Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a notforprofit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Biometrika Trust is collaborating with JSTOR to digitize, preserve and extend access to Biometrika.
http://www.jstor.org
CONTENTS.
Pae
Note.
Introductory 266
Theorem.
(1) General If the valuesof the constants
of a curvebe found by the
methodofmomenits, thefitwillbe good. . .. . . 267
oftheareaand moments
(2) On thediscovery ofa curvegivenbya seriesofisolated
observations.. ..272
(3) Comparison ofmoments 279
in a specialcase of methodofleastsquaresand method
(4) On the discovery of the area and momentswhernthe data consistof the
frequencies withiin
falling certainelementaryranges. . . . . . 282
(5) l1ustrationI. To fita curveof type
y=yO(1+a)m 1  ai)
of moments
Calculation and constantsforthe fecundity of brood
distribution
mares. 289
(6) Illustratiot1. To fita curve of type
y = yO(1+ Pe a.
Observations of Thiele. 292
(7) Illustration11. To fita curveof type
y=a sin(nx+a),
whena partonlyoftherangehas beenobserved.
or a sinecurve,
Observations ofChree. 295
(8) illustrationIV. To fitMakeham's statistics.
curveto mortality 298
Biometrika
I 26
IntroductoryNote.
FIG. 1. FIo. 2.
First Embryo: 72 Hours ofIncubation. Second Embryo: 96 Hours of Incubation.
Here ao, al, a., ... etc. will be functions of the n constants cl, c2, ... cn of the curve.
Hence theoreticallywe can findthe n c's in termsof ac,,a,, a2, ... an_i. We should
thus be able on substitutionto expressa", an+i, ... etc. in termsof ao, a,, ... ani.
Now considerthiefirstn a's as the constantsof our curveand it will be expressible
in the form:
xI.
y= ao+ oilx+
,znI
12 + ***+atnlt
f
In
+ +r (ot, a,, of.,... a,,,)
+ etc............... (i).
Next let y' be the ordinatecorresponding to x given by observation,theny y
will be the distance between the theoreticaland observedcurves at the point
corresponding to x, and our object is to make the values of thiisas small as possible
a
by proper choice of ao , a2, ... an,. This may be done by the methodof least
squares or making
f (y  yl)2 dx = a miniinum.
This obviouslygives a verv good method,if not the "best," a term incapable of
definition. The resultingequation,since y is the variable,is
f(y  y') Sydx = 0 ........... (ii).
Now Sy depends on the variation of ao, ... an, or
da. dal +f
da1 I
nl)
(dqbn+ls ____a+
+ etc.
=Sao (I dc d4+1 1 +
+d in+ daoIn+1
Xn2(2+d n+1 + )
= BO {1+ do%j(Ed))}
+ 1x +?a(~
8aa d($(ox))}
+ etc ...................(iii),
aftern termsofMaclaurin'sexpansion.
it becomes
Substitutingin (ii) and rearranging,
{f y)
Y (I + daR)dx} 8ac
But the quantitiesaO,al, a2, ... an_lare at our choice,and thereforeto satisfythis
equationl,theirvariationsmulstbe independently zero. Thus we have the following
equationsto finsda0, al, . .. a
x. dR\2 I
y y') (x + dx \= 0)
.I0
B),..........
t (iV).
(YY')(l 2 3+ da) =0
Now let A be the area, A,a1,A1.2, A,a3, A1.4, ... etc. be the first,second,
third,fourth,etc. momentsof the theoreticalcurve,and A' be the area, A'th1',
A',2', A/#'p~ A',u4',... the like moments for the observation curve, mnoments
being taken round the axis of y (which is of course any axis). Then the above
equationstmaybe written
A =A' f(y_y')Adhx
dx dR
/L1J\YYJdaX
A1u= A'lUA' YY')
A,2uA'=.' 2 "yy'')RdR
dax
AIa\ _
,u
AIf= Y _ ) dx
............d..
dR
Y (v).
A/X
= A'4l  4(YY
 dR dx
"~dR
............................................. d dR

A,u_l = A'P'nln ( Y')dRdj
.(vi)S v i.
.... ..... ............. ( I .
n= / ni
Or, the constantsof the theoreticalcurve are to be foundby equating its area
and firstn 1 momentsto the area and firstn 1 momentsof the observedcurve.
