]\gt ; On a Method of Tracing On a Method of Tracing Periodicities in a Series of Observations when the Periods are unknown By VINAYEK NARAYEU NENE , First Assistant , Government Observatory , Colaba , Bombay .
Communicated by C. CHAMBERS , F.R.S. , Superintendent of the Observatory .
Received May 30 .
Read June 1 1882 .
1 .
The object of this paper is to investigate a new method which will us to find out the period of an unknown inequality , and thence the inequality itself , from a series of observations\mdash ; such as magnetical , meteorological , or astronomical\mdash ; when they are of a periodic nature .
2 .
We assume that there is some large complex pel'iod , such that the observations\mdash ; which need not necessarily correspond to equidistant times , but must have the time similarly divided in each general period\mdash ; will be exactly repeated after such full period ; but since we have not yet found such a period , although we observations of thirty years ' duration or more , say , of barometer , we may say that such a ) riod is not yet completed , we cannot , therefore , apply Bessel 's method , which is applicable to only known periods , to find subordinate periodic inequalities .
Diurnal and annual periods which have already been discovered may be considered as subordinate periods of the large period which we are considering .
Our aim is first to ascertain the duration of subordinate periods one by one , and then to find ont the amplitude for each period .
3 .
Any observations of a periodic natnre may be represented by the equation\mdash ; , \amp ; c. . .
( 1 ) , in which , \amp ; c. , which are all unknown periods which are in an ascending order of magnitude , but the relations of with each other also unknown ; is the enomenon , and the varial ) time on which the phenomenon dopends ; , and , \amp ; c. , are constants ; and is of course equivalent to 4 .
The form given in Bessel 's paper , embodied in the " " Quarterly Weather Report of the MeteoroIogical Department Part , October to December , 1870 , is\mdash ; , \amp ; c. , which form can also be written\mdash ; * Practically we shall deal only observations that subdivide the lnrge period spoken of into intervals .
362 Mr. .
N. Nene .
On a Method of Tracing , \amp ; c. in which represents a period , after the completion of which phenomenon constantly recurs , and is the variable quantity which the phenomenon depends .
The given condition is that remains constant for values of , which are increased or by , \amp ; c. , , and , \amp ; c. , being constants , and 21 ' equivalent to 5 .
Comparing the form of equation 1 ) and ( 2 ) it will be seen that they are both alike , except in subordinate periods which are in equation ( 1 ) in an ascending order of magnitude , and apparently unconnected with each other ; and in equation ( 2 ) they are in descending order of magnitude , being connected with a proportion 1 , , \amp ; c. , in order .
If we suppose that the least common multiple of the periods , \amp ; c. , in equation ( 1 ) is the same as in equatiori ( 2 ) , then the two equations are identical except as to the inversion of the order of the terms , and in that some of the terms of ( 2 ) whose coefficients are zero will not appear in equation ( 1 ) .
6 .
It might be asked what reasons make us give preference to the form of equation ( 1 ) instead of to the form of equation ( 2 ) ; the reply is that the form of equation ( 2 ) includes all the subordinate periods that are possible whether they may really exist in a particular series of observations or not , and it is therefore confounding and complicated on account of introducing unnecessary terms in th6 equation , whereas the form of equation ( 1 ) contains only all the ullknown subordinate peri , that is , those which may really exist in a particular series of observations .
To illustrate practically the advantage of the one over the other in point of simplicity , we shall take a case in which a particular set of observations is composed of subordinate periods of 5 , 7 , and 15 days only .
This can be represented by the form of equation ( 1 ) by three terms only , while by the form of equation ( 2 ) 105 are required , although it will prove , after all the periods are found , that all of them have zero for their value except the terms whose poriods are of the full period .
On the hand , if a particuIar set of observations is really composed of subordinate periods 105 in number , , those whose periods are 105 , , \amp ; c. , the form of equation ( 1 ) will still hold good .
It really does not affect the forms but the ideas only .
Having found the equation for any kind of observations of a periodio nature , our next step is to see what results take place if certain operations be performed on the observed values of equation ( 1 ) .
7 .
Before doing this let us prove some propositions .
It has already been established in chapters on the subject of Periodicities when the Periods are iknown .
in Trigonometry that the sum of terms of the sines of hich are in arithmetical ression from the first term of the .
( 3 ) written below is ' and that beginat the second term it is , and so on .
he object of the following investigation is to find what relations exist between the series ( 3 ) and the series del.ived from il , in the manner that we are going to describe below .
Let the proposed series be ( 3 ) , from which we see that the mean of terms from the firs tcrm of the .
series is , and that the mean of terms from the second term of the series is\mdash ; ' or , by , we obtain and similarly the mean of terms from the third term of the scries is , and so on .
Thus the mean series will be\mdash ; , \amp ; c. ( 4 .
) Mr. .
N. Nene .
On a Method of Tracing Again , from this new series we see that the mean of the two from the first term of the mean series is that the mean of the two terms from the second term of the series is , and similarly the mean of two terms from the third berm of the mean series is and so on .
Thus the second mean series will be\mdash ; ( 5 ) .
Now in these results we see that the following laws hold\mdash ; I. When is an odti number the first factols of the terms of the first mean series are the same terms in order as the term of the proposed series .
II .
