]\gt ; Bimodular Afethod of Computing Logarithms .
On a Method of Destroying the Effects of slight Errors of Adjustment in Experiments of Changes of Refrangibility due to Relative Motions in the Line of Sight By E. J. STONE , F.R.S. , Director of the Radcliffe Observatory , Oxford .
Received January 17 , 1881 .
Let arrange11lents be made for the reversion of the prisms without any disturbance of the other optical arrangements , including , of course , the position of the cylindrical , if one be used .
Any slight errors of adjustment which prevent the light from the star and the comparison light from falling upon the train of prisms under the same optical circumstances , so far as mere direction is concerned , will have opposite effiects in the reversed positions of the prisms ; but the separation of the emergent lights due to relative motion will remain unchanged by the reversal of the positions of the prisms .
If , therefore , the apparent change of refrangibility due to relative motion remains unchanged by the reYersion of the prisms , all doubts about the effects of errors of adjustment will be removed .
But if the results in the reversed positions of the prisms sensibly differ , then the existing errors of adjustment must be removed , or their effects allowed for by taking a mean of the results in reversed positions , before any reliance can be fairly placed upon the determination of relative motions in the line of sight .
A reversible spectroscope was arranged by me , and made ) Mr. Simms , some ago , but I have never since had an equatoreal , with a good driving clock , under my control with which the experimemt indicated could be properly tried .
the direct prisms in use , the required reversion can be easily arranged .
I am not likely , for some time , to have the use of a good equatoreal , and I , therefore , publish the plan with the hope that some one more fortunately situated may give it a fair trial .
Tho experiment is a crucial one , and , in my opinion , should be tried .
Iy .
" " On an Improved Biraodular thod of computing Natural and Tabular Logarithms and arithms to Twelve or Sixteen Places , with very brief Tables By ALEXANDIDR J. ELLIS , B.A. , P.R.S. , .
Received January 17 , 1881 .
SECTION ruRE OF THE IODULAR METHOD AND 1TS ImPROVEhIIlNT .
The Bimodulus is a constant , which is exactly double of the modulus of any system of logarithms .
The is derived from 382 Mr. A. J. Ellis .
lmproved Bimodular Method of the familiar proposition that , when the difference of two numberr is small , the difference of their logarithms is nearly equal to tho bimodulus mnltiplied by the difference and divided by the sum of the numbers .
themselves .
The improvement here for the first time effected , prefixing a brief preparation , which makes bhe method universally applicable , and subjoining an easy correcbion depending on the ransformation of a well-known series proceeding by the odd powers of the difference divided by the sum of two numbers , the number of places obtained is greatly increased .
This method is here applied for finding the natural and tabular Iogarithms of any number to twelve places of figures by means of a table of pages for each kind of logarithm , and to sixteen places by help of a seven .
place table of .
or Briggs 's Iogarithms .
An extremely simple rule , which , so far as I know , was never before imagined , enables us to pass from the logarithm to the number , that is , to find antilogarithms from the same tables .
Although the method is applicable to any system of logarithms , and was actually first applied by me to the direct calculation of musical logarithms to the bases 2 ( octave ) , ( equal semitone ) , and ( comma ) , and appropriate tables have been constructed , I confine myseIf for brevity to natural and tabular logarithms .
The tables are constructed from existing materials , but the method is capable of constructing them independently .
SECTION \mdash ; PRINCIPLES 0F THE BIMODULAR METHOD AND 1TS Relations.\mdash ; Let and be any whole numbers of which is the smaller , and let , a proper fraction .
Let not .
Iog , and lo . . .
( 1 ) where is the modulus , and hence the bimodulus to any unspecified system of arithms marked by .
Let , 2MqMxoslash ( 2 ) .
In future and will often be called " " the numbers " " the tabulal .
number " " the difference " " the dividend ' ' the sum \ldquo ; or " " divisor and " " the quotient Now it is familiarly known that . . . . .
( 3 ) , . . . . . .
( 4 ) .
Putting in ( 4 ) the values of in terms of and from ( 2 ) we have computing Natural and Tabular Logarithms , ( 6 ) .