These results having been obtained we may at once replace a,, a, qc, ... a,"_ by
the real constantscl, c2, c8, ... cn of the theoreticalcurve,and we obtain the rule:
and accordingly
J dR dn (?) 1
+ d+"+' (0) 1
(Aon+l  A'/'n+1)
1
~d4pn+2
+4 (0)  A/kn2
(A/pkn2
+ etc.
Thus if A =A', we have the factors (n un'), (/n I'n+l) (/n+2 /'n+2), etc.
Thus besides the smallness of the factors jn?) on+ , ... etc., dependingon
oc ZI 1 ,11 2 1Z\ 11
Azo=z1zo, Azi=z1zi.
BiometrikaI 27
y5 = 014,2692 5 =*033,1918
ey7= 011,3674 76' =030,9989
77 = *009,3563 y77=*029,1253
7Y8 = 007,8925 7Y8 =027,5110
79 = *006,7858 9 =*026,1066
7y,o=005,9241 7io 024,8732
,yll= 005,2367 = 025,7807
yr11
Y12= 004,6775 7Y12 = 022,8052
713= *004,2150
714 = *003,8269
J 4P
zdx = 2h
{7zo + 14 (z4 + z8 + ... + z4p4) + 7z4p
xo
+ 32 (Z1+ z3+,5+z., ++z4,1)
JZd 3h fzO
+ Z2 + Z4 + Z8 + ZO + + Zp2 +z6p
+ a (Z1 + Z5 + Z7 +. + Z6P1)
All these rules give with increasingexactness the value of the integral,but
theysufferurnderobviousdisadvantages:
(a) The numberof elementscannot oftenbe selectedbeforehand, and if for
exampletherebe 7 or 11 or 13 a new rule has at once to be workedout.
(b) The multiplyingdifferentordinates by differentfactors is a source
fruitful
of arithmnetical
slips.
(c) The multiplyingof certainordinatesby factorsmuchlargerthanothers,
multipliesthe errormade in the determinationof certain ordinateslargely. We
do not give equal weightto all the ordinates.
Area = A0 + (1
2 z Z)  (z  zp,)} h..............................().
Aea
Area=Ac+ + 1 p (p
=Ac (763p2 3444p + 3636)
1) (p2)
5040 (p  3)
1 p(ll9p2_504p + 432)
1260 (p  2) (p 3) (p  4)
1 p(133p2  462p + 360)
+ f(z3 z2)(zp2  X3)} h......
Z3} .()
040(p 3) (p 4) (p 5)(2Z2(z
f
It is, perhaps, worthwhile exhibitingthe sort of relative exactness to be
obtained by the whole series of formuleon a special example,say d.1 for
12 or 13 ordinates. We find
Jidx
1 + X = '693,147,18.
Divergence
(a), withfourdifferences, + 000,000,2.5
(,8),withfourdifferences,  '000,000,59
+ '000,001,48
() + '000,003,28
(e) + '000,000,07
() + '000,000,04
(n) +X000,014,59
(0) + '000,000,93
(t) + '000,000,07
(IC)  *000,014,93
(X)  *000,001,26
(I",)  *000,000,12
(v)  '000,003,91
(0)  *000,000,14
(o) + '000,008,12
(7r) + '000,000,22
(p) +'000,001,27
AO = '693,580,83,or A = + '000,433,65,
AT= '692,930,49,or A =  '000,216,69.
The latter is less divergentfromthe true value than the former,but they
differby as muchas 1 in 3200 and 1 in 1600 respectivelyfromthe truevalue. On
the other hand the worstof the above quadratureformulae (Kc)and (I) give results
onlyabout 1 in 48,000 in error,whilethe best,like Boole's or Weddle's Rules, or
(t) and (,), varyfromabout 1 in 6,000,000to 1 in 17,000,000,while ({) and (7r)
are almostas good. When we are dealing with frequencywe probablynever,and
oftenwhen we are dealing with measurements, physicalor economic,we do not,
know our data with anythinglike the accuracy of 1 in 48,000. We conclude
therefore that we may expectgood resultsfrommostof these formlulhe.But some
remarkson theirrelativegoodnessmay be ofservice. In the firstplace the Euler
Maclaurin formulae(a) and (,8) withfour differences are not nearlyas good as
Mr Sheppard's new formule (t) and (,A) using,only threedifferences, and not
so good as (t) or (7r)withtwodifferences. It seems to me accordinglythat unless
we are preparedto go to great labourand calculate high differences, (5) or
(t\ (P,w,
(7r) are the best formulaeto.use, and that fornearlyall practicalpurposes(0) and
(k) are quite accurateenoughforuse. Boole's Rule (e) and Weddle's Rule (') give
splendidresults,but great care mustbe taken when we apply thein to somewhat
irregularobservationsof physical quantities and to frequencies,and not to the
evaluationof mathematicalintegrals,for in the bulk of the formulwtheyweight
and largelyweight certainordinates,and thus may tend to emphasise errorsin
particularobservations.