When is an even number the first factors of the terms of the second mean series are the same terms in order as the term of the proposed series .
III .
The second factor in each term of the first or second mean series is a common factor to its own series .
IV .
Each first factor of the terms of the first or second mean series is the middle term of the proposed series from which the corresponding mean term is derived .
Thus the investigation established , that the first meafi series Periodicities when the Periods are 361- ) with the proposed series term when is odd number , and that the second mean series is proportional with proposed series from term when is an even number .
8 .
Suppose in the proposed series that ; then since the mean of the sines of the proposed series will be zero , and consequently the first mean series or the second mean series will also be zero .
9 .
The factor or is a proper fraction .
Put , thus the first factor becomes Now , and ; and , similarJy , A A. A A. A A A A. Thus A. . .
A The number of terms on the right hand side of the equation is , and the numerical sum rdless of the ternls in the bracket is obviously less than , thus the numerical value of is less than the numerical valuc of ; and since A is numericalIy less than unit the other factor is also a proper fraction .
10 .
The owing conditions will without proof when the first and the second factors respectively will ve the positive or negalive sign : The first factor is positive when lies between and , and lies between and , or when lies between and , and lies between and . .
N. NeIle .
On a Method of Tracing The second factor is positive when lies between , and lies between and , or when lies between and , and lies between and .
The first factor is negative when lies between , and lies between and , or when lies between and , and lies between and , The second factor is negative when lies between , and ] between and , or when between and , and lies between , where and stand for zero or any integral number .
11 .
To trace the in the factor or as angle varies between and , between and , between and , \amp ; c. We shall discuss this subject in a particular case only , , when , as there is no necessity of discussing all the for our future purpose .
We have already shown that when , or when factor has the value zero , and it is easy to see that when or or , where is some whole number , the factor has also the value zero .
We have now to find what value of makes the factor a maximum when .
Put and therefore , thus the first factor becomes\mdash ; : ' remove constant factors , thus then If we put we obtain , therefore , and when Periodicities when the Periods are unknown .
il Thus when is negative .
Therefore when has a maximum value .
Thus we see that the angle changes from to , the factor diminishes from the value unity to zero .
We shall next see when we shall have a minimum value of the when changes from to Put , and consequently ; thus the first factor , remove constant factors , ' If we put we obtain , therefore Therefore when plus something ; therefore when has a minimum value .
Thus when we have a minimum value , and similarly VOL. XXXVI .
2 V. N. Nene .
On a Method of Tracing we can sbow that when we shall have maximum , and when shall have another minimum , and so on alternately ; for the tion from which these values are to be determined is where is zero or some integral number .
We have taken the first factor only in dealing with the above ject , bnt it will be seen by inspection that the same results have followed if the second factor had been taken .
These changes of factors have been worked out for a particular and curved in fig. 1 ( Plate 4 ) to give the reader a general idea them .
12 .
Now we shall repeat the equation ( 1 ) and proceed as follows : Let be a small period or interval say of an hour , a day , or a month , \amp ; c. , as the case may be , at which observations are able , then riting for ; for for , \amp ; c. , in the above equation ( and appropriately modifying the constants , \amp ; c. , and are reckoned from different epochs ) , the successive tions will be equal to the sum of the terms of the following sions\mdash ; 1st obs .
2nd obs .
\amp ; c. obs .
4th obs .
In order to save room we shall henceforth use the factor \amp ; c. , in place of , \amp ; c. , respectively .
13 .
Suppose we have a table ruled with horizontal and vertical lines ; in the first horizontal line Jet us enter the numerical values of Periodicities when the Periods are ?
mknown .
in order .
Now let successive means be taken of obser( an odd integer ) advancing to the right by one step each operation , and entering the respective means under the ' \amp ; c. , terms of the observations , thus forming a second our table ; the means in our second line taken in order will equal to the sum of the terms of the following expressions\mdash ; 1st mean , \amp ; c. 2nd mean , \amp ; c. 3rd mean , \amp ; c. , \amp ; c. If these means which are entered be subtracted from the observations column by columu , then the remain will severally be equal to the sum of the terms of the following expressions\mdash ; 1st remainder , \amp ; c. 2nd remainder , \amp ; c 3rd remainder Mr. .
N. Nene .
On a Method of \amp ; c. Let these remainders be entered in the third line of our table .
Again , if successive means be taken of of these precisely in the same manner as the means of the observations whioh were taken before , we shall have a set of second means , and similarly if these means be subtracted from the first set of remainders we shall have a second set of remainders ; and , further , if these operations continued times we shall have an rth set of means and an set remainders , and the set of means taken in order will be equal the snm of the terms of the following expressions*\mdash ; 1st mean 2nd mean \amp ; 0 .
3rd mean \amp ; 0 .
\amp ; 0 .
And similarly each of the remainders will be represented to be equal to the sum of the terms of the following expressions\mdash ; * See two next paragraphs .
Periodicities when the Periods are remainder , \amp ; c. remainder , \amp ; c. remainder , \amp ; c. , \amp ; c. Pursning the mode of entry adopted for the first mealls and first remainders , the means will occupy the line of our table and the remainders the line .
In future we shall for shortness write\mdash ; ; , \amp ; c. 14 .