And putting for and their values from ( 2 ) we find , whence ( 7 ) , And by expanding the first of these equations ( 7 ) . . . . .
( 8 ) .
Subtracting ( 8 ) from ( 3 ) , and multiplying by to find the of ( 6 ) , we have ( 9 ) , a converging series of which the limits are the first term and the first two terms .
.\mdash ; To insure being small in all cases , I have invented the rnle of preparation , nded on the fact that if be the number whose logarithm is sought , and and any two numbers of which the Iogarithms are known , such that , where is the next less number to in the table , and , the difference , is less ) the difference between two numl ) in the table , then .
In Tables I and II the difference between two consecutive numbers , and as there are 100 all the numbers lie betweeTl 1 and ; so that if is less than , the required reduction is effected .
Preparation is accomplished in two lines of simple multiplication and division , as follows:\mdash ; The given number is divided or multiplied by such a power of 10 as will leave the quotient or product as a decimal fraction between 1 and 10 .
This is effected by simply shifting the decimal point .
If the first decimal place is less than 3 times the integer ( which is always the case when the integer exceeds 3 ) , divide by the integer and divide the quotient by or .
The resuIt is less than If the first decimal place is more than twice the integer , then it is always possible , generally in several ways , to find an integer between 1 and 10 which , used as a multiplier , will give a product of which the integer is less than 13 , and the first decimal place less than the integer .
The following rule embraces every case:\mdash ; Multiply any of the numbers to by 4 ; to by 7 ; to by 5 ; to by 6 ; to by 4 ; to by 3 .
Then dividing this product by the integer the quotient is less than This preparation is very convenient also for starting Weddle 's and 384 Mr. A. J. Ellis .
Improved Bimodular Method of Hearn 's processes given by Mr. Peter Gray in the introduction to Tables for twelve-place logarithms , 1865 ( first published in 1845 ) , is also very much simpler than that proposed by Mons. Thoman in " " Tables de Iogarithmes a 27 decimales 1867 .
Interpolahon.\mdash ; The finding of is thus made dependent finding , where is a tabular number and is less 001 .
We then find , which gives the logarithm of , or the " " quotient\ldquo ; or .
The is eifected by the multiples of the bimodulus given in tables , when is not 1 , the unit , place of each multiple of placed immediately below the determining figure of , care being to preserve as many places as are necessary for the final result .
division is a single contracted division .
The resulting or has be " " corrected\ldquo ; by the equations ( 5 ) and ( 6 ) , as shown in Section III .
Completion.\mdash ; Having found , we add the logarithm of power of 10 by which we first divided , and the logarithm of any divisor , and the arithmetical complement of the Iogarithm of power of 10 or any other multiplier .
All these logarithms are in the table .
The result is the complete to the number decimal places for which the table is adapted .
rithms .
logarithm being given we have to reduce it to the logarithm of a number between 1 and .
This is most conveniently done by subtracting ffim it ( or adding to it ) the Iogarithm of the largest power of 10 , which will make the result lie between and , and afterwards subtracting the next least logarithm of integer between 1 and 10 , and then the next least logarithm of a between and 2 .
The logarithms of all these numbers are given the table .
The result will be the logarithm of a number less than We theu subtract the next less logarithm in the table of interpolation , and obtain the equivalent to the corrected quotient or ( 5 ) and ( 6 ) .
We find the correction in the same way as for the quotient , and subtract it , thus obtaining or .
Then we divide the bimodulus increased by this or , by the bimodulus decreased by or , as in ( 7 ) , and thus fiud , which is the number corresponding to the " " quotient\ldquo ; in the direct method .
For " " completion\ldquo ; has to be multiplied by the numbel .
S corresponding to all the subtracted in the preparation .
SECTION III.\mdash ; CALCULATION 0F THE BIMODULAB CORRECTIONS .
The principal peculiarity of this improved bimodular method sists in the calculatio1l of the corrections and the determination of the number of places which can be trusted in any case , as assigned in the tables .
The repetition of any digit times within the same number computing Natural and Tabular Loyarithms , represented by suffixing to the right of that digit .
