
 PO can be computed with sufficient accuracy with the aid of 
 (138) and the first of (141). Consequently provisional fractional 
 limits for the components can be determined by (143), and the 
 corresponding precision measures by the last n of equations (141). 
 Beyond this point the problem is identical with the preceding 
 case, except that the adjusted limits of precision must satisfy 
 (142) in place of (139). 
 
 The methods developed in the preceding articles are entirely 
 general and applicable to any form of the function 6, but they 
 frequently lead to complicated computations. In the present 
 article we have seen how the discussion can be simplified when the 
 function can be put in either of the particular forms represented 
 by (134) and (135). Many of the problems met with in practice 
 cannot be put in either of these special forms, but it frequently 
 happens that the treatment of the functions representing them 
 can be simplified by a suitable modification or combination of the 
 above general and particular methods. The general ideas under- 
 lying all discussions of the necessary precision of components 
 have been discussed above with sufficient fullness to show their 
 nature and significance. Their application to particular prob- 
 lems must be left to the ingenuity of the observer and computer. 
 
 81. Numerical Example. As an illustration of the fore- 
 going methods, suppose that the electromotive force of a battery 
 is to be determined, and that the precision measure of the result 
 is required to satisfy the condition 
 
 R = 0.0012 volts, (i) 
 
 T-> 
 
 within the limits T?!>i- e -> #o must lie between 0.0011 and 
 
 =b 0.0013 volt. Preliminary considerations demand that the 
 expense of the work shall be as low as is consistent with the 
 required precision. 
 
158 
 
 THE THEORY OF MEASUREMENTS [ART. 81 
 
 The given conditions are most likely to be fulfilled by some 
 form of potentiometer method. Suppose that the arrangement 
 of apparatus illustrated in Fig. 10 is adopted provisionally; and, 
 to simplify the discussion, suppose that the various parts of the 
 apparatus are so well insulated that leakage currents need not 
 be considered. The generality of the problem is not appreciably 
 affected by the latter assumption since the specified condition 
 can be easily satisfied in practice within negligible limits. With 
 what precision must the several components and correction 
 factors be determined in order that equation (i) may be satisfied? 
 
 -T&Z 
 
 FIG. 10. 
 
 L e t V = e.m.f. of tested battery BI, 
 
 Et = e.m.f. of Clark cell B 2 at time of observation, 
 t = temperature of Clark cell at time of observation, 
 Ri = resistance between 1 and 2, 
 Rz = resistance between 1 and 3, 
 
 / = current in circuit 1, 2, 3, B 3 , 1 when the key K is open, 
 5i = algebraic sum of thermo e.m.f.'s in the circuit 1, 2, 6, 
 
 G, 1 when K is closed to 6, 
 2 = algebraic sum of thermo e.m.f. 's in the circuit 1, 3, a, 
 
 G, 1 when K is closed to a, 
 Ei5 e.m.f. of Clark cell at temperature 15 C., 
 a. = mean temperature coefficient of Clark cell in the 
 neighborhood of 20 C. 
 
ART. 81] DISCUSSION OF PROPOSED MEASUREMENTS 159 
 
 When the sliding contacts 2 and 3 are so adjusted that the 
 galvanometer G shows no deflection on closing the key K to 
 either a or 6, 
 
 RI RZ 
 
 Consequently 
 
 F = (^+6 2 )|- 1 -5 1 . (ii) 
 
 -fi/2 
 
 But 
 
 (in) 
 
 Hence 
 
 F = -B 16 !l-a-15)jf- 1 + 2 f- 1 -8 1 . (iv) 
 
 KZ n>z 
 
 The resistances RI and # 2 are functions of the temperature; but, 
 
 since they represent simultaneous adjustments with the cells BI 
 
 p 
 
 and Bz and are composed of the same coils, the ratio ~ is inde- 
 
 KZ 
 
 pendent of the temperature. Thus, if R t ' and R t " represent the 
 resistances of the used coils at t C., and ft is their temperature 
 coefficient, 
 
 RS Ri(l+ fit) Ri 
 
 whatever the temperature t at which the comparison is made. 
 This advantage is due to the particular method of connection and 
 adjustment adopted, and is by no means common to all forms of 
 the potentiometer method. 
 
 Under the conditions specified above, equation (iv) may be 
 adopted as the complete expression for the discussion of precision. 
 It corresponds to equation (120) in the general treatment of the 
 problem. Suppose that the following approximate values of the 
 components, which are sufficiently close for the determination of 
 the capabilities of the method, have been obtained from the 
 normal constants of the Clark cell and a preliminary adjustment 
 of the apparatus or by computation from a known approximate 
 value of V: 
 
 #15 = 1.434 volts; a = 0.00086; 
 
 t = 20 C.; Ri = 1000 ohms; 
 
 R 2 = 1310 ohms; V = 1.1 volts. 
 
 The thermoelectromotive forces 5i and 5 2 are to some extent 
 
 due to inhomogeneity of the wires used in the construction of 
 
 the instruments and connections. For the most part, however, 
 
 (v) 
 
160 THE THEORY OF MEASUREMENTS [ART. 81 
 
 they arise from the junctions of dissimilar metals in the circuits 
 considered. Suppose that the resistances R\ and #2 are made of 
 manganin, the key K of brass, and that the copper used in the 
 galvanometer coil and the connecting wires is thermoelectrically 
 different. Both 5i and 5 2 would represent the resultant action 
 of at least six thermo-elements in series. While these effects can- 
 not be accurately specified in advance, their combined action 
 would not be likely to be greater than twenty-five microvolts per 
 degree difference in temperature between the various parts of the 
 apparatus, and it might be much less than this. Obviously 5i 
 and 6 2 are both equal to zero when the temperature of the appa- 
 ratus is uniform throughout. 
 
 By equations (133), article seventy-nine, the correction terms 
 depending on thermoelectric forces will be negligible in compar- 
 ison with the given precision measure R , when 5i and 62 satisfy 
 the conditions 
 
 . 1 #o 1 , - 1 flo 1 
 
 ' l *3'vT5E ^s'vTE' 
 
 ddi dd 2 
 
 In the present case 
 
 Ro = 0.0012 volt; q = 2; 
 dV . dV R l 
 
 sE*-- 1 ' and srsr 
 
 Consequently the above conditions become 
 
 - 5^i? . _L _ 0.00028 volt = 280 microvolts, 
 3 v 2 1 
 
 _L - 0.00037 volt = 370 microvolts. 
 0.76 
 
 From the above discussion of the possible magnitude of the thermo- 
 electromotive forces in the circuits considered, it is obvious that 
 these limits correspond to temperature differences of approxi- 
 mately ten degrees between the various parts of the apparatus. 
 Since the temperature of the apparatus can be easily maintained 
 uniform within five degrees, the last two terms in equation (iv) 
 are negligible within the limits of precision set in the present 
 problem. Hence, for the determination of the required precision 
 of the remaining components, the functional relation (iv) may be 
 taken in the form 
 
 (vi) 
 
ART. 81] DISCUSSION OF PROPOSED MEASUREMENTS 161 
 
 By equation (123), article seventy-six, the primary condition 
 for determining the necessary precision of the components is 
 
 R<? = 144 X 10~ 8 = Z>! 2 + > 2 2 + D 3 2 + > 4 2 + D<?, (vii) 
 where dV 67 67 
 
 67 
 
 (viii) 
 
 and EI, EZ, E 3 , E^ E$ are the required probable errors of EI$, a, t, 
 Ri, and Rz, respectively. 
 
 For the preliminary determination of the jE"s by the principle 
 of equal effects, equation (127), article seventy-seven, becomes 
 
 = 0.00054. (ix) 
 
 VN V5 
 
 Neglecting all factors that do not affect the differential coefficients 
 by more than one unit in the second significant figure and adopt- 
 ing the approximate values of the components given in (v), 
 
 67 R! 1000 n _ 
 j- = p- = T^ = 0.76, 
 
 d-Cns /t2 lolU 
 
 =- E 15 a = - - 0.00094, 
 it/2 
 
 = Eu = 0.0011, 
 
 it 2 
 
 (x) 
 
 Hence, by combining (viii) and (ix), or directly from equations 
 (128), article seventy-seven, 
 , 0.00054 
 
 (xi) 
 
 E z 
 
 0.76 
 0.00054 
 
 ZEI V7.V7UV/I J. VWftVj 
 
 n nnnoQS 
 
 v*y 
 
 (b) 
 (c) 
 (d) 
 (e) 
 
 5.5 
 0.00054 
 
 = 0.57 C. 
 = =b 0.49 ohm 
 = db 0.65 ohm. 
 
 0.00094 
 0.00054 
 
 - o.oon 
 
 0.00054 
 
 0.00083 
 
162 THE THEORY OF MEASUREMENTS [ART. 81 
 
 In practice the attainableness of these limits might be deter- 
 mined experimentally; but in the present case, as in most practical 
 problems, general considerations based on theory and previous 
 experience lead to equally trustworthy results. In the first place, 
 it is obvious that the temperature of the Clark cell can be easily 
 determined closer than 0.6 C. Consequently the limit (c) is easily 
 attainable and might possibly be reduced to a negligible quantity. 
 
 The constants of the normal Clark cell are known well within 
 the limits (a) and (b). But it requires very careful treatment of 
 the cell to keep Ei 6 constant within the limit (a), and new cells, 
 unless they are set up with great care and skill, are likely to vary 
 among themselves and from the normal cell by more than 0.0007 
 volt. Consequently the limit (a) is somewhat smaller than is 
 desirable in practical work of the precision considered in the 
 present problem. On the other hand, the limit (b) is very rarely 
 exceeded by either old or new cells unless they are very care- 
 lessly constructed and handled. Hence E 2 could probably be 
 reduced to the negligible limit. 
 
 With a suitable galvanometer, the nominal values of the resist- 
 ances Ri and R% can be easily adjusted within the limits (d) and 
 (e). But EI and E 5 must be considered practically as the pre- 
 cision measures of R i and R 2 . They include the calibration 
 errors of the resistances, the errors due to leakage between the 
 terminals of individual coils, and the errors due to nonuniformity 
 of temperature as well as the errors of setting of the contacts 2 
 and 3, Fig. 10. The resultant of these errors can be reduced 
 below the limits (d) and (e), but in the present case it would be 
 convenient to have somewhat larger limits in order to reduce the 
 expense of construction and calibration. 
 
 Hence, while all of the E's given by equations (xi) are within 
 attainable limits, the preliminary consideration of minimum 
 expense would be more likely to be fulfilled if the limits (a), 
 (d), and (e) were somewhat larger. Obviously the magnitude of 
 these limits can be increased without violating the primary con- 
 dition (vii) provided a corresponding decrease in the magnitudes 
 of the limits (b) and (c) is possible. 
 
 By equation (131), article seventy-nine, the separate effects D 2 
 and DZ will be simultaneously negligible if 
 n n 1 #o 1 0.0012 
 
 1/2 = DZ = 7^ = ~ ;= ^ 
 
 3 Vq 3 V2 
 
ABT.SI] DISCUSSION OF PROPOSED MEASUREMENTS 163 
 
 Hence, by equations (132), the errors of a and t will be negligible 
 
 when 0.00028 
 
 E 2 = ^~ =; 0.000051, (b') 
 
 and 
 
 00028 
 
 Ets mm ^o-3oc. (C ') 
 
 Since these limits can be reached with much greater ease than the 
 limits (a), (d), and (e), they may be adopted as final specifica- 
 tions and the corresponding Z)'s may be omitted during the deter- 
 mination of new limits for the components E 15) R 1} and R%. 
 Under these conditions, equation (ix) becomes 
 
 Hence the largest allowable limits for the errors of EM, Ri, and 
 
 RZ are OOOfiQ 
 
 ~ = 0.00091 volt, (a') 
 
 = .63 ohm, (d') 
 
 While these limits cannot be quite so easily attained as (b') and 
 (c'), they cannot be increased without violating the primary con- 
 dition (vii). Consequently they satisfy the condition of minimum 
 expense, so far as the proposed method is concerned, and may be 
 adopted as final specifications. 
 
 The fractional errors corresponding to the specified precision 
 measure of V and the above limiting errors of the components 
 
 Po = Y = 0.0011 = 0.11%, 
 Pi = ~ = db 0,00063 = 0.063%, 
 P 2 = ^ = 0.059 = =t 5.9%, 
 P 3 = y = 0.015 = db 1.5%, 
 P 4 = ~ = =fc 0.00063 = d= 0.063%, 
 P 5 = f- 5 = 0.00063 = 0.063%. 
 
164 THE THEORY OF MEASUREMENTS [ART. 81 
 
 Consequently in order to obtain a value of V that is exact within 
 0.11 per cent by the proposed method, a must be determined 
 within 5.9 per cent, t within 1.5 per cent, and E.-&, Ri, and R 2) each 
 within 0.063 per cent. These limits are all attainable in practice 
 under suitable conditions, as pointed out above. Hence the pro- 
 posed method is practicable. 
 
 If the final measurements are so devised and executed that the 
 above conditions are fulfilled, the precision of the result computed 
 from them will be within the specified limits and the expense of 
 the work will be reduced to the lowest limit compatible with the 
 proposed method. The desired result might be obtained at less 
 expense by some other method, but a decision on this point can 
 be reached only by comparing the precision requirements and 
 practicability of various methods with the aid of analyses similar 
 to the above. 
 
CHAPTER XII. 
 BEST MAGNITUDES FOR COMPONENTS. 
 
 82. Statement of the Problem. The precision of a derived 
 quantity depends on the relative magnitudes and precision of the 
 components from which it is computed, as explained in Chapter 
 VIII. Thus, if the derived quantity X Q is given in terms of the 
 components x\, x^ . . . , x q by the expression 
 
 x = F (xi, x 2 , . . . , x g ), (144) 
 
 the probable error of X Q is given by the expression 
 
 E Q * = SSEJ + S 2 2 E 2 Z + + S q *E q 2 , (145) 
 
 where the E's represent the probable errors of the x's with corre- 
 sponding subscripts, and 
 
 AF AF AF 
 
 *-& *-&'< (146) 
 
 The error E, corresponding to any directly measured com- 
 ponent, is generally, but not always, independent of the absolute 
 magnitude of that component so long as the measurements are 
 made by the same method and apparatus. For example: the 
 probable error of a single measurement with a micrometer caliper, 
 graduated to 0.01 millimeter, is approximately equal to 0.004 
 millimeter, whatever the magnitude of the object measured so 
 long as it is within the range of the instrument. Hence, when 
 the methods and instruments to be used in measuring each of 
 the components are known in advance, the probable errors EI, 
 E 2 , etc., can be determined, at least approximately, by preliminary 
 measurements on quantities of the same kind as the components 
 but of any convenient magnitude. Under these conditions the 
 E's on the right-hand side of equation (145) may be treated as 
 known constants, and, since the S's are expressible in terms of 
 Xi, x z , etc., by equations (146), the value of E corresponding to 
 the given methods cannot be changed without a simultaneous 
 change in the relative or absolute magnitudes of the components. 
 
 165 
 
166 THE THEORY OF MEASUREMENTS [ART. 82 
 
 Since equation (144) must always be fulfilled, and since the 
 value of XQ is usually fixed by the conditions of the problem, a 
 change in the magnitudes of the re's is not always possible. But 
 it frequently happens that the form of the function F is such that 
 the relative magnitudes of the components can be changed through 
 somewhat wide limits and still satisfy equation (144). Thus, if 
 a cylinder is to have a specified volume, it may be made long and 
 thin, or short and thick, and have the same volume in either case. 
 Consequently it is sometimes possible to select magnitudes for 
 the components that will give a minimum value of E and at the 
 same time satisfy equation (144). 
 
 The problem before us may be briefly stated as follows : Having 
 given definite methods and apparatus for the measurement of the 
 components of a derived quantity reo, what magnitudes of the 
 components will give a minimum value to the probable error EQ of 
 XQ and at the same time satisfy the functional relation (144)? 
 
 It can be easily seen that a practical solution of this problem 
 is not always possible. In the first place the form of the function 
 F may be such as to admit of but a single system of magnitudes 
 of the components, and consequently the value of EQ is definitely 
 fixed by equation (145). In some cases there are no real values 
 of the re's that will satisfy both (144) and the conditions for a 
 minimum of EQ. When values can be found that satisfy the 
 mathematical conditions they are not always attainable in prac- 
 tice. Finally the probable errors Ei, E 2 , etc., may not be inde- 
 pendent of the magnitudes of the corresponding components or 
 it may be impossible to determine them in advance of the final 
 measurements. 
 
 When the E's are not independent of the re's it sometimes 
 happens that the fractional errors 
 
 Pi = ? ; p * = ? '> p * = ? (147) 
 
 3/1 it/2 Xq 
 
 are constant and determinable in advance. In such cases the 
 problem may be solvable by putting (145) in the equivalent form 
 
 Ef = SfPfy? + SfPfxf +!>+ S q *P q *x q *, (148) 
 
 expressing the S's in terms of the components by equations (146), 
 and determining the values of the re's that will render (148) a 
 minimum subject to the condition (144). 
 
ART. 83] BEST MAGNITUDES FOR COMPONENTS 167 
 
 When a practicable solution of the problem is possible, it is 
 obvious that the results thus obtained are the best magnitudes 
 that can be assigned to the components, and that they should 
 be adopted as nearly as possible in carrying out the final measure- 
 ments from which X Q is to be computed. 
 
 83. General Solutions. The general conditions for a mini- 
 mum or a maximum value of E Q 2 , when XQ is treated as a constant 
 and the variables are required to satisfy the relation (144), but 
 are otherwise independent, are 
 
 dF 
 ^ A = U, 
 
 0) 
 
 where K is an arbitrary constant. By introducing the expressions 
 (145) and (146), transposing and dividing by two, equations (i) 
 become 
 
 Sl gtf 1 . + S ,g^ + ... 
 o O&1 ET 2 _j_ O 0O2 pi 2 i 
 
 1 dx 2 2 ^2 
 
 
 (149) 
 
 When the S's have been replaced by x's with the aid of equa- 
 tions (146), the q equations (149), together withj(144), are theoreti- 
 cally sufficient for the determination of all of the q + 1 unknown 
 quantities Xi, x 2 , . . . , x q , and K. However, in some cases a 
 practicable solution is not possible, and in others the components 
 or their ratios come out as the roots of equations of the second 
 or higher degree. The zero, infinite, and imaginary roots of these 
 equations have no practical significance in the present discussion 
 and need not be considered. Some of the real roots correspond to 
 a maximum, some to a minimum, and others to neither a maximum 
 nor a minimum value of E Z . In most cases the roots that corre- 
 spond to a minimum of E 2 can be selected by inspection with the 
 
168 
 
 THE THEORY OF MEASUREMENTS [ART. 83 
 
 aid of equation (145), but it is sometimes necessary to apply the 
 well-known criteria of the calculus. 
 
 Dividing equation (145) by x Q 2 and putting 
 
 XQ dX 2 ' q XQ XQ dX q 
 
 XQ XQ dXi 
 
 gives the expression 
 
 PZ = EI 
 
 X 2 
 
 XQ 
 
 (150) 
 
 + T*E* (151) 
 
 for the fractional error of XQ. Since XQ is a constant in any given 
 problem the maxima and minima of P 2 correspond to the same 
 values of the components as those of E Q 2 . Sometimes the form 
 of the function F is such that the expression (151), when expanded 
 in terms of the x's, is much simpler than (145). In such cases it 
 is much easier to determine the minima of P 2 than of E 2 . For 
 this purpose the equations of condition (i) may be put in the form 
 
 6X1 
 
 XQ dXi 
 
 KdF_ 
 
 XQ 6X2 
 
 dx, 
 
 (152) 
 
 , q XQ dX q 
 
 and by substitution and transposition we have 
 
 dTi dT% dT g 
 
 1 dxi 2 dxi 2 q dxi 
 
 dT< 
 
 (153) 
 
 When the components are required to satisfy the condition (144) 
 and a given constant value is assigned to XQ, equations (153) lead 
 to exactly the same results as equations (149). In fact either of 
 these sets of equations can be derived from the other by purely 
 algebraic methods when the $'s and T's are expressed in terms of 
 the x's. In practice one or the other of the sets will be the simpler, 
 depending on the form of the function F; and the simpler form 
 
ART. 83] BEST MAGNITUDES FOR COMPONENTS 169 
 
 can be more easily derived by direct methods as above than by 
 algebraic transformation. 
 
 In some problems the magnitude of one or more of the com- 
 ponents in the function F can be varied at will and determined 
 with such precision that their probable errors are negligible in 
 comparison with those of the other components. Variables that 
 fulfill these conditions will be called free components. Since any 
 convenient magnitude can be assigned to them, their values can 
 always be so chosen that the condition (144) will be fulfilled 
 whatever the values of the other components. Consequently the 
 latter components may be treated as independent variables in 
 determining the minima of E Q 2 or P Q 2 . 
 
 Under these conditions the E's corresponding to the free com- 
 ponents can be placed equal to zero, and either E 2 or P 2 can 
 sometimes be expressed as a function of independent variables 
 only by eliminating the free components from the S's or the T's 
 with the aid of equation (144). When this elimination can be 
 effected, the minimum conditions may be derived from equations 
 (149) or (153), as the case may be, by placing K equal to zero and 
 omitting the equations involving derivatives with respect to the 
 free components. This is evident because the remaining com- 
 ponents are entirely independent, and consequently the partial 
 derivatives of E Q 2 or P 2 with respect to each of them must vanish 
 when the values of the variables correspond to the maxima or 
 minima of these functions. When the elimination cannot be 
 accomplished, neither equations (149) nor (153) will lead to con- 
 sistent results and the problem is generally insolvable. 
 
 In practice it frequently happens that the free components are 
 factors of the function F, and are not included in any other way. 
 Under these conditions they do not occur in the T's corresponding 
 to the remaining components, since the form of equations (150) 
 is such that they are automatically eliminated. Consequently, 
 in this case, the conditions for a minimum are given at once by 
 equations (153) when K is taken equal to zero, since the derivatives 
 with respect to the free components all vanish and the correspond- 
 ing E's are negligible. It is scarcely necessary to point out that 
 the remarks in the paragraph following equations (149), except 
 for obvious changes in notation, apply with equal rigor to equa- 
 tions (153), whether K is zero or finite. The values of the x's 
 derived from these equations should never be assumed to corre- 
 spond to the minima of P 2 without further investigation. 
 
170 THE THEORY OF MEASUREMENTS [ART. 84 
 
 84. Special Cases. Suppose that the relation between the 
 derived quantity XQ and the measured components xi, # 2 , and x s 
 is given in the form 
 
 XQ = ax?* + bxj 1 * + cxj 1 *, (ii) 
 
 where a, b, c, and the n's are constants. If the probable errors 
 Ei t E z , and E 3 of the x's with corresponding subscripts are known, 
 and independent of the magnitude of the components, what mag- 
 nitudes of the components will give the least possible value to the 
 probable error E of X Q ? 
 By equations (146), 
 
 Si = arnxi^-V; S 2 = bn&^'-V; S s = c/W^-D. (iii) 
 Consequently 
 
 dSi , i\ ( <>\ ^$2 rv ^$3 _. 
 
 -!(, -I)**-*; ._ =0; = 0, 
 
 Substituting these results in equations (149) and dividing the 
 first equation by Si, the second by $ 2 , and the third by S S) the 
 conditions for a minimum value of E Q 2 become 
 
 Efari! (m - 1) xi<*-*> = K, 
 
 Dividing the second and third of these equations by the first 
 and transposing the coefficients to the second member gives the 
 ratios of the components in the form 
 
 x 2 (n ^- 2) = EJani (ni-l) 
 
 T,(tti-2) ~~ EL2Jm n (n n - IV 
 
 (HI - 1) 
 
 ~ 
 
 (n s - 
 
 These two equations together with (ii) are theoretically sufficient 
 for the determination of the best magnitudes for the three com- 
 ponents xij Xzj and x$] but it can be easily seen, from the form of 
 the equations, that a solution is not practicable for all possible 
 values of the n's. 
 
ART. 84] BEST MAGNITUDES FOR COMPONENTS 171 
 
 For example, if the n's are all equal to unity, the ratios of the 
 components given by (iv) are both indeterminate, each being 
 
 equal to ^- Consequently the problem has no solution in this 
 
 case. This conclusion might have been reached at once by 
 inspecting the value of E Q 2 given by equation (145), when the S's 
 are expressed in terms of the components. Thus, placing the n's 
 equal to unity in equations (iii) and substituting the results in 
 (145), we find 
 
 Since E<? is independent of the x's it can have no maxima or 
 minima with respect to the components. 
 
 When each of the n's equals two, equations (iv) are inde- 
 pendent of the x's, and consequently the problem is not solvable. 
 In this case (ii) becomes 
 
 XQ = 
 
 and (145) reduces to 
 E 2 = 4 
 
 Since these equations differ only in the values of the constant 
 coefficients of the x's, no magnitudes can be assigned to the com- 
 ponents that will give a minimum value to E Q 2 , and at the same 
 time satisfy the equation for XQ. 
 
 If each of the n's is placed equal to three, equation (ii) takes 
 the form 
 
 XQ = ax^ + bx 2 * + c#3 3 , (v) 
 
 and equations (iv) become 
 
 Xt~bEf' 
 
 (iv') 
 
 C# 3 2 
 
 In this case the problem can be easily solved when the numerical 
 values of the coefficients and the E's are known. As a very 
 simple illustration, suppose that 
 
 7 -f J 77T -TGI ~Ij1 XT' 
 
 a = o = c = 1, and J^i = & 2 MS &, 
 then, by (iv') and (v), 
 
 and, by (145) and (iii), 
 
172 THE THEORY OF MEASUREMENTS [ART. 84 
 
 Since a decrease in the magnitude of one of the x's involves an 
 increase in that of one or both of the others, in order to satisfy 
 equation (v), and since the fourth power of a quantity varies 
 more rapidly than the third, it is obvious that the minimum 
 value of E 2 will occur when the x's are all equal. Consequently 
 the above solution corresponds to a minimum of E 2 . 
 
 It can be easily seen that there are many other cases in which 
 equations (ii) and (iv) can be solved, and also some others in 
 which no solution is possible. The extension of the problem to 
 functions in the same form as equation (ii), but containing any 
 number of similar terms, involves only the addition of one equa- 
 tion in the form of (iv) for each added component. Obviously 
 these equations hold for negative as well as positive values of the 
 coefficients and exponents of the x's. 
 
 As a second example, consider the functional relation 
 
 x = axi n i X xf*. (vi) 
 
 In this case the solution is more easily effected by the second 
 method given in the preceding article. By equations (150) 
 
 Consequently 
 
 and equations (153) reduce to the simple form 
 
 ^ES=-K; %Ef = -K, ; (viii) 
 
 where EI and E 2 are the known constant probable errors of Xi and 
 #2. Eliminating K, we have 
 
 Consequently the problem is always solvable when n\ and n 2 
 have the same sign. When they have different signs the solu- 
 tion is imaginary. Hence there are no best magnitudes for the 
 components when the derived quantity is given as the ratio of 
 two measured quantities. 
 
ART. 85] BEST MAGNITUDES FOR COMPONENTS 173 
 
 The extension of this solution to functions involving any num- 
 ber of factors is obvious. When the exponents of all of the 
 factors have the same sign the problem is always solvable but 
 the best magnitudes thus found may not be attainable in practice. 
 If part of the exponents are positive and others are negative the 
 solution is imaginary. 
 
 85. Practical Examples. 
 I. 
 
 In many experiments the desired result depends directly upon 
 the determination of the quantity of heat generated by an electric 
 current in passing through a resistance coil. Let I represent the 
 current intensity and E the fall of potential between the terminals 
 of the coil. Then the quantity of heat H developed in t seconds 
 may be computed by the relation 
 
 JH = TEt, 
 
 where J represents the mechanical equivalent of heat. If H is 
 measured in calories, I in amperes, E in volts, and t in seconds, 
 
 y is equal to 0.239 calorie per Joule and the above relation becomes 
 
 H = 0.239 lEt. (ix) 
 
 Suppose that the conditions of the problem in hand are such 
 that H should be made approximately equal to 1000 calories. 
 Since the resistance of the heating coil is not specified it can be so 
 chosen that 7 and E may have any convenient values that satisfy 
 the relation (ix) when H has the above value. Obviously t can 
 be varied at will, by changing the time of run, and (ix) will not 
 be violated if suitable values are assigned to / and E. If the 
 instruments available for measuring /, E, and t are an ammeter 
 graduated to tenths of an ampere, a voltmeter graduated to 
 tenths of a volt, and a common watch with a seconds hand, what 
 are the best magnitudes that can be assigned to the components, 
 i.e., what magnitudes of /, E, and t will give the computed H 
 with the least probable error? 
 
 By comparing equations (ix) and (vi), it is easy to see that 
 the present problem is an application of the second special case 
 worked out in the preceding article when a third variable factor 
 Z 3 n 3 is annexed to (vi). H corresponds "to x 0) I to Xi, E to x z , t to 
 # 3 /and all of the n's in (vi) are equal to unity. Consequently 
 
174 THE THEORY OF MEASUREMENTS [ART. 85 
 
 the solution can be derived at once from three equations in the 
 form of (viii) if suitable values can be assigned to the probable 
 errors of the components. 
 
 With the available instruments, the probable errors E i} E e , and 
 E t of /, E, and t, respectively, will be practically independent of 
 the magnitude of the measured quantities so long as the range 
 of the instruments is not exceeded. Under the conditions that 
 usually prevail in such observations the following precision may 
 be attained with reasonable care: 
 
 E t = 0.05 ampere; E e = 0.05 volt; E t = 1 second. 
 
 The conditions for a minimum value of the probable error E 
 of H can be derived by exactly the same method that was used 
 in obtaining equations (viii), or these equations may be used at 
 once with proper substitutions as outlined above. Consequently 
 the best magnitudes for the components are given by the simul- 
 taneous solution of (ix) and the following three equations, 
 
 ^ 2 _ K . ^_ E? 
 ~P = ~ K > ~W~ ~ K > ~P = 
 
 Eliminating K and substituting the numerical values of the 
 probable errors we have 
 
 E_E e _. l_E t _ 
 
 I ~ E<~ L > I~ Ei~ 
 
 Consequently 
 
 E = I and t = 20 /. (x) 
 
 Substituting these results and the numerical value of H in (ix) 
 we have 
 
 1000 = 0.239 X 20 X / 3 , 
 and hence 
 
 I = 5.94 amperes 
 
 is the best magnitude to assign to the current strength under the 
 given conditions. The corresponding magnitudes for the electro- 
 motive force and time found by (x) are 
 
 E = 5.94 volts and t = 119 seconds. 
 
 If the above values of the components and their probable errors 
 are substituted in equation (151), the fractional error of H comes 
 out 
 
ART. 85] BEST MAGNITUDES FOR COMPONENTS 175 
 
 and the probable error of H is given by the relation 
 E Q = 1000 Po =15 calories. 
 
 If any other magnitudes for the components, that satisfy equa- 
 tion (ix), are used in place of the above in (151), the computed 
 value of E will be greater than fifteen calories. Consequently 
 the above solution corresponds to a minimum value of E Q . 
 
 In order to fulfill the above conditions the resistance of the 
 heating coil must be so chosen as to satisfy the relation 
 
 *- 
 
 Since our solution calls for numerically equal values of I and E, 
 the resistance R must be made equal to one ohm. 
 
 It can be easily seen that small variations in the values of the 
 components will produce no appreciable effect on the probable 
 error of H, ^ince the numerical value of E is never expressed by 
 more than two significant figures. Consequently the foregoing 
 discussion leads to the following practical suggestions regarding 
 the conduct of the experiment. The heating coil should be so 
 constructed that the heat developed in the leads is negligible in 
 comparison with that developed between the terminals of the 
 voltmeter. The resistance of the coil should be one ohm. The 
 current strength should be adjusted to approximately six amperes 
 and allowed to flow continuously for about two minutes. Under 
 these conditions the difference in potential between the terminals 
 of the coil will be about six volts. The conditions under which 
 7, E, and t are observed should be so chosen that the probable 
 errors specified above are not exceeded. 
 
 If the above suggestions are carried out in practice the value 
 of H computed from the observed values of /, E, and t by equa- 
 tion (ix) will be approximately 1000 calories, and its probable 
 error will be about fifteen calories. A more precise result than 
 this cannot be obtained with the given instruments unless the 
 probable errors of 7, E, and t can be materially decreased by 
 modifying the conditions and methods of observation. 
 
 II. 
 
 A partial discussion of the problem of finding the best magni- 
 tudes for the components involved in the measurement of the 
 strength of an electric current with a tangent galvanometer may 
 
176 THE THEORY OF MEASUREMENTS [ART. 85 
 
 be found in many laboratory manuals and textbooks. Such dis- 
 cussions are usually confined to a consideration of the error in the 
 computed current strength due to a given error in the observed 
 deflection. On the assumption, tacit or expressed, that the effects 
 of the errors of all other components are negligible it is proved 
 that the effect of the deflection error is a minimum when the 
 deflection is about forty-five degrees. Although the tangent gal- 
 vanometer is now seldom used in practice it provides an instructive 
 example in the calculation of best magnitudes since the general 
 bearings of the problem are already familiar to most students. 
 
 In order to avoid unnecessary complications, consider a simple 
 form of instrument with a compass needle whose position is 
 observed directly on a circle graduated in degrees. Suppose that 
 the needle is pivoted at the center of a single coil of N turns of 
 wire, and R centimeters mean radius. Under these conditions the 
 current strength I is connected with the observed deflection (f> by 
 the relation 
 
 where H is the horizontal intensity of a uniform external magnetic 
 field parallel to the plane of the coil. In practice the plane of the 
 coil is usually placed parallel to the magnetic meridian and H 
 is taken equal to the horizontal component of the earth's mag- 
 netism. 
 
 N is an observed component but it can be so precisely deter- 
 mined by direct counting, during the construction of the coil, 
 that its error may be considered negligible in comparison with 
 those of the other components. Furthermore it can be given any 
 desired value when an instrument is designed to meet special 
 needs, and a choice among a number of different values is possi- 
 ble in most completed instruments. Consequently the quantity 
 
 x TT may be treated as a free component, represented by A, and 
 the expression for the current strength may be written in the 
 form 7 = A#.tan0. (xi) 
 
 Comparing this expression with the general equation (144) we 
 note that / corresponds to x 0) H to x\, R to x 2 , and to z 3 . 
 
 Since A is free, the components H, R, and </> are entirely inde- 
 pendent; and any convenient magnitudes can be made to satisfy 
 
ART. 85] BEST MAGNITUDES FOR COMPONENTS 177 
 
 (xi) by suitably choosing the number of turns in the coil. Con- 
 sequently, as pointed out in article eighty-three with respect to 
 functions containing a free component as a factor, the conditions 
 for a minimum probable error of / are given by equations (153) 
 with K placed equal to zero. By making the above substitutions 
 for the x's in equations (150) and performing the differentiations 
 we have 
 
 I/' 7?' oi-r O ^ * V^^X 
 
 11 /L bill cp 
 
 Consequently 
 
 0/77 -i z\nn H^TI 
 
 ol i 1 . o J. 2 f\ OJ. 3 ~ 
 
 dH = ~H~ 2 ' dH ; ~dH = ' 
 
 *^/T7 *\ ATT "I fk T7 
 
 ?l n ^ 2 _ _ L ^ 3 _ n. 
 
 dR ~ dR R 2 ' dR ' 
 
 dTi ^ = n- dTz = 4cos2<?i> 
 
 d0 60 " 60 sin 2 2 ' 
 
 and, if the probable errors of H, R, and are represented by E\ 9 
 EZ, and #3, respectively, equations (153) become 
 
 If EI and E 2 could be made negligible, as is tacitly assumed in 
 most discussions of the present problem, the first two of equations 
 (xiii) would be satisfied whatever the values of H and R. Conse- 
 quently these components would be free and would be the only 
 independent variable involved in equation (xi). Under these 
 conditions the minimum value of the probable error of 7 corre- 
 sponds to the value of derived from the third of equations (xiii). 
 The general solution of this equation is 
 
 0= (2n-l)|> 
 
 where n represents any integer. But, since values of greater 
 than I are not attainable in practice, n must be taken equal to 
 
 unity in the present case and consequently the best magnitude 
 for the deflection is forty-five degrees. It is obvious that (xi) 
 can always be satisfied when / has any given value, and is 
 equal to forty-five degrees by suitably choosing the values of the 
 free components 2V, H, and R. 
 
178 THE THEORY OF MEASUREMENTS [ART. 85 
 
 If the fractional error of / is represented by P and the T's 
 given by equations (xii) are substituted in (151), 
 
 H 2 ' R 2 ' sin 2 20 
 Pi 2 + P 2 2 + Pa 2 , 
 
 (xiv) 
 
 = Pi 2 + P 2 2 + P 3 2 , 
 where 
 
 2 
 
 = : and 
 
 are the separate effects of the probable errors E\, EZ, and E 3) 
 respectively. If both ends of the needle are read with direct and 
 reversed current so that represents the mean of four observa- 
 tions, EZ should not exceed 0.025 or 0.00044 radians, and it might 
 be made less than this with sufficient care. Consequently, when 
 <j> is equal to forty-five degrees, 
 
 P 3 = 0.00088. 
 
 By an argument similar to that given in article seventy-nine it can 
 be proved that PI and P 2 will be simultaneously negligible when 
 they satisfy the condition 
 
 p l = P 2 = i A = 0.00021. 
 3V2 
 
 Hence, in order that the effects of E\ and E% may be negligible in 
 comparison with that of E 3 , H and R must be determined within 
 about two one-hundredths of one per cent. 
 
 With an instrument of the type considered it would seldom be 
 possible and never worth while to determine H and R with the 
 precision necessary to fulfill the above condition. In common 
 practice E\ and E 2 are generally far above the negligible limit 
 and it would be necessary to make both H and R equal to infinity 
 in order to satisfy the first two of the minimum conditions (xiii). 
 Hence there is no practically attainable minimum value of P . 
 This conclusion can also be derived directly by inspection of 
 equation (xiv). P 2 decreases uniformly as H and R are increased, 
 and becomes equal to Ps 2 when they reach infinity. 
 
 Although a minimum value of P is not attainable, the fore- 
 going discussion leads to some practical suggestions regarding 
 the design and use of the tangent galvanometer. For any given 
 values of E\, E 2 , and E 3 , the minimum value of PS occurs when <j> 
 is equal to forty-five degrees. Also PI and P% decrease as H and 
 R increase. Consequently the directive force H and the radius 
 
ART. 85] BEST MAGNITUDES FOR COMPONENTS 179 
 
 of the coil R should be made as large as is consistent with the 
 conditions under which the instrument is to be used, and the 
 number of turns N in the coil should be so chosen that the observed 
 deflection will be about forty-five degrees. 
 