The following calculations are made , ready reference , and to show how the factors of the form and which occurred in our means and remainders are obtained\mdash ; The factor for\mdash ; 1st means 1st remainders 2nd means 2nd remainders .
V. N. Nene .
On a Method of Tracing 3rd means 3rd remainders means remainders 15 .
We mnst notice that at each operation of taking means we lose values at each end of the series , thus the mean series commences in our form after observations .
For this reason supl ) strong vertical lines are drawn in our form at the observation from each end , and let us confine our attention hereafter to the series embraced between the two strong vertical lines .
16 .
Then , of the numbers occupying the first vertical column from the first of the strong vertical lines mentioned above , the following , , an observation , an individual of the first set of mearls , an indi .
vidual of the second set of means , an individual of the third set of means , \amp ; c. , an individual of the set of means , and an individual of the set of remainders , will have for their mathematical equivalents the following expressions , taken in order , viz. , 1st observation from the observation series , \amp ; c. 1st mean from the first set of means \amp ; c. 1st mean from the second set of means , \amp ; c. 1st mean from the third set of means Periodicities when the are knotvn .
, \amp ; c. mean from the set of means , \amp ; c. remainder from set of remainder and similarly the corresponding numbers from the second vertical column will be representsd by the following expressions\mdash ; 2nd observation from the observation series , \amp ; c. 2nd mean from the first set of means , \amp ; c. 2nd mean from the second set of means , \amp ; c. 2nd mean from the third set of means , \amp ; c. 2nd mean from the rth set of means , \amp ; c. 2nd remainder from the set of remainders , \amp ; c. And the corresponding numbers from the vertical column will be represented by the following expressions\mdash ; observation from the observation series , \amp ; c. mean from the first set of means , \amp ; c. Mr. .
N. Nene .
On a Method of mean from the second set of means , \amp ; mean from the third set of means , \amp ; c. mean from set of means , \amp ; c. remainder from the rth set of remainders , \amp ; c. And therefore if we use a variable integral number ( which may be zero ) as a multiple to the angle of the type , 17s can make one general expression serve to represent the numbers of any one horizontal line ; this will be for the 1st , 3rd , 5th , \amp ; c. , lines as follows:\mdash ; 1st line , observation series , \amp ; c. .3rd line , 1st mean series , \amp ; c. 5th line , 2nd mean series , \amp ; c. th line , 3rd mean series , \amp ; c. line , mean series , \amp ; c. line , series of remainders , \amp ; c. Where when is put , 1 , 2 , \amp ; c. , successively the expression in Periodicities the Periods are unknown .
case gives the vaIue of the first , second , third , \amp ; c. , numbers to the right of the firsb strong vertical line .
17 .
We shall now take the sum of all the mean series and put the by one series only , thus the sum of all the mean series is equal , \amp ; c. Suppose therefore Hence by subtraction by division 18 .
From this result we can write the sum of all the mean series\mdash ; , \amp ; c. Thus from one series of observations we get two series , one is the sum of several mean series , and the other the series of last remainders , and of course the sum of these series is equal to the observation series ; thus whenever we want the sum of all the mean series we shall get it by subtracting the last remaining series from the corresponding observation series .
We shall rewrite series below and call them series , mean series , and series respectively .
Observation serie , \amp ; c. ( 6 ) .
V. N. Nene .
On a Method of Mean series , \amp ; c. Remaining series , \amp ; c. ( 8 ) .
Throughout these three expressions a particular value of correspond to a particular vertical column , the next greater value to the next vertical column , and so on .
19 .
We have up to this point derived these last two results as general with respect to the factors of the type whioh depend on the value of , and of this we have only said that it is an odd whole number .
Onr next step is to see what conclusions we shall arrive at when wa .apply some particular value to ( 1 .
) Suppose that mc is less than , therefore it is also less than eaaeh of \amp ; c. , and \amp ; c. , are each less than 2 ; thus the factors , \amp ; c. , or their equivalents , \amp ; c. , are positive .
have shown before that each of the above factors is a proper fraction , and consequently each of their equivalents is a proper frac tion ; thus each of , \amp ; c. , is greater than one , and therefore the value of each of the factors , \amp ; c. , when is sufflciently great reduces to zero .
Suppose that mc is equal to tlJerefore it is less than each of , \amp ; c. , then , as before , we can show that is zero , and each of , \amp ; c. , is positive , and the value of or is one , and the value of each of the factors , \amp ; c. , when is sufficiently great , uces to zero .
Thus , when mc is equal to period , and the number of operations of taking successive meanl and remainders are sufficiently performed , the remaining series will be reduced to , or a subordinate series in the observation whose period is ( 3 .
) Suppose that mc is greater than and less than , and Periodicities when the Periods are unknown .
than each of , \amp ; c. , , as before , we can show that is and each of , \amp ; c. , is positive .
Let , then the factor will be equal to , and the value of each of the factors , \amp ; c. , where is sufficiently great , reduces to zero .
Thus , is greater than the period , less than , and less than the period of , and the number of operations of taking means and remainders are sufficiently performed , the reseries reduces to , or a ) series in the observation series whose period is , and the of each term of the series is greater than the corresponding of the original series : since is greater than one , and is large enough , can be made as large as we please .
20 .
Up to this point we have treated the above subject by supposing that is an odd number .