Thus , a decimal fraction beginmng with zeroes and followed by 1 , not .
to forty-eight places .
Other have used in this sense , but it is not applicable to other and conflicts with the usual notation of powers , thus 23 looks ( 230)3 , in place of Write equations ( 5 ) and ( 6 ) thus\mdash ; not . . . . .
( 10 ) , tab . . . . .
( 11 ) , , ( 12 ) , tab .
tab . . . . .
( 13 ) , tab .
Iog tab .
( 14 ) , tab .
tab . . . . . .
( 15 ) , .
( 16 ) , tab .
tab . . . . .
( 17 ) , tab .
tab . . . .
( 18 ) , tab .
tab .
2500 . . . . .
( 19 ) .
By means of these equations the corrections can be calculated from the " " quotient\ldquo ; that approximate values either with or without existing tables of logarithms , or the quotient or may be calculated to which a particular value of the first correction is due .
From these has been calculated the following table of the critical values of the first and second corrections , upon which the hole practical use of the corrections depends .
The quotients were first taken to proceed from to by steps of .
Then the values of the quotients were determined , which reduced either of the two first corrections to being variable , from which point the suffix of , or the number of initial zeroes , changed , giving critical values of the corrections .
Such quotients were then inserted in numerical order .
The approximate numbers were obtained from the quotients on the snpposition that was small enough to make not .
, and tab .
, to three places of significant figures .
The suffix of in the first correction , diminished by 1 , shows the number of places which are unaffected by that correction , that is , the number of places iu the uncorrected quotient which may be trusted without corrections .
The undiminished suffix shows a number of places which cannot be wrong by more than one unit in defect in the computing and Logaritl , ms . .
387 place .
The suffix of in the second correction , diminished by 1 , how many places of the quotient , after first are left unaffected by the second correction , that is , ' can be trusted on the first correction on natural arithms it will be seen that this never gives less than places , that is , places in addition to those determined correction .
Thus in Table I , where is never less than 3 , can always obtain sixteen places .
For tabular Iogarithms , as in II , we must first observe a critical value in the numbers themIn that table the number , whose Iogarithm is finally must be less than .
Hence , while in the upper part of preceding table of critical values , will always be 8 or more , in lower part , will always be 3 or more , so that will always 4 or more .
As far then as the quotient , the first correction only 5 .
places , and this is the largest quotient that commence with .
If the significant figures are greatel than then will be 4 , and up to the quotient we can trust places , and beyond it we can even trust 19 places .
Observe at the bottom of this table is followed by at the top ( II , second column ) , for which , also , the second corrections leave places unaifected .
But in determining the full number of places of the first correction the uncorrected quotient by equations ( 12 ) and ( 16 ) , we are , course , obliged to take so many significant places , that on result and multiplying by the proper coefficient , no error affecting full number of places should be committed .
The number of places required for this purpose is so large that if we calculated the esnlt directly , the present method of correction would be illusory .
Hence it is necessary to use common seven-place logariohmic tables which can be trusted to six places .
Consequently , we can use oulv six onificant places in the quotient for ding the correction , and we thus introduce an error not exceeding half a unit in the last place in excess or defect .
On estimating the limiting effect of this error I fiud practically that on six significant places of the uncorrected quotient to determine the first correction , we may trust all six places of the correction found .
The total number of places that can be trusted , when this error allowed for , depends on the quotient .
Let be the significant places of the quotient collverted into a decimal fraction with one unit place .
Then the real quotieIlt is , but on taking only six significant places , we use as quotient , and the error thus made in the correction may be taken as the term involving in the cube of this number divided by that is , as .
Then , putting and 100 , and finding the corresponding values of , we obtain tho critical values of the quotient where the suflix of in the error of the Mr. A. J. Ellis .
Improved Bimodular Method of correction changes .
The results are given under " " 5 .
Full tions in Tables I and II .
But although it is by no means difficult or very troublesome to the formulae ( 13 ) and ( 17 ) for finding the first correction , always inconvenient to use two tables .
It would be impossible to give a table of corrections to six figures within able limits .