 The practical limit to the magnitude of R is generally set by a 
 consideration of the cost and convenient size of the instrument. 
 Moreover when R is increased N must be increased in like ratio 
 in order to satisfy the fundamental relation (xi) without altering 
 the observed deflection or decreasing the value of H. There 
 is an indefinite limit beyond which N cannot be increased with- 
 out introducing the chance of error in counting and greatly in- 
 creasing the difficulty of determining the exact magnitude of R. 
 Above this limit E 2 is approximately proportional to R, and, as 
 can be easily seen by equation (xiv), there is no advantage to 
 be gamed by a further increase in the magnitude of R. 
 
 H can be varied by suitably placed permanent magnets, but 
 it is difficult to maintain strong magnetic fields uniform and con- 
 stant within the required limits. Even under the most favorable 
 conditions, the exact determination of H is very tedious and 
 involves relatively large errors. Consequently Pi 2 is likely to be 
 the largest of the three terms on the right-hand side of equation 
 (xiv). Under suitable conditions it can be reduced in magnitude 
 by increasing H to the limit at which the value of EI begins to 
 increase. However, such a procedure involves an increased value 
 of N in order to satisfy equation (xi), and consequently it may 
 cause an increase in E 2 owing to the relation between N and R 
 pointed out in the preceding paragraph. In such a case the gain 
 in precision due to a decreased value of PI would be nearly bal- 
 anced by an increased value of P%. 
 
 In common practice the instrument is so adjusted that H is 
 equal to the horizontal component of the earth's magnetic field 
 at the time and place of observation. Unless H is very carefully 
 determined at the exact location of the instrument, EI is likely 
 
 to be as large as 0.005 ~5 and, since the order of magnitude 
 
 Cat, 
 
 of H is about 0.2 ^r , -Pi will be approximately equal to 0.025. 
 
 cm 
 
 Hence both P 2 and P 3 will be negligible in comparison with PI if 
 they satisfy the relation 
 
 P 2 = P 3 = - ^j= = 0.0059. 
 "3 V2 
 
180 THE THEORY OF MEASUREMENTS [ART. 85 
 
 Under ordinary conditions R and < can be easily determined within 
 the above limit. Consequently, in the supposed case, 
 PO = PI = 2.5 per cent, 
 
 and it would be useless to attempt an improvement in precision 
 by adjusting the values of N, R } and <. With sufficient care in 
 determining H, PI can be reduced to such an extent that it be- 
 comes worth while to carry out the suggestions regarding the 
 design and use of the instrument given by the foregoing theory. 
 But when the value of H is assumed from measurements made in 
 a neighboring location or is taken from tables or charts the per- 
 centage error of / will be nearly equal to that of H regardless of 
 the adopted values of R and <. Under such conditions P Q can- 
 not be exactly determined but it will seldom be less than two or 
 three per cent of the measured magnitude of I. 
 
 The above problem has been discussed somewhat in detail in 
 order to illustrate the inconsistent results that are likely to be 
 obtained in determining best magnitudes when the effects of the 
 errors of some of the components are neglected. It is never 
 safe to assume that the error of a component is negligible until 
 its effect has been compared with that of the errors of the other 
 components. 
 
 III. 
 
 Figure eleven is a diagram of the apparatus and connections 
 commonly used in determining the internal resistance of a bat- 
 tery by the condenser method. G is a ballistic galvanometer, 
 C a condenser, R a known resistance, KI a charge and discharge 
 key, Kz a plug or mercury key, and B a battery to be tested. 
 
 Let Xi represent the ballistic throw of the galvanometer when 
 the condenser is charged and discharged with the key K 2 open, 
 and x z the corresponding throw when K 2 is closed. Then the 
 internal resistance R Q of the battery may be computed by the 
 relation 
 
 Ro = R ^L^l. ( XV ) 
 
 Under ordinary conditions the probable errors of x\ and x^ 
 cannot be made much less than one-half of one per cent of the 
 observed throws when a telescope, mirror, and scale are used. On 
 the other hand the probable error of R should not exceed one-tenth 
 of one per cent if a suitably calibrated resistance is used and the 
 
ART. 85] BEST MAGNITUDES FOR COMPONENTS 181 
 
 connections are carefully made. When these conditions are ful- 
 filled, it can be easily proved that the effect of the error of R is 
 negligible in comparison with that of the errors of Zi and x 2 . 
 Furthermore any convenient value can be assigned to R, such 
 
 <T 2 R 
 
 " !L -A/WVW\AAAA/ 
 
 B 
 FIG. 11. 
 
 that (xv) will be satisfied whatever the values of Xi and #2. Con- 
 sequently R may be treated as a free component and the throws 
 Xi and x z as independent variables. 
 
 For the purpose of determining the magnitudes of the com- 
 ponents R, xij and x z that correspond to a minimum value of the 
 fractional error P of RQ, we have by equations (150) and (xv) 
 
 Consequently 
 
 - X 2 ) 
 
 (xvi) 
 
 Since x\ and x 2 are independent, K must be taken equal to zero 
 in the minimum conditions (153). Hence, dividing the first two 
 equations by T i} we have 
 
 1 xi 1 
 
 1 
 
 E, 2 -^ 
 
 = o, 
 
 = 0, 
 
 (x,-xz) 2 x 2 x 2 2 (x l -xz) 2 
 
 where EI and E 2 are the probable errors of x\ and 2 , respectively. 
 
182 THE THEORY OF MEASUREMENTS [ART. 85 
 
 Multiply each of these equations by -- 1 ^ 2 and they as- 
 sume the simple form 
 
 +- * 
 
 Since #i 2 and Ez 2 are always positive, it is obvious that there 
 are no values of Xi and x% that will satisfy both of these equations 
 at the same time. Hence, when Xi and x z can be varied inde- 
 pendently, they cannot be so chosen that the fractional error P 
 will be a minimum. However, if Xz is kept constant at any as- 
 signed value, PO will pass through a minimum when Xi satisfies 
 equation (a). On the other hand if any constant value is assigned 
 to Xi the minima and maxima of P will correspond to the roots 
 of equation (b). 
 
 In practice x\ is the throw of the galvanometer needle due to 
 the electromotive force of the battery when on open circuit; and 
 it is very nearly constant, during a series of observations, when 
 suitable precautions are taken to avoid the effects of polariza- 
 tion. Both Xi and Xz can be varied by changing the capacity 
 of the condenser or the sensitiveness of the galvanometer, but 
 their ratio depends only on the ratio of R to R. Consequently, 
 if any convenient magnitude is assigned to Xi, the root of equa- 
 tion (b) that corresponds to a minimum value of PO gives the 
 best magnitude for the component Xz. 
 
 Since x\ and x 2 are similar quantities, determined with the same 
 instruments and under the same conditions, E\ is generally equal 
 
 to EZ. Hence, if we replace the ratio -- by y, equation (b) be- 
 
 ^-2^-1 = 0^ (b') 
 
 The only real root of this equation is 
 
 y = 2.2056. 
 By equations (151) and (xvi) 
 
 Putting E l = Ez = E and - = y, 
 
 Xz 
 
 Pl = y* + y* 
 E* x 2 -l 2 ' 
 
ART. 86] BEST MAGNITUDES FOR COMPONENTS 183 
 
 Since Xi is necessarily greater than x 2 , y cannot be less than unity. 
 
 P 2 
 Under this condition it can be easily proved by trial that -== 
 
 & 
 
 approaches a minimum as y approaches the value given above, 
 provided any constant value is assigned to x\. 
 Equation (xv) may be put in the form 
 R Q = R(y- 1), 
 
 and, by introducing the value of y given by the minimum condi- 
 tion (b')> we have 
 
 R = 0.83 R . 
 
 Consequently the greatest attainable precision in the determina- 
 tion of RQ will be obtained when R is made equal to about eighty 
 three per cent of RQ. If R is adjusted to this value Xi and x% will 
 satisfy equation (b), whatever the magnitude of the capacity used, 
 provided the observations are so made that E\ and E% are equal. 
 
 When the internal resistance of the battery is very low it is 
 sometimes impracticable to fulfill the above theoretical conditions 
 because the errors due to polarization are likely to more than off- 
 set the gain in precision corresponding to the theoretically best 
 magnitudes of the components. In such cases a high degree of 
 precision is not attainable, but it is generally advisable to make R 
 considerably larger than R Q in order to reduce polarization errors. 
 
 86. Sensitiveness of Methods and Instruments. The pre- 
 cision attainable in the determination of directly measured com- 
 ponents depends very largely on the sensitiveness of indicating 
 instruments and on the methods of adjustment and observation. 
 The design and construction of an instrument fixes its intrinsic 
 sensitiveness; but its effective sensitiveness, when used as an indi- 
 cating device, depends on the circumstances under which it is used 
 and is frequently a function of the magnitudes of measured quan- 
 tities and other determining factors. Thus; the intrinsic sensi- 
 tiveness of a galvanometer is determined by the number of 
 windings in the coils, the moment of the directive couple, and 
 various other factors that enter into its design and construction. 
 On the other hand its effective sensitiveness as an indicator in a 
 Wheatstone Bridge is a function of the resistances in the various 
 arms of the bridge and the electromotive force of the battery 
 used. An increase in the intrinsic sensitiveness of an instrument 
 may cause an increase or a decrease in its effective sensitiveness, 
 
184 THE THEORY OF MEASUREMENTS [ART. 86 
 
 depending on the nature of the corresponding modification in 
 design and the circumstances under which the instrument is 
 used. 
 
 By a suitable choice of the magnitudes of observed components 
 and other determining factors it is sometimes possible to increase 
 the effective sensitiveness of indicating instruments and hence 
 also the precision of the measurements. On the other hand, 
 as pointed out in Chapter XI, the precision of measurements 
 should not be greater than that demanded by the use to which 
 they are to be put. In all cases the effective sensitiveness of 
 instruments and methods should be adjusted to give a result 
 definitely within the required precision limits determined as in 
 Chapter XI. Consequently the best magnitudes for the quan- 
 tities that determine the effective sensitiveness are those that 
 will give the required precision with the least labor and expense. 
 The methods by which such magnitudes can be determined depend 
 largely on the nature of the problem in hand, and a general treat- 
 ment of them is quite beyond the scope of the present treatise. 
 Each separate case demands a somewhat detailed discussion of 
 the theory and practice of the proposed measurements and only 
 a single example can be given here for the purpose of illustration. 
 
 Since the potentiometer method of comparing electromotive 
 forces has been quite fully discussed in article eighty-one, it will 
 be taken as a basis for the illustration and we will proceed to find 
 the relation between the effective sensitiveness of the galvanom- 
 eter and the various resistances and electromotive forces involved. 
 Since the directly observed components in this method are the 
 resistances R\ and R%, the effective sensitiveness is equal to the 
 galvanometer deflection corresponding to a unit fractional devia- 
 tion of Ri or R z from the condition of balance. 
 
 From the discussion given in article eighty-one it is evident that 
 the potentiometer method could be carried out with any conven- 
 ient values of the resistances R\ and R 2 provided they are so ad- 
 
 7- 
 
 justed that the ratio - satisfies equation (ii) in the cited article. 
 tiz 
 
 The absolute magnitudes of these resistances depend on the electro- 
 motive force of the battery J5 3 and the total resistance of the cir- 
 cuit 1, 2, 3, B 3 , 1 in Fig. 10. The effective sensitiveness of the 
 method, and hence the accuracy attainable in adjusting the con- 
 tacts 2 and 3 for the condition of balance, depends on the above 
 
ABT.86] BEST MAGNITUDES FOR COMPONENTS 185 
 
 factors together with the resistance and intrinsic sensitiveness of 
 the galvanometer. 
 
 Since RI and R% are adjusted in the same way and under the 
 same conditions, the effective sensitiveness of the method is the 
 same for both. Consequently only one of them will be considered 
 in the present discussion, but the results obtained will apply with 
 equal rigor to either. The essential parts of the apparatus and 
 connections are illustrated in Fig. 12, which is the same as Fig. 10 
 with the battery B 2 and its connections omitted. 
 
 FIG. 12. 
 
 Let V = e.m.f. of battery BI, 
 
 E = e.m.f. of battery B 3 , 
 R = resistance between 1 and 2, 
 W = total resistance of the circuit 1, 2, B s , 1, 
 G = total resistance of the branch 1, G, BI, 2, 
 
 I = current through B 3) 
 
 r = current through R, 
 
 g = current through BI and G. 
 
 When the contact 2 is adjusted to the balance position 
 
 Consequently 
 
 = 0, r = 7, and 7=^ = -^ 
 
 
 (xvii) 
 
 This is the fundamental equation of the potentiometer and must 
 be fulfilled in every case of balance. Consequently E must be 
 
186 THE THEORY OF MEASUREMENTS [ART. 86 
 
 chosen larger than V because R is a part of the resistance in the 
 circuit 1, 2, B z , 1, and hence is always less than W. Equation 
 (xvii) may then be satisfied by a suitable adjustment of R. 
 
 By applying Kirchhoff's laws to the circuits 1, G, BI, 2, 1, and 
 1, 2, B 3) 1, when the contact 2 is not in the balance position, we 
 
 have 
 
 Rr-Gg= V, 
 
 and Rr + (W - R) I = E. 
 But r = I - g. 
 
 Hence RI-(R + G)g = V, 
 and WI - Rg = E. 
 
 Eliminating I and solving for g we find 
 
 WV -RE 
 
 If D is the galvanometer deflection corresponding to the current 
 g and K is the constant of the instrument 
 
 g = KD. 
 
 Most galvanometers are, or can be, provided with interchange- 
 able coils. The winding space in such coils is usually constant, 
 but the number of windings, and hence the resistance, is variable. 
 Under these conditions the resistance of the galvanometer will be 
 approximately proportional to the square of the number of turns 
 of wire in the coils used. For the purpose of the present discussion, 
 this resistance may be assumed to be equal to G since the resist- 
 ance of the battery and connecting wires in branch 1, G, BI, 2, 
 can usually be made very small in comparison with that of the 
 galvanometer. The constant K is inversely proportional to the 
 number of windings in the coils used. Consequently, as a suffi- 
 ciently close approximation for our present purpose, we have 
 
 T 
 
 v 
 
 K = T=> 
 
 VG 
 
 where T is a constant determined by the dimensions of the coils, 
 the moment of the directive couple, and various other factors 
 depending on the type of galvanometer adopted. Hence, for any 
 given instrument, 
 
ART. 86] BEST MAGNITUDES FOR COMPONENTS 187 
 
 VG 
 
 The quantity -jr is the intrinsic sensitiveness of the galvanometer. 
 
 It is equal to the deflection that would be produced by unit current 
 if the instrument followed the same law for all values of g. 
 By equation (xix) and (xviii) 
 
 VG WV-RE 
 
 T *R*-WR-WG' 
 
 The variation in D due to a change dR in R is 
 
 dD VG E(R*-WR-WG) + (WV-RE)(2R-W) 
 
 dR ' T ' (R*-WR-WGY 
 
 When the potentiometer is adjusted for a balance, D is equal to 
 zero and WV is equal to RE by equation (xvii). Hence, if d is the 
 galvanometer deflection produced when the resistance R is changed 
 from the balancing value by an amount dR, equation (xx) may 
 be put in the form 
 
 1 VVG 
 
 The fractional change in R corresponding to the total change dR 
 is 
 
 . I '-f : I 
 
 Consequently 
 
 1 VVO 
 
 ~' ' 
 
 is the galvanometer deflection corresponding to a fractional error 
 P r in the adjustment of R for balance. The coefficient of P r in 
 equation (xxi) is the effective sensitiveness of the method under 
 the given conditions. If this quantity is represented by S, equa- 
 tion (xxi) becomes 
 
 8 = SP r , 
 
 8 I 
 
 
 
 All of the quantities appearing in the right-hand member of this 
 equation may be considered as independent variables since equa- 
 tion (xvii) can always be satisfied, and hence the potentiometer 
 
188 THE THEORY OF MEASUREMENTS [ART. 86 
 
 can be balanced, when R, V, and E have any assigned values, if 
 the resistance W is suitably chosen. 
 
 If d' is the smallest galvanometer deflection that can be defi- 
 nitely recognized with the available means of observation, the frac- 
 tional error P/ of a single observation on R should not be greater 
 
 5' 
 than -~ Since the precision attainable in adj usting the potentiom- 
 
 o 
 
 eter for balance is inversely proportional to P/, it is directly pro- 
 portional to the effective sensitiveness S. By choosing suitable 
 magnitudes for the variables T, G, R, and E, it is usually possible 
 to adjust the value of S, and hence also of P/, to meet the re- 
 quirements of any problem. 
 
 From equation (xxii) it is evident that S will increase in magni- 
 tude continuously as the quantities T, R, and E decrease and that 
 it does not pass through a maximum value. The practicable in- 
 crease in S is limited by the following considerations: E must be 
 greater than V, for the reason pointed out above, and its variation 
 is limited by the nature of available batteries. Since E must 
 remain constant while the potentiometer is being balanced alter- 
 nately against V and the electromotive force of a standard cell, 
 as explained in article eighty-one, the battery B 3 must be capable 
 of generating a constant electromotive force during a considerable 
 period of time. In practice storage cells are commonly used for 
 this purpose and E may be varied by steps of about two volts by 
 connecting the required number of cells in series. Obviously E 
 should be made as nearly equal to V as local conditions permit. 
 
 When the potentiometer is balanced 
 
 V E 
 
 If R is reduced for the purpose of increasing the effective sensitive- 
 ness, W must also be reduced in like ratio, and, consequently, the 
 current 7 through the instrument will be increased. The prac- 
 tical limit to this adjustment is reached when the heating effect 
 of the current becomes sufficient to cause an appreciable change 
 in the resistances R and W. With ordinary resistance boxes this 
 limit is reached when 7 is equal to a few thousandths of an ampere. 
 Consequently, if E is about two volts, R should not be made much 
 less than one thousand ohms. Resistance coils made expressly 
 for use in a potentiometer can be designed to carry a much larger 
 
ART. 86] BEST MAGNITUDES FOR COMPONENTS 189 
 
 current so that R may be made less than one hundred ohms with- 
 out introducing serious errors due to the heating effect of the 
 current. 
 
 The constant T depends on the type and design of the galva- 
 nometer. In the suspended magnet type it can be varied some- 
 what by changing the strength of the external magnetic field, and 
 in the D'Arsonval type the same result may be attained by chang- 
 ing the suspending wires of the movable coil. The effects of the 
 vibrations of the building in which the instrument is located and 
 of accidental changes in the external magnetic field become much 
 more troublesome as T is decreased, i.e., as the intrinsic sensitive- 
 ness is increased. Consequently the practical limit to the reduc- 
 tion of T is reached when the above effects become sufficient to 
 render the observation of small values of 6 uncertain. This limit 
 will depend largely on the location of the instrument and the care 
 that is taken in mounting it. Sometimes a considerable reduc- 
 tion in T can be effected by selecting a type of galvanometer 
 suited to the local conditions. 
 
 If the quantities T 7 , R, V, and E are kept constant, S passes 
 through a maximum value when G satisfies the condition 
 
 *?' 
 
 It can be easily proved by direct differentiation that this is the 
 case when 
 
 G = 
 
 Hence, after suitable values of the other variables have been de- 
 termined as outlined above, the best magnitude for G is given by 
 equation (xxiii). Generally this condition cannot be exactly ful- 
 filled in practice unless a galvanometer coil is specially wound for 
 the purpose; but, when several interchangeable coils are available, 
 the one should be chosen that most nearly fulfills the condition. 
 In some galvanometers T and G cannot be varied independently, 
 and in such cases suitable values can be determined only by trial. 
 Since the ease and rapidity with which the observations can be 
 made increase with T, it is usually advisable to adjust the other 
 variables to give the greatest practicable value to the second 
 factor in S, and then adjust T so that the effective sensitiveness 
 
190 THE THEORY OF MEASUREMENTS [ART. 86 
 
 will be just sufficient to give the required precision in the deter- 
 mination of R. 
 
 As an illustration consider the numerical data given in article 
 eighty-one. It was proved that the specified precision require- 
 ments cannot be satisfied unless R is determined within a frac- 
 tional precision measure equal to 0.00063. Allowing one-half 
 of this to errors of calibration we have left for the allowable error 
 in adjusting the potentiometer 
 
 P r ' = 0.00031. 
 
 If a single storage cell is used at B$, E is approximately two volts, 
 and, with ordinary resistance boxes, R should be about one thou- 
 sand ohms, for the reason pointed out above. This condition is 
 fulfilled by the cited data; and, for our present purpose, it will be 
 sufficiently exact to take V equal to one volt. Hence, by equa- 
 tion (xxiii), the most advantageous magnitude for G is about 
 five hundred ohms; and, by equation (xxii), the largest practi- 
 cable value for the second factor in S is 
 
 ST = V Jf = 0.0224. 
 
 gf 1-41+0 
 
 With a mirror galvanometer of the D'Arsonval type, read by 
 telescope and scale, a deflection of one-half a millimeter can be 
 easily detected. Consequently, if we express the galvanometer 
 constant K in terms of amperes per centimeter deflection, we must 
 take 5' equal to 0.05 centimeter; and, in order to fulfill the specified 
 precision requirements, the effective sensitiveness must satisfy the 
 condition 
 
 S' 0.05 
 ~P7~00003l~ 
 
 Combining this result with the above maximum value of ST we 
 find that the intrinsic sensitiveness must be such that 
 
 0.0224 _ 
 161 
 
 Hence the galvanometer should be so constructed and adjusted 
 that 
 
 G = 500 ohms, 
 and 
 
 T 
 K = = = 6.2 X lO" 6 amperes per centimeter deflection. 
 
ART. 86] BEST MAGNITUDES FOR COMPONENTS 191 
 
 D'Arsonval galvanometers that satisfy the above specifications 
 can be very easily obtained and are much less expensive than 
 more sensitive instruments. They are so nearly dead-beat and 
 free from the effects of vibration that the adjustment of the poten- 
 tiometer for balance can be easily and rapidly carried out with 
 the necessary precision. Hence the use of such an instrument 
 reduces the expense of the measurements without increasing the 
 errors of observation beyond the specified limit. 
 
CHAPTER XIII. 
 RESEARCH. 
 
 87. Fundamental Principles. The word research, as used 
 by men of science, signifies a detailed study of some natural 
 phenomenon for the purpose of determining the relation between 
 the variables involved or a comparative study of different phe- 
 nomena for the purpose of classification. The mere execution of 
 measurements, however precise they may be, is not research. On 
 the other hand, the development of suitable methods of measure- 
 ment and instruments for any specific purpose, the estimation of 
 unavoidable errors, and the determination of the attainable limit 
 of precision frequently demand rigorous and far-reaching research. 
 As an illustration, it is sufficient to cite Michelson's determination 
 of the length of the meter in terms of the wave length of light. A 
 repetition of this measurement by exactly the same method and 
 with the same instruments would involve no research, but the 
 original development of the method and apparatus was the result 
 of careful researches extending over many years. 
 
 The first and most essential prerequisite for research in any field 
 is an idea. The importance of research, as a factor in the advance- 
 ment of science, is directly proportional to the fecundity of the 
 underlying ideas. 
 
 A detailed discussion of the nature of ideas and of the conditions 
 necessary for their occurrence and development would lead us too 
 far into the field of psychology. They arise more or less vividly 
 in the mind in response to various and often apparently trivial 
 circumstances. Their inception is sometimes due to a flash of 
 intuition during a period of repose when the mind is free to respond 
 to feeble stimuli from the subconscious. Their development and 
 execution generally demand vigorous and sustained mental effort. 
 Probably they arise most frequently in response to suggestion or 
 as the result of careful, though tentative, observations. 
 
 A large majority of our ideas have been received, in more 
 or less fully developed form, through the spoken or written dis- 
 course of their authors or expositors. Such ideas are the common 
 
 192 
 
ART. 88] RESEARCH 193 
 
 heritage of mankind, and it is one of the functions of research to 
 correct and amplify them. On the other hand, original ideas, 
 that may serve as a basis for effective research, frequently arise 
 from suggestions received during the study of generally accepted 
 notions or during the progress of other and sometimes quite differ- 
 ent investigations. 
 
 The originality and productiveness of our ideas are determined 
 by our previous mental training, by our habits of thought and 
 action, and by inherited tendencies. Without these attributes, 
 an idea has very little influence on the advancement of science. 
 Important researches may be, and sometimes are, carried out by 
 investigators who did not originate the underlying ideas. But, 
 however these ideas may have originated, they must be so thor- 
 oughly assimilated by the investigator that they supply the stim- 
 ulus and driving power necessary to overcome the obstacles that 
 inevitably arise during the prosecution of the work. The driving 
 power of an idea is due to the mental state that it produces in the 
 investigator whereby he is unable to rest content until the idea 
 has been thoroughly tested in all its bearings and definitely proved 
 to be true or false. It acts by sustaining an effective concentra- 
 tion of the mental and physical faculties that quickens his in- 
 genuity, broadens his insight, and increases his dexterity. 
 
 In order to become effective, an idea must furnish the incentive 
 for research, direct the development of suitable methods of pro- 
 cedure, and guide the interpretation of results. But it must 
 never be dogmatically applied to warp the facts of observation 
 into conformity with itself. The mind of the investigator must 
 be as ready to receive and give due weight to evidence against 
 his ideas as to that in their favor. The ultimate truth regarding 
 phenomena and their relations should be sought regardless of 
 the collapse of generally accepted or preconceived notions. From 
 this point of view, research is the process by which ideas are 
 tested in regard to their validity. 
 
 88. General Methods of Physical Research. Researches 
 that pertain to the physical sciences may be roughly classified 
 in two groups: one comprising determinations of the so-called 
 physical constants such as the atomic weights of the elements, the 
 velocity of light, the constant of gravitation, etc.; the other 
 containing investigations of physical relations such as that which 
 connects the mass, volume, .pressure, and temperature of a gas. 
 
194 THE THEORY OF MEASUREMENTS [ART. 88 
 
 The researches in the first group ultimately reduce to a careful 
 execution of direct or indirect measurements and a determination 
 of the precision of the results obtained. The general principles 
 that should be followed in this part of the work have been suffi- 
 ciently discussed in preceding chapters. Their application to prac- 
 tical problems must be left to the ingenuity and insight of the 
 investigator. Some men, with large experience, make such appli- 
 cations almost intuitively. But most of us must depend on a 
 more or less detailed study of the relative capabilities of available 
 methods to guide us in the prosecution of investigations and in 
 the discussion of results. 
 
 In general, physical constants do not maintain exactly the same 
 numerical value under all circumstances, but vary somewhat with 
 changes in surrounding conditions or with lapse of time. Thus 
 the velocity of light is different in different media and in dispersive 
 media it is a function of the frequency of the vibrations on which 
 it depends. Consequently the determination of such constants 
 should be accompanied by a thorough study of all of the factors 
 that are likely to affect the values obtained and an exact specifica- 
 tion of the conditions under which the measurements are made. 
 Such a study frequently involves extensive investigations of the 
 phenomena on which the constants depend and it should be 
 carried out by very much the same methods that apply to the 
 determination of physical relations in general. On the other 
 hand, the exact expression of a physical relation generally involves 
 one or more constants that must be determined by direct or in- 
 direct measurements. Hence there is no sharp line of division 
 between the first and second groups specified above, many re- 
 searches belonging partly to one group and partly to the other. 
 
 The occurrence of any phenomenon is usually the result of the 
 coexistence of a number of more or less independent antecedents. 
 Its complete investigation requires an exact determination of the 
 relative effect of each of the contributary causes and the develop- 
 ment of the general relation by which their interaction is expressed. 
 A determination of the nature and mode of action of all of the 
 antecedents is the first step in this process. Since it is gen- 
 erally impossible to derive useful information by observing the 
 combined action of a number of different causal factors, it becomes 
 necessary to devise means by which the effects of the several 
 factors can be controlled in such manner that they can be studied 
 
ART. 88] RESEARCH 195 
 
 separately. The success of researches of this type depends very 
 largely on the effectiveness of such means of control and the 
 accuracy with which departures from specified conditions can be 
 determined. 
 
 Suppose that an idea has occurred to us that a certain phenome- 
 non is due to the interaction of a number of different factors that 
 we will represent by A, B, C, . . . , P. This idea may involve 
 a more or less definite notion regarding the relative effects of the 
 several factors or it may comprehend only a notion that they are 
 connected by some functional relation. In either case we wish 
 to submit our idea to the test of careful research and to determine 
 the exact form of the functional relation if it exists. 
 
 The investigation is initiated by making a series of preliminary 
 observations of the phenomenon corresponding to as many vari- 
 ations in the values of the several factors as can be easily effected. 
 The nature of such observations and the precision with which they 
 should be made depend so much on the character of the problem 
 in hand that it would be impossible to give a useful general dis- 
 cussion of suitable methods of procedure. Sometimes roughly 
 quantitative, or even qualitative, observations are sufficient. In 
 other cases a considerable degree of precision is necessary before 
 definite information can be obtained. In all cases the observa- 
 tions should be sufficiently extensive and exact to reveal the gen- 
 eral nature and approximate relative magnitudes of the effects 
 produced by each of the factors. They should also serve to detect 
 the presence of factors not initially contemplated. 
 
 With the aid of the information derived from preliminary obser- 
 vations and from a study of such theoretical considerations as 
 they may suggest, means are devised for exactly controlling the 
 magnitude of each of the factors. Methods are then developed 
 for the precise measurement of these magnitudes under the con- 
 ditions imposed by the adopted means of control. This process 
 often involves a preliminary trial of several different methods 
 for the purpose of determining their relative availability and pre- 
 cision. The methods that are found to be most exact and con- 
 venient usually require some modification to adapt them to the 
 requirements of a particular problem. Sometimes it becomes 
 necessary to devise and test entirely new methods. During this 
 part of the investigation the discussions of the precision of meas- 
 urements given in the preceding chapters find constant applica- 
 
196 THE THEORY OF MEASUREMENTS [ART. 88 
 
 tion and it is largely through them that the suitableness of 
 proposed methods is determined. 
 
 After definite methods of measurement and means of control 
 have been adopted and perfected to the required degree of pre- 
 cision, the final measurements on the factors, A, B, C, . . . , P, 
 are carried out under the conditions that are found to be most 
 advantageous. Usually two of the factors, say A and B, are 
 caused to vary through as large a range of values as conditions 
 will permit while the other factors are maintained constant at 
 definite observed values. At stated intervals the progress of the 
 variation is arrested and corresponding values of A and B are 
 measured while they are kept constant. From a sufficiently 
 extended series of such observations it is usually possible to make 
 an empirical determination of the form of the functional relation 
 
 A =/i(); C,Z>, . . . ,P. constant. (i) 
 
 On the other hand, if the form of the function /i is given as a 
 theoretical deduction from the idea underlying the investigation, 
 the observations serve to test the exactness of the idea and de- 
 termine the magnitudes of the constants involved in the given 
 function. By allowing different factors to vary and making 
 corresponding measurements, the relations 
 
 A =/ 2 (C); B,D, . . , P, constant, 
 
 A =/ n (P); ,C,Z>, ., constant, 
 
 (ii) 
 
 may be empirically determined or verified. As many functions of 
 this type as there are pairs of factors might be determined, but 
 usually it is not necessary to establish more than one relation for 
 each factor. Generally it is convenient to determine one of the 
 factors as a function of each of the others as illustrated above; 
 but it is not necessary to do so, and sometimes the determination 
 of a different set of relations facilitates the investigation. 
 
 During the establishment of the relation between any two 
 factors all of the others are supposed to remain rigorously con- 
 stant. Frequently this condition cannot be exactly fulfilled with 
 available means of control, but the variations thus introduced 
 can usually be made so small that their effects can be treated as 
 constant errors and removed with the aid of the relations after- 
 wards found to exist between the factors concerned, For this 
 
ART. 88] RESEARCH 197 
 
 purpose frequent observations must be made on the factors that 
 are supposed to remain constant during the measurement of the 
 two principal variables. If the variations in these factors are not 
 very small all of the relations determined by the principal measure- 
 ments will be more or less in error and must be treated as first 
 approximations. Usually such errors can be eliminated and the 
 true relations established with sufficient precision, by a series of 
 successive approximations. However, the weight of the final 
 result increases very rapidly with the effectiveness of the means of 
 control and it is always worth while to exercise the care necessary 
 to make them adequate. 
 
 When the functions involved in equations (i) and (ii), or their 
 equivalents in terms of other combinations of factors, have been 
 determined with sufficient precision, they can usually be com- 
 bined into a single relation, in the form 
 
 or 
 
 A=F(B,C,D, . . . ,P), 
 F(A,B,C,D, ,P)=0, 
 
 (iii) 
 
 which expresses the general course of the investigated phenomenon 
 in response to variations of the factors within the limits of the 
 observations. Such generalizations may be purely empirical or 
 they may rest partly or entirely on theoretical deductions from 
 well-established principles. In either case the test of their validity 
 lies in the exactness with which they represent observed facts. 
 While an exact empirical formula finds many useful applications 
 in practical problems it should not be assumed to express the true 
 physical nature of the phenomenon it represents. In fact our 
 understanding of any phenomenon is but scanty until we can 
 represent its course by a formula that gives explicit or implicit 
 expression to the physical principles that underlie it. Conse- 
 quently a research ought not to be considered complete until the 
 investigated phenomenon has been classified and represented by a 
 function that exhibits the physical relations among its factors. 
 (i It is scat cely necessary to point out that a complete research 
 as outlined above is seldom carried out by one man and that the 
 underlying ideas very rarely originate at the same time or in the 
 same person. The preliminary relations in the form of equations 
 (i) and (ii) are frequently inspired by independent ideas and 
 worked out by different men. The exact determination of any 
 
198 THE THEORY OF MEASUREMENTS [ART. 89 
 
 one of them constitutes a research that is complete so far as it 
 goes. The establishment of the general relation that compre- 
 hends all of the others and the interpretation of its physical signifi- 
 cance are generally the result of a process of gradual growth and 
 modification to which many men have contributed. 
 
 89. Graphical Methods of Reduction. After the necessary 
 measurements have been completed and corrected for all known 
 constant errors, the form of the functions appearing in equations 
 (i) and (ii), or other equations of similar type, and the numerical 
 value of the constants involved can sometimes be determined 
 easily and effectively by graphical methods. Such methods are 
 almost universally adopted for the discussion of preliminary obser- 
 vations and the determination of approximate values of the con- 
 stants. In some cases they are the only methods by which the 
 results of the measurements can be expressed. In some other 
 cases the constants can be more exactly determined by an appli- 
 cation of the method of least squares to be described later. Usu- 
 ally, however, the general form of the functions and approximate 
 values of the constants must first be determined by graphical 
 methods or otherwise. 
 
 Let x and y represent the simultaneous values of two variable 
 factors corresponding to specified constant values of the other 
 factors involved in the phenomenon under investigation. Suppose 
 that x has been varied by successive nearly equal steps through 
 as great a range as conditions permit and that the simultaneous 
 values x and y have been measured after each of these steps while 
 the factors that they represent were kept constant. If all other 
 factors have remained constant throughout these operations, the 
 above series of measurements on x and y may be applied at once 
 to the determination of the form and constants of the functional 
 relation 
 
 This expression is of the same type as equations (i.) and (ii). 
 Consequently the following discussion applies generally to all 
 cases in which there are only two variable factors. If the sup- 
 posedly constant factors are not strictly constant during the 
 measurements, the observations on x and y will not give the true 
 form of the function in (iv) until they have been corrected for 
 the effects of the variations thus introduced. 
 
ART. 89] RESEARCH 199 
 
 As the first step in the graphical method of reduction, the 
 observations on x and y are laid off as abscissae and ordinates on 
 accurately squared paper, and the points determined by corre- 
 sponding coordinates are accurately located with a fine pointed 
 needle. The visibility of these points is usually increased by 
 drawing a small circle or other figure with its center exactly at 
 the indicated point. The scale of the plot should be so chosen 
 that the form of the curve determined by the located points is 
 easily recognized by eye. In order to bring out the desired rela- 
 tion, it is frequently necessary to adopt a different scale for ordi- 
 nates and abscissae. Usually it is advantageous to choose such 
 scales that the total variations of x and y will be represented by 
 approximately equal spaces. Thus, if the total variation of y is 
 numerically equal to about one-tenth of the corresponding vari- 
 ation of x, the i/'s should be plotted to a scale approximately ten 
 times as large as that adopted for the x's. In all cases the adopted 
 scales should be clearly indicated by suitable numbers placed at 
 equal intervals along the vertical and horizontal axes. Letters 
 or other abbreviations should be placed near the ends of the axes 
 to indicate the quantities represented. 
 
 The points thus located usually lie very nearly on a uniform 
 curve that represents the functional relation (iv). Consequently 
 the problem in hand may be solved by determining the equation 
 of this curve and the numerical value of the constants involved 
 in it. Sometimes it is impossible or inadvisable to carry out such 
 a determination in practice and in such cases the plotted curve 
 is the only available means of representing the relation between 
 the observed factors. In all cases the deviations of the located 
 points from the uniform curve represent the residuals of the 
 observations, and, consequently, indicate the precision of the 
 measurements on x and y. 
 
 The simplest case, and one that frequently occurs in practice, is 
 illustrated in Fig. 13. The plotted points lie very nearly on a 
 straight line. Consequently the functional relation (iv) takes the 
 linear form 
 
 y = Ax + B, (v) 
 
 where A is the tangent of the angle a between the line and the 
 positive direction of the x axis, and B is the intercept OP on 
 the y axis. For the determination of the numerical values of the 
 
200 
 
 THE THEORY OF MEASUREMENTS [ART. 89 
 
 constants A and B, the line should be sharply drawn in such a 
 position that the plotted points deviate from it about equally in 
 opposite directions, i.e., the sum of the positive deviations should 
 be made as nearly as possible equal to the sum of the negative 
 deviations. If this has been carefully and accurately done, the 
 constant B may be determined by a direct measurement of the 
 intercept OP in terms of the scale used in plotting the y's- 
 
 0.10 
 
 05 
 
 25 
 
 FIG. 13. 
 
 50 
 
 75 
 
 The constant A may be computed from measurements of the 
 coordinates x\ and 2/1 of any point on the line, not one of the plotted 
 points, by the relation 
 
 If the position of the line is such that the point P does not fall 
 within the limits of the plotting sheet, the coordinates, Xi, y\ and 
 2, 2/2, of two points on the line are measured. Since they must 
 satisfy equation (v), 
 
 2/i = Axi + B, 
 and 
 
 2/2 = Ax 2 + B. 
 Consequently 
 
 A = and B 
 
 X 2 
 
 The points selected for this purpose should be as widely separated 
 as possible in order to reduce the effect of errors of plotting and 
 
ART. 89] RESEARCH 201 
 
 measurement. The accuracy of these determinations is likely to 
 be greatest when the vertical and horizontal scales are so chosen 
 that the line makes an angle of approximately forty-five degrees 
 with the x axis. Space may sometimes be saved and the appear- 
 ance of the plot improved by subtracting a constant quantity, 
 nearly equal to B, from each of the y's before they are plotted. 
 