* Clearly: Area=AT
i (AT AO),nearly. This is a veryuseful formulabased on an assumption
as to parabolicsegmentslike Simpson'swhen both extremeand midordinatesare available.
Now these resultsshow us at once that with(X) we have a fitby the method
of nmomentssensiblyas good as that obtainedby the methodof least squares. Had
in the R. M. S. between
we used (t) or (,a) therecould not have been any difference
the method of momentsand the methodof least squares. There is of coursea
distinctionbetween the two methods which it is importantto bear in mind,
5 4 3 2 _1 +I ;2 43 F4 6
O 1 2 3 4 5 6 7 8 9 10
namely: the methodof least squares takes a curve which passes with the least
rootmean square deviationfrom11 observationpoints; the method of moments
BiometrikaI 28

[2czxn]
+ nflPie'Zdx
P
XO X
Z Jxp
+ n
Z0X0n Zx"'dx.
Here we must be very carefulto notice that our origin is the startof the base
elementin whichthe frequencybegins,that ZO= N is the total frequency, and Zp
is zero,and that xp is measuredto the end of the last baseelementh forwhichwe
are consideringthe frequency.Thus a lengthx$ + h,and not xp,wouldbe the total
range we should obtain by plottingthe frequenciesz as if theywere ordinates
at the middle of the elements. This process thereforetends to exaggerate the
range. As a rule it is convenientin frequencydistributionsto determinethe U's
about the mean. In this case theymay be foundfromthe ,u"s about any other
line by the formulwe
=AI0,
2
2= L2 ,_ /
3=
/4
= P4
=
L3'  3,1't,u.+ 2F(ix),
'1JU h3 + 6Al/2,U2/
1'It4 + 2(L3'
3^ 'L
+
[.
J
J = F6 +
4 6i24 l/0t1  lO0,t1'3j.2' (ix).,
/1,6 = U6  + 15,1 '244'
6u'/L'p5  20iti'3u3' + 151'4/2'  5,'1
x=l
NowbytheEulerMaclaurin
formula
FB1hdZ' B3h3
d3Z' BMhd'Z' 1
hi 
dx x + 6 d ...
[dn(Ze
d O=n(n1)(n2)
=de(n
... (n  s+l)
+1)
dnp ,Z,
La
providedn be greaterthans, otherwise
it is zero,unlessn = s, whenwe havethe
value i N.
Nowlet CAhstandforthechordalarea
(jZ0o8 + Zxs1'+ Z2, + ... + jZ 18) h,
where of course x0 0, x =h, x2= 2h, etc., if h be the base element. Then we
easily find
f Zdx2=  0 024
h
zz A+12N2 +6048 a4
h6 h8
f'Zedzx= C2h+ 1512N 6a  5a,+
. (x)
h Ah4 ,6 A8
jZx3dx= C+ 120N+ 2 0a2 7 a4+
Z4dx Ch h4
00Ch 4 h a, h+ a,..x)
j............
hA6 ,
1 A
N 720"+32~a
2Ch = A4 a'= e
v 6 1
=
+
N8, ,0 2+3024
/L2 hl a4
=6C5h
, 3r 2 h6 120 ___
ba N +504 ,9600a5+
4W3e
,an Ai (x)i......n.e A A6 . (xi).
/A
N 0 +12~6a2 14 a4'+
Al 504h
z/ A6 _ (_0'fA1h0
Sa,/ als' h
,
OC5h A6 A8 hl
P' a2 + a4
+425
I18= I )i8B8..,A8,
NCT 
as s is odd or even.
the seriesendingwitha constantor n, according,
= IfJ
Zdx= (chordalarea) by (x)
h 120
124v +3024 a4'
2 5 7I + a84 1
4= 4'V2 +270 h4+h6
64 4
2 '+Z h2' 1 8 h8 A10
h v3'+ 0h4v' 
/LV /x
p13'= V a3 + 3168a5 
, + n,3+ n4)= ae
 (n + n2
13
L2 a3 3 3+
L4
+ 1~55
These values of ycan be easily calculated. Now let h/e= p. Then if we take h
equal unity as usual, i/p will measure the fiaction e is of h. Of couirsevery
usually p = 1, but this is not necessary; thus in certaindiseases the frequencyof
cases in each of the firstfive yearsof lifeis oftenrecorded,but later onlyin five
year periods. Making these substituitions we may write our final results forthe
moments:
3
A =V  + p 75 _
fL 17203+ 30240
2^2 p474
2= V2 12 120 + 3024
P ~ = v's
f43S 1 I+.