Hence , leaving the " " full correction \ldquo ; to be found , desired , by.these formulae , I append a table of ' ' short correctionS so as to obtain twelve places of the result from Tables I and II sight .
The thirteenth place has been allowed for , so that the may be trusted , but in the " " completion " " an error of unit in the twelfth place may easily creep in unless " " full corrections\ldquo ; are used .
These " " short corrections\ldquo ; have been calculated from formulae ( 15 ) and ( 19 ) , by assuming successive values of the correction , as and so on , and calculating corresponding value of the quotients .
But in the table itseIf corrections are entered as , \amp ; c. The limiting correction reached when the corresponding quotient is the next least to that to the number .
These twelve places are fully as many as required for ordinary purposes , and for them only thirteen out of eighteen places in the tables should bs used .
SECTION TABLES AND EXAMPLES .
Table I applies to natural logarithms giving from nine bo places , according to circumstances , with no corrections , twelve with short corrections , and fourteen to sixteen places with corrections .
Rule to find the logarithm from the number .
\mdash ; Reduce the number to the form of a decimal fraction with an integer less than 10 .
Multiply and divide by such ] numbers less than 13 as will reduce the number to one less than , as shown in Section II .
Find the next less .mber in " " 1 .
Table for Tnterpolation first subtract it from the reduced number , then omit the point , and multiply by 2 , forming the " " dividend secondly , add next less number to the reduced number , and then omit the point , forming the " " divisor Divide the dividend by the divisor by simple contracted division as many places as are required .
Correct the quotient , as may be sary , by the table or formula of correction , No. 6 or 5 .
Add the arithms of the divisors and the arithmetical of the logarithms of the multipliers used in forming number , to find the full corrected logarithm .
Table II applies to Briggs 's or Tabular Logarithms , giving nine to sixteen places , according to circumstances , with no computing Nalural and Tabular Logarithms , } twelve places with short corrections , and fourteen to with corrections .
Rule.\mdash ; Proceed precisely as for rithms , exccpt of by 2 it is to multiply by by of the multiples given in No. 3 .
Tables I and II .
Rule to find the jrom the } .\mdash ; Subthe logarithm of the next power of 10 , and then , next lower rithm in the lower , and then thttt in the of the table " " 2 .
For Prcparation and ards the ncxt lower logarithm in the table for interpolation .
Considering this as an of a reduced find the correction as if it were a by No. 5 or ( instead of adding ) the correction , reduces it to the forIll of quotient or approximate ithm .
Add the resulting number to and it from the ( which is 2 for natural logarithms ) nd divide the sum by the ference .
Multiply the quotient in succession by the numbers corrcEi } ) Ollding to the Iogarithms subtracted .
The result is the number Examples , fully worked ont , explanations .
Let C99 928 Then calculating the value of 6 not .
from Wolfram 's ppended to Vega 's , and the resuIt by the modulus we find to twen tie places\mdash ; not .
tab .
These numbers serve as chccks to the correctness of the ] work .
Here form the " " ration \ldquo ; of N. where the first decimal place is more than 3 times the is multiplied by 6 to produce . . .
, a fraction of which integer 11 is less than 13 and more than twicc first ) .
Both 5 and 4 would have also answered .
divisor 11 is ated off by ) , and in the quotient the next less number tho ) for interpolation is similarly .
This leaves ) to the right of ) , with the decimal point already omitted .
Then di rence is multiplied by the bimodulus 2 , to whole of is added to the separated part , and then the point is omitted , giving .
As the difference , lics ) otwcen and , we can certainly twclve placcs without rection ( Table I , No. 4 ) , and as it lies betwoon and ) ) obtain seventeen places with full corrcctions ( Tablo I , No. .
computing Naturat and Tabutar Logarithms , preparation , is similar to that in Ex. 1 , but 5 is used as multiplier by way variety .
Tho difference being which lies between and , we cannot be certain more than ten places without correction ( Table If , No. 4 ) .
As only places are wanted , we use the short corrections and work to places .
The chief peculiarity relates to the tion of by the bimodulus by means of the multiples in Table II , No. 3 , omitting all the decimal points .