 Many physical relations are not linear in form. Perhaps none 
 of them are strictly linear when large ranges of variation are con- 
 sidered. Consequently the plotted points are more likely to lie 
 nearly on some regular curve than on a straight line. In such 
 cases the form of the functional relation (iv) is sometimes sug- 
 gested by theoretical considerations, but frequently it must be 
 determined by the method of trial and error or successive approxi- 
 mations. For this purpose the curve representing the observa- 
 tions is compared with a number of curves representing known 
 equations. The equation of the curve that comes nearest to the 
 desired form is modified by altering the numerical values of its 
 constants until it represents the given measurements within the 
 accidental errors of observation. Frequently several different 
 equations and a number of modifications of the constants must 
 be tried before satisfactory agreement is obtained. 
 
 When the desired relation does not contain more than two inde- 
 pendent constants, it can sometimes be reduced to a linear relation 
 between simple functions of x and y. Thus, the equation 
 
 y = Be~ Ax , . (vi) 
 
 represented by the curve in Fig. 14, is frequently met with in 
 physical investigations. By inverting (vi) and introducing ' log- 
 arithms, we obtain the relation 
 
 log* y = log* B - Ax. 
 
 Hence if the logarithms of the y's are laid off as ordinates against 
 the corresponding x's as abscissae, the located points will lie very 
 nearly on a straight line if the given observations satisfy the func- 
 tional relation (vi) . When this is the case, the constants A and 
 loge B may be determined by the methods developed during the 
 discussion of equation (v). If logarithms to the base ten are 
 
 used the above equation becomes 
 
 ^| 
 log y = logio B - x, 
 
202 
 
 THE THEORY OF MEASUREMENTS [ART. 89 
 
 where M is the modulus of the natural system of logarithms. In 
 
 ^ 
 this case the plot gives the values of logio B and -^ from which 
 
 the constants A and B can be easily computed. When the plotted 
 points do not lie nearer to a straight line than to any other curve, 
 y 
 
 10 
 
 \ 
 
 \ 
 
 0.5 
 
 1.0 
 
 1.5 
 
 FIG. 14. 
 
 equation (vi) does not represent the functional relation between 
 the observed factors and some other form must be tried. Many 
 of the commonly occurring forms may be treated by the above 
 method and the process is usually so simple that further illustra- 
 tion seems unnecessary. 
 
 The curve determined by plotting the x's and y's directly fre- 
 quently exhibits points of discontinuity or sharp bends as at p 
 and q in Fig. 15. Such irregularities are generally due to changes 
 in the state of the material under investigation. The nature- and 
 causes of such changes are frequently determined, or at least 
 suggested, by the location and character of such points. The 
 different branches of the curve may correspond to entirely differ- 
 ent equations or to equations in the same form but with different 
 constants. In either case the equation of each branch must be 
 determined separately. 
 
 The accuracy attainable by graphical methods depends very 
 largely on the skill of the draughtsman in choosing suitable scales 
 and executing the necessary operations. In many cases the errors 
 
ART. 90] 
 
 RESEARCH 
 
 203 
 
 due to the plot are less than the errors of observation and it would 
 be useless to adopt a more precise method of reduction. When 
 the means of control are so well devised and effective that the 
 constant errors left in the measurements are less than the errors 
 of plotting it is probably worth while to make the reductions by 
 the method of least squares, as explained in the following article. 
 y 
 
 'FiG. 15. 
 
 90. Application of the Method of Least Squares. In the 
 
 case of linear relations, expressible in the form of equation (v), 
 the best values of the constants A and B can be very easily deter- 
 mined by applying the method of least squares in the manner 
 explained in article fifty-one. However, as pointed out in the 
 preceding article, very few physical relations are strictly linear 
 when large variations of the involved factors are considered. 
 Consequently a straight line, corresponding to constants deter- 
 mined as above, usually represents only a small part of the course 
 of the investigated phenomenon. Such a line is generally a short 
 chord of the curve that represents the true relation and conse- 
 quently its direction depends on the particular range covered by 
 the observations from which it is derived. 
 
 When the measurements are extended over a sufficiently wide 
 range, the points plotted from them usually deviate from a straight 
 line in an approximately regular manner, as illustrated in Fig. 16, 
 
204 
 
 THE THEORY OF MEASUREMENTS [ART. 90 
 
 and lie very near to a continuous curve of slight curvature. Meas- 
 urements of this type can always be represented empirically by a 
 power series in the form 
 
 y = A + Bx + Cx* + . - - , (vii) 
 
 the number of terms and the signs of the constants depending on 
 the magnitude and sign of the curvature to be represented. 
 
 FIG. 16. 
 
 Since equation (vii) is linear with respect to the constants A, B, 
 C, etc., they might be computed directly from the observations 
 on x and y by the method of least squares. Usually, however, 
 the computations can be simplified by introducing approximate 
 values of the constants A and B. Thus, let A' and B' represent 
 two numerical quantities so chosen that the line 
 
 y' = A' + B'x 
 
 passes in the same general direction as the plotted points, in the 
 manner illustrated by the dotted line in pig. 16. The difference 
 between y and y' can be put in the form 
 
 y y' = (A A') + MI (B B'} -^ 4- M 2 C + . . . (viii) 
 
 MI M 2 
 
 where Afi, M 2 , etc., represent numerical constants so chosen that 
 
 *Y* s2 
 
 the quantities y - y', -=, etc., are nearly of the same order 
 
ART. 90] RESEARCH 205 
 
 of magnitude. For the sake of convenience let 
 
 (ix) 
 and 
 
 The quantities s, 6, c, etc., may be derived from the observations, 
 with the aid of the assumed constants A', B', MI, M z , etc.; and xi, 
 x z , x S} etc., are the unknowns to be computed by the method of 
 least squares. After the above substitutions, equation (viii) takes 
 the simple form 
 
 xi + bx 2 + cx 3 + = s, 
 
 which is identical with that of the observation equations (53), 
 article forty-nine. As many equations of this type may be formed 
 as there are pairs of corresponding measurements on x andj y. 
 
 The normal equations (56) may be derived from the observation 
 equations thus established, by the methods explained in articles 
 fifty and fifty-three. Their final solution for the unknowns Xi, Xz, 
 xsj etc., may be effected by Gauss's method, developed in article 
 fifty-four and illustrated in article fifty-five, or by any other con- 
 venient method. The corresponding numerical values of the 
 constants A, B, C, etc., may then be computed by equations (ix). 
 These values, when substituted in (vii) , give the required empirical 
 relation between x and y. 
 
 If a sufficient number of terms have been included in equation 
 (vii), the relation thus established will represent the given measure- 
 ments within the accidental errors of observation. The residuals, 
 computed by equations (54), article forty-nine, and arranged in 
 the order of increasing values of y, should show approximately as 
 many sign changes as sign follows. When this is not the case 
 the observed y's deviate systematically from the values given by 
 equation (vii) for corresponding x's. In such cases the number of 
 terms employed is not sufficient for the exact representation of the 
 observed phenomenon, and a new relation in the same general 
 form as the one already tested but containing more independent 
 constants should be determined. This process must be repeated 
 until such a relation is established that systematically varying 
 differences between observed and computed y's no longer occur. 
 
 The observation equations used as a basis for the numerical 
 illustration given in article fifty-five were derived from the follow- 
 
206 
 
 THE THEORY OF MEASUREMENTS [ART. 90 
 
 ing observations on the thermal expansion of petroleum by equa- 
 tions (viii) and (ix), taking 
 
 A' = 1000; B' = l; M l = 10; and M 2 = 1000. 
 
 X 
 
 temperature 
 
 volume 
 
 degrees 
 
 cc. 
 
 
 
 1000.24 
 
 20 
 
 1018.82 
 
 40 
 
 1038.47 
 
 60 
 
 1059 31 
 
 80 
 
 1081.20 
 
 100 
 
 1104.27 
 
 The computations carried out in the cited article resulted as 
 
 follows : 
 
 xi = 0.245; x 2 = - 1.0003; x 3 = 1.4022. 
 
 Hence, by equations (ix) 
 
 A = 1000.245; B = 0.89997; C = 0.0014022, 
 
 and the functional relation (vii) becomes 
 
 y = 1000.245 + 0.89997 x + 0.0014022 x\ 
 
 The residuals corresponding to this relation, computed and tab- 
 ulated in article fifty-five, show five sign changes and no sign 
 follows. Such a distribution of signs sometimes indicates that the 
 observed factors deviate periodically from the assumed functional 
 relation. In the present case, however, the number of observa- 
 tions is so small that the apparent indications of the residuals are 
 probably fortuitous. Consequently it would not be worth while 
 to repeat the computations with a larger number of terms unless 
 it could be shown by independent means that the accidental errors 
 of the observations are less than the residuals corresponding to the 
 above relation. 
 
 Any continuous relation between two variables can usually be 
 represented empirically by an expression in the form of equation 
 (vii). However, it frequently happens that the physical signifi- 
 cance of the investigated phenomenon is not suggested by such 
 an expression but is represented explicitly by a function that is not 
 linear with respect to either the variable factors or the constants 
 involved. Such functions usually contain more than two inde- 
 pendent constants and sometimes include more than two variable 
 factors. They may be expressed by the general equation 
 
 y = F(A,B,C,. ,x,z,. . ), (154) 
 
ART. 90] RESEARCH 207 
 
 where A, B, C, etc., represent constants to be determined and y t x, 
 z, etc., represent corresponding values of observed factors. 
 
 Sometimes the form of the function F is given by theoretical 
 considerations, but more frequently it must be determined, to- 
 gether with the numerical values of the constants, by the method 
 of successive approximations. In the latter case a definite form, 
 suggested by the graphical representation of the observations or 
 by analogy with similar phenomena, is assumed tentatively as a 
 first approximation. Then, by substituting a number of different 
 corresponding observations on y, x, z, etc., in (154), as many inde- 
 pendent equations as there are constants in the assumed function 
 are established. The simultaneous solution of these equations 
 gives first approximations to the values of the constants A, B, C, 
 etc. Sometimes the solution cannot be effected directly by means 
 of the ordinary algebraic methods, but it can usually be accom- 
 plished with sufficient accuracy either by trial and error or by 
 some other method of approximation. 
 
 Let A', B' ', C', etc., represent approximate values of the con- 
 stants and let 61, 5 2 , 5 3 , etc., represent their respective deviations 
 from the true values. Then 
 
 A=A' + 5 1 ; B = B' + d 2 ] C = C' + 5 3 , etc., (155) 
 and (154) may be put in the form 
 y-F\(A' + Sd, (B' + fc), (C" + .) ---- ,*,*, . - . | (x) 
 
 If the S's are so small that their squares and higher powers may 
 be neglected, expansion by Taylor's Theorem gives 
 
 y-F(A',B',C', . . . ,x,z, . . 
 dF dF , dF 
 
 ,,. . .,,,.. 
 
 By putting 
 
 y-F(A',B',C', . . . ,x,z, . . . ) = ; 
 
 (156) 
 
 and transposing, equation (xi) becomes 
 
 adi + 65 2 + c5 3 + . . . = s. (157) 
 
 As many independent equations of this type as there are sets of 
 corresponding observations on y, x, z, etc., can be formed. The 
 absolute term s and the coefficients a, 6, c, etc., in each equation 
 are computed from a single set of observations by the relations 
 
208 THE THEORY OF MEASUREMENTS [ART. 90 
 
 (156) with the aid of the approximate values A', B f , C", etc. Since 
 the resulting equations are in the same form as the observation 
 equations (53), the normal equations (56) may be found and 
 solved by the methods described in Chapter VII. The values 
 of $1, 6 2 , 5 3 , etc., thus obtained, when substituted in (155), give 
 second approximations to the values of the constants A, B, C, 
 etc. 
 
 The accuracy of the second approximations will depend on the 
 assumed form of the function F and on the magnitude of the correc- 
 tions Si, 6 2 , 6 3 , etc. If these corrections are not small, the con- 
 ditions underlying equation (xi) are not fulfilled and the results 
 obtained by the above process may deviate widely from the correct 
 values of the constants; but, except in extreme cases, they are 
 more accurate than the first approximations A', B f , C', etc. Let 
 A", B", C", etc., represent the second approximations. The 
 corresponding residuals, n, r 2 , . . . , r n , may be computed by 
 substituting different sets of corresponding observations on y, 
 x, z, etc., successively in the equation 
 
 F(A",B",C", . . . ,x,z, . . . )-y = r, (xii) 
 
 where the function F has the same form that was used in comput- 
 ing the corrections 5i, ^2, 5 3 , etc. If these residuals are of the same 
 order of magnitude as the accidental errors of the observations 
 and distributed in accordance with the laws of such errors, the 
 functional relation 
 
 y = F(A",B",C", . . . ,x,z, . . . ) (158) 
 
 is the most probable result that can be derived from the given 
 observations. 
 
 Frequently the residuals corresponding to the second approxi- 
 mations do not atisfy the above conditions. This may be due 
 to the inadequacy of the assumed form of the function F, to 
 insufficient precision of the approximations A", B", C", etc., or 
 to both of these causes. 
 
 If the form of the function is faulty, the residuals usually show 
 systematic and easily recognizable deviations from the distribu- 
 tion characteristic of accidental errors. Generally the number of 
 sign follows greatly exceeds the number of sign changes, when the 
 residuals are arranged in the order of increasing y's, and opposite 
 signs do not occur with nearly the same frequency. Sometimes 
 the nature of the fault can be determined by inspecting the order 
 
ART. 91] RESEARCH 209 
 
 of sequence of the residuals or by comparing the graph correspond- 
 ing to equation (158) with the plotted observations. After the 
 form of the function F has been rectified, by the above means or 
 otherwise, the computations must be repeated from the beginning 
 and the new form must be tested in the same manner as its prede- 
 cessor. This process should be continued until the residuals cor- 
 responding to the second approximations give no evidence that 
 the form of the function on which they depend is faulty. 
 
 When the residuals, computed by equation (xii), do not suggest 
 that the assumed form of the function F is inadequate, but are 
 large in comparison with the probable errors of the observations, 
 the second approximations are not sufficiently exact. In such 
 cases new equations in the form of (157) are derived by using A", 
 B" , C", etc., in place of A', B f , C', etc., in equations (156). The 
 solution of the equations thus formed, by the method of least 
 squares, gives the corrections 5/, 5 2 ', 5 3 ', etc., that must be applied 
 to A", B", C", etc., in order to obtain the third approximations 
 
 At tt A n I x / . T>itt ~Dir I <j / . r</n rut \ * t . 4. 
 = A -f- di ; > = n + 62 ; C = C + 03 ; etc. 
 
 These operations must be repeated until the residuals correspond- 
 ing to the last approximations are of the same order of magnitude 
 as the accidental errors of the observations. 
 
 Although an algebraic expression, that represents any given 
 series of observations with sufficient precision, can usually be de- 
 rived by the foregoing methods, such a procedure is by no means 
 advisable in all cases. In many investigations, a graphical repre- 
 sentation of the results leads to quite as definite and trustworthy 
 conclusions as the more tedious mathematical process. Conse- 
 quently the latter method is usually adopted only when the former 
 is inapplicable or fails to utilize the full precision of the observa- 
 tions. In all cases the choice of suitable methods and the estab- 
 lishment of rational conclusions is a matter of judgment and 
 experience. 
 
 91. Publication. Research does not become effective as a 
 factor in the advancement of science until its results have been 
 published, or otherwise reported, in intelligible and widely acces- 
 sible form. It is the duty as well as the privilege of the investiga- 
 tor to make such report as soon as he has arrived at definite 
 conclusions. But nothing could be more inadvisable or untimely 
 than the premature publication of observations that have not been 
 thoroughly discussed and correlated with fundamental principles. 
 
210 THE THEORY OF MEASUREMENTS [ART. 91 
 
 Until an investigation has progressed to such a point that it makes 
 some definite addition to existing ideas, or gives some important 
 physical constant with increased precision, its publication is likely 
 to retard rather than stimulate the progress of science. On the 
 other hand, free discussion of methods and preliminary results is 
 an effective molder of ideas. 
 
 The form of a published report is scarcely less important than 
 the substance. The significance of the most brilliant ideas may 
 be entirely masked by faulty or inadequate expression. Hence 
 the investigator should strive to develop a lucid and concise style 
 that will present his ideas and the observations that support 
 them in logical sequence. Above all things he should remember 
 that the value of a scientific communication is measured by the 
 importance of the underlying ideas, not by its length. 
 
 The author's point of view, the problem he proposes to solve, 
 and the ideas that have guided his work should be clearly defined. 
 Theoretical considerations should be rigorously developed in so 
 far as they have direct bearing on the work in hand. But general 
 discussions that can be found in well-known treatises or in easily 
 accessible journals should be given by reference, and the formulae 
 derived therein assumed without further proof whenever their 
 rigor is not questioned. However, the author should always 
 explain his own interpretation of adopted formulae and point out 
 their significance with respect to his observations. Due weight 
 and credit should be given to the ideas and results of other workers 
 in the same or closely related fields, but lengthy descriptions of 
 their methods and apparatus should be avoided. Explicit refer- 
 ence to original sources is usually sufficient. 
 
 The methods and apparatus actually used in making the re- 
 ported observations, should be concisely described, with the aid 
 of schematic diagrams whenever possible. Well-known methods 
 and instruments should be described only in so far as they have 
 been modified to fulfill special purposes. Detailed discussion of 
 all of the methods and instruments that have been found to be 
 inadequate are generally superfluous, but the difficulties that have 
 been overcome should be briefly pointed out and explained. The 
 precautions adopted to avoid constant errors should be explicitly 
 stated and the processes by which unavoidable errors of this 
 type have been removed from the measurements should be clearly 
 described. The effects likely to arise from such errors should be 
 
ART. 91] RESEARCH 211 
 
 considered briefly and the magnitude of applied corrections should 
 be stated. 
 
 Observations and the results derived from them should be 
 reported in such form that their significance is readily intelligible 
 and their precision easily ascertainable. In many cases graphical 
 methods of representation are the most suitable provided the 
 points determined by the observations are accurately located 
 and marked. The reproduction of a large mass of numerical data 
 is thus avoided without detracting from the comprehensiveness 
 of the report. When such methods do not exhibit the full pre- 
 cision of the observations or when they are inapplicable on account 
 of the nature of the problem in hand, the original data should be 
 reproduced with sufficient fullness to substantiate the conclusions 
 drawn from them. In such cases the significance of the obser- 
 vations and derived results can generally be most convincingly 
 brought out by a suitable tabulation of numerical data. An 
 estimate of the precision attained should be made whenever the 
 results of the investigation can -be expressed numerically. 
 
 Final conclusions should be logically drawn, explicitly stated, 
 and rigorously developed in their theoretical bearings. They 
 express a culmination of the author's ideas relative to the inves- 
 tigated phenomena and invite criticism of their exactness and 
 rationality. Unless they are amply substantiated by the obser- 
 vations and theoretical considerations brought forward in their 
 support, and constitute a real addition to scientific knowledge, 
 they are likely to receive scant recognition. 
 
TABLES. 
 
 The following tables contain formulae and numerical data that 
 will be found useful to the student in applying the principles 
 developed in the preceding chapters. The four figure numerical 
 tables are amply sufficient for the computation of errors, but more 
 extensive tables should be used in computing indirectly measured 
 magnitudes whenever the precision of the observations warrants 
 the use of more than four significant figures. 
 
 The references placed under some of the tables indicate the 
 texts from which they were adapted. 
 
 TABLE I. DIMENSIONS OF UNITS. 
 
 Units. 
 
 Dimensions. 
 
 Fundamental. 
 
 Length, mass, time 
 
 Length, force, time. 
 
 Length 
 
 [L] 
 [M] 
 [T] 
 [LMT-*] 
 
 m 
 
 M 
 
 [L-W] 
 [Llr*\ 
 
 [LT-i] 
 
 pNj 
 
 [LT-*\ 
 [T- 2 ] 
 [LMT~ l ] 
 [L*M] 
 [LW7 7 - 1 ] 
 [L*MT-*] 
 [L-W7 1 - 2 ] 
 [LW7 7 - 2 ] 
 
 [Lwr- 8 ] 
 
 [L] 
 [L-iFT*] 
 [T] 
 [F] 
 [V] 
 [If] 
 [L-*FT*] 
 [LL-i] 
 [LT-i] 
 [T- 1 ] 
 [LT-*] 
 [T - 2] 
 
 [FT] 
 [LFT*] 
 [LFT] 
 [LF] 
 [L-*F] 
 [LF] 
 [LFT- 1 ] 
 
 Mass 
 
 Time 
 
 Force 
 
 Area 
 
 Volume 
 
 Density 
 
 Angle 
 
 Velocity, linear 
 
 Velocity, angular 
 
 Acceleration, linear 
 
 Acceleration, angular 
 
 Momentum 
 
 Moment of inertia 
 
 Moment of momentum 
 
 Torque 
 
 Pressure 
 
 Energy, work 
 
 Power 
 
 
 212 
 
TABLES 
 
 213 
 
 TABLE II. CONVERSION FACTORS. 
 
 Length Units. 
 
 Logarithm. 
 
 1 centimeter (cm.) _ = 0. 393700 inch 1 . 5951654 
 
 " " = 0. 0328083 foot 2. 5159842 
 
 " = 0. 0109361 yard 2. 0388629 
 
 1 meter (m.) = 1000 millimeters 3. 0000000 
 
 " = 100 centimeters 2. 0000000 
 
 " = 10 decimeters. 1.0000000 
 
 1 kilometer (km.) = 1000 meters 3. 0000000 
 
 = 0. 621370 mile 1. 7933503 
 
 " = 3280. 83 feet 3. 5159842 
 
 1 inch (in.) = 2. 540005 centimeters 0. 4048346 
 
 1 foot (ft.) =12 inches 1.0791812 
 
 = 30. 4801 centimeters 1 . 4840158 
 
 1 yard (yd.) = 36 inches 1. 5563025 
 
 " =3 feet 0. 4771213 
 
 " = 91.4402 centimeters 1.9611371 
 
 1 mile (ml.) = 5280 feet 3 . 7226339 
 
 " = 1760 yards 3. 2455127 
 
 = 1609. 35 meters 3.2066497 
 
 = 0.868392 knot (U. S.) 1.9387157 
 
 Mass Units. 
 
 1 gram (g.) = 1000 milligrams 3. 0000000 
 
 " = 100 centigrams 2. 0000000 
 
 " = 10 decigrams 1. 0000000 
 
 = 0.0352740 ounce (av.) 2.5474542 
 
 " = 0. 00220462 pound (av.) 3. 3433342 
 
 = 0. 000068486 slugg 5. 8355997 
 
 1 kilogram (kg.) = 1000 grams 3. 0000000 
 
 1 ounce (oz.) (av.) = 28. 3495 grams 1. 4525458 
 
 = 0. 062500 pound (av.) 2. 7958800 
 
 " =0.0019415 slugg 3.2881455 
 
 1 pound (Ib.) (av.) = 16 ounces (av.) 1.2041200 
 
 " = 453. 5924277 grams 2. 6566658 
 
 = 0.0310646 slugg 2.4922655 
 
 1 slugg (sg.) = 32. 191 pounds (av.) 1. 5077345 
 
 = 515.06 ounces (av.) 2.7118545 
 
 = 14601. 6 grams 4. 1644003 
 
 1 short ton (tn.) = 2000 pounds (av.) 3. 3010300 
 
 = 907. 185 kilograms 2. 9576958 
 
 " =62. 129 sluggs 1 . 7932955 
 
214 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE II. CONVERSION FACTORS (Concluded}. 
 
 Force Units. 
 
 The following gravitational units are expressed in terms of the earth's 
 attraction at London where the acceleration due to gravity is 32.191 ft. /sec. 2 
 
 or 981.19 cm./sec Logarithm. 
 
 1 dyne = 1 . 01917 milligram's wt 0. 0082469 
 
 " = 0. 00101917 gram's wt 3 . 0082469 
 
 " =2.2469 X 10- 6 pound's wt 6.3515811 
 
 1 gram's wt. = 981.19 dynes 2. 9917531 
 
 1 kilogram's wt. = 1000 gram's wt 3. 0000000 
 
 = 98. 119 X 10 4 dynes 5.9917531 
 
 = 2.20462 pound's wt 0. 3433342 
 
 1 pound's wt. =0. 45359 kilogram's wt 1 . 6566658 
 
 = 44.506 X 10 4 dynes 5.6484189 
 
 1 pound's wt. (local) = 0/32.191 pound's wt. at London. 
 
 g = local acceleration due to gravity in ft./secT 2 . 
 
 Mean Solar Time Units. 
 
 1 second (s.) = 0. 016667 minute 2. 2218487 
 
 " = 0. 00027778 hour 4. 4436975 
 
 = 0.000011574 day 5.0634863 
 
 1 minute (m.) = 60 seconds 1 . 7781513 
 
 " =0.016667 hour 2.2218487 
 
 = 0.00069444 day 4.8416375 
 
 1 hour (h.) = 3600 seconds 3. 5563025 
 
 = 60 minutes 1. 7781513 
 
 " = 0. 041667 day 2. 6197888 
 
 1 day (d.) = 86400 seconds 4. 9365137 
 
 = 1440 minutes 3. 1583625 
 
 " =24 hours 1.3802112 
 
 1 mean solar unit = 1 . 00273791 sidereal units 0. 0011874 
 
 Angle Units. 
 
 1 circumference = 360 degrees 2. 5563025 
 
 = 2 TT radians 0. 7981799 
 
 " = 6.28319 radians 0. 7981799 
 
 1 degree () = 0. 017453 radian 2. 2418774 
 
 = 60 minutes 1. 7781513 
 
 = 3600 seconds 3. 5563025 
 
 1 minute (') =2. 9089 X 10- 4 radians 4 . 4637261 
 
 = 0.016667 degree 2.2218487 
 
 = 60 seconds 1. 7781513 
 
 1 second (') = 4.8481 X KH 5 radians 6 . 6855749 
 
 = 2. 7778 X 10- 4 degrees 4. 4436975 
 
 = 0. 01667 minute 2. 2218487 
 
 1 radian = 57.29578 degrees 1. 7581226 
 
 = 3437.7468 minutes 3. 5362739 
 
 = 206264.8 seconds . . 5. 3144251 
 
TABLES 215 
 
 TABLE III. TRIGONOMETRICAL RELATIONS. 
 
 a 3 . a 5 t <t\, 
 
 sma = a 777 +T? (!) 
 
 (2n-l)! 
 
 cos 2 a = 
 
 1 cos 2 a 
 
 2 cosec a 
 
 _ . ce a cos a tan a 
 
 = 2 sin ^ cos tr 
 
 2 2 cot a sec a 
 tan a 1 
 
 - = cos a tan a 
 
 Vl+tan 2 a VI + cot 2 a 
 = sin /3 cos (|8 a) cos /3 sin (/3 a) 
 = cos /3 sin (0 + a) sin /8 cos (/3 + a). 
 
 l/l 
 a =y 
 
 cos a 
 2~~ 
 
 2 tan a 
 
 sin 2 a = 2 sin a cos a = ., 
 
 1 + tan 2 a. 
 
 sin 2 a = 1 cos 2 a = \ (cos 2 a 1). 
 
 sin (a j8) = sin a cos cos a sin 0. 
 
 sin a =fc sin /3 = 2 sin (a d= /8) cos |( =F 0). 
 
 sin 2 a + sin 2 /3 = 1 cos (a + /3) cos (a /8). 
 
 sin 2 a sin 2 = cos 2 /3 cos 2 a = sin (a + 0) sin (a /8). 
 
 V 1 + sin a = sin | a + cos a. 
 
 VI sin a = (sin | a cos \ a). 
 
 cos 
 
 cos 2 i a sin 2 | a 
 cot a 
 
 V 1 + tan 2 a V 1 + cot 2 a 
 sin a cot a 1 
 
 = sin a cot a 
 
 COS ^ a = 
 
 tan o; cosec a sec a 
 = cos cos (a + /8) + sin sin (a + 0) 
 = cos /? cos ((8 - a) + sin ft sin (0 a). 
 
 1 + cos a 
 
216 THE THEORY OF MEASUREMENTS 
 
 TABLE III. TRIGONOMETRICAL RELATIONS (Continued). 
 
 cos 2 a = 2 cos 2 a 1 = 1 2 sin 2 a 
 
 1 - tan 2 a 
 
 = cos 2 a sm 2 a. = ^ 
 
 1 + tan 2 a 
 
 cos 2 a. = 1 - sin 2 a = (cos 2 a + 1). 
 
 cos (a d= 0) = cos a cos T sin a sin 0. 
 
 cos a + cos = 2 cos 5 (a + 0) cos H 0)- 
 
 cos a cos = 2 sin (a -f 0) sin | (a 0) . 
 
 cos 2 a + cos 2 = 1 + cos (a + 0) cos (a - 0). 
 
 cos 2 a cos 2 = sin 2 sin 2 a = sin (a + 0) sin (a 0). 
 
 cos 2 a sin 2 = cos (a + 0) cos (a 0) = cos 2 sin 2 a. 
 
 sin a + cos a = V 1 + sin 2 a. 
 
 sin a cos a = Vi sin 2 a. 
 
 sin 2 a + cos 2 a. = 1. 
 
 sin 2 a cos 2 a: = cos 2 or. 
 
 tan a = a + | a 3 + - r 2 5 CK 5 + 3^5 a 7 + . . . w > a> TT 
 
 sin a. sin 2 a 1 cos 2 
 
 cos a 1 + cos 2 a sin 2 a 
 
 V'l cos 2 a _ 4 / 
 1+ cos 2 a " V 
 
 cos 2 a VI sin 2 a 
 
 = Vsec 2 a I 
 
 tan 2 a = 
 
 cosec a: Vcosec 2 a-l 
 
 = cot a 2 cot 2 a 
 cot a 
 
 sin (a + 0) + sin ( 0) _ cos (a 0) cos (a + 0) 
 cos (a + 0) + cos (a 0) sin (a + 0) - sin (a - 0) 
 
 2 tan a 2 cot a 2 
 
 1 tan 2 a cot 2 a 1 cot a tan a 
 
 tan f a. = - ; - = cosec a cot or. 
 1 + sec a 
 
 ( . R\ tan a tan _ cos 2 cos 2 a 
 * W * 1 T tan a tan ~ sin 2 =F sin 2 a 
 
 sin (a 0) 
 
 tan a tan = 
 
 cos a cos 
 
TABLES 217 
 
 TABLE III. TRIGONOMETRICAL RELATIONS (Concluded). 
 
 Ill 2 
 
 cot a. = -- - a j= a 3 ^r-= a 5 TT > a > IT 
 a: 3 45 olo 
 
 cos a _ sin 2 a _ 1 + cos 2 a 
 
 sin a ~~ 1 cos 2 a "~ sin 2 a 
 
 V/ 
 
 1 + cos 2 o: _ cos a vl sin 2 a 
 
 1 cos 2 a Vl cos 2 a 
 
 = tan a. + 2 cot 2 a. 
 tan a. 
 
 _ 1 tan 2 a. _ cot 2 a 1 cot a tan a 
 " 2 tan a 2cota ~^~ 
 
 cot - a = (1 + sec a) cot a 
 
 2<-. v j. | kjv/v; <-*. y vv/u c*. : 
 
 cosec a cot a 
 
 1 =F tan a tan /? cot cot =F 1 
 
 cot (a d= 0) = 
 
 tan a tan cot d= cot a 
 
 sin 
 
 TABLE IV. SERIES. 
 
 Taylor's Theorem. 
 
 /(*+&)=/(*) + AT (*) + ^/" (*)+;+ ^/W (x) + 
 
 f(x + h, y + k, 
 
 where u = f (x, y, z). 
 
 Maclaurin's Series. 
 
 /(0) + f /' (0) + !/" (0) + + fj/N (0). 
 
218 THE THEORY OF MEASUREMENTS 
 
 TABLE IV. SERIES (Concluded}. 
 
 Binomial Theorem. 
 
 = xm + rn x ^ ly + m(n^ xm _^ + 
 
 . . . , * (m - 1) . . . (m - n + 1) ^- y> 
 
 when m is a positive integer, also when m is negative or fractional and 
 x > y. When x < y and m is fractional or negative the series must be 
 taken in the form 
 
 (x + y) m = y m + j y m -*x+ v ^ *' y *-'z + 
 
 m (m - 1) . . . (m - n + 1) 
 
 n! 
 Fourier's Series. 
 
 j- / \ It it ""E i t 2 7TX , 3 7TX . 
 
 / (x) = - 6 + &i cos H &2 cos - + 6 3 cos H 
 
 C C C 
 
 . TTX . . 2irX . . STTX , 
 
 + 01 sin \- a 2 sin h a 3 sin f- 
 
 c c c 
 
 where 
 
 1 r + c ,/ v WTTX . 
 >m = ~ I / (*) COS - dx, 
 
 C / c t/ 
 
 1 f +c r/ v . m-n-x , 
 m = - \ f(x) sm dx, 
 
 C / c ^ 
 
 2 / c , , , . WTTX , 
 
 = - I / W sin - " 
 
 C /o C 
 
 provided / (x) is single valued, uniform, and continuous, and c > x > 
 c. For values of x lying between zero and c the function may be ex- 
 panded in the form 
 
 , / x . TTX . . 2-JTX . . 3 TTX , 
 
 f (x) = 0,1 sin -- \-a-i sin -- H a 3 sin --- (- , 
 where a 
 
 Also f(x) =^60 + 61 cosy 4-6 2 cos + 6 3 cos 
 
 2 r c - / x WTTX , 
 
 where b m = - I / (x) cos - ax. 
 
 C JQ C 
 
 General Series. 
 
 xloga (x log a) 2 (x log a) 3 (x log a) n 
 
 ~~ ~~ ~~ ~ 
 
 .- :>} 
 
TABLES 
 
 219 
 
 TABLE V. DERIVATIVES. 
 U, F, W any functions; a, 6, c constants. 
 
 dx 
 
 F 2 
 
 S : ***St^T? 
 
 axx 
 
 a , log a e. 
 
 _log a x= , 
 
 dU 
 
 a . i at; 
 
 _ logaC7 . = __ 
 
 V dx 
 
 = a x log a. 
 
 dx 
 
 d 
 dx ( 
 
 a 
 
 dx 
 
 sm x = cos x. 
 ax 
 
 . r , . 
 
 sm aC7 = a cos ac7 ^ , 
 ax ax 
 
 a l 
 
 tan x = r = sec 2 x: 
 ax cos 2 x 
 
 cos x = sm x. 
 ax 
 
 a -i 
 
 cot x = . , = cosec 2 x. 
 ax sin 2 x 
 
 sec x = tan x sec x; 
 
 oX 
 
 cosec x = cot x cosec x. 
 ax 
 
 log sinx = cotx; 
 
 log cos x = tan x. 
 
 ox 
 
 The following expressions for the derivatives of inverse functions hold 
 for angles in the first and third quadrants. For angles in the second and 
 fourth quadrants the signs should be reversed. 
 
 ax 
 
 tan- 1 x = = 
 
 ax i 
 
 . 
 
 T- cos- 1 x = 
 dx 
 
 i 
 
220 THE THEORY OF MEASUREMENTS 
 
 TABLE VI. SOLUTION OF EQUATIONS. 
 
 The following algebraic expressions for the roots of equations of the 
 second, third, and fourth degrees are in the form given by Merriman. 
 (Merriman and Woodward, "Higher Mathematics"; Wiley and Sons, 
 1896.) 
 
 The Quadratic Equation. 
 Reduce to the form 
 
 x 2 
 Then the two roots are 
 
 x\ = a + a? 6; z 2 = a Va 2 b 
 
 The Cubic Equation. 
 Reduce to the form 
 
 = 0. 
 
 Compute the following auxiliary quantities : 
 
 B = - a 2 + 6; C = a 3 - f ab + c; 
 
 Then the three roots are 
 
 xi=-a + (si + s 2 ), _ 
 
 x z =-a -Mi+s 2 ) +| V-_3( Sl -s 2 ), 
 x 3 = - a - HSI + s 2 ) - | V- 3 (si - s a ). 
 
 When B 3 + C 2 is negative the roots are all real but they cannot be de- 
 termined numerically by the above formulae owing to the complex nature 
 of si and s 2 . In such cases the numerical values of the roots can be deter- 
 mined only by some method of approximation. 
 
 The Quartic Equation. 
 Reduce to the form 
 
 z 4 + 4az 3 + 66z 2 + 4cz + d = 0. 
 Compute the following auxiliary quantities : 
 
 g = a*-b; h = 6 3 + c 2 -2abc + dg; fc = |ac - 6 2 - |d; 
 
 I = I (h + V^TF')* + 1 (h - VF+^)*; 
 
 u = g + l', v = 2g-l; w = 4u* + 3k - 12gl. 
 Then the four roots are 
 
 xi = a + ^u + Vy + 
 
 a u 
 
 in which the signs are to be used as written provided that 2 a 3 3 ab + c 
 is a negative number; but if this is positive all radicals except Vw are to 
 be changed in sign. 
 
 The above expressions are irreducible when h z + k* is a negative number. 
 In this case the given equation has either four imaginary roots or four real 
 roots that can be determined numerically only by some method of approxi- 
 mation. 
 
TABLES 221 
 
 TABLE VII. APPROXIMATE FORMULA. 
 
 In the following formulae, a, /3, 5, etc., represent quantities so small that 
 their squares, higher powers, and products are negligible in comparison with 
 unity. The limit of negligibility depends on the particular problem in 
 hand. Most of the formulae give results within one part in one million 
 when the variables are equal to or less than 0.001. 
 
 1. (l+a) n =l+n; (1 -a) n = 1 - na. 
 
 4. 
 
 6 l = 1 --' , l = 1 +- 
 ' Vl+ n' Vl -a n 
 
 7. 
 
 9. (x + a 
 
 When the angle a, expressed in radians, is small in comparison with unity 
 a first approximation gives 
 
 10. sin a = a', sin (x a) = sin x a cos x. 
 
 11. cos a = 1; cos (x a) = cos x =F a sin x. 
 
 12. tan a = a] tan (x d= a) = tana; ^ 
 The second approximation gives 
 
 13. sin a = a -TT ; sin 2 a = a 2 1 ^r- 
 
 o \ o 
 
 a 2 
 
 14. cos a = 1 -5- ; COS 2 a = 1 a 2 . 
 
 3 / o \ 
 
 15. tana = a + ^-| tan 2 a = a 2 ( 1 + ^ a 2 V 
 
 (Kohlrausch, "Praktische Physik.") 
 
222 THE THEORY OF MEASUREMENTS 
 
 TABLE VIII. NUMERICAL CONSTANTS. 
 
 Logarithm . 
 
 Base of Naperian logarithms: e = 2. 7182818 ........ 0. 4342945 
 
 Modulus of Naperian log.: M = ^ = 2.30259 ........... 0.3622157 
 
 Modulus of common log.: = log e = 0. 4342945 ......... 1. 6377843 
 
 Circumference ,.. 1415 9265 . 0. 4971499 
 
 Diameter 
 
 2?r = 6.28318530 .............. 0.7981799 
 
 - =0.3183099 . 1.5028501 
 
 7T 
 
 Tr 2 = 9.8696044 . . ............. 0.9942998 
 
 V^ = 1.7724539 ............... 0.2485749 
 
 | = 0.7853982 ............... 1.8950899 
 
 5 =0.5235988 . 1.7189986 
 o 
 
 w = Precision constant; k = Unit error; A = Average error; 
 M = Mean error; E = Probable error. 
 
