Early  Journal  Content  on  JSTOR,  Free  to  Anyone  in  the  World  
This  article  is  one  of  nearly  500,000  scholarly  works  digitized  and  made  freely  available  to  everyone  in  
the  world  by  JSTOR.    

Known  as  the  Early  Journal  Content,  this  set  of  works  include  research  articles,  news,  letters,  and  other  
writings  published  in  more  than  200  of  the  oldest  leading  academic  journals.  The  works  date  from  the  
mid-­‐seventeenth  to  the  early  twentieth  centuries.    

 We  encourage  people  to  read  and  share  the  Early  Journal  Content  openly  and  to  tell  others  that  this  
resource  exists.    People  may  post  this  content  online  or  redistribute  in  any  way  for  non-­‐commercial  
purposes.  

Read  more  about  Early  Journal  Content  at  http://about.jstor.org/participate-­‐jstor/individuals/early-­‐

journal-­‐content.    

JSTOR  is  a  digital  library  of  academic  journals,  books,  and  primary  source  objects.  JSTOR  helps  people  
discover,  use,  and  build  upon  a  wide  range  of  content  through  a  powerful  research  and  teaching  
platform,  and  preserves  this  content  for  future  generations.  JSTOR  is  part  of  ITHAKA,  a  not-­‐for-­‐profit  
organization  that  also  includes  Ithaka  S+R  and  Portico.  For  more  information  about  JSTOR,  please  
contact  support@jstor.org.  
 HAVING SMALL DETERMINANTS 249

.24561, .35381, .36518

.62433, .90274, .91143
.17145, .23642, .24375
X .34622, .35381, .36518_
.89318, .90274, .91143
.22431, .23642, .24375
If in the numeratordeterminanttwo thirdsof the thirdcolumnis subtractedfrom
the firstcolumn,and if the second columnis subtractedfromthe third,the value
of the determinantis not altered. In the denominatorlet the middle column
be subtractedfromeach of the others,and then the thirdcolumn added to the
first. The resultingexpressionforx is
.00216, .35381, .01137
.01671, .90274, .00869
.00895, .23642, .00733
X +.00378, .35381, .01137
-.00087, .90274, .00869
-.00478, - .23642, .00733

The computation can now be made with four-placetables containingall the

numbers required on two pages. In computingx it is found in two or three
minutesthat to threefiguresits value is - 1.03,whichis as faras it is determinate.
TlEF UNIVERSITY OF CHICAGO,
May 19, 1913.

THE ACCURACY OF INTERPOLATION IN A FIVE-PLACE TABLE

OF LOGARITHMS OF SINES.
By A. M. KENYON and G. JAMES, Purdue University.

In a table which gives the logarithmsof sines for each minuteof angle, the
logarithmsof sines of angles to tenthsof minutesare obtained by interpolation.
The name OrdinaryInterpolationwill be used to denote the process based on
the assumption:
(1) As the angle increasesfromm' to (m + 1)' the incrementof the logarithm
of its sine varies as the incrementof the angle; that is,
log sin (m + n/10)' = log sin m' + (n/10)(logsin(m + 1)' - log sin i')
(n = 1, 2, 3, ... 9).
The errorintroducedby ordinaryinterpolationarises fromtwo sources:
(a) The essentialerror,due to the assumption(1).
(b) The tabularerror,due to the fact that the tabulated values of log sin m'
and of log sin(m + 1)' are in error.
250 ACCURACY OF INTERPOLATION IN A FIVE-PLACE

The essentialerrorincreasesas the angle decreasesand makes ordinaryinter-

polation unreliableforsmall angles. The tabular erroris irregularbut nowhere
in the table does it exceed, in absolute value, one unit in the fifthplace.
To make this clear we need to note that wheneveran irrationalnumberis
approximatedby a rationalnumbercorrectto n decimal places, an erroris intro-
duced which in absolute value is less than 5/10n+1. Thus if we writesin 450 -

.70711, the resultis too large by an errorbetween .000003 and .000004.

Let log sin m' = a, log sin(m + 1)' = b and let a-e1, b-e2, be the five
place values of these same logarithms. Then by ordinaryinterpolation(assumed
to be a correctprocesssince we seek here onlythe tabular error),the interpolated
value of log sin(m + n/10)' would be
c = a + (n/10)(b- a) = c -e3

wherecl representsthe fiveplace value of c.