By simple inspection it will be seen that when is an even number the results will be quite similar to the one when is an odd number , but the labour of operations of taking successive means will be twofold .
Because the mean of values from which the mean is taken will not correspond to any of the values in the lot , but will correspond to the middle of some two consecutive values , and for this reason the mean should be written on the line instead of in the column on our form ; and after the operations of these first means are performed , we must take the mean of two consecntive means found out first , at both sides of any particular column , and put it in the column .
Then these second means thus found are to be subtracted from the corresponding values from which the two sets of means are derived .
Similarly from these remainders new means and new remainders are to be found precisely in the same manner as spoken of above .
In order to keep our language the same is an odd or an even number , we shall always , in the case where is an even number , give the name of first set of means , second set of means , \amp ; c. , to those means found by two of the operations described above .
21 .
This suggests that we must always prefer as an odd number , except when there is real need of preferring as an even number , since the labour of operations , when is an even number , is greater than when is an odd number .
It will be seen further on that there is sometimes real need of taking as an even number .
'378 Mr. V. N. Nene .
On a Method of Tracing 22 .
The three results which we have found above are very portant , and we shall now show their utility , and discnss some cases from them which we have not yet discussed , by taking particular examples by which we can show our method of unknown periods .
Suppose we have at our disposal a long series of daily means of barometer observations , and we suspeot in it period of 25 days .
The first thing that we shall do is , that we shallrake a series of daily means and curve it on a curve form , and see the period of duration of turning points between successive or between successive minima .
Suppose the duration of the period was from 5 to 10 days , then we shall take the least period of 5 that is , we shall adopt 5 as our working value of , and work processes as already discussed with the whole series of daily until generally each mean of the last mean series is less than 2 ( say ) the last place of figures .
* We shall take then for further treatment sum of all the mean series instead of the original series .
The sum all the mean series will be found directly by subtracting the remaining series from the first daily mean series .
We shall take series for a second set of operations .
The number we shall take operations is 25 nearly 3 , but we shall take 35 the operations are simpler for odd number than for even number .
Now we shall have 35 as our working value of , by which we shall tinue the operations , say up to 5th , and then curve the 5th series on a curve form .
If there be only one period in the series from about 17 to 35 days , we shall begin to get simple halmonic waves in the fifth or in some further remaining series ; but if there be moro than one period the will be compound , about which we shall discuss prssently .
But first suppose that there is only one period .
If we do not find definitely the duration of simple wave curve in the 5th remaining series , then we shall continue the process on to the 6th remaining series and curve on the same curve form and with the same zero line , until the duration of simple wave curve is definitely found .
It may here be mentioned , that if the period is not of an exact number of days , we should have a long series to begin with , so that we shall take some exact number of waves from the last rema , ining series and find the number of days in them , and divide the days by the number of waves , and the result will be the number of days in the period .
We must also remark that if there should be no always have some limit of accuracy of observing the instruments and ducing the observed values ; for example , we observe barometer to a thousandth of an inch , and also reduce daily means to a nearest thousandth of inch .
By less than two in the last place of figures , iu this case , we mean less two thousandths of an inch of barometer , and so on .
Periodicities when the Periods un known .
harmonic period of exact duration\mdash ; 25 days\mdash ; that we have suspecting and searching for , the process have been perwill still tell us if there be any simple harmonic period of duration between ( say ) 20 and SO days .
The only difference shall have is in these cases that we shall have to increase number of operations to find the remaining series .
In the foltable we shall give a few numerical resnlts of the factors\mdash ; ' and ) The object of introducing the above table is only to give a rough idea to the reader of the rapidity with which the numerical values of these and such similal .
factors increase after a certain number of operations of the remaining series .
23 .
Suppose we have suspected only the period of 25 days , but in reality there is not any period between 5 and 35 days ; the result will be , as investigated before , that the range of remaining series will be Iessened and lessened , until each term of the last remaining series will be less than two in the last place of the figures .
24 .
Suppose in the daily mean series we have a real existence of the two periods ( say ) 25 and 20 days , we shall still find the existence of the period of 25 days .
Because the numerical value ( regardless of sign ) of the factor is greater than the numerical value Mr. .
N. Nene .
On a Method of Tracing ( regardless of sign ) of the factor ( refer to the curve factors ) and consequently can be made as large as we in comparison with when is large enough , where stands for the former factor and stands for the latter factor .
Thns if the compound result in the last remaining series be curved out on a curve form maxima and minima of 25 days ' period be apparent .
Similarly if any number of compound periods be mixed in the last remaining series , still by sufficiently increasing the number of operations the 25 days ' period will be apparent .
In conclusion we can say that among the periods which may have mixed in the emaining series the one of which the factor of the type has the greatest numerical value ( regardless of sign ) will be apparent , whether the period is of 25 days or not .
25 .
The only case in which the period will not be rendered distinctly aprarent is , that when the two periods are such that their factors of the first mean series are equal to one another , and then the two variations will continue combined in the same proportions as in the original observations .
26 .
We have in the above example stated that when we find systematic movements in the daily mean curve , of which the duration ia from 5 to 10 days , we must first work the seriss with five day operations , \amp ; c. It is not , however , essential , when the period which we are seeking is of a considerable range compared with that of the shorter period , then the period be found without it ; but when the range is very small the first process of smoothing is quite essential .