" P
3
Ya 6
96on+
i 5074 ______
4 42
/ ^2+240 1P276
P'72 . 1P440 +((v)
__ '4........(xvi).
=
6 V+48 288+ 3168
of vital
are known; my fourthdeals with the representation
the observations
statisticsby Makeham'scurveand myfifthwiththe deductionof the curveof
errors offrequency.
frompartialobservations
(5) Illustration I. On themeanvariability and distributionoffecundity in
2000 thoroughbred broodmares.
By fecundity of the mareis heremeantthe ratioof the numberof yearling
foalsshe has actuallyproduce(d to the numnber of her opportunities.The base
elementsweretaken+ on eithersideof 0, I, , 16,... It, 1. Thusfecundity
'I
from0 to 1 was dividedinto 16 grades,respectively denotedbya, b,c, d, ... 1,m,
n,p, q. The data wereextractedfromthe studbooks, everymarehavinghad at
least8 or morecoverings.
I proposein thisfirstillustration to go throughthe wholeof the workof
dealingwiththe frequenicy distribution as it inaybe unfainiliar to manyof my
readers,and yet it is reallyverysimple. I shall firstsupposethe curveto have
highcontactat theterminals oftherange,andworkoutthev'sand deducetheit's
bySheppard'scorrections: see p. 287. This,however, is notin thiscase legitimate
frommereinspection of the curve,and therefore we ouight to startbyusing(xiv).
Workingout the ,u's in the latter way also we can comparethe resultsactually
to go as faras p,
obtained. It willbe sufficient
Grade Frequencyx x zx zx2 zx3 zx'
a 0 9  0 + 0  0 + 0
b 2 8  16 + 128 1024 + 8192
c 75 7  525 + 3675  2572,5 + 180075
d 11.5 6  69 + 414  2484 + 14904
e 215 5  1075 + 537.5  26875 + 134375
f 55 4 220 + 880 3520 + 14080
g 1045  3  3135 + 9405  28215 + 84645
A 182 2  364 + 728  1456 + 2912
i 271P5 1 2715 + 2715  2715 + 2715
j 315 0 1414 16837
k 337 +1 + 337 + 337 + 337 + 337
1 2935 +2 + 587 + 1174 + 2348 + 4696
nt 204 +3 + 612 + 1836 + 5508 + 16524
n 127 +4 + 508 + 2032 + 8128 + 32512
p 49 +5 + 245 + 1225 + 6125 + 30625
q 19 +6 + 114 + 684 + 4104 + 24624
+ 989 + 9713
'= .4945 Y2p=57775 V3 = 4 8565 )'= 94 7935
It will thus be seen that the determinationof the four,u's about the mean is
neithera long nor a difficult process. I will now proceedto findthemde novoby
applying,(viii).
Grade Frequency Z x Zx Zx2 Zx'
I. II. II1. IV.
a 0
2000 0 0 0 0
b 2
1998 1 1998 1998 1998
c 75
1990.5 2 3981 7962 15924
d 11 5
1979 3 5937 17811 53433
e 215
19575 4 7830 31320 125280
f 55
19025 5 95125 475625 237812,5
g 1045
1798 6 10788 64728 388368
h 182
1616 7 11312 79184 554288
i 2715
13445 8 10756 86048 688384
j 315
0295 9 9265,5 833895 7505055
k 337
6925 10 6925 69250 692500
1 2935
399 11 4389 48279 531069
rni 204
195 12 2340 28080 336960
n 127
68 13 884 11492 149396
p 49
19 14 266 3724 52136
q 19
0 15 0 0 0
Using (ix):
1.2= 5*449,637,
/3=
 3>472,590,
/4= 90 747,442.
The divergence between these results and those given by Mr Sheppard's
process is verysmall and solely due to the arithmeticbeing cut offat the sixth
place of decimals in the multiplication. Thus #4above really ends with 6, and
is sensible in the fourthplace of decimals of p whenwe multiply
this difference
/' by 6#42..