 4p = 0.31831 ................. 1.5028501 
 
 ^ = 0.39894 ................. 1.6009101 
 
 ^ = 0.26908 ................. 1.4298888 
 
 ^ = 1.25331 ................. 0.0980600 
 
 A. 
 
 f = 0.84535 ................. 1.9270387 
 
 A. 
 
 = 0.67449 ................. 1.8289787 
 
TABLES 
 
 223 
 
 TABLE IX. EXPONENTIAL FUNCTIONS. 
 
 X 
 
 logic (e*) 
 
 e* 
 
 e* 
 
 X 
 
 log 10 (O 
 
 e' 
 
 e~' 
 
 0.0 
 
 0.00000 
 
 1.0000 
 
 1.000000 
 
 5.0 
 
 2.17147 
 
 148.41 
 
 0.006738 
 
 0.1 
 
 0.04343 
 
 1.1052 
 
 0.904837 
 
 5.1 
 
 2.21490 
 
 164.02 
 
 0.006097 
 
 0.2 
 
 0.08686 
 
 1.2214 
 
 0.818731 
 
 5.2 
 
 2.25833 
 
 181.27 
 
 0.005517 
 
 0.3 
 
 0.13029 
 
 1.3499 
 
 0.740818 
 
 5.3 
 
 2.30176 
 
 200.34 
 
 0.004992 
 
 0.4 
 
 0.17372 
 
 1.4918 
 
 0.670320 
 
 5.4 
 
 2.34519 
 
 221.41 
 
 0.004517 
 
 0.5 
 
 0.21715 
 
 1.6487 
 
 0.606531 
 
 5.5 
 
 2.38862 
 
 244.69 
 
 0.004087 
 
 0.6 
 
 0.26058 
 
 1.8221 
 
 0.548812 
 
 5.6 
 
 2.43205 
 
 270.43 
 
 0.003698 
 
 0.7 
 
 0.30401 
 
 2.0138 
 
 0.496585 
 
 5.7 
 
 2.47548 
 
 298.87 
 
 0.003346 
 
 0.8 
 
 0.34744 
 
 2.2255 
 
 0.449329 
 
 5.8 
 
 2.51891 
 
 330.30 
 
 0.003028 
 
 0.9 
 
 0.39087 
 
 2.4596 
 
 0.406570 
 
 5.9 
 
 2.56234 
 
 365.04 
 
 0.002739 
 
 1.0 
 
 0.43429 
 
 2.7183 
 
 0.367879 
 
 6.0 
 
 2.60577 
 
 403.43 
 
 0.002479 
 
 1.1 
 
 0.47772 
 
 3.0042 
 
 0.332871 
 
 6.1 
 
 2.64920 
 
 445.86 
 
 0.002243 
 
 1.2 
 
 0.52115 
 
 3.3201 
 
 0.301194 
 
 6.2 
 
 2.69263 
 
 492.75 
 
 0.002029 
 
 1.3 
 
 0.56458 
 
 3.6693 
 
 0.272532 
 
 6.3 
 
 2.73606 
 
 544.57 
 
 0.001836 
 
 1.4 
 
 0.60801 
 
 4.0552 
 
 0.246597 
 
 6.4 
 
 2.77948 
 
 601.85 
 
 0.001662 
 
 .5 
 
 0.65144 
 
 4.4817 
 
 0.223130 
 
 6.5 
 
 2.82291 
 
 665.14 
 
 0.001503 
 
 .6 
 
 0.69487 
 
 4.9530 
 
 0.201897 
 
 6.6 
 
 2.86634 
 
 735.10 
 
 0.001360 
 
 .7 
 
 0.73830 
 
 5.4739 
 
 0.182684 
 
 6.7 
 
 2.90977 
 
 812.41 
 
 0.001231 
 
 .8 
 
 0.78173 
 
 6.0496 
 
 0.165299 
 
 6.8 
 
 2.95320 
 
 897.85 
 
 0.001114 
 
 .9 
 
 0.82516 
 
 6.6859 
 
 0.149569 
 
 6.9 
 
 2.99663 
 
 992.27 
 
 0.001008 
 
 2.0 
 
 0.86859 
 
 7.3891 
 
 0.135335 
 
 7.0 
 
 3.04006 
 
 1096.6 
 
 0.000912 
 
 2.1 
 
 0.91202 
 
 8.1662 
 
 0.122456 
 
 7.1 
 
 3.08349 
 
 1212.0 
 
 0.000825 
 
 2.2 
 
 0.95545 
 
 9.0250 
 
 0.110803 
 
 7.2 
 
 3.12692 
 
 1339.4 
 
 0.000747 
 
 2.3 
 
 0.99888 
 
 9.9742 
 
 0.100259 
 
 7.3 
 
 3.17035 
 
 1480.3 
 
 0.000676 
 
 2.4 
 
 1.04231 
 
 11.023 
 
 0.090718 
 
 7.4 
 
 3.21378 
 
 1636.0 
 
 0.000611 
 
 2.5 
 
 1.08574 
 
 12.182 
 
 0.082085 
 
 7.5 
 
 3.25721 
 
 1808.0 
 
 0.000553 
 
 2.6 
 
 1.12917 
 
 13.464 
 
 0.074274 
 
 7.6 
 
 3.30064 
 
 1998.2 
 
 0.000500 
 
 2.7 
 
 1 . 17260 
 
 14.880 
 
 0.067206 
 
 7.7 
 
 3.34407 
 
 2208.3 
 
 0.000453 
 
 2.8 
 
 1.21602 
 
 16.445 
 
 0.060810 
 
 7.8 
 
 3.38750 
 
 2440.6 
 
 0.000410 
 
 2.9 
 
 1.25945 
 
 18.174 
 
 0.055023 
 
 7.9 
 
 3.43093 
 
 2697.3 
 
 0.000371 
 
 3.0 
 
 1.30288 
 
 20.086 
 
 0.049787 
 
 8.0 
 
 3.47436 
 
 2981.0 
 
 0.000335 
 
 3.1 
 
 1.34631 
 
 22.198 
 
 0.045049 
 
 8.1 
 
 3.51779 
 
 3294.5 
 
 0.000304 
 
 3.2 
 
 1.38974 
 
 24.533 
 
 0.040762 
 
 8.2 
 
 3.56121 
 
 3641.0 
 
 0.000275 
 
 3.3 
 
 1.43317 
 
 27.113 
 
 0.036883 
 
 8.3 
 
 3.60464 
 
 4023.9 
 
 0.000249 
 
 3.4 
 
 1.47660 
 
 29.964 
 
 0.033373 
 
 8.4 
 
 3.64807 
 
 4447.1 
 
 0.000225 
 
 3.5 
 
 1.52003 
 
 33.115 
 
 0.030197 
 
 8.5 
 
 3.69150 
 
 4914.8 
 
 0.000203 
 
 3.6 
 
 1.56346 
 
 36.598 
 
 0.027324 
 
 8.6 
 
 3.73493 
 
 5431.7 
 
 0.000184 
 
 3.7 
 
 1.60689 
 
 40.447 
 
 0.024724 
 
 8.7 
 
 3.77836 
 
 6002.9 
 
 0.000167 
 
 3.8 
 
 1.65032 
 
 44.701 
 
 0.022371 
 
 8.8 
 
 3.82179 
 
 6634.2 
 
 0.000151 
 
 3.9 
 
 1.69375 
 
 49.402 
 
 0.020242 
 
 8.9 
 
 3.86522 
 
 7332.0 
 
 0.000136 
 
 4.0 
 
 1.73718 
 
 54.598 
 
 0.018316 
 
 9.0 
 
 3.90865 
 
 8103.1 
 
 0.000123 
 
 4.1 
 
 .78061 
 
 60.340 
 
 0.016573 
 
 9.1 
 
 3.95208 
 
 8955.3 
 
 0.000112 
 
 4.2 
 
 .82404 
 
 66.686 
 
 0.014996 
 
 9.2 
 
 3.99551 
 
 9897.1 
 
 0.000101 
 
 4.3 
 
 .86747 
 
 73.700 
 
 0.013569 
 
 9.3 
 
 4.03894 
 
 10938. 
 
 0.000091 
 
 4.4 
 
 .91090 
 
 81.451 
 
 0.012277 
 
 9.4 
 
 4.08237 
 
 12088. 
 
 0.000083 
 
 4.5 
 
 .95433 
 
 90.017 
 
 0.011109 
 
 9.5 
 
 4.12580 
 
 13360. 
 
 0.000075 
 
 4.6 
 
 .99775 
 
 99.484 
 
 0.010052 
 
 9.6 
 
 4.16923 
 
 14765. 
 
 0.000068 
 
 4.7 
 
 2.04118 
 
 109.95 
 
 0.009095 
 
 9.7 
 
 4.21266 
 
 16318. 
 
 0.000061 
 
 4.8 
 
 2.08461 
 
 121.51 
 
 0.008230 
 
 9.8 
 
 4.25609 
 
 18034. 
 
 0.000055 
 
 4.9 
 
 2.12804 
 
 134.29 
 
 0.007447 
 
 9.9 
 
 4.29952 
 
 19930. 
 
 0.000050 
 
 5.0 
 
 2.17147 
 
 148.41 
 
 0.006738 
 
 10.0 
 
 4.34294 
 
 22026. 
 
 0.000045 
 
 Taken from Glaisher's "Tables of the Exponential Function," Trans. Cambridge Phil. Soc., 
 vol. xiii, 1883. This volume also contains a " Table of the Descending Exponential to Twelve 
 or Fourteen Places of Decimals," by F. W. Newman. 
 
224 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE X. EXPONENTIAL FUNCTIONS. 
 Value of e x<t and er x<i and their logarithms. 
 
 X 
 
 <? 
 
 log e 2 
 
 e~* 2 
 
 log e'* z 
 
 0.1 
 
 1.0101 
 
 0.00434 
 
 0.99005 
 
 1.99566 
 
 0.2 
 
 1.0408 
 
 0.01737 
 
 0.96079 
 
 1.98263 
 
 0.3 
 
 1.0942 
 
 0.03909 
 
 0.91393 
 
 1.96091 
 
 0.4 
 
 .1735 
 
 0.06949 
 
 0.85214 
 
 .93051 
 
 0.5 
 
 .2840 
 
 0.10857 
 
 0.77880 
 
 .89143 
 
 0.6 
 
 .4333 
 
 0.15635 
 
 0.69768 
 
 .84365 
 
 0.7 
 
 .6323 
 
 0.21280 
 
 0.61263 
 
 .78720 
 
 0.8 
 
 .8965 
 
 0.27795 
 
 0.52729 
 
 .72205 
 
 0.9 
 
 2.2479 
 
 0.35178 
 
 0.44486 
 
 .64822 
 
 1.0 
 
 2.7183 
 
 0.43429 
 
 0.36788 
 
 .56571 
 
 1.1 
 
 3.3535 
 
 0.52550 
 
 0.29820 
 
 .47450 
 
 1.2 
 
 4.2207 
 
 0.62538 
 
 0.23693 
 
 .37462 
 
 1.3 
 
 5.4195 
 
 0.73396 
 
 0.18452 
 
 .26604 
 
 1.4 
 
 7.0993 
 
 0.85122 
 
 0.14086 
 
 .14878 
 
 1.5 
 
 9.4877 
 
 0.97716 
 
 0.10540 
 
 .02284 
 
 1.6 
 
 1.2936X10 
 
 1.11179 
 
 0. 77305 XlO- 1 
 
 2.88821 
 
 1.7 
 
 1.7993X10 
 
 1.25511 
 
 0. 55576 XlO- 1 
 
 2.74489 
 
 1.8 
 
 2.5534x10 
 
 1.40711 
 
 0. 39164 XlO- 1 
 
 2.59289 
 
 1.9 
 
 3.6966X10 
 
 1.56780 
 
 0.27052 XlO- 1 
 
 2.43220 
 
 2.0 
 
 5.4598X10 
 
 1.73718 
 
 0.18316 XlO- 1 
 
 2.26282 
 
 2.1 
 
 8.2269x10 
 
 1.91524 
 
 0.12155 XlO- 1 
 
 2.08476 
 
 22 
 
 1.2647X10 2 
 
 2.10199 
 
 0.79071 XlO- 2 
 
 3.89801 
 
 2.3 
 
 1.9834X10 2 
 
 2.29742 
 
 0.50417 XlO- 2 
 
 3.70258 
 
 2.4 
 
 3.1735X10 2 
 
 2.50154 
 
 0.31511 XlO- 2 
 
 3.49846 
 
 2.5 
 
 5.1801X10 2 
 
 2.71434 
 
 0.19305 XlO- 2 
 
 3.28566 
 
 2.6 
 
 8.6264X10 2 
 
 2.93583 
 
 0.1 1592 XlO- 2 
 
 3.06417 
 
 2.7 
 
 1.4656X10 3 
 
 3.16601 
 
 0.68232X10- 3 
 
 4.83399 
 
 2.8 
 
 2.5402X10 3 
 
 3.40487 
 
 0.39367X10- 3 
 
 4.59513 
 
 2.9 
 
 4.4918X10 3 
 
 3.65242 
 
 0.22263X10-3 
 
 4.34758 
 
 3.0 
 
 8.1031X10 3 
 
 3.90865 
 
 0.12341X10- 3 
 
 4.09135 
 
 3.1 
 
 1.4913X10 4 
 
 4.17357 
 
 0.67055x10-* 
 
 5.82643 
 
 3.2 
 
 2.8001X10 4 
 
 4.44718 
 
 0.35713 XlO- 4 
 
 5.55282 
 
 3.3 
 
 5.3637X10 4 
 
 4.72947 
 
 0.18644 XlO- 4 
 
 5.27053 
 
 3.4 
 
 1.0482X10 5 
 
 5.02044 
 
 0.95403 XlO- 5 
 
 6.97956 
 
 3.5 
 
 2.0898X10 5 
 
 5.32011 
 
 0.47851 XlO- 5 
 
 6.67989 
 
 3.6 
 
 4.2507X10 5 
 
 5.62846 
 
 0.23526 XlO- 5 
 
 6.37154 
 
 3.7 
 
 8.8204X10 5 
 
 5.94549 
 
 0.1 1337 XlO- 5 
 
 6.05451 
 
 3.8 
 
 1.8673X10 6 
 
 6.27121 
 
 0.53554 XlO- 6 
 
 7.72879 
 
 3.9 
 
 4.0329X10 6 
 
 6.60562 
 
 0.24796 XlO- 6 
 
 ?. 39438 
 
 4.0 
 
 8.8861X10 6 
 
 6.94871 
 
 0.11254X10- 6 
 
 7.05129 
 
 41 
 
 1.9975X10 7 
 
 7.30049 
 
 0.50062 XlO- 7 
 
 .69951 
 
 4.2 
 
 4.5809X10 7 
 
 7.66095 
 
 0.21830 XlO- 7 
 
 S. 33905 
 
 4.3 
 
 1.0718X10 8 
 
 8.03011 
 
 0.93302 XlO- 8 
 
 9.96989 
 
 4.4 
 
 2.5582X10 8 
 
 8.40794 
 
 0.39089 XlO- 8 
 
 9.59206 
 
 4.5 
 
 6.2296X10 8 
 
 8.79446 
 
 0.16052X10- 8 
 
 9.20554 
 
 4.6 
 
 1.5476X10 9 
 
 9.18967 
 
 0.64614X10- 9 
 
 10.81033 
 
 4.7 
 
 3.9226X10 9 
 
 9.59357 
 
 0.25494X10- 9 
 
 10.40643 
 
 4.8 
 
 1.0143X10 10 
 
 10.00615 
 
 0.98594 XlO- 10 
 
 11.99385 
 
 4.9 
 
 2.6755X10 10 
 
 10.42741 
 
 0.37376 XlO- 10 
 
 11.57259 
 
 5.0 
 
 7.2005X10 10 
 
 10.85736 
 
 0.13888 XlO- 10 
 
 11.14264 
 
TABLES 
 
 225 
 
 TABLE XI. VALUES OF THE PROBABILITY INTEGRAL. 
 
 t 
 
 P* 
 
 Diff. 
 
 t 
 
 ^A 
 
 Diff. 
 
 1 
 
 ^A 
 
 Diff 
 
 t 
 
 ^A 
 
 Diff. 
 
 0.00 
 
 0.00000 
 
 1 IOC 
 
 0.50 
 
 0.52050 
 
 074 
 
 1.00 
 
 0.84270 
 
 A 1 
 
 1.50 
 
 0.96611 
 
 
 0.01 
 
 0.01128 
 
 1 I ! 
 
 1 100 
 
 0.51 
 
 0.52924 
 
 o/ 1 
 
 O/3f> 
 
 1.01 
 
 0.84681 
 
 411 
 A AO 
 
 1.51 
 
 0.96728 
 
 
 0.02 
 0.03 
 0.04 
 0.05 
 0.06 
 0.07 
 0.08 
 0.09 
 
 0.02256 
 0.03384 
 0.04511 
 0.05637 
 0.06762 
 0.07886 
 0.09008 
 0.10128 
 
 liZo 
 
 1128 
 1127 
 1126 
 1125 
 1124 
 1122 
 1120 
 
 1 1 1C 
 
 0.52 
 0.53 
 0.54 
 0.55 
 0.56 
 0.57 
 0.58 
 0.59 
 
 0.53790 
 0.54646 
 0.55494 
 0.56332 
 0.57162 
 0.57982 
 0.58792 
 0.59594 
 
 ODD 
 
 856 
 848 
 838 
 830 
 820 
 810 
 802 
 
 7QO 
 
 1.02 
 1.03 
 1.04 
 1.05 
 1.06 
 1.07 
 1.08 
 1.09 
 
 0.85084 
 0.85478 
 0.85865 
 0.86244 
 0.86614 
 0.86977 
 0.87333 
 0.87680 
 
 403 
 394 
 387 
 379 
 370 
 363 
 356 
 347 
 
 O/M 
 
 1.52 
 1.53 
 1.54 
 1.55 
 1.56 
 1.57 
 1.58 
 1.59 
 
 0.96841 
 0.96952 
 0.97059 
 0.97162 
 0.97263 
 0.97360 
 0.97455 
 0.97546 
 
 113 
 
 111 
 
 107 
 103 
 101 
 97 
 95 
 91 
 
 OA 
 
 0.10 
 0.11 
 0.12 
 
 0.11246 
 0.12362 
 0.13476 
 
 1 1 J.O 
 
 1116 
 1114 
 1111 
 
 0.60 
 0.61 
 0.62 
 
 0.60386 
 0.61168 
 0.61941 
 
 /y^ 
 
 782 
 
 773 
 
 7j 
 
 1.10 
 1.11 
 1.12 
 
 0.88021 
 0.88353 
 0.88679 
 
 O41 
 
 332 
 326 
 
 010 
 
 1.60 
 1.61 
 1.62 
 
 0.97635 
 0.97721 
 0.97804 
 
 89 
 86 
 83 
 
 Qf\ 
 
 0.13 
 0.14 
 0.15 
 0.16 
 0.17 
 
 0.14587 
 0.15695 
 0.16800 
 0.17901 
 0.18999 
 
 1 1 i. i 
 1108 
 1105 
 1101 
 1098 
 
 1 AQ~ 
 
 0.63 
 0.64 
 0.65 
 0.66 
 0.67 
 
 0.62705 
 0.63459 
 0.64203 
 0.64938 
 0.65663 
 
 < D^ 
 
 754 
 744 
 735 
 725 
 
 71 t\ 
 
 1.13 
 1.14 
 1.15 
 1.16 
 1.17 
 
 0.88997 
 0.89308 
 0.89612 
 0.89910 
 0.90200 
 
 OIo 
 
 311 
 304 
 
 298 
 290 
 
 oo/i 
 
 1.63 
 .64 
 .65 
 .66 
 .67 
 
 0.97884 
 0.97962 
 0.98038 
 0.98110 
 0.98181 
 
 oU 
 
 78 
 76 
 72 
 71 
 
 f*Q 
 
 0.18 
 0.19 
 0.20 
 0.21 
 0.22 
 0.23 
 0.24 
 0.25 
 0.26 
 0.27 
 0.28 
 0.29 
 0.30 
 0.31 
 0.32 
 
 0.20094 
 0.21184 
 0.22270 
 0.23352 
 0.24430 
 0.25502 
 0.26570 
 0.27633 
 0.28690 
 0.29742 
 0.30788 
 0.31828 
 0.32863 
 0.33891 
 0.34913 
 
 iuyo 
 1090 
 1086 
 1082 
 1078 
 1072 
 1068 
 1083 
 1057 
 1052 
 1046 
 1040 
 1035 
 1028 
 1022 
 
 1 A1 K. 
 
 0.68 
 0.69 
 0.70 
 0.71 
 0.72 
 0.73 
 0.74 
 0.75 
 0.76 
 0.77 
 0.78 
 0.79 
 0.80 
 0.81 
 0.82 
 
 0.66378 
 0.67084 
 0.67780 
 . 68467 
 0.69143 
 0.69810 
 0.70468 
 0.71116 
 0.71754 
 0.72382 
 0.73001 
 0.73610 
 0.74210 
 0.74800 
 0.75381 
 
 < 10 
 706 
 696 
 687 
 676 
 667 
 658 
 648 
 638 
 628 
 619 
 609 
 600 
 590 
 581 
 
 CT1 
 
 1.18 
 1.19 
 1.20 
 1.21 
 1.22 
 1.23 
 1.24 
 1.25 
 1.26 
 1.27 
 1.28 
 1.29 
 1.30 
 1.31 
 1.32 
 
 0.90484 
 0.90761 
 0.91031 
 0.91296 
 0.91553 
 0.91805 
 0.92051 
 0.92290 
 0.92524 
 0.92751 
 0.92973 
 0.93190 
 0.93401 
 0.93606 
 0.93807 
 
 Zo4 
 
 277 
 270 
 265 
 257 
 252 
 246 
 239 
 234 
 227 
 222 
 217 
 211 
 205 
 201 
 
 .68 
 .69 
 .70 
 .71 
 .72 
 .73 
 1.74 
 1.75 
 1.76 
 1.77 
 1.78 
 1.79 
 1.80 
 1.81 
 1.82 
 
 0.98249 
 0.98315 
 0.98379 
 0.98441 
 0.98500 
 0.98558 
 0.98613 
 0.98667 
 0.98719 
 0.98769 
 0.98817 
 0.98864 
 0.98909 
 0.98952 
 0.98994 
 
 Do 
 
 66 
 64 
 62 
 59 
 58 
 55 
 54 
 52 
 50 
 48 
 47 
 45 
 43 
 42 
 
 0.33 
 0.34 
 0.35 
 0.36 
 0.37 
 0.38 
 0.39 
 
 0.35928 
 0.36936 
 0.37938 
 0.38933 
 0.39921 
 0.40901 
 0.41874 
 
 lUio 
 1008 
 1002 
 995 
 988 
 980 
 973 
 
 Qf?r 
 
 0.83 
 0.84 
 0.85 
 0.86 
 0.87 
 0.88 
 0.89 
 
 0.75952 
 0.76514 
 0.77067 
 0.77610 
 0.78144 
 0.78669 
 0.79184 
 
 571 
 562 
 553 
 543 
 534 
 525 
 515 
 
 CA*7 
 
 .33 
 .34 
 .35 
 .36 
 .37 
 .38 
 .39 
 
 0.94002 
 0.94191 
 0.94376 
 0.94556 
 0.94731 
 0.94902 
 0.95067 
 
 195 
 189 
 185 
 180 
 175 
 171 
 165 
 
 1 JO 
 
 1.83 
 1.84 
 1.85 
 1.86 
 1.87 
 1.88 
 1.89 
 
 0.99035 
 0.99074 
 0.99111 
 0.99147 
 0.99182 
 0.99216 
 0.99248 
 
 41 
 39 
 37 
 36 
 35 
 34 
 
 32 
 01 
 
 0.40 
 0.41 
 
 0.42839 
 0.43797 
 
 yoo 
 
 958 
 
 nr:n 
 
 0.90 
 0.91 
 
 0.79691 
 0.80188 
 
 oU/ 
 497 
 
 4P.Q 
 
 .40 
 .41 
 
 0.95229 
 0.95385 
 
 ItW 
 
 156 
 i ^. 
 
 1.90 
 1.91 
 
 0.99279 
 0.99309 
 
 ol 
 
 30 
 
 on 
 
 0.42 
 
 0.44747 
 
 you 
 n/io 
 
 0.92 
 
 0.80677 
 
 "oy 
 
 A>-r{\ 
 
 1.42 
 
 0.95538 
 
 100 
 
 1 AO 
 
 1.92 
 
 3.99338 
 
 zy 
 
 oo 
 
 0.43 
 0.44 
 
 0.45689 
 0.46623 
 
 y4z 
 934 
 
 QOC 
 
 0.93 
 0.94 
 
 0.81156 
 0.81627 
 
 479 
 471 
 4fi9 
 
 1.43 
 1.44 
 
 0.95686 
 0.95830 
 
 148 
 144 
 140 
 
 1.93 
 1.94 
 
 3.99366 
 3.99392 
 
 28 
 26 
 
 9fi 
 
 0.45 
 
 0.47548 
 
 t/^O 
 QIC 
 
 0.95 
 
 0.82089 
 
 '\j 
 4CQ 
 
 1.45 
 
 0.95970 
 
 J.TAJ 
 
 IOC 
 
 1.95 
 
 3.99418 
 
 ^O 
 
 OK 
 
 0.46 
 0.47 
 0.48 
 0.49 
 0.50 
 
 0.48466 
 0.49375 
 0.50275 
 0.51167 
 0.52050 
 
 7 j.o 
 
 909 
 900 
 892 
 883 
 
 0.96 
 0.97 
 0.98 
 0.99 
 1.00 
 
 0.82542 
 0.82987 
 0.83423 
 0.83851 
 0.84270 
 
 ^too 
 445 
 436 
 
 428 
 419 
 
 1.46 
 1.47 
 1.48 
 1.49 
 1.50 
 
 0.96105 
 0.96237 
 0.96365 
 0.96490 
 0.96611 
 
 J.OO 
 
 132 
 
 128 
 125 
 121 
 
 1.96 
 1.97 
 1.98 
 1.99 
 2.00 
 
 3.99443 
 3.99466 
 3.99489 
 3.99511 
 3.99532 
 
 ^O 
 
 23 
 23 
 22 
 21 
 
 
 
 
 
 
 
 
 
 
 oo 
 
 .00000 
 
 
 (Chauvenet, " Spherical and Practical Astronomy.") 
 
226 THE THEORY OF MEASUREMENTS 
 
 TABLE XII. VALUES OF THE PROBABILITY INTEGRAL. 
 
 3 
 
 
 
 1 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 0.0 
 
 .00000 
 
 .00538 
 
 .01076 
 
 .01614 
 
 .02152 
 
 .02690 
 
 .03228 
 
 .03766 
 
 . 04303 
 
 .04840 
 
 0.1 
 
 .05378 
 
 .05914 
 
 .06451 
 
 .06987 
 
 .07523 
 
 .08059 
 
 .08594 
 
 .09129 
 
 .09663 
 
 . 10197 
 
 0.2 
 
 .10731 
 
 .11264 
 
 .11796 
 
 . 12328 
 
 . 12860 
 
 . 13391 
 
 . 13921 
 
 . 14451 
 
 . 14980 
 
 . 15508 
 
 0.3 
 
 .16035 
 
 . 16562 
 
 . 17088 
 
 . 17614 
 
 . 18138 
 
 . 18662 
 
 .19185 
 
 . 19707 
 
 .20229 
 
 .20749 
 
 0.4 
 
 .21268 
 
 .21787 
 
 .22304 
 
 .22821 
 
 .23336 
 
 .23851 
 
 .24364 
 
 .24876 
 
 .25388 
 
 .25898 
 
 0.5 
 
 .26407 
 
 .26915 
 
 .27421 
 
 .27927 
 
 .28431 
 
 .28934 
 
 .29436 
 
 .29936 
 
 .30435 
 
 .30933 
 
 0.6 
 
 .31430 
 
 .31925 
 
 .32419 
 
 .32911 
 
 .33402 
 
 .33892 
 
 .34380 
 
 .34866 
 
 .35352 
 
 .35835 
 
 0.7 
 
 .36317 
 
 .36798 
 
 .37277 
 
 .37755 
 
 .38231 
 
 .38705 
 
 .39178 
 
 .39649 
 
 .40118 
 
 .40586 
 
 0.8 
 
 .41052 
 
 .41517 
 
 .41979 
 
 .42440 
 
 . 42899 
 
 . 43357 
 
 . 43813 
 
 . 44267 
 
 .44719 
 
 .45169 
 
 0.9 
 
 .45618 
 
 .46064 
 
 .46509 
 
 .46952 
 
 .47393 
 
 . 47832 
 
 . 48270 
 
 . 48605 
 
 .49139 
 
 .49570 
 
 .0 
 
 .50000 
 
 .50428 
 
 .50853 
 
 .51277 
 
 .51699 
 
 .52119 
 
 .52537 
 
 .52952 
 
 .53366 
 
 .53778 
 
 .1 
 
 .54188 
 
 .54595 
 
 .55001 
 
 .55404 
 
 .55806 
 
 .56205 
 
 .56602 
 
 . 56998 
 
 .57391 
 
 .57782 
 
 .2 
 
 .58171 
 
 .58558 
 
 .58942 
 
 .59325 
 
 .59705 
 
 .60083 
 
 .60460 
 
 . 60833 
 
 .61205 
 
 .61575 
 
 .3 
 
 .61942 
 
 .62308 
 
 .62671 
 
 .63032 
 
 .63391 
 
 .63747 
 
 .64102 
 
 .64454 
 
 .64804 
 
 .65152 
 
 .4 
 
 .65498 
 
 .65841 
 
 .66182 
 
 .66521 
 
 .66858 
 
 .67193 
 
 .67526 
 
 .67856 
 
 .68184 
 
 .68510 
 
 .5 
 
 .68833 
 
 .69155 
 
 .69474 
 
 .69791 
 
 .70106 
 
 .70419 
 
 .70729 
 
 .71038 
 
 .71344 
 
 .71648 
 
 .6 
 
 .71949 
 
 .72249 
 
 .72546 
 
 .72841 
 
 .73134 
 
 .73425 
 
 .73714 
 
 .74000 
 
 .74285 
 
 .74567 
 
 .7 
 
 .74847 
 
 .75124 
 
 .75400 
 
 .75674 
 
 .75945 
 
 .76214 
 
 .76481 
 
 .76746 
 
 .77009 
 
 .77270 
 
 .8 
 
 .77528 
 
 .77785 
 
 .78039 
 
 .78291 
 
 .78542 
 
 .78790 
 
 .79036 
 
 .79280 
 
 .79522 
 
 .79761 
 
 .9 
 
 .79999 
 
 .80235 
 
 .80469 
 
 .80700 
 
 .80930 
 
 .81158 
 
 .81383 
 
 .81607 
 
 .81828 
 
 .82048 
 
 2.0 
 
 .82266 
 
 .82481 
 
 .82695 
 
 .82907 
 
 .83117 
 
 .83324 
 
 .83530 
 
 .83734 
 
 .83936 
 
 .84137 
 
 2.1 
 
 .84335 
 
 .84531 
 
 .84726 
 
 .84919 
 
 .85109 
 
 .85298 
 
 .85486 
 
 .85671 
 
 .85854 
 
 .86036 
 
 2.2 
 
 .86216 
 
 .86394 
 
 .86570 
 
 .86745 
 
 .86917 
 
 .87088 
 
 .87258 
 
 .87425 
 
 .87591 
 
 .87755 
 
 2.3 
 
 .87918 
 
 .88078 
 
 .88237 
 
 .88395 
 
 .88550 
 
 .88705 
 
 .88857 
 
 .89008 
 
 : 89157 
 
 .89304 
 
 2.4 
 
 .89450 
 
 .89595 
 
 .89738 
 
 .89879 
 
 .90019 
 
 .90157 
 
 .90293 
 
 .90428 
 
 .90562 
 
 .90694 
 
 25 
 
 .90825 
 
 .90954 
 
 .91082 
 
 .91208 
 
 .91332 
 
 .91456 
 
 .91578 
 
 .91698 
 
 .91817 
 
 .91935 
 
 2.6 
 
 .92051 
 
 .92166 
 
 .92280 
 
 .92392 
 
 .92503 
 
 .92613 
 
 .92721 
 
 .92828 
 
 .92934 
 
 .93038 
 
 2.7 
 
 .93141 
 
 .93243 
 
 .93344 
 
 .93443 
 
 .93541 
 
 .93638 
 
 .93734 
 
 .93828 
 
 .93922 
 
 .94014 
 
 2.8 
 
 .94105 
 
 .94195 
 
 .94284 
 
 .94371 
 
 .94458 
 
 .94543 
 
 .94627 
 
 .94711 
 
 .94793 
 
 .94874 
 
 2.9 
 
 .94954 
 
 .95033 
 
 .95111 
 
 .95187 
 
 .95263 
 
 .95338 
 
 .95412 
 
 .95485 
 
 .95557 
 
 .95628 
 
 3 
 
 .95698 
 
 .96346 
 
 96910 
 
 .97397 
 
 .97817 
 
 .98176 
 
 .98482 
 
 .98743 
 
 .98962 
 
 .99147 
 
 4 
 
 .99302 
 
 .99431 
 
 .99539 
 
 .99627 
 
 .99700 
 
 .99760 
 
 .99808 
 
 .99848 
 
 .99879 
 
 .99905 
 
 5 
 
 .99926 
 
 .99943 
 
 .99956 
 
 .99966 
 
 .99974 
 
 .99980 
 
 .99985 
 
 . 99988 
 
 .99991 
 
 .99993 
 
 TABLE XIII. CHAUVENET'S CRITERION. 
 
 N 
 
 T 
 
 N 
 
 r 
 
 AT 
 
 r 
 
 3 
 
 2.05 
 
 13 
 
 3.07 
 
 23 
 
 3.40 
 
 4 
 
 2.27 
 
 14 
 
 3.11 
 
 24 
 
 3.43 
 
 5 
 
 2.44 
 
 15 
 
 3.15 
 
 25 
 
 3.45 
 
 6 
 
 2.57 
 
 16 
 
 3.19 
 
 30 
 
 3.55 
 
 7 
 
 2.67 
 
 17 
 
 3.22 
 
 40 
 
 3.70 
 
 8 
 
 2.76 
 
 18 
 
 3.26 
 
 50 
 
 3.82 
 
 9 
 
 2.84 
 
 19 
 
 3.29 
 
 75 
 
 4.02 
 
 10 
 
 2.91 
 
 20 
 
 3.32 
 
 100 
 
 4.16 
 
 11 
 
 2.97 
 
 21 
 
 3.35 
 
 200 
 
 4.48 
 
 12 
 
 3.02 
 
 22 
 
 3.38 
 
 500 
 
 4.90 
 
TABLES 
 
 227 
 
 TABLE XTV. FOR COMPUTING PROBABLE ERRORS BY FORMULA 
 
 (31) AND (32). 
 
 AT 
 
 0.6745 
 
 0.6745 
 
 AT 
 
 0.6745 
 
 0.6745 
 
 iV 
 
 VJv^T 
 
 VN(N-l) 
 
 iM 
 
 vim 
 
 v# (AT- i) 
 
 
 
 
 40 
 
 0.1080 
 
 0.0171 
 
 
 
 
 41 
 
 0.1066 
 
 0.0167 
 
 2 
 
 0.6745 
 
 0.4769 
 
 42 
 
 0.1053 
 
 0.0163 
 
 3 
 
 0.4769 
 
 0.2754 
 
 43 
 
 0.1041 
 
 0.0159 
 
 4 
 
 0.3894 
 
 0.1947 
 
 44 
 
 0.1029 
 
 0.0155 
 
 5 
 
 0.3372 
 
 0.1508 
 
 45 
 
 0.1017 
 
 0.0152 
 
 6 
 
 0.3016 
 
 0.1231 
 
 46 
 
 0.1005 
 
 0.0148 
 
 7 
 
 0.2754 
 
 0.1041 
 
 47 
 
 0.0994 
 
 0.0145 
 
 8 
 
 0.2549 
 
 0.0901 
 
 48 
 
 0.0984 
 
 0.0142 
 
 9 
 
 0.2385 
 
 0.0795 
 
 49 
 
 0.0974 
 
 0.0139 
 
 10 
 
 0.2248 
 
 0.0711 
 
 50 
 
 0.0964 
 
 0.0136 
 
 11 
 
 0.2133 
 
 0.0643 
 
 51 
 
 0.0954 
 
 0.0134 
 
 12 
 
 0.2029 
 
 0.0587 
 
 52 
 
 0.0944 
 
 0.0131 
 
 13 
 
 0.1947 
 
 0.0540 
 
 53 
 
 0.0935 
 
 0.0128 
 
 14 
 
 0.1871 
 
 0.0500 
 
 54 
 
 0.0926 
 
 0.0126 
 
 15 
 
 0.1803 
 
 0.0465 
 
 55 
 
 0.0918 
 
 0.0124 
 
 16 
 
 0.1742 
 
 0.0435 
 
 56 
 
 0.0909 
 
 0.0122 
 
 17 
 
 0.1686 
 
 0.0409 
 
 57 
 
 0.0901 
 
 0.0119 
 
 18 
 
 0.1636 
 
 0.0386 
 
 58 
 
 0.0893 
 
 0.0117 
 
 19 
 
 0.1590 
 
 0.0365 
 
 59 
 
 0.0886 
 
 0.0115 
 
 20 
 
 0.1547 
 
 0.0346 
 
 60 
 
 0.0878 
 
 0.0113 
 
 21 
 
 0.1508 
 
 0.0329 
 
 61 
 
 0.0871 
 
 0.0111 
 
 22 
 
 0.1472 
 
 0.0314 
 
 62 
 
 0.0864 
 
 0.0110 
 
 23 
 
 0.1438 
 
 0.0300 
 
 63 
 
 0.0857 
 
 0.0108 
 
 24 
 
 0.1406 
 
 0.0287 
 
 64 
 
 0.0850 
 
 0.0106 
 
 25 
 
 0.1377 
 
 0.0275 
 
 65 
 
 0.0843 
 
 0.0105 
 
 26 
 
 0.1349 
 
 0.0265 
 
 66 
 
 0.0837 
 
 0.0103 
 
 27 
 
 0.1323 
 
 0.0255 
 
 67 
 
 0.0830 
 
 0.0101 
 
 28 
 
 0.1298 
 
 0.0245 
 
 68 
 
 0.0824 
 
 0.0100 
 
 29 
 
 0.1275 
 
 0.0237 
 
 69 
 
 0.0818 
 
 0.0098 
 
 30 
 
 0.1252 
 
 0.0229 
 
 70 
 
 0.0812 
 
 0,0097 
 
 31 
 
 0.1231 
 
 0.0221 
 
 71 
 
 0.0806 
 
 0.0096 
 
 32 
 
 0.1211 
 
 0.0214 
 
 72 
 
 0.0800 
 
 0.0094 
 
 33 
 
 0.1192 
 
 0.0208 
 
 73 
 
 0.0795 
 
 0.0093 
 
 34 
 
 0.1174 
 
 0.0201 
 
 74 
 
 0.0789 
 
 0.0092 
 
 35 
 
 0.1157 
 
 0.0196 
 
 75 
 
 0.0784 
 
 0.0091 
 
 36 
 
 0.1140 
 
 0.0190 
 
 80 
 
 0.0759 
 
 0.0085 
 
 37 
 
 0.1124 
 
 0.0185 
 
 85 
 
 0.0736 
 
 0.0080 
 
 38 
 
 0.1109 
 
 0.0180 
 
 90 
 
 0.0713 
 
 0.0075 
 
 39 
 
 0.1094 
 
 0.0175 
 
 100 
 
 0.0678 
 
 0.0068 
 
 (Merriman, " Least Squares. ") 
 
228 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE XV. FOR COMPUTING PROBABLE ERRORS BY FORMULAE (34). 
 