It is nob also essential that we should stop the operations when the limit of the mean series is reduced to less than 2 in the last of figures in the series , but we may stop at any stage of the process after considering what reduction will take place in the range of the period we are seeking .
This is merely a process for the series of smooth-flowing numbers without sensibly reducing the range of the larger periods .
27 .
The process of smoothing series generally in practice is to take the mean of two consecutive numbers , and put it down below and on the line between the two numbers , and so on with each pair of numbers in succession .
Next from this new series the same thing is repeated up to a certain number of opera ions .
One effoct of this process is to reduce the range of each simple harmonic variation of Periodicities when Periods are unknown .
it 81 the original series is composed .
The extent of this reduction range is shown in the following expression for the series that is after repetitions of such smoothing operations:\mdash ; observation series* co \amp ; c. It will be seen at that as the period is smaller and smaller , we have greater and greater reduction and vice 28 .
We shall , however , give a clue when a series contains enormous irregularities , to the distribution of each number amongst the adjacent columns , that is effocted by our two methods of smoothing .
We shall give this by supposing that we have a series whose terms , of which each of the first and last 6 has the value zero , and the 7th has the .
The resuIt of six smoothing operations of the second method gives 1 , 6 , 15 , 20 , 15 , 6 , 1 .
From this we see that the reduction for the middle term which occupies the same column as the number 64 in ths original set is rapid , and the distribution amongst the adjacent terms is smooth-flowing and of a convergent character .
29 .
We shall now apply our first method to a series whose terms are 37 , of which each of the first and last 18 has the value zero , and the 19th has the value 348 .
Suppose we have a number 7 for a working value of to starb the operations , we shall after three operations get a series of 19 terms whose values are .
From this we see also that the reduction is rapid for the middle term , and the distribution amongst the adjacent terms is in smooth-flowing waves which grow smaller and smaller and ultimately vanish .
Method of finding the Amplitude of a Vo , riation of known Period .
30 .
The processes which we are going to apply for this purpose are * This may be easily understood if we put in the factor of the series 4 in paragraph 7 .
When the factor becomes the first operation , and in the same way it becomes , i. e. , for the second operation , thus for operations it becomes .
This may be taken as a type of the factor for each expression of series of the type Mr. .
N. Nene .
On a Mrthod of Tracing slightly diiferent from the preceding ones : we shall , therefore , of all apply them to the general form and then to particular 31 .
Suppose that where is a known period and is integral number .
Then using as working value Jet us find before first mean series from the observation series , the first of remainders , and the second mean series ; but instead of the terms of the second mean series from the corresponding of the first series of remainders to get the second series of Jet us add them in the same order and call this series the difference and sum series .
Again from this series Jet us find , in same manner as stated above , the second , bhe third , \amp ; c. , and difference and sum series .
The difference and sum series as will presently appear , be of the following form\mdash ; , \amp ; c. ( 9 ) .
32 .
The following calculations are made for ready reference and show how the factors of the form , which occurred in our difference sum series , are obtained .
The factor for 1st means The factor for 1st remainders The factor for 2nd means The factor for 1st diiference and mean series .
The factor for 3rd means The factor for 3rd remainders The factor for 4th means Periodicities when the Periods are unknown .
factor for 2nd difference and sum series Thus it can be shown that the factor for difference and sum series will be 33 .
Now the last factors of all the terms in the series ( 9 ) a term of the type , so that all these factors are positive , and , as before , it is easy to see that each of them has a tendency to vanish as increases , except in the cases where the terms of the type have the value zero .
The only terms of the type , whose values will be zero , are those which are formed from the sines of the angJes , \amp ; c. , because these angles ] are some integral multiple of .
Thus when is sufficiently great the series ( 9 ) wiJl be reduced to\mdash ; , \amp ; c. If we curved now the first values in order out of the last horizontal line of our form which forms our difference and sunt series , it will be a variation with its full amplitude the known period , and if the next values be curved in the same way we shall have a repetition of the same curve , and so on .
34 .
Although we have shown that the factors of the form when not zero , and is sufficiently great , reduce to zero , yet the reduction of the factors which are formed for the periods less than mc is so much slower than the factors which are formed for the periods greater than mc , that it is almost impracticable to continue the process to the limit .
We shall therefore in what follows , investigate a particular case of a known period , and we shall find it advantageous to use partly our method and partly the usual method of combining the series with the known period .
35 .
Suppose we have a series of daily means of barometer observations in which we havs detected a period ( say ) of exactly 36 days , by the method already described .
We shall , of all , in the following table , give factors of the kind in series ( 9 ) for different values of and differe1Jt values of the periods of the type K. VOL. XXXVI .
2 .
N. Nene .
On a Method of Tracing In the above table , the periods less than mc of the type which we have selected are those that have maxima or minima values of their first factors of the type so that their factors of the type will have the greatest possible reduction , that is , factors for intermediate periods will reduce at slower rates .
36 .
We also see that when is equal to 5 , we have the factor corresponding to the period 72 days reduced to , and from this we can infer that the factors corresponding to greater periods than 72 days will be reduced to less than .