But by ordinaryinterpolationfroma fiveplace table, the resultwill be
C2= a - ec + (n/10)(b -e2- a + el)
= a + (n/1
0) (b - a) - [ec+ (n/lO)(e2- el)]

= cl -[3 + el + (n/10)(e2- el)].

Whence the fiveplace tabular erroris

C- C2 = (/10)e1j + (n/10)e2 + e3, wheren +n- = 10,
and obviouslythe absolute value of this erroris less than .00001.
Actual trial of a verylarge numberof cases (over 1,200) shows that ordinary
interpolationbelow 20 gives the result correctto five places in less than 30 per
cent. of all cases; that it gives an errorof one in the fifthplace in about 50 per
cent.,and an errorof two or morein the fifthplace in about 20 per cent. of cases.
There is no case above 10 30' wherethe total erroris greaterthan one unit in the
fifthplace. Above 20 it gives the resultcorrectto fiveplaces in from70 per cent.
to 83 per cent. of all cases and in no case triedis the errorgreaterthan one unit
in the fifthplace.
The name LogarithmicInterpolationwill be used to denote the process based
on the followingassumption:
(2) For sufficiently small angles,the sine varies as the angle; which gives,if
m and n denote measuresof angles in the same unit,

log sin(m + n/10) = log sin m + log(m + n/10) - log m.

Here the essentialerrordecreases with the angle; the tabular erroras before
is irregularbut nowherein the table exceeds,in absolute value, one unit in the
fifthplace.
Actual trial of all cases (to tenthsof minutes)shows that in no case below 5?
45' does logarithmicinterpolationintroducea total errorgreaterthan one in the
 TABLE OF LOGARITHMS OF SINES 251

fifthplace.' Below 5? it gives the result correctto five places in about 60 per
cent. of all cases and shows an errorof one in the fifthplace in about 40 per cent.
of cases.
Below 2?, logarithmicinterpolationis certainlypreferableto ordinaryinter-
polation for threereasons: (1) it gives the correctresult in a larger numberof
cases; (2) it gives an errorin the fifthplace in a smallernumberof cases; (3) it
never gives an errorgreaterthan one in the fifthplace.
Above 2?, ordinaryinterpolationis preferableforthe same threereasons. All
cases between 10 30' and 2? 15' were examined.
An examinationof 200 cases in the neighborhoodof 84? shows that ordinary
interpolationwill give the result correctto five places in about 83 per cent.,
and will show an errorof one in the fifthplace in 17 per cent. of cases.
Below is a table of resultsof the statisticalinvestigation.

Per Cent. of Cases Per Cent. of Cases Per Cent. of Cases

GivingResult Correct Showingan Errorof Showingan Errorof
Numberof to 5 Places. Onaeinathe 5th Two or More inathe
Interval ofAngle. Cases Ex- Place. FifthPlace.
amined.
Ordinary. rithmic. Ordinary. Lthga Ordinary. rithaic. 2

1 /_10 45 Fails 60 Large 40 0

10 -10 30 97 33 57 10 0
10 30'-1? 45' All 55 63 45 37 0 0
1? 45'-20 All 63 69 37 31 0 0
20 -20 15' All 71 62 29 38 0 0
2 15'-30 55 78 60 22 40 0 0
30 -40 55 80 20 0 0
40 -50 0 0
50 -5? 45' All 66 34 0 0
50 45'-60 All _ 64 36 s

A THEOREM ABOUT ISOGONAL CONJUGATES.

By DAVID F. BARROW, Harvard University.

As an introductionlet us recall two well-knowntheorems of elementary

plane geometry:
THEOREM I. Givena triangleA1A2A3(Fig. 1) and any point P not a vertex,
each ofthethreelines thusdrawnin thebisector
and reflect
join P to thethreevertices
of theangle at thecorrespondingvertex. The threereflectedlineswillmeetin a point
P', called theisogonalconjugateof P.
THEOREM II. Givena triangleand a point markedat randomon each side.
If threecirclesbe drawn,one through each vertexand thetwoadjacentmarkedpoints,
these three circles meet in a common point.
place entersat 50
1 The firsterrorof two in the fifth 53'.7
2 All cases examinedforthiscolumn.
place enters.
3At 5? 53'.7 thefirsterrorof2 in thefifth