 N 
 
 0.8453 
 
 0.8453 
 
 N 
 
 0.8453 
 
 0.8453 
 
 ^N(N - 1) 
 
 N^N-1 
 
 VN(N - 1) 
 
 N^W=1 
 
 
 
 
 40 
 
 0.0214 
 
 0.0034 
 
 
 
 
 41 
 
 0.0209 
 
 0.0033 
 
 2 
 
 0.5978 
 
 0.4227 
 
 42 
 
 0.0204 
 
 0.0031 
 
 3 
 
 0.3451 
 
 0.1993 
 
 43 
 
 0.0199 
 
 0.0030 
 
 4 
 
 0.2440 
 
 0.1220 
 
 44 
 
 0.0194 
 
 0.0029 
 
 5 
 
 0.1890 
 
 0.0845 
 
 45 
 
 0.0190 
 
 0.0028 
 
 6 
 
 0.1543 
 
 0.0630 
 
 46 
 
 0.0186 
 
 0.0027 
 
 7 
 
 0.1304 
 
 0.0493 
 
 47 
 
 0.0182 
 
 0.0027 
 
 8 
 
 0.1130 
 
 0.0399 
 
 48 
 
 0.0178 
 
 0.0026 
 
 9 
 
 0.0996 
 
 0.0332 
 
 49 
 
 0.0174 
 
 0.0025 
 
 10 
 
 0.0891 
 
 0.0282 
 
 50 
 
 0.0171 
 
 0.0024 
 
 11 
 
 0.0806 
 
 0.0243 
 
 51 
 
 0.0167 
 
 0.0023 
 
 12 
 
 0.0736 
 
 0.0212 
 
 52 
 
 0.0164 
 
 0.0023 
 
 13 
 
 0.0677 
 
 0.0188 
 
 53 
 
 0.0161 
 
 0.0022 
 
 14 
 
 0.0627 
 
 0.0167 
 
 54 
 
 0.0158 
 
 0.0022 
 
 15 
 
 0.0583 
 
 0.0151 
 
 55 
 
 0.0155 
 
 0.0021 
 
 16 
 
 0.0546 
 
 0.0136 
 
 56 
 
 0.0152 
 
 0.0020 
 
 17 
 
 0.0513 
 
 0.0124 
 
 57 
 
 0.0150 
 
 0.0020 
 
 18 
 
 0.0483 
 
 0.0114 
 
 58 
 
 0.0147 
 
 0.0019 
 
 19 
 
 0.0457 
 
 0.0105 
 
 59 
 
 0.0145 
 
 0.0019 
 
 20 
 
 0.0434 
 
 0.0097 
 
 60 
 
 0.0142 
 
 0.0018 
 
 21 
 
 0.0412 
 
 0.0090 
 
 61 
 
 0.0140 
 
 0.0018 
 
 22 
 
 0.0393 
 
 0.0084 
 
 62 
 
 0.0137 
 
 0.0017 
 
 23 
 
 0.0376 
 
 0.0078 
 
 63 
 
 0.0135 
 
 0.0017 
 
 24 
 
 0.0360 
 
 0.0073 
 
 64 
 
 0.0133 
 
 0.0017 
 
 25 
 
 0.0345 
 
 0.0069 
 
 65 
 
 0.0131 
 
 0.0016 
 
 26 
 
 0.0332 
 
 0.0065 
 
 66 
 
 0.0129 
 
 0.0016 
 
 27 
 
 0.0319 
 
 0.0061 
 
 67 
 
 0.0127 
 
 0.0016 
 
 28 
 
 0.0307 
 
 0.0058 
 
 68 
 
 0.0125 
 
 0.0015 
 
 29 
 
 0.0297 
 
 0.0055 
 
 69 
 
 0.0123 
 
 0.0015 
 
 30 
 
 0.0287 
 
 0.0052 
 
 70 
 
 0.0122 
 
 0.0015 
 
 31 
 
 0.0277 
 
 0.0050 
 
 71 
 
 0.0120 
 
 0.0014 
 
 32 
 
 0.0268 
 
 0.0047 
 
 72 
 
 0.0118 
 
 0.0014 
 
 33 
 
 0.0260 
 
 0.0045 
 
 73 
 
 0.0117 
 
 0.0014 
 
 34 
 
 0.0252 
 
 0.0043 
 
 74 
 
 0.0115 
 
 0.0013 
 
 35 
 
 0.0245 
 
 0.0041 
 
 75 
 
 0.0113 
 
 0.0013 
 
 36 
 
 0.0238 
 
 0.0040 
 
 80 
 
 0.0106 
 
 0.0012 
 
 37 
 
 0.0232 
 
 0.0038 
 
 85 
 
 0.0100 
 
 0.0011 
 
 38 
 
 0.0225 
 
 0.0037 
 
 90 
 
 0.0095 
 
 0.0010 
 
 39 
 
 0.0220 
 
 0.0035 
 
 100 
 
 0.0085 
 
 0.0008 
 
 (Merriman, "Least Squares.") 
 
TABLES 
 
 229 
 
 TABLE XVI. SQUARES OP NUMBERS. 
 
 n 
 
 
 
 i 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 Diff. 
 
 1.0 
 
 1.000 
 
 1.020 
 
 1.040 
 
 1.061 
 
 1.082 
 
 1.103 
 
 1.124 
 
 1.145 
 
 1.166 
 
 1.188 
 
 22 
 
 1.1 
 
 1.210 
 
 1.232 
 
 1.254 
 
 1.277 
 
 1.300 
 
 1.323 
 
 1.346 
 
 1.369 
 
 1.392 
 
 1.416 
 
 24 
 
 1.2 
 
 1.440 
 
 1.464 
 
 1.488 
 
 1.513 
 
 1.538 
 
 1.563 
 
 1.588 
 
 1.613 
 
 1.638 
 
 1.664 
 
 26 
 
 1.3 
 
 1.690 
 
 1.716 
 
 1.742 
 
 1.769 
 
 1.796 
 
 1.823 
 
 1.850 
 
 1.877 
 
 1.904 
 
 1.932 
 
 28 
 
 1.4 
 
 1.960 
 
 1.988 
 
 2.016 
 
 2.045 
 
 2.074 
 
 2.103 
 
 2.132 
 
 2.161 
 
 2.190 
 
 2.220 
 
 30 
 
 1.5 
 
 2.250 
 
 2.280 
 
 2.310 
 
 2.341 
 
 2.372 
 
 2.403 
 
 2.434 
 
 2.465 
 
 2.496 
 
 2.528 
 
 32 
 
 1.6 
 
 2.560 
 
 2.592 
 
 2.624 
 
 2.657 
 
 2.690 
 
 2.723 
 
 2.756 
 
 2.789 
 
 2.822 
 
 2.856 
 
 34 
 
 1.7 
 
 2.890 
 
 2.924 
 
 2.958 
 
 2.993 
 
 3.028 
 
 3.063 
 
 3.098 
 
 3.133 
 
 3.168 
 
 3.204 
 
 36 
 
 1.8 
 
 3.240 
 
 3.276 
 
 3.312 
 
 3.349 
 
 3.386 
 
 3.423 
 
 3.460 
 
 3.497 
 
 3.534 
 
 3.572 
 
 38 
 
 1.9 
 
 3.610 
 
 3.648 
 
 3.686 
 
 3.725 
 
 3.764 
 
 3.803 
 
 3.842 
 
 3.881 
 
 3.920 
 
 3.960 
 
 40 
 
 2.0 
 
 4.000 
 
 4.040 
 
 4.080 
 
 4.121 
 
 4.162 
 
 4.203 
 
 4.244 
 
 4.285 
 
 4.326 
 
 4.368 
 
 42 
 
 2.1 
 
 4.410 
 
 4.452 
 
 4.494 
 
 4.537 
 
 4.580 
 
 4.623 
 
 4.666 
 
 4.709 
 
 4.752 
 
 4.796 
 
 44 
 
 2.2 
 
 4.840 
 
 4.884 
 
 4.928 
 
 4.973 
 
 5.018 
 
 5.063 
 
 5.108 
 
 5.153 
 
 5.198 
 
 5.244 
 
 46 
 
 23 
 
 5.290 
 
 5.336 
 
 5.382 
 
 5.429 
 
 5.476 
 
 5.523 
 
 5.570 
 
 5.617 
 
 5.664 
 
 5.712 
 
 48 
 
 2.4 
 
 5.760 
 
 5.808 
 
 5.856 
 
 5.905 
 
 5.954 
 
 6.003 
 
 6.052 
 
 6.101 
 
 6.150 
 
 6.200 
 
 50 
 
 25 
 
 6.250 
 
 6.300 
 
 6.350 
 
 6.401 
 
 6.452 
 
 6.503 
 
 6.554 
 
 6.605 
 
 6.656 
 
 6.708 
 
 52 
 
 2.6 
 
 6.760 
 
 6.812 
 
 6.864 
 
 6.917 
 
 6.970 
 
 7.023 
 
 7.076 
 
 7.129 
 
 7.182 
 
 7.236 
 
 54 
 
 27 
 
 7.290 
 
 7.344 
 
 7.398 
 
 7.453 
 
 7.508 
 
 7.563 
 
 7.618 
 
 7.673 
 
 7.728 
 
 7.784 
 
 56 
 
 2.8 
 
 7.840 
 
 7.896 
 
 7.952 
 
 8.009 
 
 8.066 
 
 8.123 
 
 8.180 
 
 8.237 
 
 8.294 
 
 8.352 
 
 58 
 
 2.9 
 
 8.410 
 
 8.468 
 
 8.526 
 
 8.585 
 
 8.644 
 
 8.703 
 
 8.762 
 
 8.821 
 
 8.880 
 
 8.940 
 
 60 
 
 3.0 
 
 9.000 
 
 9.060 
 
 9.120 
 
 9.181 
 
 9.242 
 
 9.303 
 
 9.364 
 
 9.425 
 
 9.486 
 
 9.548 
 
 62 
 
 3.1 
 
 9.610 
 
 9.672 
 
 9.734 
 
 9.797 
 
 9.860 
 
 9.923 
 
 9.986 
 
 10.05 
 
 10.11 
 
 10.18 
 
 6 
 
 3.2 
 
 10.24 
 
 10.30 
 
 10.37 
 
 10.43 
 
 10.50 
 
 10.56 
 
 10.63 
 
 10.69 
 
 10.76 
 
 10.82 
 
 7 
 
 3.3 
 
 10.89 
 
 10.96 
 
 11.02 
 
 11.09 
 
 11.16 
 
 11.22 
 
 11.29 
 
 11.36 
 
 11.42 
 
 11.49 
 
 7 
 
 3.4 
 
 11.56 
 
 11.63 
 
 11.70 
 
 11.76 
 
 11.83 
 
 11.90 
 
 11.97 
 
 12.04 
 
 12.11 
 
 12.18 
 
 7 
 
 3.5 
 
 12.25 
 
 12.32 
 
 12.39 
 
 12.46 
 
 12.53 
 
 12.60 
 
 12.67 
 
 12.74 
 
 12.82 
 
 12.89 
 
 7 
 
 3.6 
 
 12.96 
 
 13.03 
 
 13.10 
 
 13.18 
 
 13.25 
 
 13.32 
 
 13.40 
 
 13.47 
 
 13.54 
 
 14.62 
 
 7 
 
 3.7 
 
 13.69 
 
 13.76 
 
 13.84 
 
 13.91 
 
 13.99 
 
 14.06 
 
 14.14 
 
 14.21 
 
 14.29 
 
 14.36 
 
 8 
 
 3.8 
 
 14.44 
 
 14.52 
 
 14.59 
 
 14.67 
 
 14.75 
 
 14.82 
 
 14.90 
 
 14.98 
 
 15.05 
 
 15.13 
 
 8 
 
 3.9 
 
 15.21 
 
 15.29 
 
 15.37 
 
 15.44 
 
 15.52 
 
 15.60 
 
 15.68 
 
 15.76 
 
 15.84 
 
 15.92 
 
 8 
 
 4.0 
 
 16.00 
 
 16.08 
 
 16.16 
 
 16.24 
 
 16.32 
 
 16.40 
 
 16.48 
 
 16.56 
 
 16.65 
 
 16.73 
 
 8 
 
 4.1 
 
 16.81 
 
 16.89 
 
 16.97 
 
 17.06 
 
 17.14 
 
 17.22 
 
 17.31 
 
 17.39 
 
 17.47 
 
 17.65 
 
 8 
 
 4.2 
 
 17.64 
 
 17.72 
 
 17.81 
 
 17.89 
 
 17.98 
 
 18.06 
 
 18.15 
 
 18.23 
 
 18.32 
 
 18.40 
 
 9 
 
 4.3 
 
 18.49 
 
 18.58 
 
 18.66 
 
 18.75 
 
 18.84 
 
 18.92 
 
 19.01 
 
 19.10 
 
 19.18 
 
 19.27 
 
 9 
 
 4.4 
 
 19.36 
 
 19.45 
 
 19.54 
 
 19.62 
 
 19.71 
 
 19.80 
 
 19.89 
 
 19.98 
 
 20.07 
 
 20.16 
 
 9 
 
 4.5 
 
 20.25 
 
 20.34 
 
 20.43 
 
 20.52 
 
 20.61 
 
 20.70 
 
 20.79 
 
 20.88 
 
 20.98 
 
 21.07 
 
 9 
 
 4.6 
 
 21.16 
 
 21.25 
 
 21.34 
 
 21.44 
 
 21.53 
 
 21.62 
 
 21.72 
 
 21.81 
 
 21.90 
 
 22.00 
 
 9 
 
 4.7 
 
 22.09 
 
 22.18 
 
 22.28 
 
 22.37 
 
 22.47 
 
 22.56 
 
 22.66 
 
 22.75 
 
 22.85 
 
 22.94 
 
 10 
 
 4.8 
 
 23.04 
 
 23.14 
 
 23.23 
 
 23.33 
 
 23.43 
 
 23.52 
 
 23.62 
 
 23.72 
 
 23.81 
 
 23.91 
 
 10 
 
 4.9 
 
 24.01 
 
 24.11 
 
 24.21 
 
 24.30 
 
 24.40 
 
 24.50 
 
 24.60 
 
 24.70 
 
 24.80 
 
 24.90 
 
 10 
 
 5.0 
 
 25.00 
 
 25.10 
 
 25.20 
 
 25.30 
 
 25.40 
 
 25.50 
 
 25.60 
 
 25.70 
 
 25.81 
 
 25.91 
 
 10 
 
 5.1 
 
 26.01 
 
 26.11 
 
 26.21 
 
 26.32 
 
 26.42 
 
 26.52 
 
 26.63 
 
 26.73 
 
 26.83 
 
 26.94 
 
 10 
 
 5.2 
 
 27.04 
 
 27.14 
 
 27.25 
 
 27.35 
 
 27.46 
 
 27.56 
 
 27.67 
 
 27.77 
 
 27.88 
 
 27.98 
 
 11 
 
 5.3 
 
 28.09 
 
 28.20 
 
 28.30 
 
 28.41 
 
 28.52 
 
 28.62 
 
 28.73 
 
 28.84 
 
 28.94 
 
 29.05 
 
 11 
 
 5.4 
 
 29.16 
 
 29.27 
 
 29.38 
 
 29.48 
 
 29.59 
 
 29.70 
 
 29.81 
 
 29.92 
 
 30.03 
 
 30.14 
 
 11 
 
 n 
 
 
 
 1 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 Diff. 
 
 (Merriman, "Least Squares.") 
 
230 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE XVI. SQUARES OF NUMBERS (Concluded). 
 
 n 
 
 
 
 i 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 Diff. 
 
 5.5 
 
 30.25 
 
 30.36 
 
 30.47 
 
 30.58 
 
 30.69 
 
 30.80 
 
 30.91 
 
 31.02 
 
 31.14 
 
 31.25 
 
 11 
 
 5.6 
 
 31.36 
 
 31.47 
 
 31.58 
 
 31.70 
 
 31.81 
 
 31.92 
 
 32.04 
 
 32.15 
 
 32.26 
 
 32.38 
 
 11 
 
 5.7 
 
 32.49 
 
 32.60 
 
 32 72 
 
 32.83 
 
 32.95 
 
 33.0633.18 
 
 33.29 
 
 33.41 
 
 33.52 
 
 12 
 
 5.8 
 
 33.64 
 
 33.76 
 
 33.87 
 
 33.99 
 
 34.11 
 
 34.2234.34 
 
 34.46 
 
 34.57 
 
 34.69 
 
 12 
 
 5.9 
 
 34.81 
 
 34.93 
 
 35.05 
 
 35.16 
 
 35.28 
 
 35.40 
 
 35.52 
 
 35.64 
 
 35.76 
 
 35.88 
 
 12 
 
 6.0 
 
 36.00 
 
 36.12 
 
 36.24 
 
 36.36 
 
 36.48 
 
 36.60 
 
 36.72 
 
 36.84 
 
 36.97 
 
 37.09 
 
 12 
 
 6.1 
 
 37.21 
 
 37.33 
 
 37.45 
 
 37.58 
 
 37.70 
 
 37.82 
 
 37.95 
 
 38.07 
 
 38.19 
 
 38.32 
 
 12 
 
 6.2 
 
 38.44 
 
 38.56 
 
 38.69 
 
 38.81 
 
 38.94 
 
 39.06 
 
 39.19 
 
 39.31 
 
 39.44 
 
 39.56 
 
 13 
 
 6.3 
 
 39.69 
 
 39.82 
 
 39.94 
 
 40.07 
 
 40.20 
 
 40.32 
 
 40.45 
 
 40.58 
 
 40.70 
 
 40.83 
 
 13 
 
 6.4 
 
 40.96 
 
 41.09 
 
 41.22 
 
 41.34 
 
 41.47 
 
 41.60 
 
 41.73 
 
 41.86 
 
 41.99 
 
 42.12 
 
 13 
 
 6.5 
 
 42.25 
 
 42.38 
 
 42.51 
 
 42.64 
 
 42.77 
 
 42.90 
 
 43.03 
 
 43.16 
 
 43.30 
 
 43.43 
 
 13 
 
 6.6 
 
 43.56 
 
 43.69 
 
 43.82 
 
 43.96 
 
 44.09 
 
 44.22 
 
 44.36 
 
 44.49 
 
 44.62 
 
 44.76 
 
 13 
 
 6.7 
 
 44.89 
 
 45.02 
 
 45.16 
 
 45.29 
 
 45.43 
 
 45.56 
 
 45.70 
 
 45.83 
 
 45.97 
 
 46.10 
 
 14 
 
 6.8 
 
 46.24 
 
 46.38 
 
 46.51 
 
 46.65 
 
 46.79 
 
 46.92 
 
 47.06 
 
 47.20 
 
 47.33 
 
 47.47 
 
 14 
 
 6.9 
 
 47.61 
 
 47.75 
 
 47.89 
 
 48.02 
 
 48.16 
 
 48.30 
 
 48.44 
 
 48.58 
 
 48.72 
 
 48.86 
 
 14 
 
 7.0 
 
 49.00 
 
 49.14 
 
 49.28 
 
 49.42 
 
 49.56 
 
 49.70 
 
 49.84 
 
 49.98 
 
 50.13 
 
 50.27 
 
 14 
 
 7.1 
 
 50.41 
 
 50.55 
 
 50.69 
 
 50.84 
 
 50.98 
 
 51.12 
 
 51.27 
 
 51.41 
 
 51.55 
 
 51.70 
 
 14 
 
 7.2 
 
 51.84 
 
 51.98 
 
 52.13 
 
 52.27 
 
 52.42 
 
 52.56 
 
 52.71 
 
 52.85 
 
 53.00 
 
 53.14 
 
 15 
 
 7.3 
 
 53.29 
 
 53.44 
 
 53.58 
 
 53.73 
 
 53.88 
 
 54.02 
 
 54.17 
 
 54.32 
 
 54.46 
 
 54.61 
 
 15 
 
 7.4 
 
 54.76 
 
 54.91 
 
 55.06 
 
 55.20 
 
 55.35 
 
 55.50 
 
 55.65 
 
 55.80 
 
 55.95 
 
 56.10 
 
 15 
 
 7.5 
 
 56.25 
 
 56.40 
 
 "56.55 
 
 56.70 
 
 56.85 
 
 57.00 
 
 57.15 
 
 57.30 
 
 57.46 
 
 57.61 
 
 15 
 
 7.6 
 
 57.76 
 
 57.91 
 
 58.06 
 
 58.22 
 
 58.37 
 
 58.52 
 
 58.68 
 
 58.83 
 
 58.98 
 
 59.14 
 
 15 
 
 7.7 
 
 59.29 
 
 59.44 
 
 59.60 
 
 59.75 
 
 59.91 
 
 60.06 
 
 60.22 
 
 60.37 
 
 60.53 
 
 60.68 
 
 16 
 
 7.8 
 
 60.84 
 
 61.00 
 
 61.15 
 
 61.31 
 
 61.47 
 
 61.62 
 
 61.78 
 
 61.94 
 
 62.09 
 
 62.25 
 
 16 
 
 7.9 
 
 62.41 
 
 62.57 
 
 62.73 
 
 62.88 
 
 63.04 
 
 63.20 
 
 63.36 
 
 63.52 
 
 63.68 
 
 63.84 
 
 16 
 
 8.0 
 
 64.00 
 
 64.16 
 
 64.32 
 
 64.48 
 
 64.64 
 
 64.80 
 
 64.96 
 
 65.12 
 
 65.29 
 
 65.45 
 
 16 
 
 8.1 
 
 65.61 
 
 65.77 
 
 65.93 
 
 66.10 
 
 66.26 
 
 66.42 
 
 66.59 
 
 66.75 
 
 66.91 
 
 67.08 
 
 16 
 
 8.2 
 
 67.24 
 
 67.40 
 
 67.57 
 
 67.73 
 
 67.90 
 
 68.06 
 
 68.23 
 
 68.39 
 
 68.56 
 
 68.72 
 
 17 
 
 8.3 
 
 68.89 
 
 69.06 
 
 69.22 
 
 69.39 
 
 69.56 
 
 69.72 
 
 69.89 
 
 70.06 
 
 70.22 
 
 70.39 
 
 17 
 
 8.4 
 
 70.56 
 
 70.73 
 
 70.90 
 
 71.06 
 
 71.23 
 
 71.40 
 
 71.57 
 
 71.74 
 
 71.91 
 
 72.08 
 
 17 
 
 8.5 
 
 72.25 
 
 72.42 
 
 72.59 
 
 72.76 
 
 72.93 
 
 73.10 
 
 73.27 
 
 73.44 
 
 73.62 
 
 73.79 
 
 17 
 
 8.6 
 
 73.96 
 
 74.13 
 
 74.30 
 
 74.48 
 
 74.65 
 
 74.82 
 
 75.00 
 
 75.17 
 
 75.34 
 
 75.52 
 
 17 
 
 8.7 
 
 75.69 
 
 75.86 
 
 76.04 
 
 76.21 
 
 76.39 
 
 76.56 
 
 76.74 
 
 76.91 
 
 77.09 
 
 77.26 
 
 18 
 
 8.8 
 
 77.44 
 
 77.62 
 
 77.79 
 
 77.97 
 
 78.15 
 
 78.32 
 
 78.50 
 
 78.68 
 
 78.85 
 
 79.03 
 
 18 
 
 8.9 
 
 79.21 
 
 79.39 
 
 79.57 
 
 79.74 
 
 79.92 
 
 80.10 
 
 80.28 
 
 80.46 
 
 80.64 
 
 80.82 
 
 18 
 
 9.0 
 
 81.00 
 
 81.18 
 
 81.36 
 
 81.54 
 
 81.72 
 
 81.90 
 
 82.08 
 
 82.26 
 
 82.45 
 
 82.63 
 
 18 
 
 9.1 
 
 82.81 
 
 82.99 
 
 83.17 
 
 83.36 
 
 83.54 
 
 83.72 
 
 83.91 
 
 84.09 
 
 84.27 
 
 84.46 
 
 18 
 
 9.2 
 
 84.64 
 
 84.82 
 
 85.01 
 
 85.19 
 
 85.38 
 
 85.56 
 
 85.75 
 
 85.93 
 
 86.12 
 
 86.30 
 
 19 
 
 9.3 
 
 86.49 
 
 86.68 
 
 86.86 
 
 87.05 
 
 87.24 
 
 87.42 
 
 87.61 
 
 87.80 
 
 87.98 
 
 88.17 
 
 19 
 
 9.4 
 
 88.36 
 
 88.55 
 
 88.74 
 
 88.92 
 
 89.11 
 
 89.30 
 
 89.49 
 
 89.68 
 
 89.87 
 
 90.06 
 
 19 
 
 9.5 
 
 90.25 
 
 90.44 
 
 90.63 
 
 90.82 
 
 91.01 
 
 91.20 
 
 91.39 
 
 91.58 
 
 91.78 
 
 91.97 
 
 19 
 
 9.6 
 
 92.16 
 
 92.35 
 
 92.54 
 
 92.74 
 
 92.93 
 
 93.12 
 
 93.32 
 
 93.51 
 
 93.70 
 
 93.90 
 
 19 
 
 9.7 
 
 94.09 
 
 94.28 
 
 94.48 
 
 94.67 
 
 94.87 
 
 95.06 
 
 95.26 
 
 95.45 
 
 95.65 
 
 95.84 
 
 20 
 
 9.8 
 
 96.04 
 
 96.24 
 
 96.43 
 
 96.63 
 
 96.83 
 
 97.02 
 
 97.22 
 
 97.42 
 
 97.61 
 
 97.81 
 
 20 
 
 9.9 
 
 98.01 
 
 98.21 
 
 98.41 
 
 98.60 
 
 98.80 
 
 99.00 
 
 99.20 
 
 99.40 
 
 99.60 
 
 99.80 
 
 20 
 
 n 
 
 
 
 l 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 Diff. 
 
TABLES 
 TABLE XVII. LOGARITHMS; 1000 TO 1409. 
 
 231 
 
 
 
 
 1 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 100 
 
 0000 
 
 0004 
 
 0009 
 
 0013 
 
 0017 
 
 0022 
 
 0026 
 
 0030 
 
 0035 
 
 0039 
 
 101 
 
 0043 
 
 0048 
 
 0052 
 
 0056 
 
 0060 
 
 0065 
 
 0069 
 
 0073 
 
 0077 
 
 0082 
 
 102 
 
 0086 
 
 0090 
 
 0095 
 
 0099 
 
 0103 
 
 0107 
 
 0111 
 
 0116 
 
 0120 
 
 0124 
 
 103 
 
 0128 
 
 0133 
 
 0137 
 
 0141 
 
 0145 
 
 0149 
 
 0154 
 
 0158 
 
 0162 
 
 0166 
 
 104 
 
 0170 
 
 0175 
 
 0179 
 
 0183 
 
 0187 
 
 0191 
 
 0195 
 
 0199 
 
 0204 
 
 0208 
 
 105 
 
 0212 
 
 0216 
 
 0220 
 
 0224 
 
 0228 
 
 0233 
 
 0237 
 
 0241 
 
 0245 
 
 0249 
 
 106 
 
 0253 
 
 0257 
 
 0261 
 
 0265 
 
 0269 
 
 0273 
 
 0278 
 
 0282 
 
 0286 
 
 0290 
 
 107 
 
 0294 
 
 0298 
 
 0302 
 
 0306 
 
 0310 
 
 0314 
 
 0318 
 
 0322 
 
 0326 
 
 0330 
 
 108 
 
 0334 
 
 0338 
 
 0342 
 
 0346 
 
 0350 
 
 0354 
 
 0358 
 
 0362 
 
 0366 
 
 0370 
 
 109 
 
 0374 
 
 0378 
 
 0382 
 
 0386 
 
 0390 
 
 0394 
 
 0398 
 
 0402 
 
 0406 
 
 0410 
 
 110 
 
 0414 
 
 0418 
 
 0422 
 
 0426 
 
 0430 
 
 0434 
 
 0438 
 
 0441 
 
 0445 
 
 0449 
 
 111 
 
 0453 
 
 0457 
 
 0461 
 
 0465 
 
 0469 
 
 0473 
 
 0477 
 
 0481 
 
 0484 
 
 0488 
 
 112 
 
 0492 
 
 0496 
 
 0500 
 
 0504 
 
 0508 
 
 0512 
 
 0515 
 
 0519 
 
 0523 
 
 0527 
 
 113 
 
 0531 
 
 0535 
 
 0538 
 
 0542 
 
 0546 
 
 0550 
 
 0554 
 
 0558 
 
 0561 
 
 0565 
 
 114 
 
 0569 
 
 0573 
 
 0577 
 
 0580 
 
 0584 
 
 0588 
 
 0592 
 
 0596 
 
 0599 
 
 0603 
 
 115 
 
 0607 
 
 0611 
 
 0615 
 
 0618 
 
 0622 
 
 0626 
 
 0630 
 
 0633 
 
 0637 
 
 0641 
 
 116 
 
 0645 
 
 0648 
 
 0652 
 
 0656 
 
 0660 
 
 0663 
 
 0667 
 
 0671 
 
 0674 
 
 0678 
 
 117 
 
 0682 
 
 0686 
 
 0689 
 
 0693 
 
 0697 
 
 0700 
 
 0704 
 
 0708 
 
 0711 
 
 0715 
 
 118 
 
 0719 
 
 0722 
 
 0726 
 
 0730 
 
 0734 
 
 0737 
 
 0741 
 
 0745 
 
 0748 
 
 0752 
 
 119 
 
 0755 
 
 0759 
 
 0763 
 
 0766 
 
 0770 
 
 0774 
 
 0777 
 
 0781 
 
 0785 
 
 0788 
 
 120 
 
 0792 
 
 0795 
 
 0799 
 
 0803 
 
 0806 
 
 0810 
 
 0813 
 
 0817 
 
 0821 
 
 0824 
 
 121 
 
 0828 
 
 0831 
 
 0835 
 
 0839 
 
 0842 
 
 0846 
 
 0849 
 
 0853 
 
 0856 
 
 0860 
 
 122 
 
 0864 
 
 0867 
 
 0871 
 
 0874 
 
 0878 
 
 0881 
 
 0885 
 
 0888 
 
 0892 
 
 0896 
 
 123 
 
 0899 
 
 0903 
 
 0906 
 
 0910 
 
 0913 
 
 0917 
 
 0920 
 
 0924 
 
 0927 
 
 0931 
 
 124 
 
 0934 
 
 0938 
 
 0941 
 
 0945 
 
 0948 
 
 0952 
 
 0955 
 
 0959 
 
 0962 
 
 0966 
 
 125 
 
 0969 
 
 0973 
 
 0976 
 
 0980 
 
 0983 
 
 0986 
 
 0990 
 
 0993 
 
 0997 
 
 1000 
 
 126 
 
 1004 
 
 1007 
 
 1011 
 
 1014 
 
 1017 
 
 1021 
 
 1024 
 
 1028 
 
 1031 
 
 1035 
 
 127 
 
 1038 
 
 1041 
 
 1045 
 
 1048 
 
 1052 
 
 1055 
 
 1059 
 
 1062 
 
 1065 
 
 1069 
 
 128 
 
 1072 
 
 1075 
 
 1079 
 
 1082 
 
 1086 
 
 1089 
 
 1092 
 
 1096 
 
 1099 
 
 1103 
 
 129 
 
 1106 
 
 1109 
 
 1113 
 
 1116 
 
 1119 
 
 1123 
 
 1126 
 
 1129 
 
 1133 
 
 1136 
 
 130 
 
 1139 
 
 1143 
 
 1146 
 
 1149 
 
 1153 
 
 1156 
 
 1159 
 
 1163 
 
 1166 
 
 1169 
 
 131 
 
 1173 
 
 1176 
 
 1179 
 
 1183 
 
 1186 
 
 1189 
 
 1193 
 
 1196 
 
 1199 
 
 1202 
 
 132 
 
 1206 
 
 1209 
 
 1212 
 
 1216 
 
 1219 
 
 1222 
 
 1225 
 
 1229 
 
 1232 
 
 1235 
 
 133 
 
 1239 
 
 1242 
 
 1245 
 
 1248 
 
 1252 
 
 1255 
 
 1258 
 
 1261 
 
 1265 
 
 1268 
 
 134 
 
 1271 
 
 1274 
 
 1278 
 
 1281 
 
 1284 
 
 1287 
 
 1290 
 
 1294 
 
 1297 
 
 1300 
 
 135 
 
 1303 
 
 1307 
 
 1310 
 
 1313 
 
 1316 
 
 1319 
 
 1323 
 
 1326 
 
 1329 
 
 1332 
 
 136 
 
 1335 
 
 1339 
 
 1342 
 
 1345 
 
 1348 
 
 1351 
 
 1355 
 
 1358 
 
 1361 
 
 1364 
 
 137 
 
 1367 
 
 1370 
 
 1374 
 
 1377 
 
 1380 
 
 1383 
 
 1386 
 
 1389 
 
 1392 
 
 1396 
 
 138 
 
 1399 
 
 1402 
 
 1405 
 
 1408 
 
 1411 
 
 1414 
 
 1418 
 
 1421 
 
 1424 
 
 1427 
 
 139 
 
 1430 
 
 1433 
 
 1436 
 
 1440 
 
 1443 
 
 1446 
 
 1449 
 
 1452 
 
 1455 
 
 1458 
 
 140 
 
 1461 
 
 1464 
 
 1467 
 
 1471 
 
 1474 
 
 1477 
 
 1480 
 
 1483 
 
 1486 
 
 1489 
 
 (Bottomley, "Four Fig. Math. Tables.") 
 
232 
 
 THE THEORY OF MEASUREMENTS 
 
 * TABLE XVIII. LOGARITHMS. 
 