So that for all practical purposes we shall make the operations 5 , and assume practicalJy that in the fifth difference and sum series we have reduced all series of the periods 72 and greater than 72 days to zero , and retained in their full magnitude all the series of the less than 36 days , whilst the of periods intermediate between 36 and 72 days will be partially obliterated .
Such a series may be represented by , \amp ; c. In the first pair of brackets of the above expression , we have arranged those terms of the simple harmonic series whose periods are less than the period of 36 days , but not any of the period of tho form where is any integral number ; in the second pair of brackets of the expression , we have arranged those terms of Periodicities when the Periods are unknown .
gimple harmonic series whose periods are of the form , that is , it contains all the harmonics of 36 days ; and in the third pair of brackets of the expression , we have arranged those terms of the simple harmonic series whose periods are greater than the period 36 days .
Suppose in the fifth difference and series , or in the last horizontal line of our form , we have values where is some large integral number .
Suppose we have another table ruIed with more than horizontal and 36 vertical lines .
In the first horizontal line Jet us enter 36 numerical values from the commencement of the series in order , in the second horizontal line the next 36 numerical values , and so on , in the Mth horizontal line the last 36 llumerical values .
Let the sum and mean of each vertical column be taken in the and horizontal lines respectively ; then these means are the subject of our presenb investigation .
37 .
The general term of the simple harmonic series in the first horizontal line is\mdash ; ; the general term of the same simple harmonic series in the same vertical column of the secon .
horizontal line is\mdash ; ; the general term of the same simple harmonic series in the same vertical column of the third horizontal line is\mdash ; ; the general term of the same simple harmonic series in the same vertical coJumn of the Mth horizontal lin is\mdash ; ; thus the mean of this vertical column will be\mdash ; then the genelal series may be written\mdash ; Thus the last horizonta , lino of means in our new form , which is composed of 36 values , will be of the form\mdash ; Mr. .
N. Nene .
On a Method of Tracing , \amp ; c. And it can be shown that the last factors of the terms in the first and the third pairs of brackets , where is sufficiently great , reduce to zero ( see curve of factors in fig. 1 ) .
Thus the S6 mean values found in the last horizontal line of our new form will represent the variation of the , 36-day period .
38 .
Next suppose that the period which we have detected is not an exact number of days but ( say ) of days , we shall still take the nearest integral number , , 36 , as our working value of , and by it find the 5th difference and sum series as already described .
In , case the amplitude of the variation of the period of will be slightly reduced in the 5th difference and sum series .
The reduction of the 1st simple harmonic variation that is of the period of days will be ( 1\mdash ; 99954 ) , that of days will be , that of will be , and so on .
The next thing is to show how the sel.ies is to be combined , which we are going to show in the following manner .
39 .
In the first horizontal line of our new form , we shall enter first 36 values .
In the second horizontal line 8 values , from the beginning of the next 37 values , will be entered up to the 8th vertical column , but in the 9th vertical column we shall enter the mean of 9th and values out of the 37 values intended for the second horizonta ] line ; in the rest of the horizontal line the remaining values will be entered .
The object of combining the 9th and 10th values and entering the mean result in the 9th vertical column of the second borizontal line is that , since we are dealing with exact days only , must enter the values in the columns in such a manner , that the Periodicities when the Periods are unknown .
cf distribution must not be greater than half a day , and so the mean of the 9th and 10th values , which was found more appropriate to 9th column than any other column , was entered in the 9th column .
8imilarly up to the horizontal line , the values will be entered by operly combining two values for entering in ons coIumn as many times as need be in the course of entering the series in the manner stated above .
Also the sums and means of each vertical column are laken in the same way as in the preceding example .
We will not go into details , as in the preceding example , but can simply say that the last numbers of our new form will be the iation of the period of days with its almost full magnitude , and for 36 points , the intervals between which from point to point will be days .
40 .
We should remark that the process of difference and sum series has also an obvious in separating the observation series into any number of difference and sum series without affecting the series by the process .
By affecting the eeries by the process , we mean that in the former process for the series and mean series , if we take any arbitrary mber as our working value of , and if there should be periodicities lying between and , we shall get the psriodicities mixed up in both the series with a greater amplitude than the original series .
separations are very useful for the first trials when suspecting approximate periods .
Appbication of the Process anl Detection of Period of about Months .
41 .
For this purpose we have taken means of barometric pressure at Bombay , from 1847 to 1872 and 1873 to 1878 , thus forming our Table I. The part of monthly means is also given in Table I , page 11 , of the volume entitled " " The Meteorology of ths Bombay Presidency by Charles Chambers , ] ] , F.R.S. , Superintendent of the Colaba Observatory , Bombay ; and the last part at the foot of Tables I to , pages 58 to 63 of the " " Bombay Magnetioal and MeteoroIogical Observations , 18 to 1878 The description of the barometer and mode of observation , \amp ; c. , are fully given in both the volumes .
With the numbers in Table I , we have performed one operation to get the first difference and sum series by taking 12 as our working value of , and the result is entered in Table II .
numbers in Table II are subtracted from the correspouding numbers in Table 1 and entered in Tabl ' , III .
At the foot of Table II we have entered the mean for January in the 24 years , 1848 to 1871 .
the mean .
February in the 24 years , the mean for March , \amp ; c. , \amp ; c. , up to December .
