]\gt ; On the Computation of certain On the Computation of the Harmonic Components of Series representing a Phenomenon recurrmg Dail.y and Yearly Periods By Lieut. eneral R. STRACHEy , R.E. , F.B.S. Received April \mdash ; Read May 13 , 1886 .
1 .
Following the notation commonly used , the general expression for the harmonic components of the successive terms of a serles representing a riodically recurring phenomenon , observed at equal intervals of time , is\mdash ; 2nz\amp ; c. , where is the observed value of any term in question ; is the angular equivaleot of the time interval between the observations ; is the number of intervals from the commencement of the period to the time when the term occurs ; is the mean vain of all the terms for the whole period .
Then if represents the sum of the terms in the above series which involve the sum of the terms involving , , , , \amp ; c. , Computah.onfvr a Daily Period .
9 .
For a daily period of 24 hourly intervals , and consequently , ; ; ; ; whence , disregarding the terms involving multiples of g1eater , and ; but 62 Lieut.-General Strachey .
On , , , and re Wherefore ; and and In Jike manner , , and ; and also , and , whence ; and The snccessive values of , and , thus obtained will give with a considerable degree of accuracy , the coefficients , and the entire series of harmonic components of the observed quantity .s .
Ib will at once be seen that\mdash ; , nearly .
and .
Computation of certain Harmonic } ponents .
The equations lving the coefficients , arising from the se , ries observed quantities , are usually solved to the method of least squares ; and , writing for , and for , and so on , the resulting values of thus obtained are as follows : \mdash ; To these may be added sin15 SO 4 .
The values of the coefficients may , however , be obtained otherwise , in a form which is somewhat simpler computation , and not sensibly less accurate .
Since Lieut.-General Strachey .
On the , and and putting for the sum of the series of quantities to and for the sum of , it follows that and In like manner , Substituting for , their values in terms of , \amp ; c. , we have\mdash ; And Computation of certain Harmonic Components .
The expressions before given for , and , are very readily computed , the multiplier for , being the same as that for The computation of , and , may be rendered somewhat easier as follows:\mdash ; 5 .
Assumin that the probable errors in the observed quantities are all equal , and that ( e ) represents the error in a pair of observations combined ( corresponding to the quantities ( d ) and ) , then probable error of all the coefficients calculated by the formulae ( 2 ) will be The errors , calculated in the manner now proposed from equations ( 3 ) , will be , \mdash ; Probable error of or .
or .
or . . .
VOL. XLII .
Lieut.-General Strachey .
On the The results obtained by the two methods of culation will therefore have no sensible ference of accuracy , and as the method now proposed is believed to be both simpler and less liable to arithmetical error it may without objection be preferred .
The preliminary computations on both systems , consisting of combinations of the obsel.ved quantities by addition and subtraction , are identical up to a certain point , but the formulae ( 2 ) involve more frequent use of tables , and greater chance of error in algebraical signs , in the final operations .
It may be added that much additional labour is often needlessly created by employing the hourly differences from the mean value , instead of the hourly values themselves , which are obviously sufficient for the computation of the coeffioients .
The probable errors of the values of the coefficients obtained fxom the equations ( 1 ) , will be sensibly larger than those above stated , and on the same assumptions will be as follows:\mdash ; For and ) .
, , and .
, , and .
, , .
The probable error of a pair of observations being rather less than of that of a single observation , the greatest possible error of the coefficients thus fonnd will only be about of the probable error of one of the original observations , and when great precision is not aimed at the results thns obtained may suffice , and will not be found to differ materially from those got by the more tedious methods of calculation .
6 .
If the original expression for the value of is ransformed into the series it follows that , and so with all the other terms of the series .
The most convenient method of computing the values of and is as follows:\mdash ; ; from this may at once be obtained , and T. 7 .
The foregoing discussion assumes that the series of quantities dealt with is truly recul.rent , tIlat is to say , that the 25th observation will be exactly coincident with the 1st , or .
In fact this will rarely be the case , and it becomes necessary to ascertain what effect any non-periodic change , or want of coincidence even the beginning and end of the series will have on the values of the several coefficients .
In the absence of any knowledge of the law which determines such a non-periodic change , it may be assumed to be uniform for the period over which the observations extend .