Now let us see if Mr Sheppard's process is in this case justified; let us no
longerput the a's zero,i.e. no longersuppose high contactat the high fecundity
end of the curve. We have
n,= 19/2000, n2= 49/2000, n,= 127/2000, n4= 204/2000.
Hence we findfrom(xv)
72=*035,729, "ye=080,344, PY4=*136,750, rye='080,625.
This leads us by (xvi) to
/4'= V1 000,1089, V2/ + 000)2525;
IS =VS'+Vi+O000,1l50,
3 /4'= vJ ^32+ 000,1886,
or the P"s are onlyinfluencedin the fourthplace of decimals. Substitutingthe
values of vl',v,',v.' and v,'about one end of the rangejust found,we have
'= *494,391,#p= 5'694,4195,
/A= 4'733,026, /h4 = 91933,7284,
whichlead by (ix) to /2= 5449,997,
,43= 3471,085,
/4= 90'745,703.
We see that modificationsare in the thirdplace ofdecimalsof F,3and , and in the
of
fourth p2. Clearly we are in assumiinghigh contact
not justifiedtheoretically
end of the frequency
at the high fecuindity curve,but for mostpracticalproblems
Sheppard'scorrectionswould in this case have been quite sufficient.The actual
slope of the tangentto the frequencycurveat the end of the rangewould be
d2Z
2=Y2 = 035,729,
which is of course fairlysmall. Thus if there be not high contact at one end of
the curve,but the slope of the tangent be not large, Sheppard's correctionswill
still give the substantialpart of the requiredcorrection. If, however,as in the
mortalitydue to variousdiseases the curve meets the axis at a considerableangle,
we must endeavourto determinein some such mainer as the above the value of
the correctiveterms.
292
to the above data. The values ofthe constantsin termsofthe momentsare given
in Phil. Tranm.Vol. 186, A., pp. 367 et seq.,and we find
The mean being at j + '4945, and the mode, which is the origin,at j + '7970.
Here j denotes a fecundityof 9/15),and 1/15 is the unit of fecundity. Fig. 2
showsthat we have a veryreasonablefit,a curvewhicheffectively representsthe
phenomenon.
150~~~~~~~~~~~~~~
i'ecundity.
FIG .  Fe?undityof 2000 Brood Mares, 8 or more coverings.
chVaratef4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 2228
Frequency    3 7 35 101 89 94 70 46 30 1 5 4 5 1   
Y=YOf1+ ) e a
Hence we deducet
Y98762 t(i +?
y= ~74449)
1S.7530
The distance fromthe mean to the mode, which is the origin,is *5413. Thus
the modal value is at 113187,and the range startsat 3,2931.
The ordinatescorresponding to the obseivationsare given in columntwo of the
Table in Art. 13 of this memoir. Fig. 3 shows a reasonable fit. The Table
comparesthese results with the successive parabolas up to the sixth and shows
how a well selected curve with three constantscan easily be superiorto one with
seven constants+.
* Thiele; Forelaesningerover almindeligIagttagelseslaere,p. 12.
t The formuleforp, a, and yoin termsof the momentsare given,Phil. Trans.,Vol. 186, A, p. 373.
 This point is of special importance,for objectionshave been raised against the skewfrequency
curvesjust referredto on the groundthat theygive betterfitsthan the normal curvebecause theyhave
one or twomoreconstants as the case may be. This is true,but theyalso give betterfitsthan some
other curves with double their numberof constants!
apo~~
~~x,
co'
(
~    r1~ 
S S R a O o
fiouaflb9,d
Now either of the curves in IllustrationsI. and II. is a good example of the
impossibilityof using the method of least squares for systematiccurvefitting.
The reader need onlyattempt to write down the typeequations,which must be
solved to findthe constants,and he will realise the siinplyappalling amount of
lengthyapproximationswhichmustbe carriedout even afterroughvalues of these
constantshave been guessed by some one or othermeans.
But it is not onlyalgebraic and exponentialcurves forwhich the methodof
curves. I will now illustrate
least squares fails; it fails also for trigonometrical
this in the verysimplestcase possible.
(7) III.
Illu,stration Let it be requiredto fitthe simplestsinecurve
y = a sin (nx + a)
to the aneroidbarometerobservationsin the Illustrationin ? 3. Let us writethe
equation in the form
y= A sin nx + B cos nx............... (xvii).