* These means form the annual variation of barometric speaking , we should have taken the means of thirty years , 1848 to 1877 , instead of 1848 to 1871 ; but first of all we have taken for our discussion the 388 Mr. .
N. Nene .
On a Method of Tracing pressure for the 24 years .
If we add the constant , the mean of 26 years , 1847 to 1872 , to the 12 numbers in our Table then these numbers will be found almost identical with the at the foot of Table I , given in the Meteorology of the Presidency .
The 12 numbers at the foot of Table II were snbtracted in order from each of the rows of 12 numbers from January , and ending with December for each of the years to 1877 of the same table , and the remainders entered in Table Again , with the numbers in Table , we have performed operations , taking 3 as our working value of to get the difference and sum series , which we have given in Table , then numbers in Table were subtracted from the corresponding in Table , and entered in Table VI .
We need not say that the snm of the numbers iu Tables III , and VI , from July , 1848 , to June , 1876 , and successive repetitions corresponding numbers of annual variation at the foot of Table II , taken in order , will be equal to the corresponding numbers in Table I. The numbers in Tables and are curved in figs. 2 and 3 tively .
We presume that the curve of fig. 2 is composed of periods of more than 12 months , and that of fig. 3 of less than 12 months and greater than three months .
By looking at the oscillations of of fig. 2 we suspect that the periods there are of 11 or 12 years , years , and 2 years .
By looking at the of curve of fig. 3 , we see that amongst the periods there is a predominance of a !
) -monthly period .
This we have ascertained the following manmer:\mdash ; We ha first of all marked by a ( thus ) all the minima that have their time-intelval from one minimum to next minimum 8 to , or any integral multiple of 8 to 11 months , and then filled up the rest 8 to 11-monthly minima that were less marked by a mark ( thus ) .
It is remarkable that the oscillations are more prominent and of greater amplitude from the year 1849 ] than from the year 1863 to 1876 .
If we count ths marks of minima commencing from July , 1848 , and ending with November , 1876 , they will be found to be 37 ; thus there are 36 periods in 340 months , i.e. , months , as an average duration for the period .
42 .
Again , we find that there are intermediate minima which are of less amplitude than already stated , and the duration of which is also from 8 to 11 months .
These also we have marked by marks and ) , as in the preceding example .
In the present oase both these series from 1847 to 1872 , and have made all the operations we are going to describe .
But afterwards it was thought proper to add to tha above result the six years 187$ to 1878 , which were availabJe .
But as the work was far advanced it was worth while to go into the trouble of doing the work over again for the sake including those six years .
Periodicities when the Periods are unknown .
are made on the upper side of the curve , while in the preceding on the lower side of it , to distinguish them from each other .
43 .
We find from May , 1858 , to October , 1866 , that the occurrences of these minima have disappeared .
From December , 1848 , to day , 1858 , there are 12 periods , that is months for a period , and from October , 1866 , to April , 1877 , there are periods , i.e. , months for a period .
this we might guess that to form such a curve there must be two prominent periods , which may be as ( say ) 5 and months .
44 .
To clear this doubt , we have further taken the numbers in Table , and performed five operations to get the fifth differenee sum series by taking 5 as our working value of .
These numbers we have given in Table and curved in fig. 4 .
Then the numbers in Table were subtracted from the corresponding numbers in Table VI and entered in Table VIII , and curved in dotted lines with the corresponding curve of fig. 3 , and with the same scale and on the same zero line .
Now , by looking at the oscillations of the dotted curve , it appears that it is of a simple harmonic character , and it confirms the -monbhly period as already described .
But in both these curves ( continuous and dotted ) there are a few exceptions to confirmation of the -monthly period .
These exceptional periods are\mdash ; from March , 1857 , to December , 1859 ; Jnne , 1863 , to September , 1866 ; April , 1868 , to January , 1871 ; and November , 1872 , to the end .
45 .
We shall now turn to fig. 4 .
By looking at the oscillations of this curve we again see that amongst the periods there is a great predominance of a -monthly period .
This we have ascertained as in the of -monthly period .
If we count all the marks made on the curve , we have 65 ; thus there are 64 periods in 302 months June , 1850 , to August , 187.5 ) , and , i.e. , nearly months , * which is an average duration for the period .
We should now here that , although in the continuous curve of fig. 3 we found that the occurrences of the subordinate minima have disappeared irom May , 1858 , to October , 1866 , still they in the -monthly period curve of fig. 4 .
Application of the ilTethod to find the Duration of an exact Period .
46 .
The had taken a small series of means of barometer observations , derived from hourly tabulations of barograph , rected to standard barometer No. 58 , from 1st December , 1875 , to December , 1876 , for trial , when the method of detecting any * The Yiter found worth noticing that the period of months is most an exact multiple of the -day period which he has found further on in 50 .
Mr. .
N. Nene .
On a Method of mple harmonic waves of oximately known period was in infancy .
The aim was to see whether on barometer observations was any influence of lunar period , the duration of which is 30 days .
The trial which we are going to describe was made in earlier period of 1878 , when the raph tabulation was under con .
sideration .
Daily means of ograph tabulations , corrected standard barometer No. 58 , for this same period , may be found in " " Bombay Magnetical and Observations , 1871 to 1878 pp. 60 and 61 .
The difference between the daily means , corrected uucorrected to standard instrument , is very small , and not from day to day more than ( say ) about two or three thousandths an inch .