 
 
 
 1 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 & 
 
 9 
 
 123 
 
 456 
 
 789 
 
 10 
 
 0000 
 
 0043 
 
 0086 
 
 0128 
 
 0170 
 
 O2I2 
 
 0253 
 
 0294 
 
 0334 
 
 0374 
 
 4812 
 
 17 21 2 5 
 
 29 33 37 
 
 11 
 
 12 
 13 
 
 0414 
 0792 
 
 "39 
 
 0453 
 0828 
 
 "73 
 
 0492 
 0864 
 1206 
 
 0531 
 0899 
 1239 
 
 0569 
 
 0934 
 1271 
 
 0607 
 9 6 9 
 I33 
 
 0645 
 100^ 
 
 1335 
 
 0682 
 1038 
 1367 
 
 0719 
 1072 
 1399 
 
 0755 
 1106 
 
 1430 
 
 4811 
 3 7io 
 3 6 10 
 
 15 19 23 
 14 17 21 
 
 13 16 10 
 
 26 30 34 
 24 28 31 
 23 26 29 
 
 21 24 27 
 20 22 25 
 
 18 21 24 
 
 14 
 15 
 16 
 
 1461 
 1761 
 2041 
 
 1492 
 1790 
 2068 
 
 iffi 
 
 2095 
 
 1553 
 
 1847 
 
 2122 
 
 1584 
 1875 
 2148 
 
 i6i<: 
 1903 
 2175 
 
 164^: 
 
 1931 
 
 22OI 
 
 1673 
 1959 
 2227 
 
 1703 
 1987 
 
 2253 
 
 1732 
 2014 
 2279 
 
 3 6 9 
 36 8 
 
 3 5 8 
 
 12 15 18 
 ii 14 17 
 ii 13 16 
 
 17 
 18 
 19 
 
 2304 
 
 $1 
 
 2330 
 2577 
 2810 
 
 2355 
 2601 
 
 2833 
 
 2380 
 2625 
 2856 
 
 2405 
 2648 
 2878 
 
 2430 
 2672 
 2900 
 
 2455 
 2695 
 
 2923 
 
 2480 
 2718 
 2945 
 
 2504 
 2742 
 2967 
 
 2529 
 2765 
 2989 
 
 2 57 
 
 2 5 7 
 247 
 
 10 12 15 
 
 9 12 14 
 9 " I 2 
 
 17 2O 22 
 
 16 19 21 
 16 18 20 
 
 20 
 
 3010 
 
 3032 
 
 3054 
 
 375 
 
 3096 
 
 3"8 
 
 3139 
 
 3160 
 
 3181 
 
 3201 
 
 24 6 
 
 8 ii 13 
 
 15 17 19 
 
 21 
 22 
 23 
 
 3222 
 3424 
 3617 
 
 3243 
 3444 
 3636 
 
 3263 
 3464 
 3655 
 
 3284 
 3483 
 3674 
 
 3304 
 3502 
 3692 
 
 3324 
 3522 
 
 37" 
 
 3345 
 354i 
 3729 
 
 3365 
 3560 
 
 3747 
 
 3385 
 3579 
 3766 
 
 3404 
 3598 
 3784 
 
 2 4 6 
 24 6 
 2 4 6 
 
 8 10 12 
 8 10 12 
 
 7 9 ii 
 
 14 16 18 
 H 15 17 
 J 3 15 '7 
 
 24 
 25 
 26 
 
 3802 
 3979 
 415 
 
 3820 
 
 3997 
 4166 
 
 3838 
 4014 
 4183 
 
 3856 
 
 403 1 
 4200 
 
 3874 
 4048 
 4216 
 
 3892 
 4065 
 4232 
 
 3909 
 4082 
 4249 
 
 3927 
 4099 
 4265 
 
 3945 
 4116 
 4281 
 
 3962 
 
 4133 
 4298 
 
 245 
 
 235 
 235 
 
 7 9 ii 
 7 9 10 
 7 8 10 
 
 12 14 16 
 
 12 14 15 
 II 13 15 
 
 27 
 28 
 29 
 
 43H 
 4472 
 4624 
 
 4330 
 4487 
 
 4639 
 
 4346 
 4502 
 
 4654 
 
 4362 
 45 l8 
 4669 
 
 4378 
 4533 
 4683 
 
 4393 
 4548 
 4698 
 
 4409 
 45 6 4 
 47 J 3 
 
 4425 
 4579 
 4728 
 
 444 
 4594 
 4742 
 
 445 6 
 4609 
 
 4757 
 
 2 3 5 
 
 2 3 5 
 1 3 4 
 
 689 
 689 
 6 7 9 
 
 II 13 14 
 II 12 \i 
 10 12 13 
 
 30 
 
 477 1 
 
 4786 
 
 4800 
 
 4814 
 
 4829 
 
 4843 
 
 4857 
 
 4871 
 
 4886 
 
 4900 
 
 i 3 4 
 
 6 7 9 
 
 10 ii 13 
 
 31 
 32 
 33 
 
 4914 
 
 505i 
 5185 
 
 4928 
 5065 
 5198 
 
 4942 
 5079 
 5211 
 
 4955 
 5092 
 
 5224 
 
 4969 
 5105 
 5237 
 
 4983 
 5"9 
 5250 
 
 4997 
 5i|2 
 5263 
 
 5011 
 
 5*45 
 5276 
 
 5024 
 
 5159 
 5289 
 
 5 38 
 5172 
 5302 
 
 3 4 
 3 4 
 3 4 
 
 6 7 8 
 
 HI 
 
 10 II 12 
 9 II 12 
 9 10 12 
 
 34 
 35 
 36 
 
 5315 
 544i 
 55 6 3 
 
 5328 
 5453 
 5575 
 
 5340 
 5465 
 5587 
 
 5353 
 5478 
 5599 
 
 5366 
 5490 
 5611 
 
 5378 
 5502 
 5 6 23 
 
 5391 
 554 
 5 6 35 
 
 5403 
 5527 
 5647 
 
 54i6 
 
 5539 
 5658 
 
 5428 
 
 555i 
 5670 
 
 3 4 
 2 4 
 2 4 
 
 !.:; 
 
 5 6 7 
 
 9 10 ii 
 9 10 ii 
 8 10 ii 
 
 37 
 38 
 39 
 
 5682 
 5798 
 59" 
 
 5 6 94 
 5809 
 5922 
 
 5705 
 5821 
 
 5933 
 
 5717 
 5832 
 5944 
 
 5729 
 5843 
 5955 
 
 5740 
 
 58 II 
 5966 
 
 5977 
 
 5763 
 5877 
 5988 
 
 577 C 
 5999 
 
 5786 
 
 5899 
 6010 
 
 2 3 
 2 3 
 2 3 
 
 5 6 7 
 
 5 6 7 
 4 5 7 
 
 8 9 10 
 8 9 10 
 8 9 10 
 
 40 
 
 6021 
 
 6031 
 
 6042 
 
 6o53 
 
 6064 
 
 6075 
 
 6085 
 
 6096 
 
 6107 
 
 6117 
 
 2 3 
 
 4 5 6 
 
 8 9 10 
 
 41 
 42 
 43 
 
 6128 
 6232 
 6335 
 
 6138 
 6243 
 6345 
 
 6149 
 6253 
 6355 
 
 6160 
 6263 
 6365 
 
 6170 
 6274 
 6375 
 
 6180 
 6284 
 6385 
 
 6191 
 6294 
 6395 
 
 6201 
 6304 
 6405 
 
 6212 
 
 6314 
 6415 
 
 6222 
 6 3 2 5 
 6425 
 
 2 3 
 2 3 
 
 2 3 
 
 4 5 6 
 4 5 6 
 4 5 6 
 
 7 8 9 
 7 8 9 
 7 8 9 
 
 44 
 45 
 46 
 
 6435 
 6532 
 6628 
 
 6444 
 6542 
 6637 
 
 6454 
 ^6 
 
 6464 
 6561 
 6656 
 
 6474 
 
 657 1 
 6665 
 
 6484 
 6580 
 6675 
 
 6 493 
 6590 
 6684 
 
 6 53 
 6599 
 
 6513 
 6609 
 6702 
 
 6522 
 6618 
 6712 
 
 2 3 
 
 I 2 3 
 
 I 2 3 
 
 4 5 6 
 
 4 5 6 
 456 
 
 7 8 9 
 7 8 9 
 7 7 8 
 
 47 
 48 
 49 
 
 6721 
 6812 
 6902 
 
 6730 
 6821 
 6911 
 
 6739 
 683O 
 6920 
 
 6749 
 6839 
 6928 
 
 6758 
 6848 
 
 6937 
 
 6767 
 6857 
 6946 
 
 6776 
 6866 
 6955 
 
 6785 
 
 ?2 5 
 6964 
 
 6794 
 6884 
 6972 
 
 6803 
 6893 
 6981 
 
 I 2 3 
 I 2 3 
 I 2 3 
 
 4 5 5 
 4 4 5 
 445 
 
 6 7 8 
 678 
 678 
 
 50 
 
 6990 
 
 6998 
 
 7007 
 
 7016 
 
 7024 
 
 733 
 
 7042 
 
 7050 
 
 7059 
 
 7067 
 
 I 2 3 
 
 3 4 5 
 
 678 
 
 51 
 52 
 53 
 
 7076 
 7160 
 7243 
 
 7084 
 7168 
 7251 
 
 7093 
 7177 
 
 7259 
 
 7101 
 
 7185 
 7267 
 
 7110 
 7193 
 7275 
 
 7118 
 7202 
 7284 
 
 7126 
 7210 
 7292 
 
 7i35 
 7218 
 7300 
 
 7H3 
 7226 
 7308 
 
 7152 
 7235 
 73i6 
 
 I 2 3 
 122 
 I 2 2 
 
 3 4 5 
 3 4 5 
 345 
 
 678 
 
 6 7 7 
 667 
 
 54 
 
 7324 
 
 7332 
 
 7340 
 
 7348 
 
 735 6 
 
 7364 
 
 7372 
 
 7380 
 
 7388 
 
 7396 
 
 I 2 2 
 
 3 4 5 
 
 667 
 
 * From Bottomley's Four Figure Mathematical Tables, by courtesy of The Macmillan Company. 
 
TABLES 
 
 233 
 
 TABLE XVIII. LOGARITHMS (Concluded). 
 
 
 
 
 1 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 7 
 
 8 
 
 9 
 
 1 23 
 
 456 
 
 789 
 
 55 
 
 7404 
 
 .7412 
 
 7490 
 566 
 ^642 
 
 74i9 
 
 7427 
 
 7435 
 
 7443 
 
 745i 
 
 7459 
 
 7466 
 
 7474 
 
 122 
 
 3 4 5 
 
 5 6 7 
 
 56 
 57 
 58 
 
 7482 
 7559 
 7 6 34 
 
 7497 
 7574 
 7649 
 
 7505 
 
 7582 
 
 7657 
 
 75*3 
 7589 
 7664 
 
 7520 
 
 7597 
 7672 
 
 7528 
 7604 
 7679 
 
 7536 
 7612 
 7686 
 
 7543 
 7619 
 7694 
 
 755i 
 7627 
 7701 
 
 2 2 
 2 2 
 I 2 
 
 345 
 3 4 5 
 344 
 
 5 6 7 
 5 6 7 
 5 6 7 
 
 59 
 60 
 61 
 
 7709 
 7782 
 7853 
 
 7716 
 
 7789 
 7860 
 
 7723 
 7796 
 7868 
 
 773i 
 7803 
 7875 
 
 7738 
 7810 
 7882 
 
 7745 
 7818 
 7889 
 
 Ws 
 
 7896 
 
 7760 
 7832 
 7903 
 
 7767 
 
 7839 
 7910 
 
 7774 
 7846 
 7917 
 
 2 
 2 
 2 
 
 344 
 344 
 344 
 
 5 6 7 
 566 
 5 6 6 
 
 62 
 63 
 64 
 
 7924 
 
 7993 
 8062 
 
 7931 
 8000 
 8069 
 
 7938 
 8007 
 
 8075 
 
 7945 
 8014 
 8082 
 
 7952 
 8021 
 8089 
 
 79^Q 
 
 8096 
 
 7966 
 8035 
 8102 
 
 7973 
 8041 
 8109 
 
 7980 
 8048 
 8116 
 
 7987 
 8055 
 8122 
 
 2 
 2 
 2 
 
 334 
 334 
 334 
 
 566 
 5 5 6 
 5 5 6 
 
 65 
 
 8129 
 
 8136 
 
 8142 
 
 8149 
 
 8156 
 
 8162 
 
 8169 
 
 8176 
 
 8182 
 
 8189 
 
 2 
 
 334 
 
 5 5 6 
 
 66 
 67 
 68 
 
 .8195 
 8261 
 
 8325 
 
 8202 
 8267 
 833i 
 
 8209 
 8274 
 8338 
 
 8215 
 8280 
 8344 
 
 8222 
 8287 
 8351 
 
 8228 
 8293 
 8357 
 
 8235 
 8299 
 
 8363 
 
 8241 
 8306 
 8370 
 
 8248 
 8312 
 8376 
 
 8254 
 
 8319 
 8382 
 
 2 
 
 2 
 2 
 
 334 
 334 
 334 
 
 5 5 6 
 
 5 5 \ 
 4 5 6 
 
 69 
 70 
 71 
 
 8388 
 8451 
 8513 
 
 8395 
 8457 
 8519 
 
 8401 
 8463 
 8525 
 
 8407 
 8470 
 8531 
 
 8414 
 8476 
 8537 
 
 8420 
 8482 
 8543 
 
 8426 
 8488 
 8549 
 
 8432 
 8494 
 8555 
 
 8439 
 8500 
 8561 
 
 8445 
 8506 
 8567 
 
 2 
 2 
 2 
 
 234 
 234 
 234 
 
 4 5 6 
 4 5 6 
 4 5 5 
 
 72 
 
 73 
 
 74 
 
 ~75~ 
 
 8573 
 8633 
 8692 
 
 8579 
 8639 
 8698 
 
 8585 
 8645 
 8704 
 
 8591 
 8651 
 8710 
 
 8597 
 8657 
 8716 
 
 8603 
 866 3 
 8722 
 
 8609 
 8669 
 8727 
 
 8615 
 8675 
 8733 
 
 8621 
 8681 
 8739 
 
 8627 
 8686 
 8745 
 
 2 
 2 
 
 2 
 
 234 
 234 
 234 
 
 455 
 455 
 4 5 5 
 
 875i 
 
 8756 
 
 8762 
 
 8768 
 
 8774 
 
 8779 
 
 8785 
 
 8791 
 
 8797 
 
 8802 
 
 2 
 
 233 
 
 4 5 5 
 
 76 
 77 
 78 
 
 8808 
 8865 
 8921 
 
 8814 
 8871 
 8927 
 
 8820 
 8876 
 8932 
 
 8825 
 8882 
 8938 
 
 8831 
 8887 
 8943 
 
 8837 
 8893 
 8949 
 
 8842 
 8899 
 8954 
 
 8848 
 8904 
 8960 
 
 8854 
 8910 
 8965 
 
 8859 
 
 8915 
 8971 
 
 2 
 2 
 2 
 
 233 
 233 
 233 
 
 4 5 5 
 4 4 5 
 4 4 5 
 
 445 
 4 4 5 
 445 
 
 79 
 80 
 81 
 
 8976 
 9031 
 9085 
 
 8982 
 9036 
 9090 
 
 8987 
 9042 
 9096 
 
 8993 
 9047 
 9101 
 
 8998 
 
 9053 
 9106 
 
 9004 
 9058 
 9112 
 
 9009 
 9063 
 9117 
 
 9015 
 9069 
 9122 
 
 9020 
 9074 
 9128 
 
 9025 
 9079 
 9U3 
 
 2 
 
 2 
 2 
 
 233 
 233 
 233 
 
 82 
 83 
 84 
 
 9138 
 9191 
 
 9243 
 
 9H3 
 9196 
 9248 
 
 9149 
 9201 
 9253 
 
 9154 
 9206 
 9258 
 
 9159 
 9212 
 
 9263 
 
 9165 
 
 9217 
 9269 
 
 9170 
 9222 
 9274 
 
 9175 
 9227 
 
 9279 
 
 9180 
 9232 
 9284 
 
 9186 
 9238 
 9289 
 
 2 
 2 
 2 
 
 233 
 233 
 233 
 
 4 4 5 
 445 
 445 
 
 85 
 
 9294 
 
 9299 
 
 9304 
 
 9309 
 
 93i5 
 
 9320 
 
 9325 
 
 9330 
 
 9335 
 
 9340 
 
 I 2 
 
 233 
 
 445 
 
 86 
 87 
 88 
 
 9345 
 9395 
 9445 
 
 935 
 9400 
 
 945 
 
 9355 
 9405 
 9455 
 
 9360 
 9410 
 9460 
 
 9365 
 94i5 
 9465 
 
 9370 
 9420 
 9469 
 
 9375 
 9425 
 9474 
 
 938o 
 943 
 9479 
 
 9385 
 9435 
 9484 
 
 9390 
 9440 
 9489 
 
 I 2 
 O 
 
 
 233 
 223 
 223 
 
 4 4 5 
 344 
 344 
 
 89 
 90 
 91 
 
 9494 
 9542 
 9590 
 
 9499 
 9547 
 9595 
 
 954 
 9552 
 9600 
 
 959 
 9557 
 9605 
 
 95*3 
 9562 
 9609 
 
 9518 
 95 66 
 9614 
 
 9523 
 957i 
 9619 
 
 9528 
 9576 
 9624 
 
 9533 
 9628 
 
 9538 
 9586 
 
 9633 
 
 O 
 
 O 
 
 223 
 223 
 223 
 
 344 
 344 
 344 
 
 92 
 93 
 94 
 
 ~95~ 
 
 9638 
 9685 
 973i 
 
 9643 
 9689 
 
 9736 
 
 9647 
 9694 
 974i 
 
 9652 
 9699 
 9745 
 
 9657 
 973 
 975 
 
 9661 
 9708 
 
 9754 
 
 9666 
 
 97 i 3 
 9759 
 
 9671 
 9717 
 9763 
 
 9675 
 9722 
 9768 
 
 9680 
 9727 
 9773 
 
 
 O 
 O 
 
 223 
 223 
 223 
 
 344 
 344 
 344 
 
 9777 
 
 9782 
 
 9786 
 
 9791 
 
 9795 
 
 9800 
 
 9805 
 
 9809 
 
 9814 
 
 9818 
 
 
 
 223 
 
 344 
 
 96 
 97 
 98 
 
 9823 
 9868 
 9912 
 
 9827 
 9872 
 9917 
 
 9832 
 9877 
 9921 
 
 9836 
 9881 
 9926 
 
 9841 
 9886 
 9930 
 
 9845 
 9890 
 
 9934 
 
 9850 
 9894 
 9939 
 
 9854 
 9899 
 9943 
 
 9859 
 9903 
 9948 
 
 9863 
 9908 
 9952 
 
 O 
 O 
 O 
 
 223 
 223 
 223 
 
 344 
 344 
 344 
 
 99 
 
 995 6 
 
 9961 
 
 9965 
 
 9969 
 
 9974 
 
 9978 
 
 9983 
 
 9987 
 
 9991 
 
 9996 
 
 I I 
 
 223 
 
 334 
 
234 
 
 THE THEORY OF MEASUREMENTS 
 
 * TABLE XIX. NATURAL SINES. 
 
 
 0' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 SO' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 
 
 oooo 
 
 0017 
 
 oo35 
 
 0052 
 
 0070 
 
 0087 
 
 0105 
 
 OI22 
 
 0140 
 
 oi57 
 
 369 
 
 12 15 
 
 1 
 
 2 
 3 
 
 0175 
 0349 
 0523 
 
 0192 
 
 0366 
 0541 
 
 0209 
 0384 
 0558 
 
 0227 
 0401 
 0576 
 
 0244 
 0419 
 0593 
 
 0262 
 0436 
 0610 
 
 0279 
 
 0454 
 0628 
 
 0297 
 0471 
 0645 
 
 0314 
 0488 
 o663 
 
 0332 
 0506 
 0680 
 
 369 
 369 
 369 
 
 12 I 5 
 12 I 5 
 12 I 5 
 
 4 
 5 
 6 
 
 ~7~ 
 8 
 9 
 
 0698 
 0872 
 1045 
 
 7!5 
 0889 
 1063 
 
 0732 
 0906 
 1080 
 
 0750 
 0924 
 1097 
 
 0767 
 0941 
 "15 
 
 0785 
 0958 
 1132 
 
 0802 
 0976 
 1149 
 
 0819 
 
 0993 
 1167 
 
 0837 
 ion 
 1184 
 
 0854 
 1028 
 
 I2OI 
 
 369 
 369 
 369 
 
 12 I 5 
 12 14 
 
 12 14 
 
 1219 
 
 1392 
 1564 
 
 1236 
 1409 
 1582 
 
 1253 
 1426 
 
 1599 
 
 1271 
 
 1444 
 1616 
 
 1288 
 1461 
 1633 
 
 1305 
 1478 
 1650 
 
 J 323 
 
 1495 
 1668 
 
 1340 
 
 \&1 
 
 1357 
 1530 
 1702 
 
 !374 
 J 547 
 1719 
 
 369 
 369 
 369 
 
 12 14 
 12 14 
 12 14 
 
 10 
 
 1736 
 
 !754 
 
 1771 
 
 1788 
 
 1805 
 
 1822 
 
 1840 
 
 1857 
 
 1874 
 
 1891 
 
 369 
 
 12 14 
 
 11 
 12 
 13 
 
 1908 
 2079 
 
 2250 
 
 1925 
 2096 
 2267 
 
 1942 
 2113 
 
 2284 
 
 1959 
 2130 
 2300 
 
 1977 
 2147 
 2317 
 
 1994 
 2164 
 2334 
 
 2OII 
 
 2181 
 
 235 I 
 
 2028 
 2198 
 2368 
 
 2045 
 2215 
 2385 
 
 2062 
 2232 
 2402 
 
 369 
 369 
 368 
 
 II I 4 
 II 14 
 
 II I 4 
 
 14 
 15 
 16 
 
 TT 
 18 
 19 
 
 2419 
 2588 
 2756 
 
 2436 
 2605 
 2773 
 
 2453 
 2622 
 2790 
 
 2470 
 2639 
 2807 
 
 2487 
 2656 
 2823 
 
 2504 
 2672 
 2840 
 
 2521 
 2689 
 
 2857 
 
 2538 
 2706 
 2874 
 
 2554 
 2723 
 2890 
 
 257i 
 2740 
 2907 
 
 368 
 368 
 368 
 
 II 14 
 II I 4 
 II 14 
 
 2924 
 3090 
 3256 
 
 2940 
 3io7 
 3272 
 
 2957 
 3123 
 3289 
 
 2974 
 3 J 4 
 3305 
 
 2990 
 3156 
 3322 
 
 3007 
 3i73 
 3338 
 
 3024 
 3190 
 
 3355 
 
 3040 
 3206 
 
 337 1 
 
 3057 
 3223 
 3387 
 
 3074 
 3239 
 3404 
 
 3 6 8 
 368 
 
 3 5 8 
 
 II 14 
 II 14 
 II 14 
 
 20 
 
 3420 
 
 3437 
 
 3453 
 
 3469 
 
 3486 
 
 3502 
 
 35i8 
 
 3535 
 
 3551 
 
 35 6 7 
 
 3 5 8 
 
 II 14 
 
 21 
 22 
 23 
 
 ~24~ 
 25 
 26 
 
 3584 
 3746 
 3907 
 
 3600 
 3762 
 3923 
 
 3616 
 3778 
 3939 
 
 3633 
 3795 
 3955 
 
 3 6 49 
 3811 
 
 397 1 
 
 3665 
 3827 
 3987 
 
 3681 
 
 3843 
 4003 
 
 3697 
 3859 
 4019 
 
 37H 
 3875 
 4035 
 
 3730 
 3891 
 405 l 
 
 3 5 8 
 3 5 8 
 3 5 8 
 
 II 14 
 II 14 
 II 14 
 
 4067 
 
 4226 
 
 4384 
 
 4083 
 4242 
 4399 
 
 4099 
 4258 
 4415 
 
 4H5 
 
 4274 
 443i 
 
 4131 
 
 4289 
 4446 
 
 4147 
 
 435 
 4462 
 
 4163 
 432i 
 4478 
 
 4179 
 4337 
 4493 
 
 4195 
 4352 
 459 
 
 4210 
 4368 
 4524 
 
 3 5 8 
 3 5 8 
 3 5 8 
 
 II 13 
 II 13 
 
 10 13 
 
 27 
 28 
 29 
 
 4540 
 
 4695 
 4848 
 
 4555 
 4710 
 4863 
 
 457 i 
 4726 
 4879 
 
 4586 
 474i 
 4894 
 
 4602 
 
 475 6 
 4909 
 
 4617 
 4772 
 4924 
 
 4633 
 4787 
 4939 
 
 4648 
 4802 
 4955 
 
 4664 
 4818 
 497 
 
 4679 
 4833 
 4985 
 
 3 5 8 
 3 5 8 
 
 3 5 8 
 
 10 13 
 
 10 13 
 10 13 
 
 30 
 
 5000 
 
 5015 
 
 53o 
 
 545 
 
 5060 
 
 5075 
 
 5090 
 
 5 I0 5 
 
 5120 
 
 5135 
 
 3 5 8 
 
 10 13 
 
 31 
 32 
 33 
 
 5150 
 5299 
 5446 
 
 5*65 
 53H 
 546i 
 
 5180 
 5329 
 5476 
 
 5195 
 5344 
 5490 
 
 5210 
 5358 
 5505 
 
 5225 
 5373 
 5519 
 
 5240 
 5388 
 5534 
 
 5255 
 5402 
 
 5548 
 
 5270 
 5417 
 5563 
 
 5284 
 5432 
 5577 
 
 2 5 7 
 257 
 2 5 7 
 
 IO 12 
 10 12 
 IO 12 
 
 34 
 35 
 36 
 
 5592 
 5736 
 5878 
 
 5606 
 
 575 
 5892 
 
 5621 
 
 5764 
 5906 
 
 5635 
 5779 
 5920 
 
 5650 
 5793 
 5934 
 
 5664 
 5807 
 5948 
 
 5678 
 5821 
 5962 
 
 5693 
 5835 
 5976 
 
 577 
 5850 
 5990 
 
 572i 
 5864 
 6004 
 
 257 
 
 2 5 7 
 2 5 7 
 
 IO 12 
 IO 12 
 
 9 12 
 
 37 
 38 
 39 
 
 6018 
 
 6157 
 6293 
 
 6032 
 6170 
 6307 
 
 6046 
 6184 
 6320 
 
 6060 
 6198 
 6334 
 
 6074 
 6211 
 6347 
 
 6088 
 6225 
 6361 
 
 6101 
 6239 
 6374 
 
 6115 
 6252 
 6388 
 
 6129 
 6266 
 6401 
 
 6143 
 6280 
 6414 
 
 257 
 2 5 7 
 247 
 
 9 12 
 9 ii 
 9 ii 
 
 40 
 
 6428 
 
 6441 
 
 6 455 
 
 6468 
 
 6481 
 
 6494 
 
 6508 
 
 6521 
 
 6534 
 
 6 547 
 
 247 
 
 9 ii 
 
 41 
 42 
 43 
 
 6561 
 6820 
 
 6 574 
 6704 
 
 6833 
 
 6587 
 6717 
 6845 
 
 6600 
 6730 
 6858 
 
 6613 
 
 6743 
 6871 
 
 6626 
 6756 
 6884 
 
 6639 
 6769 
 6896 
 
 6652 
 6782 
 6909 
 
 6665 
 6794 
 6921 
 
 6678 
 6807 
 6934 
 
 247 
 2 4 6 
 246 
 
 9 ii 
 
 9 " 
 8 ii 
 
 44 
 
 6947 
 
 6959 
 
 6972 
 
 6984 
 
 6997 
 
 7009 
 
 7022 
 
 7034 
 
 7046 
 
 7059 
 
 246 
 
 8 10 
 
 * From Bottomley's Four Figure Mathematical Tables, by courtesy of The Macmillan Company. 
 
TABLES 
 TABLE XIX. NATURAL SINES (Concluded). 
 
 235 
 
 
 0' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 3O' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 45 
 
 7071 
 
 7083 
 
 7096 
 
 7108 
 
 7120 
 
 7 J 33 
 
 7H5 
 
 7i57 
 
 7169 
 
 7181 
 
 246 
 
 8 10 
 
 46 
 
 47 
 48 
 
 7 J 93 
 73*4 
 743i 
 
 7206 
 7325 
 
 7443 
 
 7218 
 7337 
 
 7455 
 
 7230 
 
 7349 
 7466 
 
 7242 
 73 61 
 
 7478 
 
 7254 
 7373 
 7490 
 
 7266 
 
 7385 
 7501 
 
 7278 
 7396 
 75i3 
 
 7290 
 7408 
 7524 
 
 7302 
 7420 
 7536 
 
 246 
 246 
 246 
 
 8 10 
 8 10 
 8 10 
 
 49 
 50 
 51 
 
 7547 
 7660 
 7771 
 
 7558 
 7672 
 7782 
 
 757 
 7683 
 7793 
 
 758i 
 7694 
 7804 
 
 7593 
 7705 
 7815 
 
 7604 
 7716 
 7826 
 
 7 6l 5 
 
 7727 
 7837 
 
 7627 
 7738 
 7848 
 
 7638 
 7749 
 7859 
 
 7649 
 7760 
 7869 
 
 2 4 6 
 246 
 2 4 5 
 
 8 9 
 7 9 
 7 9 
 
 52 
 53 
 54 
 
 7880 
 7986 
 8090 
 
 7891 
 
 7997 
 8100 
 
 7902 
 8007 
 8111 
 
 7912 
 8018 
 8121 
 
 7923 
 8028 
 8131 
 
 7934 
 8039 
 8141 
 
 7944 
 8049 
 8151 
 
 7955 
 8059 
 8161 
 
 7965 
 8070 
 8171 
 
 7976 
 8080 
 8181 
 
 2 4 5 
 235 
 2 3 5 
 
 7 9 
 7 9 
 7 8 
 
 55 
 
 8192 
 
 8202 
 
 8211 
 
 8221 
 
 8231 
 
 8241 
 
 8251 
 
 8261 
 
 8271 
 
 8281 
 
 2 3 5 
 
 7 8 
 
 56 
 57 
 58 
 
 8290 
 
 8387 
 8480 
 
 8300 
 8396 
 8490 
 
 8310 
 8406 
 8499 
 
 8320 
 
 8415 
 8508 
 
 8329 
 8425 
 8517 
 
 8339 
 8434 
 8526 
 
 8348 
 8443 
 8536 
 
 8358 
 8453 
 8545 
 
 8368 
 8462 
 8554 
 
 8377 
 8471 
 
 8563 
 
 2 3 5 
 2 3 5 
 2 3 5 
 
 6 8 
 6 8 
 6 8 
 
 59 
 60 
 61 
 
 8572 
 8660 
 8746 
 
 8581 
 8669 
 8755 
 
 8590 
 8678 
 8763 
 
 8599 
 8686 
 8771 
 
 8607 
 8695 
 8780 
 
 8616 
 8704 
 8788 
 
 8625 
 8712 
 8796 
 
 8634 
 8721 
 8805 
 
 8643 
 8729 
 8813 
 
 8652 
 8738 
 8821 
 
 i 3 4 
 i 3 4 
 i 3 4 
 
 6 7 
 2 ? 
 
 62 
 63 
 64 
 
 8829 
 8910 
 8988 
 
 8838 
 8918 
 8996 
 
 8846 
 8926 
 9003 
 
 8854 
 8934 
 9011 
 
 8862 
 8942 
 9018 
 
 8870 
 8949 
 9026 
 
 8878 
 8957 
 9033 
 
 8886 
 8965 
 9041 
 
 8894 
 
 8973 
 9048 
 
 8902 
 8980 
 9056 
 
 i 3 4 
 i 3 4 
 i 3 4 
 
 1 I 
 
 5 6 
 
 65 
 
 9063 
 
 9070 
 
 9078 
 
 9085 
 
 9092 
 
 9100 
 
 9107 
 
 9114 
 
 9121 
 
 9128 
 
 I 2 4 
 
 5 6 
 
 66 
 67 
 68 
 
 9135 
 9205 
 9272 
 
 9M3 
 9212 
 9278 
 
 915 
 9219 
 9285 
 
 9157 
 9225 
 9291 
 
 9164 
 9232 
 9298 
 
 9171 
 9239 
 934 
 
 9178 
 9245 
 93" 
 
 9184 
 9252 
 9317 
 
 9191 
 9259 
 9323 
 
 9198 
 9265 
 9330 
 
 I 2 3 
 I 2 3 
 I 2 3 
 
 5 6 
 4 6 
 
 4 5 
 
 69 
 70 
 71 
 
 9336 
 9397 
 9455 
 
 9342 
 9403 
 9461 
 
 9348 
 9409 
 9466 
 
 9354 
 94i5 
 9472 
 
 936i 
 9421 
 9478 
 
 9367 
 9426 
 
 9483 
 
 9373 
 9432 
 9489 
 
 9379 
 9438 
 9494 
 
 9385 
 9444 
 9500 
 
 939i 
 9449 
 955 
 
 2 3 
 2 3 
 2 3 
 
 4 5 
 4 5 
 4 5 
 
 72 
 73 
 
 74 
 
 95 11 
 95 6 3 
 9613 
 
 95 l6 
 9568 
 9617 
 
 952i 
 9573 
 9622 
 
 95 2 7 
 9578 
 9627 
 
 9532 
 9583 
 9632 
 
 9537 
 9588 
 9636 
 
 9542 
 9593 
 9641 
 
 9548 
 9598 
 9646 
 
 9553 
 9603 
 9650 
 
 9558 
 9608 
 
 9655 
 
 2 3 
 
 2 2 
 2 2 
 
 4 4 
 3 4 
 3 4 
 
 75 
 
 9659 
 
 9664 
 
 9668 
 
 9673 
 
 9677 
 
 9681 
 
 9686 
 
 9690 
 
 9694 
 
 9699 
 
 I 2 
 
 3 4 
 
 76 
 
 77 
 78 
 
 9703 
 9744 
 9781 
 
 9707 
 9748 
 9785 
 
 9711 
 975i 
 9789 
 
 9715 
 
 9755 
 9792 
 
 9720 
 
 9759 
 9796 
 
 9724 
 97 6 3 
 9799 
 
 9728 
 9767 
 9803 
 
 9732 
 
 977 
 9806 
 
 9736 
 9774 
 9810 
 
 9740 
 9778 
 9813 
 
 2 
 2 
 2 
 
 3 3 
 3 3 
 2 3 
 
 79 
 80 
 81 
 
 9816 
 9848 
 9877 
 
 9820 
 
 9851 
 9880 
 
 9823 
 9854 
 9882 
 
 9826 
 
 9857 
 9885 
 
 9829 
 9860 
 9888 
 
 9833 
 9863 
 9890 
 
 9836 
 9866 
 9893 
 
 9839 
 9869 
 
 9895 
 
 9842 
 9871 
 9898 
 
 9845 
 9874 
 9900 
 
 I 2 
 
 O 
 O 
 
 2 3 
 
 2 2 
 2 2 
 
 82 
 83 
 84 
 
 9903 
 9925 
 9945 
 
 9905 
 9928 
 
 9947 
 
 9907 
 9930 
 9949 
 
 9910 
 9932 
 995i 
 
 9912 
 9934 
 995 2 
 
 9914 
 9936 
 9954 
 
 9917 
 9938 
 995 6 
 
 9919 
 9940 
 9957 
 
 992i 
 9942 
 9959 
 
 9923 
 9943 
 9960 
 
 O 
 O 
 
 
 2 2 
 I 2 
 I I 
 
 85 
 
 9962 
 
 9963 
 
 9965 
 
 9966 
 
 9968 
 
 9969 
 
 9971 
 
 9972 
 
 9973 
 
 9974 
 
 001 
 
 I I 
 
 86 
 87 
 88 
 
 9976 
 9986 
 9994 
 
 9977 
 9987 
 9995 
 
 9978 
 9988 
 
 9995 
 
 9979 
 9989 
 9996 
 
 9980 
 9990 
 9996 
 
 998i 
 9990 
 
 9997 
 
 9982 
 9991 
 9997 
 
 9983 
 9992 
 9997 
 
 9984 
 9993 
 9998 
 
 9985 
 9993 
 9998 
 
 I 
 O O O 
 O O O 
 
 I I 
 I I 
 
 O O 
 
 89 
 
 9998 
 
 9999 
 
 9999 
 
 9999 
 
 9999 
 
 I'OOO 
 
 nearly. 
 
 rooo 
 
 nearly. 
 
 rooo 
 
 nearly. 
 
 I'OOO 
 nearly. 
 
 I'OOO 
 
 nearly. 
 
 O O O 
 
 O O 
 
236 
 
 THE THEORY OF MEASUREMENTS 
 
 * TABLE XX. NATURAL COSINES. 
 
 
 O' 
 
 & 
 
 12' 
 
 18' 
 
 24' 
 
 3O' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 
 
 I '000 
 
 I'OOO 
 
 nearly. 
 
 rooo 
 
 nearly. 
 
 rooo 
 
 nearly. 
 
 rooo 
 
 nearly. 
 
 9999 
 
 9999 
 
 9999 
 
 9999 
 
 9999 
 
 o o o 
 
 
 
 1 
 
 2 
 3 
 
 9998 
 
 9994 
 9986 
 
 9998 
 
 999 8 
 
 9993 
 9984 
 
 9997 
 9992 
 9983 
 
 9997 
 9991 
 9982 
 
 9997 
 9990 
 9981 
 
 9996 
 9990 
 9980 
 
 9996 
 9989 
 9979 
 
 9995 
 9988 
 
 9978 
 
 9995 
 9987 
 9977 
 
 000 
 
 o o o 
 
 O O I 
 
 
 
 I I 
 I I 
 
 4 
 5 
 6 
 
 9976 
 9962 
 9945 
 
 9974 
 9960 
 
 9943 
 
 9973 
 9959 
 9942 
 
 9972 
 9957 
 9940 
 
 9971 
 
 995 6 
 9938 
 
 9969 
 9954 
 9936 
 
 9968 
 9952 
 9934 
 
 9966 
 
 9951 
 9932 
 
 9965 
 9949 
 9930 
 
 9963 
 9947 
 9928 
 
 o o 
 
 I 
 O I 
 
 I I 
 
 I 2 
 I 2 
 
 7 
 8 
 9 
 
 9925 
 9903 
 9877 
 
 9923 
 9900 
 
 9874 
 
 9921 
 9898 
 9871 
 
 9919 
 
 9895 
 9869 
 
 9917 
 
 9893 
 9866 
 
 9914 
 9890 
 9863 
 
 9912 
 9888 
 9860 
 
 9910 
 9885 
 9857 
 
 9907 
 9882 
 9854 
 
 9905 
 9880 
 
 9851 
 
 I 
 O I 
 I I 
 
 2 2 
 2 2 
 2 2 
 
 10 
 
 9848 
 
 9845 
 
 9842 
 
 9839 
 
 9836 
 
 9833 
 
 9829 
 
 9826 
 
 9823 
 
 9820 
 
 112 
 
 2 3 
 
 11 
 12 
 13 
 
 9816 
 9781 
 9744 
 
 9813 
 9778 
 9740 
 
 9810 
 9774 
 9736 
 
 9806 
 
 977 
 9732 
 
 9803 
 9767 
 9728 
 
 9799 
 9763 
 9724 
 
 9796 
 
 9759 
 9720 
 
 9792 
 9755 
 9715 
 
 9789 
 
 9751 
 9711 
 
 9785 
 9748 
 9707 
 
 112 
 I I 2 
 I I 2 
 
 2 3 
 
 3 3 
 3 3 
 
 14 
 15 
 16 
 
 973 
 9659 
 9613 
 
 9699 
 
 9655 
 9608 
 
 9694 
 9650 
 9603 
 
 9690 
 9646 
 9598 
 
 9686 
 9641 
 9593 
 
 9681 
 9636 
 9588 
 
 9677 
 9632 
 9583 
 
 9673 
 9627 
 
 9578 
 
 9668 
 9622 
 9573 
 
 9664 
 9617 
 9568 
 
 I I 2 
 122 
 122 
 
 3 4 
 3 4 
 3 4 
 
 17 
 18 
 19 
 
 95 6 3 
 95 11 
 
 9455 
 
 9558 
 955 
 9449 
 
 9553 
 9500 
 
 9444 
 
 9548 
 9494 
 9438 
 
 9542 
 9489 
 9432 
 
 9537 
 9483 
 9426 
 
 9532 
 9478 
 9421 
 
 9527 
 9472 
 
 94i5 
 
 95 21 
 9466 
 9409 
 
 95i6 
 9461 
 9403 
 
 I 2 3 
 
 i 2 3 
 
 I 2 3 
 
 4 4 
 4 5 
 4 5 
 
 20 
 
 9397 
 
 939i 
 
 9385 
 
 9379 
 
 9373 
 
 9367 
 
 9361 
 
 9354 
 
 9348 
 
 9342 
 
 I 2 3 
 
 4 5 
 
 21 
 22 
 23 
 
 9336 
 9272 
 9205 
 
 9330 
 9265 
 9198 
 
 9323 
 9259 
 9191 
 
 9317 
 9252 
 9184 
 
 93" 
 9245 
 9178 
 
 934 
 9239 
 9171 
 
 9298 
 9232 
 9164 
 
 9291 
 9225 
 9157 
 
 9285 
 9219 
 915 
 
 9278 
 9212 
 9H3 
 
 I 2 3 
 I 2 3 
 I 2 3 
 
 4 5 
 4 6 
 5 6 
 
 24 
 25 
 26 
 
 9135 
 9063 
 8988 
 
 9128 
 9056 
 8980 
 
 9121 
 9048 
 8973 
 
 9114 
 9041 
 8965 
 
 9107 
 9033 
 8957 
 
 9100 
 9026 
 8949 
 
 9092 
 9018 
 8942 
 
 9085 
 9011 
 8934 
 
 9078 
 9003 
 8926 
 
 9070 
 8996 
 8918 
 
 I 2 4 
 
 i 3 4 
 i 3 4 
 
 5 6 
 5 6 
 5 6 
 
 27 
 28 
 29 
 
 8910 
 8829 
 8746 
 
 8902 
 8821 
 8738 
 
 8894 
 8813 
 8729 
 
 8886 
 8805 
 8721 
 
 8878 
 8796 
 8712 
 
 8870 
 8788 
 8704 
 
 8862 
 8780 
 8695 
 
 8854 
 8771 
 8686 
 
 8846 
 
 8763 
 8678 
 
 8838 
 
 8755 
 8669 
 
 i 3 4 
 i 3 4 
 i 3 4 
 
 5 7 
 6 7 
 6 7 
 
 30 
 
 8660 
 
 8652 
 
 8643 
 
 8634 
 
 8625 
 
 8616 
 
 8607 
 
 8599 
 
 8590 
 
 8581 
 
 1 3 4 
 
 6 7 
 
 31 
 32 
 33 
 
 8572 
 8480 
 
 8387 
 
 8563 
 8471 
 8377 
 
 8462 
 8368 
 
 8545 
 8453 
 8358 
 
 8536 
 8443 
 8348 
 
 8526 
 8434 
 8339 
 
 8517 
 8425 
 8329 
 
 8508 
 
 8415 
 8320 
 
 8499 
 8406 
 8310 
 
 8490 
 8396 
 8300 
 
 2 3 5 
 2 3 5 
 235 
 
 6 8 
 
 6 8 
 6 8 
 
 34 
 35 
 36 
 
 8290 
 8192 
 8090 
 
 8281 
 8181 
 8080 
 
 8271 
 8171 
 8070 
 
 8261 
 8161 
 8059 
 
 8251 
 8151 
 8049 
 
 8241 
 8141 
 8039 
 
 8231 
 8131 
 8028 
 
 8221 
 8121 
 8018 
 
 8211 
 8111 
 8007 
 
 8202 
 8100 
 7997 
 
 2 3 5 
 2 3 5 
 235 
 
 7 8 
 7 8 
 7 9 
 
 37 
 38 
 39 
 
 7986 
 7880 
 7771 
 
 7976 
 7869 
 7760 
 
 7965 
 7859 
 7749 
 
 7955 
 7848 
 7738 
 
 7944 
 7837 
 7727 
 
 7934 
 7826 
 7716 
 
 7923 
 78i5 
 775 
 
 7912 
 7804 
 7694 
 
 7902 
 
 7793 
 7683 
 
 7891 
 7782 
 7672 
 
 245 
 245 
 246 
 
 7 9 
 7 9 
 7 9 
 
 40 
 
 7660 
 
 7649 
 
 7638 
 
 7627 
 
 7 6l 5 
 
 7604 
 
 7593 
 
 758i 
 
 757 
 
 7559 
 
 2 4 6 
 
 8 9 
 
 41 
 42 
 43 
 
 7547 
 7431 
 73H 
 
 7536 
 7420 
 7302 
 
 7524 
 7408 
 7290 
 
 7513 
 7396 
 7278 
 
 75 01 
 
 73 fl 
 7266 
 
 7490 
 7373 
 7254 
 
 7478 
 736i 
 7242 
 
 7466 
 7349 
 7230 
 
 7455 
 7337 
 7218 
 
 7443 
 7325 
 7206 
 
 246 
 246 
 2 4 6 
 
 8 10 
 8 10 
 8 10 
 
 44 
 
 7'93 
 
 7181 
 
 7169 
 
 7157 
 
 7H5 
 
 7133 
 
 7120 
 
 7108 
 
 7096 
 
 7083 
 
 2 4 6 
 
 8 10 
 
 N.B. Numbers in difference-columns to be subtracted, not added. 
 * From Bottomley'g Four Figure Mathematical Tables, by courtesy of The Macmillan Company. 
 