Further it will be con- venient to refer the change to the middle of the period , or to the mean value of the series with reference to which the periodic variations are being considered , so that the non-periodic deviations will be equal and affected by opposite signs , at equal intervals on either side of the middle of the period .
Hence assuming that is the whole non-periodic change with its proper sign , so that , the correction of any observation in order to eliminate the non-periodic change which affects it , will For the mean value . . . . . . . . . . . .
For , , For , , For the vaIues from the equations ( 3 ) they will For , For , , , , , \ldquo ; , , 8 .
As it may happen that the values of the initial observation and of that corresponding to the hour immediately after the end of a series of mean hourly valnes , are not known , an approximate value of may obtained either graphically or by calculation as follows:\mdash ; If the mean values for two or three hours preceding midnight , say * As the multipliers are here the same as those which enter into the expressions ( 3 ) , the correc-ions may most be applied to the quantities within the brackets in thoso expressions .
Computation of certain Harmonic Components .
from 21 hours to 23 hours , be laid down graphically , and to the right of them the mean values from hours hours , the ordinate of hours being made coincident wibh that of 24 hours , the curves through the points thus fixed would , if prolonged to meet another , be continuous if there were no non-periodic disturbance .
If they are not thus continuous half the distance between them on the ordinate of hours will be the ntity c The may be obtained by calculation , perhaps with trouble .
Assuming that the third differences of the quantities ( u ) will vanish , when freed from the non-periodical variation ; and that the corrected series of observed values immediately before and after midnight is or , ; and \mdash ; It may also be noticed that when a series of mean values for several days is being dealt with , the quantity will be the difference between the first observed value amd the value for the hour immedintely the last of the series , divided by the number of complets in the series .
Computation for a Yearly Period .
9 .
The computation of the harmonic ccmponents of a yearly series daily values , or of 73 five-day mean values , such as are now usually calculated for the pl.incipal meteorological elements , would extremely laborious if conducted in the usual manner , and it is fore desirable to contrive some approximate method which shall not involve excessive arithmetical operations .
For meteorological purposes the following mode of procedure will , it is believed , supply what is needed , and it may possibly be employed conveniently in other cases .
When the number of the terms of a series is exactly divisible by 2 , 4 , 6 , and 8 , it will be readily seen that the new method of computation proposed in the case of a daily series of 24 hourly terms , will be applicable in its general form ; all that is necessary by way of modification being the introduction of suitable changes in the combination of the terms , and the adoption of other multipliers in place of those given in the equations ( 3 ) .
The series of five-day means for a year consishs of 73 terms , and the above-mentioned condition is therefore not directly complied with by it .
It may , however , be easily transformed by interpolation iuto Lieut.-General Strachey .
On the a series of 72 ierms , which comply with the nditions , in the following manner:\mdash ; If the original series of the 73 five-day means be represented by the numbers ( 0 ) , ( 1 ) , ( 2 ) , ( 3 ) , \amp ; c. , . . . .
( 72 ) , it will be apparent that the first , rm ( 0 ) Yesponds with noon of the 3rd of the year , the subsequent each other at five-day intervals .
The new interpolated series of 72 terms should begin at hours of the 1st day of the year , and will be designated by the numbers , 1 , 2 , 3 , \amp ; c. , . . . .
71 .
Then if and represent , respectively , the mean for 24 hours of the first day of the year under computation , and the last day of the poeceding year , the interpolated series will be\mdash ; , , , .
Moreover , for a series of 72 terms the valus of in the fundamental equation will be ; and designating the sums of the and sines , respectively , of the successive multiples of between and by , and , and applying a similar notation to the quantities before designat , ed by the series , we shall have\mdash ; to [ siu to ] to , to to , to to , to , to to to to Computation of certain Harmonic Componen ts .
Substituting for the quantities , their values in terms of the series of 73 five-day means , we obtain after reduction the values of coefficients .
In the reducsd expressions the following symbols are employed:\mdash ; to ( 5 ) to [ ( 0 ) to ( 5 ) ] [ ( 6 ) to ( 8 ) ] ( 66 ) Corresponding difference ( 11 ) [ ( 61 ) to ( 63 ) ] , , , , [ ( 12 ) to ( 17 ) ] to ( 60 ) , , , , [ ( 18 ) to ] to ( 54 ) , , , , to ( 26 ) ( 48 ) , , , , [ ( 27 ) to ( 29 ) ] to ( 45 ) , , , , to ( 35 ) [ ( 36 ) to ( 42 ) ] , , , , to ( 3 ) [ ( 69 ) to ( 72 ) ] to ( 12 ) to ( 67 ) [ ( 14 ) to ( 21 ) ] to ( 58 ) [ ( 23 ) to ( 30 ) ] to ( 49 ) to ( 35 ) to ( 40 ) . .