Then the three typeequationsto find n, A and B, arising fromapplying the
methodof least squares,are the following:
AS (cos2 nx) + IBS (sin 2nx) = S (y cos nw),
I4AS(sin 2nx) + BS (sin2 nx) = S (y sin nx),
AS (yx sin nx)  BS (yx cos nx) = j (A2  B2) S (x sin 2nx)  ABS (x cos 2nx).
Here S denotes a suminationwith regard to the eleven values of x and y given
on p. 279; afterthese have been substitutedin the summations, we must eliminate
A and B, and we shall then have an equation to determine n. Afterwardsthe
values of A and B must be foundby substituting the value of n in the two first
equations. We mlayleave this as an exercise to those readerswho have faithin
the methodofleast squares applied to curvefitting!*
Now let us turn to the method of moments. There are three constantsto
be found,so we must find the area and the firsttwo momentsof the observa
tionsand of the theoreticalcurve.
Taking the origin at the middle of the range 21, and writingM0, Ml, Ma
forthe area and firsttwo moments,we have
sin nl
MO=2B M1=2A (lcos nl+ sin nl)
I
n ( n n' )
M2= 2B sinnl 21cosnl 2 sinnl
n n2 n
* The equation to findn is intractableeven if we place the originat the centreof the range and
evaluateby trigonometrythe summationsnot involvingy or x outsidethe trigonometrical terms.
___
Ml
B 21si A 2l sin .................. (xix).
'.8 4 __  _. 2_ 4
FI 4.
curve y = cots
and thetranscendental
wereroughlyplottedand observedto intersectaboutz = 1'2. Using Newton's
I foundwithMissM. A. Iewenz'said:
methodofapproximation,
z= 11867, z= 11844, z= 1184,4132,
exact. Thisgave
whichlastvalueis practically
n='215,348= 12020' 19",
A = '213,545, B = 1'276,288.
Whencetherequiredcurveis
y= '213,545sin( 215,348x)+ 1P276,288cos('215,348x),
or y = 1P29403 x
sin1(12020' 19") + 80030' 5"I,
thelatterformallowingofeasycalculation oftheordinates.
We have
x Oboervedy alculaed y
5 '382 '417
 4 '674 '669
 3 '923 *891
 2 1104 1o071
 1 1'214 1'201
0 1'273 1'276
+ 1 1'270 1'292
+2 1'215 1'250
+3 1'137 1'148
+4 '989 .993
+5 '819 '793
The rootmeansquareerrorof the ordinatesis '0233. Thus the fitis byno
meansso goodas thatof the parabolaof the thirdordery= a. + a1x+ a,x + aV
Biometrikai D0
.' = kszx
* Journalof Institute
of Actuaries,Vol. xxii. p. 200, or G. King: Instituteof Actuaries Textbook,
Vol. ii. p. 79 et seq., especiallyp. 82.
L = K + S 7 + Ge ............ (xxvii),
where
e= c' ................ .. (xxviii).
AP, Lxdx.
it
Or if
a, = 12ApF/l2
....................................(xxxi),
= S+ 6S 'h_ nsin n)
alS+G {n ...(xxxii), (xi)
al ~ '}................................
=f+"Lx2dx.
Ap,u
Or if
a2 ,=12Ap/3 .................. .............. (xxxiii),
(sinhn 2 coshn 2 sinhn)
a2 =K+3G .n.na.+.nS. ....a. (xxxiv),
Ap8= Lxldx.
Or if
a.= 80A/14. ....... (xxxv),
Scoshn 
3 sinhn +6 coshn 6 sinhn}.(x i)
ofs S
+IOG ....(xv)
s
2 )
From (xxxii) and (xxxvi) we have:
coshn  24 sinhn 60 coshn 60 sinhn
as. a, = G {4 + ... (xxxvii).
802
hN no No
D, (dn D2 dn Do (xli).
Writing
Y, = 2n3+ 383n'+ 30n .(xlii),
Y2= ,8ns+ 12m'+ 3,8n+ 30..(xliii),
we have
Y
t +
 y. .(xliv),
D=Y2Y1, N=Y2+ Y1
dD dY2 dY1 dN dY2 dY(
dndn dn dn ' dn .............. (xlv),
where
dY1 = 6n'+ 6/3n+ 30, dY2 + 24n
dn dn
3,82 3/ 3+
+24+3j
..
. (xlvi).
lv)
* Calculatedless observedvalues.
Mult mai mult decât documente.
Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.
Anulați oricând.