TABLES 
 
 237 
 
 TABLE XX. NATURAL COSINES (Concluded). 
 
 
 O' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 30' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 45 
 
 7071 
 
 759 
 
 7046 
 
 734 
 
 7022 
 
 7009 
 
 6997 
 
 6984 
 
 6972 
 
 6959 
 
 246 
 
 8 10 
 
 46 
 
 47 
 48 
 
 6947 
 6820 
 6691 
 
 6934 
 6807 
 6678 
 
 6921 
 6794 
 6665 
 
 6909 
 6782 
 6652 
 
 6896 
 
 6769 
 6639 
 
 6884 
 6756 
 6626 
 
 6871 
 
 6 743 
 6613 
 
 6858 
 6730 
 6600 
 
 6845 
 6717 
 6587 
 
 6833 
 6704 
 
 6574 
 
 246 
 2 4 6 
 
 247 
 
 8 ii 
 9 u 
 9 ii 
 
 49 
 50 
 51 
 
 6561 
 6428 
 6293 
 
 6 547 
 6414 
 6280 
 
 6534 
 6401 
 6266 
 
 6521 
 6388 
 6252 
 
 6508 
 
 6374 
 6239 
 
 6494 
 6361 
 6225 
 
 6481 
 
 6347 
 6211 
 
 6468 
 
 6334 
 6198 
 
 6 455 
 6320 
 6184 
 
 6441 
 6307 
 6170 
 
 247 
 247 
 2 5 7 
 
 9 ii 
 9 ii 
 9 ii 
 
 52 
 53 
 54 
 
 6l 57 
 6018 
 5878 
 
 6i43 
 6004 
 5864 
 
 6129 
 5990 
 5850 
 
 6115 
 5976 
 5835 
 
 6101 
 
 5962 
 5821 
 
 6088 
 5948 
 58-07 
 
 6074 
 5934 
 5793 
 
 6060 
 5920 
 5779 
 
 6046 
 5906 
 57 6 4 
 
 6032 
 5892 
 5750 
 
 2 5 7 
 257 
 257 
 
 9 12 
 9 12 
 9 12 
 
 55 
 
 5736 
 
 572i 
 
 5707 
 
 5693 
 
 5678 
 
 5664 
 
 5650 
 
 5635 
 
 5621 
 
 5606 
 
 2 5 7 
 
 10 12 
 
 56 
 57 
 58 
 
 5592 
 5446 
 5299 
 
 5577 
 5432 
 5284 
 
 55 6 3 
 54i7 
 5270 
 
 5548 
 5402 
 
 5255 
 
 5534 
 5388 
 5240 
 
 55i9 
 5373 
 5225 
 
 5505 
 5358 
 5210 
 
 5490 
 5344 
 5 J 95 
 
 5476 
 5329 
 5180 
 
 546i 
 53H 
 5 l6 5 
 
 2 5 7 
 2 5 7 
 257 
 
 10 12 
 10 12 
 10 12 
 
 59 
 60 
 61 
 
 5 Z 5 
 5000 
 4848 
 
 5i35 
 4985 
 4833 
 
 5120 
 4970 
 4818 
 
 5105 
 
 4955 
 4802 
 
 5090 
 4939 
 4787 
 
 575 
 4924 
 4772 
 
 5060 
 4909 
 475 6 
 
 5045 
 4894 
 474i 
 
 5030 
 4879 
 4726 
 
 5i5 
 4863 
 4710 
 
 3 5 8 
 3 5 8 
 3 5 8 
 
 10 13 
 10 13 
 10 13 
 
 62 
 63 
 64 
 
 4695 
 4540 
 4384 
 
 4679 
 
 4524 
 4368 
 
 4664 
 459 
 4352 
 
 4648 
 4493 
 4337 
 
 4633 
 4478 
 
 4321 
 
 4617 
 4462 
 4305 
 
 4602 
 4446 
 4289 
 
 4586 
 443i 
 4274 
 
 457 1 
 4415 
 4258 
 
 4555 
 4399 
 4242 
 
 3 5 8 
 3 5 8 
 3 5 8 
 
 10 13 
 10 13 
 
 II 13 
 
 65 
 
 4226 
 
 4210 
 
 4195 
 
 4179 
 
 4163 
 
 4 J 47 
 
 4131 
 
 4"5 
 
 4099 
 
 4083 
 
 3 5 8 
 
 II 13 
 
 66 
 67 
 68 
 
 4067 
 3907 
 3746 
 
 405 I 
 3891 
 3730 
 
 4035 
 3875 
 37H 
 
 4019 
 3859 
 3697 
 
 4003 
 
 3843 
 3681 
 
 3987 
 3827 
 3665 
 
 397 1 
 3811 
 
 3 6 49 
 
 3955 
 3795 
 3633 
 
 3939 
 3778 
 3616 
 
 3923 
 3762 
 3600 
 
 3 5 8 
 3 5 8 
 3 5 8 
 
 II 14 
 II 14 
 II 14 
 
 69 
 70 
 
 71 
 
 3584 
 3420 
 3256 
 
 3567 
 3404 
 3239 
 
 355i 
 3387 
 3223 
 
 3535 
 337i 
 3206 
 
 35i8 
 
 3355 
 3190 
 
 3502 
 3338 
 3173 
 
 3486 
 3322 
 3156 
 
 3469 
 3305 
 3140 
 
 3453 
 3289 
 3123 
 
 3437 
 3272 
 3 J 07 
 
 3 5 8 
 3 5 8 
 3 6 8 
 
 II 14 
 II 14 
 II 14 
 
 72 
 73 
 
 74 
 
 3090 
 2924 
 2756 
 
 374 
 2907 
 2740 
 
 3057 
 2890 
 2723 
 
 3040 
 2874 
 2706 
 
 3024 
 
 2857 
 2689 
 
 3007 
 2840 
 2672 
 
 2990 
 2823 
 2656 
 
 2974 
 2807 
 2639 
 
 2957 
 2790 
 2622 
 
 2940 
 
 2773 
 2605 
 
 368 
 368 
 368 
 
 II 14 
 
 II 14 
 
 II 14 
 
 75 
 
 2588 
 
 257i 
 
 2554 
 
 2538 
 
 2521 
 
 2504 
 
 2487 
 
 2470 
 
 2453 
 
 2436 
 
 368 
 
 II 14 
 
 76 
 
 77 
 78 
 
 2419 
 2250 
 2079 
 
 2402 
 2233 
 2062 
 
 2385 
 2215 
 2045 
 
 2368 
 2198 
 2028 
 
 2351 
 2181 
 
 2OII 
 
 2334 
 2164 
 1994 
 
 2317 
 2147 
 1977 
 
 2300 
 2130 
 1959 
 
 2284 
 2113 
 1942 
 
 2267 
 2096 
 1925 
 
 368 
 369 
 3 6 9 
 
 II 14 
 II 14 
 
 II 14 
 
 79 
 80 
 81 
 
 1908 
 1736 
 i5 6 4 
 
 1891 
 1719 
 
 '547 
 
 1874 
 1702 
 1530 
 
 1857 
 1685 
 1513 
 
 1840 
 1668 
 1495 
 
 1822 
 1650 
 1478 
 
 1805 
 
 1633 
 1461 
 
 1788 
 1616 
 
 1444 
 
 1771 
 
 1599 
 1426 
 
 *754 
 1582 
 1409 
 
 3 6 9 
 3 6 9 
 369 
 
 12 14 
 12 14 
 12 14 
 
 82 
 83 
 84 
 
 1392 
 1219 
 1045 
 
 1374 
 
 I2OI 
 1028 
 
 1357 
 1184 
 
 IOII 
 
 1340 
 1167 
 
 0993 
 
 1323 
 1149 
 0976 
 
 1305 
 1132 
 0958 
 
 1288 
 
 i"5 
 
 0941 
 
 1271 
 1097 
 0924 
 
 1253 
 1080 
 0906 
 
 1236 
 1063 
 0889 
 
 369 
 369 
 369 
 
 12 I 4 
 12 I 4 
 12 14 
 
 85 
 
 0872 
 
 0854 
 
 0837 
 
 0819 
 
 0802 
 
 0785 
 
 0767 
 
 0750 
 
 0732 
 
 0715 
 
 3 6 9 
 
 12 I 5 
 
 86 
 87 
 88 
 
 0698 
 0523 
 0349 
 
 0680 
 0506 
 0332 
 
 o663 
 0488 
 03H 
 
 0645 
 0471 
 0297 
 
 0628 
 
 0454 
 0279 
 
 0610 
 0436 
 0262 
 
 0593 
 0419 
 0244 
 
 0576 
 0401 
 0227 
 
 0558 
 0384 
 0209 
 
 0541 
 0366 
 0192 
 
 369 
 369 
 369 
 
 12 15 
 12 15 
 12 15 
 
 89 
 
 oi75 
 
 0157 
 
 0140 
 
 0122 
 
 0105 
 
 0087 
 
 0070 
 
 0052 
 
 0035 
 
 0017 
 
 369 
 
 12 15 
 
 iV.B. Numbers in difference-columns to be subtracted, not added. 
 
238 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE XXI. NATURAL TANGENTS. 
 
 
 O' 
 
 & 
 
 12' 
 
 18' 
 
 24' 
 
 3O' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 
 
 oooo 
 
 0017 
 
 0035 
 
 0052 
 
 0070 
 
 0087 
 
 0105 
 
 OI22 
 
 0140 
 
 oi57 
 
 369 
 
 12 14 
 
 1 
 
 2 
 3 
 
 0175 
 
 0349 
 0524 
 
 0192 
 0367 
 0542 
 
 0209 
 0384 
 0559 
 
 0227 
 0402 
 577 
 
 0244 
 0419 
 0594 
 
 0262 
 
 0437 
 0612 
 
 0279 
 
 0454 
 0629 
 
 0297 
 0472 
 0647 
 
 0314 
 
 0489 
 0664 
 
 0332 
 0507 
 0682 
 
 369 
 369 
 369 
 
 12 I 5 
 12 15 
 12 I 5 
 
 4 
 5 
 6 
 
 0699 
 0875 
 
 1051 
 
 0717 
 0892 
 1069 
 
 0734 
 0910 
 1086 
 
 0752 
 0928 
 1104 
 
 0769 
 0945 
 
 1122 
 
 0787 
 0963 
 "39 
 
 0805 
 0981 
 "57 
 
 0822 
 99 8 
 
 "75 
 
 0840 
 1016 
 1192 
 
 0857 
 1033 
 
 I2IO 
 
 369 
 369 
 369 
 
 12 I 5 
 12 I 5 
 12 I 5 
 
 7 
 8 
 9 
 
 1228 
 
 1405 
 1584 
 
 1246 
 
 1423 
 1602 
 
 1263 
 1441 
 1620 
 
 1281 
 
 H59 
 1638 
 
 1299 
 
 H77 
 1655 
 
 1317 
 
 H95 
 1673 
 
 1334 
 1512 
 1691 
 
 1352 
 1530 
 1709 
 
 1370 
 1548 
 1727 
 
 1388 
 1566 
 1745 
 
 369 
 369 
 369 
 
 12 I 5 
 12 I 5 
 12 I 5 
 
 10 
 
 1763 
 
 1781 
 
 1799 
 
 1817 
 
 1835 
 
 1853 
 
 1871 
 
 1890 
 
 1908 
 
 1926 
 
 369 
 
 12 I 5 
 
 11 
 12 
 13 
 
 1944 
 
 2126 
 
 2309 
 
 1962 
 
 2144 
 2327 
 
 1980 
 2162 
 
 2345 
 
 1998 
 2180 
 2364 
 
 2016 
 2199 
 
 2382 
 
 2035 
 2217 
 2401 
 
 2053 
 
 2235 
 2419 
 
 2071 
 
 2254 
 2438 
 
 2089 
 2272 
 2456 
 
 2IO7 
 2290 
 2475 
 
 369 
 369 
 369 
 
 12 I 5 
 
 12 I 5 
 12 I 5 
 
 14 
 15 
 16 
 
 2493 
 2679 
 2867 
 
 2512 
 
 2698 
 2886 
 
 2 53 
 
 2717 
 2905 
 
 2549 
 2736 
 
 2924 
 
 2568 
 2754 
 2943 
 
 2586 
 
 2773 
 2962 
 
 2605 
 2792 
 2981 
 
 2623 
 2811 
 3000 
 
 2642 
 2830 
 3019 
 
 2661 
 2849 
 3038 
 
 369 
 369 
 369 
 
 12 l6 
 
 13 16 
 13 16 
 
 17 
 18 
 19 
 
 3057 
 3249 
 3443 
 
 3076 
 3269 
 3463 
 
 3096 
 3288 
 3482 
 
 3"5 
 
 3307 
 3502 
 
 3134 
 3327 
 3522 
 
 3i53 
 3346 
 354i 
 
 3172 
 3365 
 356i 
 
 3 J 9i 
 
 3385 
 358i 
 
 3211 
 
 3404 
 3600 
 
 3230 
 3424 
 3620 
 
 3 6 10 
 3 6 10 
 3 6 10 
 
 13 16 
 13 16 
 13 17 
 
 20 
 
 3640 
 
 3659 
 
 3679 
 
 3699 
 
 37 J 9 
 
 3739 
 
 3759 
 
 3779 
 
 3799 
 
 3819 
 
 3 7 I0 
 
 13 17 
 
 21 
 22 
 23 
 
 3839 
 4040 
 
 4245 
 
 3859 
 4061 
 4265 
 
 3879 
 4081 
 4286 
 
 3899 
 4101 
 
 4307 
 
 3919 
 4122 
 
 4327 
 
 3939 
 4142 
 
 4348 
 
 3959 
 4163 
 
 4369 
 
 3979 
 4183 
 4390 
 
 4000 
 4204 
 44" 
 
 4O2O 
 4224 
 4431 
 
 3 7 I0 
 3 7 I0 
 3 7 10 
 
 13 17 
 14 17 
 14 17 
 
 24 
 25 
 26 
 
 4452 
 4663 
 4877 
 
 4473 
 4684 
 
 4899 
 
 4494 
 4706 
 4921 
 
 45i5 
 
 4727 
 
 4942 
 
 4536 
 4748 
 4964 
 
 4557 
 477 
 4986 
 
 4578 
 479i 
 5008 
 
 4599 
 4813 
 5029 
 
 4621 
 4834 
 5051 
 
 4642 
 4856 
 
 573 
 
 4 7 10 
 4 7 ii 
 4 7 ii 
 
 14 18 
 14 18 
 15 18 
 
 27 
 28 
 29 
 
 5095 
 5317 
 
 '5543 
 
 5"7 
 5340 
 5566 
 
 5139 
 
 5362 
 5589 
 
 5161 
 
 5384 
 5612 
 
 5184 
 5407 
 5635 
 
 5206 
 5430 
 5658 
 
 5228 
 5452 
 5681 
 
 5250 
 5475 
 5704 
 
 5272 
 5498 
 5727 
 
 5295 
 5520 
 575 
 
 4 7 ii 
 4 8 ii 
 
 4 8 12 
 
 15 18 
 15 19 
 15 19 
 
 30 
 
 '5774 
 
 5797 
 
 5820 
 
 5844 
 
 5867 
 
 5890 
 
 59H 
 
 5938 
 
 596i 
 
 5985 
 
 4 8 12 
 
 16 20 
 
 31 
 32 
 33 
 
 6009 
 6249 
 6494 
 
 6032 
 6273 
 6519 
 
 6056 
 6297 
 6544 
 
 6080 
 6322 
 6569 
 
 6104 
 6346 
 6594 
 
 6128 
 
 6371 
 6619 
 
 6152 
 
 6395 
 6644 
 
 6176 
 6420 
 6669 
 
 6200 
 
 6445 
 6694 
 
 6224 
 6469 
 6720 
 
 4 8 12 
 
 4 8 12 
 4 8 13 
 
 16 20 
 16 20 
 
 17 21 
 
 34 
 35 
 36 
 
 ' 6 745 
 7002 
 7265 
 
 6771 
 7028 
 7292 
 
 6796 
 7054 
 7319 
 
 6822 
 7080 
 7346 
 
 6847 
 7107 
 
 7373 
 
 6873 
 7 ! 33 
 7400 
 
 6899 
 7*59 
 7427 
 
 6924 
 7186 
 7454 
 
 6950 
 7212 
 748i 
 
 6976 
 
 7239 
 7508 
 
 4 9 13 
 4 9 13 
 5 9 H 
 
 17 21 
 18 22 
 
 18 23 
 
 37 
 38 
 39 
 
 7536 
 7813 
 8098 
 
 7563 
 7841 
 8127 
 
 7590 
 8156 
 
 7618 
 7898 
 8185 
 
 7646 
 7926 
 8214 
 
 7673 
 7954 
 8243 
 
 7701 
 7983 
 8273 
 
 7729 
 8012 
 8302 
 
 7757 
 8040 
 
 8332 
 
 7785 
 8069 
 8361 
 
 5 9 H 
 5 I0 M 
 5 10 15 
 
 18 23 
 19 24 
 20 24 
 
 40 
 
 8391 
 
 8421 
 
 8451 
 
 8481 
 
 8511 
 
 8541 
 
 857i 
 
 8601 
 
 8632 
 
 8662 
 
 5 1 '5 
 
 20 25 
 
 41 
 42 
 43 
 
 8693 
 9004 
 9325 
 
 8724 
 9036 
 9358 
 
 8754 
 9067 
 
 9391 
 
 8785 
 9099 
 9424 
 
 8816 
 9131 
 
 9457 
 
 8847 
 9163 
 9490 
 
 8878 
 9195 
 9523 
 
 8910 
 9228 
 9556 
 
 8941 
 9260 
 9590 
 
 8972 
 
 9293 
 9623 
 
 5 10 16 
 5 " 16 
 6 ii 17 
 
 21 26 
 
 21 27 
 
 22 28 
 
 44 
 
 9657 
 
 9691 
 
 97 2 5 
 
 9759 
 
 9793 
 
 9827 
 
 9861 
 
 9896 
 
 9930 
 
 9965 
 
 6 ii 17 
 
 23 29 
 
 * From Bottomley's Four Figure Mathematical Tables, by courtesy of The Macmillan Company. 
 
TABLES 
 
 239 
 
 TABLE XXI. NATURAL TANGENTS (Concluded). 
 
 
 0' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 3O' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 45 
 
 I -0000 
 
 0035 
 
 0070 
 
 0105 
 
 0141 
 
 0176 
 
 O2I2 
 
 0247 
 
 0283 
 
 0319 
 
 6 12 18 
 
 24 30 
 
 46 
 47 
 48 
 
 1-0355 
 1-0724 
 1-1106 
 
 0392 
 0761 
 
 "45 
 
 0428 
 
 0799 
 1184 
 
 0464 
 0837 
 1224 
 
 0501 
 
 0875 
 1263 
 
 0538 
 0913 
 
 1303 
 
 0575 
 0951 
 1343 
 
 0612 
 0990 
 1383 
 
 0649 
 1028 
 1423 
 
 0686 
 1067 
 1463 
 
 6 12 18 
 6 13 19 
 7 13 20 
 
 25 3i 
 25 32 
 26 33 
 
 49 
 50 
 51 
 
 1504 
 1918 
 
 2349 
 
 1544 
 1960 
 
 2393 
 
 1585 
 
 2OO2 
 
 2437 
 
 1626 
 
 2045 
 2482 
 
 1667 
 2088 
 2527 
 
 1708 
 2131 
 
 2572 
 
 1750 
 2174 
 2617 
 
 1792 
 2218 
 2662 
 
 1833 
 2261 
 2708 
 
 1875 
 2305 
 
 2753 
 
 7 H 21 
 
 7 14 22 
 
 8 15 23 
 
 28 34 
 29 36 
 30 38 
 
 52 
 53 
 54 
 
 2799 
 3270 
 3764 
 
 2846 
 3319 
 3814 
 
 2892 
 3367 
 3865 
 
 2938 
 34i6 
 3916 
 
 2985 
 3465 
 3968 
 
 3032 
 35H 
 4019 
 
 3079 
 3564 
 4071 
 
 3127 
 
 3613 
 4124 
 
 3i75 
 3663 
 4176 
 
 3222 
 
 37i3 
 4229 
 
 8 16 23 
 8 16 25 
 9 17 26 
 
 3i 39 
 33 4i 
 
 34 43 
 
 55 
 
 4281 
 
 4335 
 
 4388 
 
 444 2 
 
 4496 
 
 4550 
 
 4605 
 
 4659 
 
 4715 
 
 4770 
 
 9 18 27 
 
 36 45 
 
 56 
 
 57 
 58 
 
 4826 
 5399 
 
 6003 
 
 4882 
 5458 
 6066 
 
 4938 
 5517 
 6128 
 
 4994 
 5577 
 6191 
 
 5051 
 5637 
 6255 
 
 5108 
 
 5697 
 6319 
 
 5166 
 
 mi 
 
 5224 
 5818 
 6447 
 
 5282 
 5880 
 6512 
 
 5340 
 594i 
 6577 
 
 10 19 29 
 10 20 30 
 
 II 21 32 
 
 38 48 
 40 50 
 43 53 
 
 59 
 60 
 61 
 
 6643 
 7321 
 
 8040 
 
 6709 
 
 739i 
 8115 
 
 6775 
 7461 
 8190 
 
 6842 
 7532 
 8265 
 
 6909 
 7603 
 8341 
 
 6977 
 7675 
 8418 
 
 7045 
 7747 
 8495 
 
 7113 
 
 7820 
 
 8572 
 
 7182 
 
 7893 
 8650 
 
 725 1 
 7966 
 8728 
 
 ii 23 34 
 12 24 36 
 13 26 38 
 
 45 5 6 
 48 60 
 
 5 1 6 4 
 
 62 
 63 
 64 
 
 1-8807 
 1-9626 
 2-0503 
 
 8887 
 9711 
 0594 
 
 8967 
 
 9797 
 0686 
 
 9047 
 
 9883 
 0778 
 
 9128 
 9970 
 0872 
 
 9210 
 0057 
 0965 
 
 9292 
 0145 
 1060 
 
 9375 
 0233 
 "55 
 
 9458 
 0323 
 1251 
 
 9542 
 
 041; 
 1348 
 
 14 27 41 
 15 29 44 
 16 31 47 
 
 55 68 
 
 58 73 
 63 78 
 
 65 
 
 2-1445 
 
 1543 
 
 1642 
 
 1742 
 
 1842 
 
 1943 
 
 2045 
 
 2148 
 
 2251 
 
 2355 
 
 1 7 34 5 1 
 
 68 85 
 
 66 
 67 
 68 
 
 2-2460 
 2-3559 
 2'475 i 
 
 2566 
 3673 
 4876 
 
 2673 
 
 3789 
 5002 
 
 2781 
 3906 
 5129 
 
 2889 
 4023 
 
 5257 
 
 2998 
 4142 
 5386 
 
 3109 
 4262 
 
 5517 
 
 3220 
 4383 
 5649 
 
 3332 
 454 
 5782 
 
 3445 
 4627 
 59i6 
 
 18 37 55 
 20 40 60 
 22 43 65 
 
 74 92 
 79 99 
 87 108 
 
 69 
 70 
 71 
 
 2-6051 
 27475 
 2-9042 
 
 6187 
 7625 
 9208 
 
 6325 
 7776 
 
 9375 
 
 6464 
 7929 
 9544 
 
 6605 
 8083 
 
 97H 
 
 6746 
 8239 
 
 9887 
 
 6889 
 8397 
 0061 
 
 734 
 8556 
 0237 
 
 7179 
 8716 
 0415 
 
 7326 
 8878 
 
 0595 
 
 24 47 7i 
 26 52 78 
 
 29 58 87 
 
 95 "8 
 104 130 
 
 "5 !44 
 
 72 
 73 
 74 
 
 3-0777 
 3-2709 
 
 3-4874 
 
 0961 
 2914 
 5 I0 5 
 
 1146 
 3122 
 5339 
 
 1334 
 3332 
 5576 
 
 1524 
 3544 
 5816 
 
 1716 
 
 3759 
 6059 
 
 1910 
 
 3977 
 6305 
 
 2106 
 4197 
 6554 
 
 '23 5 
 4420 
 6806 
 
 2506 
 4646 
 7062 
 
 32 64 96 
 36 72 108 
 
 41 82 122 
 
 129 161 
 144 180 
 162 203 
 
 75 
 
 3-732I 
 
 7583 
 
 7848 
 
 8118 
 
 8391 
 
 8667 
 
 8947 
 
 9232 
 
 9520 
 
 9812 
 
 46 94 139 
 
 i 86 232 
 
 76 
 
 77 
 78 
 
 4-0108 
 4-33I5 
 4-7046 
 
 0408 
 3662 
 
 7453 
 
 0713 
 4015 
 7867 
 
 IO22 
 
 4374 
 8288 
 
 1335 
 4737 
 8716 
 
 l6 53 
 5 I0 7 
 9152 
 
 1976 
 5483 
 9594 
 
 2303 
 5864 
 0045 
 
 2635 
 6252 
 
 0504 
 
 2972 
 6646 
 0970 
 
 53 107 i 60 
 62 124 186 
 73 146 219 
 
 214 267 
 248 310 
 
 292 365 
 
 79 
 80 
 81 
 
 5-I446 
 5-67I3 
 6-3138 
 
 1929 
 
 7297 
 3859 
 
 2422 
 7894 
 4596 
 
 2924 
 8502 
 5350 
 
 3435 
 9124 
 6122 
 
 3955 
 9758 
 6912 
 
 4486 
 0405 
 7920 
 
 5026 
 
 5578 
 
 6140 
 2432 
 0264 
 
 87 175 262 
 
 350 437 
 
 1066 
 8548 
 
 1742 
 9395 
 
 Difference-columns 
 cease to be useful, owing 
 to the rapidity with 
 which the value of the 
 tangent changes. 
 
 82 
 83 
 84 
 
 r"54 
 8-1443 
 9-5H4 
 
 2066 
 2636 
 9-677 
 
 3002 
 3863 
 9-845 
 
 3962 
 5126 
 
 IO-O2 
 
 4947 
 6427 
 
 10-20 
 
 5958 
 7769 
 10-39 
 
 6996 
 9152 
 10-58 
 
 8062 
 
 0579 
 10-78 
 
 9158 
 2052 
 10-99 
 
 0285 
 3572 
 
 11-20 
 
 85 
 
 n-43 
 
 11-66 
 
 11-91 
 
 12-16 
 
 12-43 
 
 12-71 
 
 13-00 
 
 13-30 
 
 13-62 
 
 I3-95 
 
 86 
 87 
 88 
 
 14-30 
 19-08 
 28-64 
 
 14-67 
 
 I9-74 
 30-14 
 
 15-06 
 20-45 
 31-82 
 
 I5-46 
 21-20 
 
 3J69 
 
 15-89 
 
 22-02 
 35-8o 
 
 16-35 
 22-90 
 38-19 
 
 16-83 
 23-86 
 40-92 
 
 I7-34 
 24-90 
 44-07 
 
 17-89 
 26-03 
 
 47-74 
 
 18-46 
 27-27 
 52-08 
 
 89 
 
 57'29 
 
 63-66 
 
 71-62 
 
 81-85 
 
 95-49 
 
 114-6 
 
 143-2 
 
 191-0 
 
 286-5 
 
 573-0 
 
240 
 
 THE THEORY OF MEASUREMENTS 
 
 * TABLE XXII. NATURAL COTANGENTS. 
 
 
 O' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 30' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 Difference-columns 
 not useful here, owing 
 to the rapidity with 
 which the value of the 
 cotangent changes. 
 
 
 
 Inf. 
 
 573-o 
 
 286-5 
 
 191-0 
 
 143-2 
 
 114-6 
 
 95'49 
 
 81-85 
 
 71-62 
 
 63-66 
 
 1 
 
 2 
 3 
 
 57-29 
 28-64 
 19-08 
 
 52-08 
 27-27 
 18-46 
 
 4774 
 26-03 
 
 17-89 
 
 44-07 
 24-90 
 17-34 
 
 40-92 
 23-86 
 16-83 
 
 38-19 
 22-90 
 
 i6'35 
 
 35-80 
 22-02 
 15-89 
 
 33-69 
 
 2 1 -2O 
 
 31-82 
 20-45 
 15-06 
 
 19-74 
 14-67 
 
 4 
 5 
 6 
 
 14-30 
 ii'43 
 9-5I44 
 
 I3-95 
 
 II'2O 
 3572 
 
 13-62 
 10-99 
 2052 
 
 13-3 
 10-78 
 
 0579 
 
 13-00 
 10-58 
 
 9152 
 
 12-71 
 10-39 
 
 7769 
 
 12-43 
 
 10-20 
 6427 
 
 I2'l6 
 10-02 
 5126 
 
 11-91 
 
 9-845 
 3863 
 
 u-66 
 9-677 
 2636 
 
 7 
 8 
 9 
 
 8-1443 
 7'"54 
 6-3138 
 
 0285 
 0264 
 2432 
 
 9158 
 
 9395 
 1742 
 
 8062 
 8548 
 1066 
 
 6996 
 7920 
 0405 
 
 5958 
 6912 
 97S8 
 
 4947 
 6122 
 9124 
 
 3962 
 
 5350 
 8502 
 
 3002 
 4596 
 7894 
 
 2066 
 3859 
 7297 
 
 10 
 
 5-67I3 
 
 6140 
 
 5578 
 
 5026 
 
 4486 
 
 3955 
 
 3435 
 
 2924 
 
 2422 
 
 1929 
 
 123 
 
 4 5 
 
 11 
 12 
 13 
 
 4-7046 
 4-33I5 
 
 0970 
 6646 
 2972 
 
 0504 
 6252 
 2635 
 
 0045 
 5864 
 2303 
 
 9594 
 5483 
 1976 
 
 9152 
 5107 
 1653 
 
 8716 
 4737 
 1335 
 
 8288 
 
 4374 
 
 1022 
 
 7867 
 4015 
 0713 
 
 7453 
 3662 
 0408 
 
 74 148 222 
 
 63 125 i 88 
 53 107 160 
 
 296 370 
 252 314 
 214 267 
 
 14 
 15 
 16 
 
 4-0108 
 J4874 
 
 9812 
 7062 
 4646 
 
 9520 
 6806 
 4420 
 
 9232 
 6554 
 4197 
 
 8947 
 6305 
 
 3977 
 
 8667 
 6059 
 3759 
 
 5816 
 3544 
 
 8118 
 5576 
 3332 
 
 7848 
 
 5339 
 3122 
 
 7583 
 5105 
 29H 
 
 46 93 139 
 
 41 82 122 
 
 36 72 108 
 
 i 86 232 
 163 204 
 144 180 
 
 17 
 18 
 19 
 
 3-2709 
 
 3-0777 
 2-9042 
 
 2506 
 
 595 
 8878 
 
 2305 
 
 0415 
 8716 
 
 2106 
 0237 
 8556 
 
 1910 
 0061 
 8397 
 
 1716 
 
 9887 
 8239 
 
 5 2 4 
 9714 
 8083 
 
 1334 
 9544 
 7929 
 
 1146 
 
 9375 
 7776 
 
 0961 
 9208 
 7625 
 
 32 64 96 
 
 29 58 87 
 26 52 78 
 
 129 161 
 
 "5 *44 
 104 130 
 
 2*7475 
 
 7326 
 
 7179 
 
 734 
 
 6889 
 
 6746 
 
 6605 
 
 6464 
 
 6325 
 
 6187 
 
 24 47 7i 
 
 95 "8 
 
 21 
 22 
 23 
 
 2-6051 
 2-475 * 
 2-3559 
 
 5916 
 4627 
 
 3445 
 
 5782 
 454 
 3332 
 
 5649 
 4383 
 3220 
 
 5517 
 4262 
 3109 
 
 5386 
 4142 
 2998 
 
 5257 
 4023 
 2889 
 
 3906 
 2781 
 
 5002 
 
 3789 
 2673 
 
 4876 
 3673 
 2566 
 
 22 43 65 
 20 40 60 
 18 37 55 
 
 87 108 
 
 79 99 
 74 92 
 
 24 
 25 
 26 
 
 ~27~ 
 28 
 29 
 
 2-2460 
 2-1445 
 2-0503 
 
 2355 
 1348 
 
 0413 
 
 2251 
 1251 
 
 0323 
 
 2148 
 "55 
 0233 
 
 2045 
 1060 
 
 0145 
 
 1943 
 0965 
 
 0057 
 
 1842 
 0872 
 9970 
 
 1742 
 0778 
 988 3 
 
 1642 
 0686 
 
 9797 
 
 1543 
 0594 
 97" 
 
 17 34 5i 
 16 31 47 
 15 29 44 
 
 68 85 
 63 78 
 58 73 
 
 1-9626 
 1-8807 
 1-8040 
 
 9542 
 8728 
 7966 
 
 9458 
 8650 
 
 7893 
 
 9375 
 8572 
 7820 
 
 9292 
 8495 
 7747 
 
 9210 
 
 8418 
 7675 
 
 9128 
 8341 
 7603 
 
 9047 
 8265 
 
 753 2 
 
 8967 
 8190 
 7461 
 
 8887 
 8115 
 739i 
 
 14 27 41 
 i3 26 38 
 12 24 36 
 
 55 68 
 5 1 64 
 48 60 
 
 30 
 
 1-7321 
 
 7251 
 
 7182 
 
 7"3 
 
 745 
 
 6977 
 
 6909 
 
 6842 
 
 6775 
 
 6709 
 
 ii 23 34 
 
 45 56 
 
 31 
 32 
 33 
 
 1-6643 
 1-6003 
 1-5399 
 
 6577 
 5340 
 
 6512 
 5880 
 5282 
 
 6447 
 5818 
 
 5224 
 
 6383 
 
 mi 
 
 6319 
 5697 
 5108 
 
 6255 
 5637 
 5051 
 
 6191 
 5577 
 4994 
 
 6128 
 
 5517 
 4938 
 
 6066 
 
 5458 
 4882 
 
 II 21 32 
 10 20 30 
 
 10 19 29 
 
 43 53 
 40 5 
 38 48 
 
 34 
 35 
 36 
 
 1-4826 
 1-4281 
 1-3764 
 
 4770 
 4229 
 3713 
 
 4715 
 4176 
 3663 
 
 4659 
 4124 
 
 3613 
 
 4605 
 4071 
 3564 
 
 4550 
 4019 
 35H 
 
 4496 
 3968 
 3465 
 
 4442 
 3916 
 
 4388 
 3865 
 3367 
 
 4335 
 3814 
 3319 
 
 9 18 27 
 9 17 26 
 8 16 25 
 
 36 45 
 34 43 
 33 4i 
 
 37 
 38 
 39 
 
 1-3270 
 1-2799 
 1-2349 
 
 3222 
 
 2753 
 2305 
 
 2708 
 2261 
 
 3^27 
 2662 
 2218 
 
 3079 
 2617 
 2174 
 
 3032 
 2572 
 2131 
 
 2985 
 2527 
 2088 
 
 2938 
 2482 
 2045 
 
 2892 
 2437 
 
 2OO2 
 
 2846 
 
 2393 
 1960 
 
 8 16 23 
 8 15 23 
 
 7 14 22 
 
 3 1 39 
 
 30 38 
 29 36 
 
 40 
 
 1-1918 
 
 1875 
 
 1833 
 
 1792 
 
 !75o 
 
 1708 
 
 1667 
 
 1626 
 
 1585 
 
 1544 
 
 7 *4 21 
 
 28 34 
 
 41 
 42 
 43 
 
 1-1504 
 1-1106 
 1-0724 
 
 1463 
 1067 
 0686 
 
 1423 
 1028 
 0649 
 
 1383 
 0990 
 0612 
 
 1343 
 0951 
 0575 
 
 1303 
 0913 
 0538 
 
 1263 
 0875 
 0501 
 
 1224 
 
 0837 
 0464 
 
 1184 
 
 0799 
 0428 
 
 "45 
 0761 
 0392 
 
 7 13 20 
 6 13 19 
 6 12 18 
 
 26 33 
 25 32 
 25 31 
 
 44 
 
 1-0355 
 
 0319 
 
 0283 
 
 0247 
 
 O2I2 
 
 0176 
 
 0141 
 
 0105 
 
 0070 
 
 0035 
 
 6 12 18 
 
 24 30 
 
 N.B. Numbers in difference-columns to be subtracted, not added. 
 * From Bottomley's Four Figure Mathematical Tables, by courtesy of The Macmillan Company. 
 
TABLES 
 
 241 
 
 TABLE XXII. NATURAL COTANGENTS (Concluded). 
 