And let then to to , . .
, , 72 Lieut.-General Strachey .
On the , The reciprocals of the numerical coefficients of the several 's and 's in the above equations will be the multipliers of the quantities represented by the right-hand portions of those equations , which will be obtained by suitable combinations of the five-day means .
The multipliers are\mdash ; For and , , and , , , and , , .
A table calculated to give multiples of the quantity evidently suffice for the other coefficients , with the small corrections shown above .
To provide for cases ( which are likely to be frequent ) in which the series of terms is not truly recurrent , corrections similar to those equired for the 24-term series mnst be applied .
As before , representing by and respectively the mean values for the first day of the period under computation , and the last day of the preceding year , and by and respectively the mean values for the first day of the following year and the last day of the year under computation ; the quantity that must be added bo the first term of the series of 72 terms , corresponding to the initial midnight of the year , to make it equal to the first term of the next yearly series , will be , and it will be found that the corrections for the several coefficients will be as follows:\mdash ; For . . . .
, , \ldquo ; .
For , , For , or the mean of the whole serics , . . . .
A similar method of computation might be adopted in the case of Computation of eertain Harmonic Components .
a yearly period of 365 days , which could be transformed by interpolation into a series of 360 days , each term of which would correspond to days days .
But the irregularities of a series of daily quantities , in meteorological discussions , would be so great , even when dealing with the mean of many years , that no practical advantage would be obtained by employing such a series ; the computations would be more troublesome , and the five-day means will be preferable .
10 .
The annexed Forms are proposed as supplying methods of computation from the formulae contained in the foregoing discussion , which shall involve the least practicable quantity of arithmetical operations .
Form 1 differs little from what is believed to be the ordinary method at present .
Ib requires tables of multiples of the sines of and Form 2 is for the proposed new mothod of computation .
It requires tables of multiples of the special multipliers applicabls to the several orders of harmonic coefficients .
Form 3 requires no comment except to mention that the value of the angles which correspond to the hours of maximum will facilitate the graphic representation of the several components , and appear to characterise them better than angles T. Forms 4 and 5 call for no special remark .
But they indicate the degree of divergence that exists between the approximate values of coefficients obtained from them , and those got by the more exact methods .
The figures employed are the mean hourly temperatures for one year for the month of June , at Greenwich .
If the mean values for a series of years had been dealt with , the results would not have differed one with another by so much as one-tenth of a degree Fahrenheit .
Form 6 shows the method of computation from 73 five-day mean values .
The figures employed are mean temperatures at sberg for 24 years ( taken from Bessel 's paper , the translation of which is given in the ' Quarterly Weather Report ' of the Meteorological Office , Parb Iy , 1870 , page [ 29 ] ) , which were selected in order to test the agreement of the ] with those given by Bessel , calculated by the method of least squares .
The values of the coofficients by the two methods are virtnally identical with the exception of , in which I feel satisfied that soms error has been committed in Bessel 's calculation ; to verify this , however , would involve a very tedious computation which I have not thought it worth while to undertake .
It should be observed that Bessel 's coefficients , being calculated with reference to a period commencing with the first of the five-day means , that is noon on the 3rd day of the year , have to be modified to adjust 74 Lieut.-General Straohey .
On the them to a period commencing with the initial midnight of the year .
In order to render the form complete , as a type of the proposed system of computation , values for the first and last days of the year have been interpolated , and a small correotion has been assumed to bs quired ; but these have no practical effect on the numerical results .
For this computabion a tabls of multiples of the multiplier ( m ) is requlred , and to save trotlbls a table of multiples of has also been drawn out .
NOTE .
\mdash ; The Tables to will be printed by the ological Office .
Computation of certain Harmonic Components .
Lieut.-General Strachey .
On the on monic Ific c of the ] st Toul of -Day Icans .
Computation of certain Harmonic Components .
VO1 .
ILII .