The treatment which was given to these daily means the results which followed from it may be considered satisfactory .
47 .
In order to get the series in smooth-flowing numbers , we have taken 15 as our working value of , and found first mean ssries .
Then this series has been taken for further operation , and by taking 45 our working value of , we have found first , second , \amp ; c. , up to fifth series of remainders .
The first , third , .
and fifth series of remainders ars curved in fig. 5 in plain , dotted , and interrupted lines respectively , with the same scale and on the same line .
Looking at the plain curve , we see that there nine turnmg-points of , the dura tion thereof being moderately uniform .
If we take the first minimum in the middle of the small gap that occurs in the curve , we have in 298 days eight periods .
Thus an average duration of one period is about 37 days .
Again , if we count the days from the second minimum , we have days in seven waves , and thus an average duration is still about 37 days .
Thus , instead of the lunar period of 30 days , we a 37-day period .
From this it is clear that if there be lunar riod in the series at all , the amplitude of it must be much smaller than the 37-day period .
48 .
The object of drawing the three curves on the same zero line and with the same scale is to show graphically the process of swelling the waves as the operations iucrease and of their appearance , as in the case of the third wave , which disappears in the first series of remainders .
The object is also to impress the necessity of making smooth-flowing numbers before taking them for opel.ation ; as in the case of second , third , fifth , and enth waves , all the irregular movements in the series of remainders are found to appear in the third and fifth series of remainders .
With these remarks we conclude without giving the numbers from which the present curves are formed .
Second 49 .
In order to test the reality of the period of about 37 days on a large scale , we have taken a series of weekly means of barometsr observations at Bombay , commencing from 1st January , 1848 , and Periodicities when the are unknown .
with the 2nd January , 1852 , as the year 1848 was the first year when the system of hourly observations began interruption at the Colaba Observatory .
The weekly means chosen in place of daily means two reasons , first that as observations were made on Sundays and a few holidays , no daily means were available during the period ; second , to comparatively the labour of calculations of the processes .
On occasions when there were atmospheric disturbances , or for similar reasons , observations were made on Sundays and none on Mondays .
For this and similar other reasons the weekly means were calculated from all the daily means from Sunday to Saturday .
These means are given in Table IX , and curved in fig. 6 in an interrupted line .
With these weekly means we have performed 5 operations to get the fifth series of remainders .
The numbers in the first and fifth series of remainders are also given in Tables X and XI respectively , and curved with the interrupted line of fig. 6 , and with the same scale in dotted and contiIluous lines respectively .
50 .
Looking at the continuous , it will be seen that from 24th week of 1848 to 8th week of 1849 , we have 7 waves of simple harmonic character , and equidistaut in duration .
Thus there are 36 weeks , or days in 7 periods , so that the average duration of a period is exactly 36 days .
Again , from 8th week of 1849 to 24th week of 1850 there are 13 waves , also almost equidistant in duration .
Thus there are 68 weeks , or days in 13 periods , so that the average duration of a period is nearly days .
Again , from 24th week of 1850 to 34th week of 1851 there are , we presume , 12 periods .
It will be observed that out of these there are 7 periods , , 1st and 7th to 12th , whose waves are quite prominent ; with regard to the remaining 5 periods their waves seem to be almost obscured by irregularities .
Thus there are 63 weeks , or days , in 12 periods , so that the average duration of a period is nearly days .
If we now add all the weeks and periods together , to et the mean duration of the period we have 167 weeks and 32 periods , efore the average duration of bhe period is nearly days .
51 .
We should hers remark that as we are dealing with weekly values , there is a possibility of an error , either too much or too little , of one week in the whole period of 167 weeks .
If we take 166 weeks for 32 periods , we get an average duration of days ; and if we take 168 weeks , we get days .
Thus the period we have found out is approximate to the extent above mentioned .
We shall , therefore , say that the period is approximately days instead of days .
52 .
The object of drawing three curves of weekly means on the same zero line , and with the same scale , is the same as in the preceding example of the curves of daily means .
The formation into regulal Mr. .
N. Nene .
On a Method of Tracing waves of five weekly period of the continuous curve from the corresponding to the interrupted curve for the periods from week of 1848 , and 13th to week of 1849 is worth It should also ) noticed that the irregular movements that the terrupted curve for the period from week of 1850 to week of 1851 , also exist in the continuous line curve .
53 .
Let us now combine the result by taking .
days as an age duration of the period to get an average range or amplitude of the period , as shown in Table showing the variations of -day period for several of periods and for the whole period .
These results are curved in figures 7 to 11 respectively .
It must be mentioned here that the ranges which we see in these results are not exact , but enlarged by our process and nearly , is the to which our ] are enlarged .
But we must also notice that the facts that the period is not an integral number of weeks , that our method of combination is in consequence rather rough , and that the number of points in the is too small , tend to reduce the range of the period .
If the process had been applied to the continuous daily means instead of to weekly means the results would have been of a larger and equal amplitude of several groups of the period .
54 .
The writer feels himself under great obligabion to his superior , C. Chambers , Esq. , F.R.S. , Superintendent of the Bombay Observa tory , for the warm interest he took in carefully reading this tract , and in giving here and there some very valuable suggestions .