 
 O' 
 
 6' 
 
 12 
 
 18' 
 
 24' 
 
 3O' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 45 
 
 ro 
 
 0-9965 
 
 0-9930 
 
 0-9896 
 
 0-9861 
 
 0-9827 
 
 0-9793 
 
 '9759 
 
 0-9725 
 
 0-9691 
 
 6 ii 17 
 
 23 29 
 
 46 
 47 
 48 
 
 9657 
 9325 
 9004 
 
 9623 
 
 9293 
 8972 
 
 9590 
 9260 
 
 8941 
 
 955 6 
 9228 
 8910 
 
 9523 
 9i95 
 8878 
 
 9490 
 9163 
 8847 
 
 9457 
 9131 
 8816 
 
 9424 
 9099 
 8785 
 
 939i 
 9067 
 
 8754 
 
 9358 
 9036 
 8724 
 
 6 ii 17 
 5 ii 10 
 5 10 16 
 
 22 28 
 21 27 
 21 26 
 
 49 
 50 
 51 
 
 8693 
 8391 
 8098 
 
 8662 
 8361 
 8069 
 
 8632 
 8332 
 8040 
 
 8601 
 8302 
 8012 
 
 8571 
 8273 
 7983 
 
 8541 
 8243 
 
 7954 
 
 8511 
 8214 
 7926 
 
 8481 
 8185 
 7898 
 
 8451 
 8156 
 7869 
 
 8421 
 8127 
 7841 
 
 5 10 i5 
 
 5 10 15 
 5 I0 M 
 
 20 25 
 20 24 
 19 24 
 
 52 
 53 
 54 
 
 7813 
 7536 
 7265 
 
 7785 
 7508 
 
 7239 
 
 7757 
 748i 
 7212 
 
 7729 
 
 $3 
 
 7701 
 
 7427 
 7i59 
 
 7 6 73 
 7400 
 
 7133 
 
 7646 
 
 7373 
 7107 
 
 7618 
 7346 
 7080 
 
 7590 
 73i9 
 754 
 
 75 6 3 
 7292 
 7028 
 
 5 9 H 
 
 5 9 H 
 4 9 13 
 
 18 23 
 18 23 
 
 18 22 
 
 55 
 
 7002 
 
 6976 
 
 6950 
 
 6924 
 
 6899 
 
 6873 
 
 6847 
 
 6822 
 
 6796 
 
 6771 
 
 4 9 13 
 
 I 7 21 
 
 56 
 57 
 58 
 
 >6 745 
 6494 
 6249 
 
 6720 
 6469 
 6224 
 
 6694 
 
 6445 
 6200 
 
 6669 
 6420 
 6176 
 
 6644 
 
 6395 
 6152 
 
 6619 
 
 637 1 
 6128 
 
 6594 
 6346 
 6104 
 
 6569 
 6322 
 6080 
 
 6544 
 6297 
 6056 
 
 6519 
 6273 
 6032 
 
 4 8 13 
 4 8 12 
 4 8 12 
 
 17 21 
 
 16 20 
 16 20 
 
 59 
 60 
 61 
 
 6009 
 '5774 
 '5543 
 
 5985 
 5750 
 5520 
 
 596i 
 5727 
 5498 
 
 5938 
 574 
 5475 
 
 59H 
 
 5681 
 
 5452 
 
 5890 
 5658 
 5430 
 
 5867 
 5^35 
 5407 
 
 5844 
 5612 
 
 5384 
 
 5820 
 5589 
 5362 
 
 5797 
 5566 
 5340 
 
 4 8 12 
 4 8 12 
 4 8 ii 
 
 16 20 
 
 15 !9 
 
 15 !9 
 
 62 
 63 
 64 
 
 5317 
 595 
 4877 
 
 5295 
 573 
 4856 
 
 5272 
 505 1 
 4834 
 
 5250 
 5029 
 
 4813 
 
 5228 
 5008 
 479i 
 
 5206 
 4986 
 4770 
 
 5184 
 4964 
 4748 
 
 5161 
 
 4942 
 4727 
 
 5*39 
 4921 
 4706 
 
 5"7 
 4899 
 4684 
 
 4 7 ii 
 
 4 7 ii 
 4 7 ii 
 
 15 18 
 15 18 
 14 18 
 
 65 
 
 4663 
 
 4642 
 
 4621 
 
 4599 
 
 4578 
 
 4557 
 
 4536 
 
 4515 
 
 4494 
 
 4473 
 
 4 7 10 
 
 14 18 
 
 66 
 67 
 68 
 
 "445 2 
 4245 
 4040 
 
 443i 
 4224 
 4020 
 
 4411 
 4204 
 4000 
 
 4390 
 4183 
 3979 
 
 4369 
 4163 
 3959 
 
 4348 
 4142 
 
 3939 
 
 4327 
 4122 
 
 3919 
 
 4307 
 4101 
 
 3899 
 
 4286 
 4081 
 3879 
 
 4265 
 4061 
 3859 
 
 371 
 3 7 10 
 3 7 I0 
 
 14 17 
 14 17 
 13 17 
 
 69 
 70 
 
 71 
 
 3839 
 3640 
 
 '3443 
 
 3819 
 3620 
 
 3424 
 
 3799 
 3600 
 
 3404 
 
 3779 
 358i 
 3385 
 
 3759 
 356i 
 3365 
 
 3739 
 354i 
 3346 
 
 3719 
 3522 
 3327 
 
 3699 
 3502 
 
 3307 
 
 3679 
 3482 
 3288 
 
 3659 
 3463 
 3269 
 
 3 7 10 
 3 6 10 
 3 6 10 
 
 13 17 
 13 17 
 13 16 
 
 72 
 73 
 74 
 
 3249 
 3057 
 2867 
 
 3230 
 3038 
 2849 
 
 3211 
 3019 
 2830 
 
 3i9i 
 3000 
 2811 
 
 3172 
 2981 
 2792 
 
 3153 
 
 2962 
 
 2773 
 
 3134 
 2943 
 2754 
 
 3"5 
 2924 
 2736 
 
 3096 
 2905 
 2717 
 
 2698 
 
 3 6 10 
 
 369 
 369 
 
 13 16 
 13 16 
 13 16 
 
 75 
 
 2679 
 
 2661 
 
 2642 
 
 2623 
 
 2605 
 
 2586 
 
 2568 
 
 2549 
 
 2530 
 
 2512 
 
 369 
 
 12 16 
 
 76 
 77 
 78 
 
 2493 
 2309 
 2126 
 
 2475 
 2290 
 2107 
 
 2456 
 2272 
 2089 
 
 2438 
 
 2254 
 2071 
 
 2419 
 2235 
 2053 
 
 2401 
 2217 
 2035 
 
 2382 
 
 2199 
 2016 
 
 2364 
 2180 
 1998 
 
 2345 
 2162 
 1980 
 
 2327 
 2144 
 1962 
 
 369 
 3 6 9 
 369 
 
 12 15 
 12 15 
 
 12 I 5 
 
 79 
 80 
 81 
 
 1944 
 1763 
 1584 
 
 1926 
 
 '745 
 1566 
 
 1908 
 1727 
 1548 
 
 1890 
 1709 
 1530 
 
 1871 
 1691 
 1512 
 
 1853 
 1673 
 H95 
 
 1835 
 l6 55 
 H77 
 
 1817 
 1638 
 H59 
 
 1799 
 1620 
 1441 
 
 1781 
 1602 
 1423 
 
 369 
 369 
 369 
 
 12 I 5 
 12 I 5 
 12 I 5 
 
 82 
 83 
 84 
 
 1405 
 1228 
 1051 
 
 1388 
 
 I2IO 
 1033 
 
 1370 
 1192 
 1016 
 
 1352 
 
 "75 
 0998 
 
 1334 
 "57 
 0981 
 
 1317 
 "39 
 0963 
 
 1299 
 
 1122 
 0945 
 
 1281 
 1104 
 0928 
 
 1263 
 1086 
 0910 
 
 1246 
 1069 
 0892 
 
 369 
 369 
 3 6 9 
 
 12 15 
 12 15 
 12 15 
 
 85 
 
 0875 
 
 08 57 
 
 0840 
 
 0822 
 
 0805 
 
 0787 
 
 0769 
 
 0752 
 
 0734 
 
 0717 
 
 3 6 9 
 
 12 I 5 
 
 86 
 87 
 88 
 
 0699 
 0524 
 0349 
 
 0682 
 0507 
 0332 
 
 0664 
 0489 
 3i4 
 
 0647 
 0472 
 0297 
 
 0629 
 
 0454 
 0279 
 
 0612 
 
 0437 
 0262 
 
 0594 
 0419 
 0244 
 
 0577 
 0402 
 0227 
 
 0559 
 0384 
 0209 
 
 0542 
 0367 
 0192 
 
 369 
 
 3 6 9 
 369 
 
 12 15 
 12 15 
 12 I 5 
 
 89 
 
 oi75 
 
 0157 
 
 0140 
 
 0122 
 
 0105 
 
 0087 
 
 OC>7O 
 
 0052 
 
 0035 
 
 0017 
 
 369 
 
 12 14 
 
 N.B. Numbers in difference-columns to be subtracted, not added. 
 
242 
 
 THE THEORY OF MEASUREMENTS 
 
 TABLE XXIII. RADIAN MEASURE. 
 
 
 0' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 30' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 123 
 
 4 5 
 
 
 
 0.0000 
 
 0017 
 
 0035 
 
 0052 
 
 0070 
 
 0087 
 
 0105 
 
 0122 
 
 0140 
 
 0157 
 
 369 
 
 12 15 
 
 1 
 
 0.0175 
 
 0192 
 
 0209 
 
 0227 
 
 0244 
 
 0262 
 
 0279 
 
 0297 
 
 0314 
 
 0332 
 
 369 
 
 12 15 
 
 2 
 
 0.0349 
 
 0367 
 
 0384 
 
 0401 
 
 0419 
 
 0436 
 
 0454 
 
 0471 
 
 0489 
 
 0506 
 
 369 
 
 12 15 
 
 3 
 
 0.0524 
 
 0541 
 
 0559 
 
 0576 
 
 0593 
 
 0611 
 
 0628 
 
 0646 
 
 0663 
 
 0681 
 
 369 
 
 12 15 
 
 4 
 
 0.0698 
 
 0716 
 
 0733 
 
 0750 
 
 0768J 0785 
 
 0803 
 
 0820 
 
 0838 
 
 0855 
 
 369 
 
 12 15 
 
 5 
 
 0.0873 
 
 0890 
 
 0908 
 
 0925 
 
 0942 
 
 0960 
 
 0977 
 
 0995 
 
 1012 
 
 1030 
 
 369 
 
 12 15 
 
 6 
 
 0.1047 
 
 1065 
 
 1082 
 
 1100 
 
 1117 
 
 1134 
 
 1152 
 
 1169 
 
 1187 
 
 1204 
 
 369 
 
 12 15 
 
 7 
 
 0.1222 
 
 1239 
 
 1257 
 
 1274 
 
 1292 
 
 1309 
 
 1326 
 
 1344 
 
 1361 
 
 1379 
 
 369 
 
 12 15 
 
 8 
 
 0.1396 
 
 1414 
 
 1431 
 
 1449 
 
 1466 
 
 1484 
 
 1501 
 
 1518 
 
 1536 
 
 1553 
 
 369 
 
 12 15 
 
 9 
 
 0.1571 
 
 1588 
 
 1606 
 
 1623 
 
 1641 
 
 1658 
 
 1676 
 
 1693 
 
 1710 
 
 1728 
 
 369 
 
 12 15 
 
 10 
 
 0.1745 
 
 1763 
 
 1780 
 
 1798 
 
 1815 
 
 1833 
 
 1850 
 
 1868 
 
 1885 
 
 1902 
 
 369 
 
 12 15 
 
 11 
 
 0.1920 
 
 1937 
 
 1955 
 
 1972 
 
 1990 
 
 2007 
 
 2025 
 
 2042 
 
 2059 
 
 2077 
 
 369 
 
 12 15 
 
 12 
 
 0.2094 
 
 2112 
 
 2129J2147 
 
 2164 
 
 2182 
 
 2199 
 
 2217 
 
 2234 
 
 2251 
 
 369 
 
 12 15 
 
 13 
 
 0.2269 
 
 2286 
 
 230412321 
 
 2339 
 
 2356 
 
 2374 
 
 2391 
 
 2409 
 
 2426 
 
 369 
 
 12 15 
 
 14 
 
 0.2443 
 
 2461 
 
 2478 2496 
 
 2513 
 
 2531 
 
 2548 
 
 2566 
 
 2583 
 
 2601 
 
 369 
 
 12 15 
 
 15 
 
 0.2618 
 
 2635 
 
 2653 
 
 2670 
 
 2688 
 
 2705 
 
 2723 
 
 2740 
 
 2758 
 
 2775 
 
 369 
 
 12 15 
 
 16 
 
 0.2793 
 
 2810 
 
 2827 
 
 2845 
 
 2862 
 
 2880 
 
 2897 
 
 2915 
 
 2932 
 
 2950 
 
 369 
 
 12 15 
 
 17 
 
 0.2967 
 
 2985 
 
 3002 
 
 3019 
 
 3037 
 
 3054 
 
 3072 
 
 3089 
 
 3107 
 
 3124 
 
 369 
 
 12 15 
 
 18 
 
 0.3142 
 
 3159 
 
 3176 
 
 3194 
 
 3211 
 
 3229 
 
 3246 
 
 3264 
 
 3281 
 
 3299 
 
 369 
 
 12 15 
 
 19 
 
 0.3316 
 
 3334 
 
 3351 
 
 3368 
 
 3386 
 
 3403 
 
 3421 
 
 3438 
 
 3456 
 
 3473 
 
 369 
 
 12 15 
 
 20 
 
 0.3491 
 
 3508 
 
 3526 
 
 3543 
 
 3560 
 
 3578 
 
 3595 
 
 3613 
 
 3630 
 
 3648 
 
 369 
 
 12 15 
 
 21 
 
 0.3665 
 
 3683 
 
 3700 
 
 3718 
 
 3735 
 
 3752 
 
 3770 
 
 3787 
 
 3805 
 
 3822 
 
 369 
 
 12 15 
 
 22 
 
 0.3840 
 
 3857 
 
 3875 
 
 3892 
 
 3910 
 
 3927 
 
 3944 
 
 3962 
 
 3979 
 
 3997 
 
 369 
 
 12 15 
 
 23 
 
 0.4014 
 
 4032 
 
 4049 
 
 4067 
 
 4084 
 
 4102 
 
 4119 
 
 4136 
 
 4154 
 
 4171 
 
 369 
 
 12 15 
 
 24 
 
 0.4189 
 
 4206 
 
 4224 
 
 4241 
 
 4259 
 
 4276 
 
 4294 
 
 4311 
 
 4328 
 
 4346 
 
 369 
 
 12 15 
 
 25 
 
 0.4363 
 
 4381 
 
 4398 
 
 4416 
 
 4433 
 
 4451 
 
 4468 
 
 4485 
 
 4503 
 
 4520 
 
 369 
 
 12 15 
 
 26 
 
 0.4538 
 
 4555 
 
 4573 
 
 4590 
 
 4608 
 
 4625 
 
 4643 
 
 4660 
 
 4677 
 
 4695 
 
 369 
 
 12 15 
 
 27 
 
 0.4712 
 
 4730 
 
 4747 
 
 4765 
 
 4782 
 
 4800 
 
 4817 
 
 4835 
 
 4852 
 
 4869 
 
 369 
 
 12 15 
 
 28 
 
 0.4887 
 
 4904 
 
 4922 
 
 4939 
 
 4957 
 
 4974 
 
 4992 
 
 5009 
 
 5027 
 
 5044 
 
 369 
 
 12 15 
 
 29 
 
 0.5061 
 
 5079 
 
 5096 
 
 5114 
 
 5131 
 
 5149 
 
 5166 
 
 5184 
 
 5201 
 
 5219 
 
 369 
 
 12 15 
 
 30 
 
 0.5236 
 
 5253 
 
 5271 
 
 5288 
 
 5306 
 
 5323 
 
 5341 
 
 5358 
 
 5376 
 
 5393 
 
 369 
 
 12 15 
 
 31 
 
 0.5411 
 
 5428 
 
 5445 
 
 5463 
 
 5480 
 
 5498 
 
 5515 
 
 5533 
 
 5550 
 
 5568 
 
 369 
 
 12 15 
 
 32 
 
 0.5585 
 
 5603 
 
 5620 
 
 5637 
 
 5655 
 
 5672 
 
 5690 
 
 5707 
 
 5725 
 
 5742 
 
 369 
 
 12 15 
 
 33 
 
 0.5760 
 
 5777 
 
 5794 
 
 5812 
 
 5829 
 
 5847 
 
 5864 
 
 5882 
 
 5899 
 
 5917 
 
 369 
 
 12 15 
 
 34 
 
 0.5934 
 
 5952 
 
 5969 
 
 5986 
 
 6004 
 
 6021 
 
 6039 
 
 6056 
 
 6074 
 
 6091 
 
 369 
 
 12 15 
 
 35 
 
 0.6109 
 
 6126 
 
 6144 
 
 6161 
 
 6178 
 
 6196 
 
 6213 
 
 6231 
 
 6248 
 
 6266 
 
 369 
 
 12 15 
 
 36 
 
 0.6283 
 
 6301 
 
 6318 
 
 6336 
 
 6353 
 
 6370 
 
 6388 
 
 6405 
 
 6423 
 
 6440 
 
 369 
 
 12 15 
 
 37 
 
 0.6458 
 
 6475 
 
 6493 
 
 6510 
 
 6528 
 
 6545 
 
 6562 
 
 6580 
 
 6597 
 
 6615 
 
 369 
 
 12 15 
 
 38 
 
 0.6632 
 
 6650 
 
 6667 
 
 6685 
 
 6702 
 
 6720 
 
 6737 
 
 6754 
 
 6772 
 
 6789 
 
 369 
 
 12 15 
 
 39 
 
 0.6807 
 
 6824 
 
 6842 
 
 6859 
 
 6877 
 
 6894 
 
 6912 
 
 6929 
 
 6946 
 
 6964 
 
 369 
 
 12 15 
 
 40 
 
 0.6981 
 
 6999 
 
 7016 
 
 7034 
 
 7051 
 
 7069 
 
 7086 
 
 7103 
 
 7121 
 
 7138 
 
 369 
 
 12 15 
 
 41 
 
 0.7156 
 
 7173 
 
 7191 
 
 7208 
 
 7226 
 
 7243 
 
 7261 
 
 7278 
 
 7295 
 
 7313 
 
 369 
 
 12 15 
 
 42 
 
 0.7330 
 
 7348 
 
 7365 
 
 7383 
 
 7400 
 
 7418 
 
 7435 
 
 7453 
 
 7470 
 
 7487 
 
 369 
 
 12 15 
 
 43 
 
 0.7505 
 
 7522 
 
 7540 
 
 7557 
 
 7575 
 
 7592 
 
 7610 
 
 7627 
 
 7645 
 
 7662 
 
 369 
 
 12 15 
 
 44 
 
 0.7679 
 
 7697 
 
 7714 
 
 7732 
 
 7749 
 
 7767 
 
 7784 
 
 7802 
 
 7819 
 
 7837 
 
 369 
 
 12 15 
 
 (Bottomley, " Four Fig. Math. Tables.") 
 
TABLES 
 TABLE XXIII. RADIAN MEASURE (Concluded). 
 
 243 
 
 
 0' 
 
 6' 
 
 12' 
 
 18' 
 
 24' 
 
 30' 
 
 36' 
 
 42' 
 
 48' 
 
 54' 
 
 1 2 3 
 
 4 5 
 
 45 
 
 0.7854 
 
 7871 
 
 7889 
 
 7906 
 
 7924 
 
 7941 
 
 7959 
 
 7976 
 
 7994 
 
 8011 
 
 369 
 
 12 15 
 
 46 
 
 0.8029 
 
 8046 
 
 8063 
 
 8081 
 
 8098 
 
 8116 
 
 8133 
 
 8151 
 
 8168 
 
 8186 
 
 369 
 
 12 15 
 
 47 
 
 0.8203 
 
 8221 
 
 8238 
 
 8255 
 
 8273 
 
 8290 
 
 8308 
 
 8325 
 
 8343 
 
 8360 
 
 369 
 
 12 15 
 
 48 
 
 0.8378 
 
 8395 
 
 8412 
 
 8430 
 
 8447 
 
 8465 
 
 8482 
 
 8500 
 
 8517 
 
 8535 
 
 369 
 
 12 15 
 
 49 
 
 0.8552 
 
 8570 
 
 8587 
 
 8604 
 
 8622 
 
 8639 
 
 8657 
 
 8674 
 
 8692 
 
 8709 
 
 369 
 
 12 15 
 
 50 
 
 0.8727 
 
 8744 
 
 8762 
 
 8779 
 
 8796 
 
 8814 
 
 8831 
 
 8849 
 
 8866 
 
 8884 
 
 369 
 
 12 15 
 
 51 
 
 0.8901 
 
 8919 
 
 8936 
 
 8954 
 
 8971 
 
 8988 
 
 9006 
 
 9023 
 
 9041 
 
 9058 
 
 369 
 
 12 15 
 
 52 
 
 0.9076 
 
 9093 
 
 9111 
 
 9128 
 
 9146 
 
 9163 
 
 9180 
 
 9198 
 
 9215 
 
 9233 
 
 369 
 
 12 15 
 
 53 
 
 0.9250 
 
 9268 
 
 9285 
 
 9303 
 
 9320 
 
 9338 
 
 9355 
 
 9372 
 
 9390 
 
 9407 
 
 369 
 
 12 15 
 
 54 
 
 0.9425 
 
 9442 
 
 9460 
 
 9477 
 
 9495 
 
 9512 
 
 9529 
 
 9547 
 
 9564 
 
 9582 
 
 369 
 
 12 15 
 
 55 
 
 0.9599 
 
 9617 
 
 9634 
 
 9652 
 
 9669 
 
 9687 
 
 9704 
 
 9721 
 
 9739 
 
 9756 
 
 369 
 
 12 15 
 
 56 
 
 0.9774 
 
 9791 
 
 9809 
 
 9826 
 
 9844 
 
 9861 
 
 9879 
 
 9896 
 
 9913 
 
 9931 
 
 369 
 
 12 15 
 
 57 
 
 0.9948 
 
 9966 
 
 9983 
 
 0001 
 
 0018 
 
 0036 
 
 0053 
 
 0071 
 
 0088 
 
 0105 
 
 369 
 
 12 15 
 
 58 
 
 1.0123 
 
 0140 
 
 0158 
 
 0175 
 
 0193 
 
 0210 
 
 0228 
 
 0245 
 
 0263 
 
 0280 
 
 369 
 
 12 15 
 
 59 
 
 1.0297 
 
 0315 
 
 0332 
 
 0350 
 
 0367 
 
 0385 
 
 0402 
 
 0420 
 
 0437 
 
 0455 
 
 369 
 
 12 15 
 
 60 
 
 1.0472 
 
 0489 
 
 0507 
 
 0524 
 
 0542 
 
 0559 
 
 0577 
 
 0594 
 
 0612 
 
 0629 
 
 369 
 
 12 15 
 
 61 
 
 1.0647 
 
 0664 
 
 0681 
 
 0699 
 
 0716 
 
 0734 
 
 0751 
 
 0769 
 
 0786 
 
 0804 
 
 369 
 
 12 15 
 
 62 
 
 1.0821 
 
 0838 
 
 0856 
 
 0873 
 
 0891 
 
 0908 
 
 0926 
 
 0943 
 
 0961 
 
 0978 
 
 369 
 
 12 15 
 
 63 
 
 1.0996 
 
 1013 
 
 1030 
 
 1048 
 
 1065 
 
 1083 
 
 1100 
 
 1118 
 
 1135 
 
 1153 
 
 369 
 
 12 15 
 
 64 
 
 1.1170 
 
 1188 
 
 1205 
 
 1222 
 
 1240 
 
 1257 
 
 1275 
 
 1292 
 
 1310 
 
 1327 
 
 369 
 
 12 15 
 
 65 
 
 1.1345 
 
 1362 
 
 1380 
 
 1397 
 
 1414 
 
 1432 
 
 1449 
 
 1467 
 
 1484 
 
 1502 
 
 369 
 
 12 15 
 
 66 
 
 1.1519 
 
 1537 
 
 1554 
 
 1572 
 
 1589 
 
 1606 
 
 1624 
 
 1641 
 
 1659 
 
 1676 
 
 369 
 
 12 15 
 
 67 
 
 1.1694 
 
 1711 
 
 1729 
 
 1746 
 
 1764 
 
 1781 
 
 1798 
 
 1816 
 
 1833 
 
 1851 
 
 369 
 
 12 15 
 
 68 
 
 1.1868 
 
 1886 
 
 1903 
 
 1921 
 
 1938 
 
 1956 
 
 1973 
 
 1990 
 
 2008 
 
 2025 
 
 369 
 
 12 15 
 
 69 
 
 1.2043 
 
 2060 
 
 2078 
 
 2095 
 
 2113 
 
 2130 
 
 2147 
 
 2165 
 
 2182 
 
 2200 
 
 369. 
 
 12 15 
 
 70 
 
 1.2217 
 
 2235 
 
 2252 
 
 2270 
 
 2287 
 
 2305 
 
 2322 
 
 2339 
 
 2357 
 
 2374 
 
 369 
 
 12 15 
 
 71 
 
 1.2392 
 
 2409 
 
 2427 
 
 2444 
 
 2462 
 
 2479 
 
 2497 
 
 2514 
 
 2531 
 
 2549 
 
 369 
 
 12 15 
 
 72 
 
 1.2566 
 
 2584 
 
 2601 
 
 2619 
 
 2636 
 
 2654 
 
 2671 
 
 2689 
 
 2706 
 
 2723 
 
 369 
 
 12 15 
 
 73 
 
 1.2741 
 
 2758 
 
 2776 
 
 2793 
 
 2811 
 
 2828 
 
 2846 
 
 2863 
 
 2881 
 
 2898 
 
 369 
 
 12 15 
 
 74 
 
 1.2915 
 
 2933 
 
 2950 
 
 2968 
 
 2985 
 
 3003 
 
 3020 
 
 3038 
 
 3055 
 
 3073 
 
 369 
 
 12 15 
 
 75 
 
 1.3090 
 
 3107 
 
 3125 
 
 3142 
 
 3160 
 
 3177 
 
 3195 
 
 3212 
 
 3230 
 
 3247 
 
 369 
 
 12 15 
 
 76 
 
 1 . 3265 
 
 3282 
 
 3299 
 
 3317 
 
 3334 
 
 3352 
 
 3369 
 
 3387 
 
 $404 
 
 3422 
 
 369 
 
 12 15 
 
 77 
 
 1 3439 
 
 3456 
 
 3474 
 
 3491 
 
 3509 
 
 3526 
 
 3544 
 
 3561 
 
 3579 
 
 3596 
 
 369 
 
 12 15 
 
 78 
 
 1.3614 
 
 3631 
 
 3648 
 
 3666 
 
 3683 
 
 3701 
 
 3718 
 
 3736 
 
 3753 
 
 3771 
 
 369 
 
 12 15 
 
 79 
 
 1.3788 
 
 3806 
 
 3823 
 
 3840 
 
 385& 
 
 3875 
 
 3893 
 
 3910 
 
 3928 
 
 3945 
 
 369 
 
 12 15 
 
 80 
 
 1.3963 
 
 3980 
 
 3998 
 
 4015 
 
 4032 
 
 4050 
 
 4067 
 
 4085 
 
 4102 
 
 4120 
 
 369 
 
 12 15 
 
 81 
 
 1.4137 
 
 4155 
 
 4172 
 
 4190 
 
 4207 
 
 4224 
 
 4242 
 
 4259 
 
 4277 
 
 4294 
 
 369 
 
 12 15 
 
 82 
 
 1.4312 
 
 4329 
 
 4347 
 
 4364 
 
 4382 
 
 4399 
 
 4416 
 
 4434 
 
 4451 
 
 4469 
 
 369 
 
 12 15 
 
 83 
 
 1.4486 
 
 4504 
 
 4521 
 
 4539 
 
 4556 
 
 4573 
 
 4591 
 
 4608 
 
 4626 
 
 4643 
 
 369 
 
 12 15 
 
 84 
 
 1.4661 
 
 4678 
 
 4696 
 
 4713 
 
 4731 
 
 4748 
 
 4765 
 
 4783 
 
 4800 
 
 4818 
 
 369 
 
 12 15 
 
 85 
 
 1.4835 
 
 4853 
 
 4870 
 
 4888 
 
 4905 
 
 4923 
 
 4940 
 
 4957 
 
 4975 
 
 4992 
 
 369 
 
 12 15 
 
 86 
 
 1.5010 
 
 5027 
 
 5045 
 
 5062 
 
 5080 
 
 5097 
 
 5115 
 
 5132 
 
 5149 
 
 5167 
 
 369 
 
 12 15 
 
 87 
 
 1.5184 
 
 5202 
 
 5219 
 
 5237 
 
 5254 
 
 5272 
 
 5289 
 
 5307 
 
 5324 
 
 5341 
 
 369 
 
 12 15 
 
 88 
 
 1.5359 
 
 5376 
 
 5394 
 
 5411 
 
 5429 
 
 5446 
 
 5464 
 
 5481 
 
 5499 
 
 5516 
 
 369 
 
 12 15 
 
 89 
 
 1.5533 
 
 5551 
 
 5568 
 
 5586 
 
 5603 
 
 5621 
 
 5638 
 
 5656 
 
 5673 
 
 5691 
 
 369 
 
 12 15 
 
INDEX. 
 
 A. 
 
 Absolute measurements, 5. 
 Accidental errors, axioms of, 29. 
 
 errors, criteria of, 121. 
 
 errors, definition of, 26 
 
 errors, law of, 29, 35. 
 Adjusted effects, 149. 
 Adjustment of the angles about a 
 point, 81. 
 
 of the angles of a plane triangle, 93. 
 
 of instruments, 15, 183. 
 
 of measurements, 21, 42, 63, 72. 
 Applications of the method of least 
 
 squares, 203. 
 
 Arithmetical mean, characteristic 
 errors of, 51. 
 
 mean, principle of, 29. 
 
 mean, properties of, 42. 
 Average error, defined, 44. 
 Axioms of accidental errors, 29. 
 
 B. 
 
 Best magnitudes for components, 
 fundamental principles, 165. 
 general solutions, 167. 
 practical examples, 173. 
 special cases, 170. 
 
 C. 
 
 Characteristic errors, defined, 44. 
 
 errors, computation of, 53, 57, 66, 
 71, 99, 101, 112, 114. 
 
 errors of the arithmetical mean, 51. 
 
 errors, relations between, 49. 
 Chauvenet's criterion, 127. 
 Computation checks for normal equa- 
 tions, 83. 
 Conditioned measurements, 17. 
 
 quantities, determination of, 92. 
 
 Constant errors, elimination of, 117. 
 
 errors, defined, 23. 
 Conversion factor, defined, 3. 
 
 factor, determination of, 8. 
 Correction factors, defined, 131. 
 Criteria of accidental errors, 121. 
 Criticism of published results, proper 
 
 basis for, 117. 
 
 Curves, use of, in reducing observa- 
 tions, 198. 
 
 D. 
 
 Dependent measurements, 17. 
 Derived measurements, defined, 12. 
 measurements, precision of, 135. 
 quantities, defined, 95. 
 quantities, errors of, 99. 
 units, 4. 
 
 Dimensions of units, 5. 
 Direct measurements, defined, 11. 
 measurements, precision of, 130. 
 Discussion of completed observa- 
 tions, 117. 
 of proposed measurements, general 
 
 problem, 145. 
 
 of proposed measurements, prelim- 
 inary considerations, 144. 
 of proposed measurements, primary 
 condition, 146. 
 
 E. 
 
 Effective sensitiveness of instru- 
 ments, 183. 
 
 Equal effects, principle of, 147. 
 Equations, observation, 74. 
 
 normal, 75. 
 Error, average, 44. 
 
 fractional, 101. 
 
 mean, 46. 
 
 probable, 47. 
 
 245 
 
246 
 
 INDEX 
 
 Error, Continued. 
 
 unit, 31. 
 
 weighted, 67. 
 Errors, accidental, 26. 
 
 characteristic, 44. 
 
 constant, 23. 
 
 definition of, 18. 
 
 of adjusted measurements, 105. 
 
 of derived quantities, 99. 
 
 of multiples of a measured quan- 
 tity, 98. 
 
 of the algebraic sum of a number 
 of terms, 95. 
 
 of the product of a number of 
 factors, 102. 
 
 percentage, 104. 
 
 personal, 25. 
 
 propagation of, 95. 
 
 systematic, 118. 
 
 systems of, 33. 
 Examples, see Numerical examples. 
 
 F. 
 
 Fractional error, defined, 101. 
 
 error of the product of a number 
 
 of factors, 102. 
 Free components, 169. 
 Functional relations, determination 
 
 of, 15, 195, 198, 203. 
 Fundamental units, 4. 
 
 G. 
 
 Gauss's method for the solution of 
 
 normal equations, 84. 
 General mean, 63. 
 
 principles, 1. 
 Graphical methods of reduction, 198. 
 
 I. 
 
 Independent measurements, 17. 
 Indirect measurements, 11. 
 Intrinsic sensitiveness of instru- 
 ments, 183. 
 
 Law of accidental errors, 29, 35. 
 
 Laws of science, 2. 
 
 Least squares, method of, 72. 
 
 M. 
 
 Mathematical constants, use of, in 
 
 computations, 153. 
 Mean error, defined, 46. 
 Measurement, defined, 2. 
 Measurements, absolute, 5. 
 adjustment of, 21, 42, 63, 72. 
 derived, 12. 
 direct, 11. 
 
 discussion of, 117, 144. 
 independent, dependent, and con- 
 ditioned, 17. 
 indirect, 11. 
 
 precision of, 19, 130, 135. 
 weights of, 61. 
 
 Method of least squares, applica- 
 tions of, 203. 
 
 of least squares, fundamental prin- 
 ciples of, 72. 
 Mistakes, 26. 
 
 N. 
 
 Negligible components, 154. 
 
 effects, 151. 
 
 Normal equations, computation 
 checks for, 83. 
 
 equations, derivation of, 75. 
 
 equations, solution by determi- 
 nants, 114. 
 
 equations, solution by Gauss's 
 method, 84. 
 
 equations, solutions by indetermi- 
 nate multipliers, 105. 
 
 equations, solution with two in- 
 dependent variables, 78. 
 Numeric, defined, 2. 
 Numerical examples: 
 
 Adjustment of angles about a point, 
 81. 
 
 Adjustment of angles of a plane 
 triangle, 93. 
 
 Application of Chauvenet's crite- 
 rion, 129. 
 
 Best magnitudes for components, 
 173, 175, 180. 
 
 Characteristic errors of direct 
 measurements, 56, 70. 
 
INDEX 
 
 247 
 
 Numerical examples Continued. 
 
 Coefficient of linear expansion, 78. 
 
 Discussion of proposed measure- 
 ment, 157. 
 
 Effective sensitiveness of potenti- 
 ometer, 190. 
 
 Errors of a derived quantity, 101. 
 
 Fractional errors, 101. 
 
 Precision of completed measure- 
 ment, 140. 
 
 Probable errors of adjusted meas- 
 urements, 113, 115. 
 
 Probable error of general mean, 69. 
 
 Propagation of errors, 101. 
 
 Solution of normal equations by 
 Gauss's method, 88. 
 
 Weighted direct measurement, 69. 
 
 O. 
 
 Observation, denned, 15. 
 
 equations, 74. 
 
 standard, 62. 
 Observations, record of, 16. 
 
 report of, 211. 
 
 representation of, by curves, 198. 
 
 P. 
 
 Percentage errors, 104. 
 Personal equation, 26. 
 
 errors, 25. 
 
 Physical tables, use of, 138. 
 Precision constant, 35. 
 Precision of derived measurements, 
 
 135. 
 
 of direct measurements, 130. 
 of measurement, denned, 19. 
 Precision measure, denned, 132. 
 Preliminary considerations for select- 
 ing methods of measurement, 
 144. 
 
 Primary condition, 146. 
 Principle of the arithmetical mean, 
 
 29. 
 
 of equal effects, 147. 
 Probability curve, 32. 
 function, 34. 
 
 Probability curve Continued. 
 function, comparison with experi- 
 ence, 40. 
 integral, 37. 
 of large residuals, 124. 
 of residuals, 30. 
 principles of, 28. 
 Probable error, denned, 47. 
 
 error of adjusted measurements, 
 
 111, 112, 116. 
 
 error of the arithmetical mean, 53. 
 error of direct measurements, com- 
 putation of, 54, 55, 57. 
 error of the general mean, 66, 68. 
 error of a single observation, 54, 
 
 68, 108. 
 
 error of a standard observation, 62. 
 Propagation of errors, 95. 
 Publication, 209. 
 
 R. 
 
 Research, fundamental principles, 
 192. 
 
 general methods, 193. 
 Residuals, defined, 27. 
 
 distribution of, 29. 
 
 probability of, 30, 124. 
 
 S. 
 
 Sensitiveness of methods and instru- 
 ments, 183. 
 
 Separate effects of errors, 133, 135. 
 Setting of instruments, 15. 
 Sign-changes, defined, 123. 
 Sign-follows, defined, 123. 
 Significant figures, use of, 19, 58. 
 Slugg, defined, 9. 
 
 Special functions, treatment of, 155. 
 Standard observation, defined, 62. 
 Systematic errors, defined, 118. 
 Systems of errors, 33. 
 of units, 7. 
 
 T. 
 
 Tables, list of, ix. 
 Transformation of units, 8. 
 Treatment of special functions, 155. 
 
248 INDEX 
 
 U. W. 
 
 Unit error, 31. Weighted errors, 67. 
 
 Units, c.g.s. system, 7. mean, 63. 
 
 dimensions of, 5. Weights of adjusted measurements, 
 
 engineer's system, 7. 105, 112, 114. 
 
 fundamental and derived, 4. of direct measurements, 61. 
 
 systems in general use, 7. 
 
 transformation of, 8. 
 Use of physical tables, 138. 
 
 significant figures, 19, 58. 
 
RETURN CIRCULATION DEPARTMENT 
 
 TO ^ 202 Main Library 
 
 LOAN PERIOD 1 
 HOME USE 
 
 2 
 
 3 
 
 4 
 
 5 
 
 6 
 
 ALL BOOKS MAY BE RECALLED AFTER 7 DAYS 
 
 Renewals and Recharges may be made 4 days prior to the due date. 
 
 Books may be Renewed by calling 642-3405. 
 
 SEPoPVf 
 
 AS STAMPED BELOW 
 
 mmt \f 5 J3J 
 
 % 
 
 o 
 
 
 AUG041989! 
 
 i| 
 
 
 Rec'd UC6 A/M/S 
 
 
 
 AUG 1 7 i968 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 FORM NO. DD6 7 
 
 UNIVERSITY OF CALIFORNIA, BERKELEY 
 
 BERKELEY, CA 94720 
 
 $ 
 
GENERAL LIBRARY - U.C. BERKELEY 
 
 , 
 
 THE UNIVERSITY OF CALIFORNIA LIBRARY 
 
 ,..-* ,