The integer of the multiple is placed under the figure of the mllltiplicand , and the multiple is then written out as far as necessary , ecting the point , but regulating the ] figure .
It is best to write in the integer , as in line , to preclude error .
As the quotient must with , only ten significant places are wanted , and hence eleven places in the divisor , the four underlined are therefore rejected .
The correction is found from Table II , No. 6 , as } to a quotient between and .
The rest is sufficiently explained in the notes .
The result is accidentally correct to thirteen places .
Hers is the given tab .
to eighteen places .
We first subtract Illog 10 , or the chaY'acteristic .
Next , if the remainder were greater than any logarithm in the lower part of IL No. 2 , we should subtract that .
But in this case it is not , and hence we proceed to the upper part of No. 2 , and subtract the next less , or tab .
This completes the preparation , as the difference , lies between the tab .
logs of and in No , 1 , the table for interpolation .
Hence , subtracting tab .
, we find tab .
, of which number has to be found .
Now , the formula ( 7 ) applies only to an uncorrected ) .
, which cannot differ from tab .
in the three first significant figures .
In the direct process , tab .
is found from tab .
by adding the correction found by Table II , No. @ .
Hence we have only to subtract this correction , which is cal.ulated from the same first six significant figures in both cases , as shown in the example .
Having found this uncorrected tab .
, we add it to and tract it from , the bimodulus , obtaining and respectively , and thus find .
Now , tab .
cannot be greater than the greatest diflerence between two tab .
logs in Table II , No. 1 , " " for rpolakiox , \ldquo ; that is , it cannot be greater than aIzd hence than iiJl modulus .
IlJenee the result of this division must he ss in any case . . .
, and must be greater than 1 , hence it commemce with .
As modulus divides out , tlxis conclusion bolds for all systems of logarithms .
As the last divisor in ; contracted division must have two digits for safety , it follows that the number of digits in the quotient , will be one less than the number of digits in the divisor , that is , the of deeinnal places in the given logarithm .
And as the first of these digits is } whole number , Mr. A. J. Ellis .
On the Potential Radix it follows that the number of decimal places in the quotient be two less than in the givenlogarithm .
Moreover , as the last place is always approximate , it follows that the number cannot found with certainty to more than three decimal places less than number of decimal places in the given .
logarithm .
Hence , in present case , although tab .
is known to eighteen ] aces of deci maJs , is known with certainty only to fifteen places of decimals ( and sixteen digits ) .
But the error in the next place ( or digit ) will not probably exceed one unit .
Having found , we have to multipIy it in succession by the bers corresponding to the logarithms subtracted in the preparations in this example , 1014 , , and .
This is most readily done in the way sufficiently explained by the notes in the example .
The resulting number is accidentally correct to seventeen digits , but only sixteen can be used with certainty .
Hence , if we use this bimodular method of finding and anti-logarithms , we should always find the logarithms to two or three places of decimals more than we require digits in the final number to be found .
V. " " On the Potential Radix as Means of Calculating Logarithms to any Required Number of Decimal Places , Vith a Summary of all Preceding Methods Chronologically Aranged .
, By ALEXANDER J. ELLIS , B.A. , F.R.S. , F.S.A. Received January 17 , 1881 .
In the tables ached to my ) " " On an Improved Bimodular Method of Computing arithms , \amp ; c ( ' ' Proc. Roy .
Soc vol. 31 , p. S81 ) , the logarithms used were all taken direct , or immediately calculated , from the tables of Wolframm and Gray .
But a complete method of calculating logarithms should be independent of extraneous aid and be applicable to the first construction of tables of logarithms .
I shall here show that my improved bimodular method is capable of furnishing a practical means of calculating natural logarithms , and hence arithms to any base and to any number of places of decimals .
By the term ) numerical radioe I shall understand a table of the numbers , with their corresponding natural logarithms , where varies from 1 to 9 , means a series of zeroes , and varies from 1 to any required number .
The word Radix in this sense is adopted from R. Flower , , mentioned below .
By the term radix I meall a similar table of , and the negatives of their corresponding logarithms .
When these radixes ( forming au English plural , as radices would be misleading ) have been

