A Geometrical Approximation to the Roots of Numbers

Author(s): J. M. Barbour
Source: The American Mathematical Monthly, Vol. 64, No. 1 (Jan., 1957), pp. 1-9
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2309077
Accessed: 15-02-2016 15:43 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to The American
Mathematical Monthly.

http://www.jstor.org

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
 A GEOMETRICAL APPROXIMATION TO THE ROOTS
OF NUMBERS*
J. M. BARBOUR, MichiganState University
Origin of problem. By equal temperamentis meant the division of the
musical octave into twelveequal parts or semitones.Since the ratio of the octave
is 2: 1, the ratio of theequally temperedsemitoneis 21/12:1. The piano-tuner,who
operates with intervalsof fiveand seven semitones,must be cognizant of their
ratios, but only to the extentthat he must know how fast one of these intervals
should beat in a givenpartofthe keyboard.He tunesby ear, notby mathematics.
His problemis different in kind fromthat of the 16th centurymusical theorist,
who wished to show players upon the lute and viol how to place the fretsupon
their instrumentsin order to have equal semitones. For the latter, therefore,
equal temperamentwas the geometricalproblemof insertingeleven mean pro-
portionalsbetween 2 and 1. A more general,but equivalent, problemwas the
division of the stringof a monochordinto equal semitones.By formingunisons
with the pitches thus obtained, the stringsof a harpsichordor the pipes of an
organ could be tuned.
Since 12 = 3 X 2 X 2, the problem of equal temperamentis essentially the
problemof the duplicationof the cube, a problemwhichthe ancient Greekswere
unable to solve by Euclidean methods. Several of the methods which they did
findsuitable,such as a circleand a variable secant, a circleand a parabola, etc.,
were repeated by writersof the 16thand 17thcenturies.The most popular of all
was the mesolabium,a mechanical device with slidingparallelograms,ascribed
to Archimedes [1].
Mersenne's approximation.Pere Mersenne gave an ingeniousand excellent
geometricalapproximationfor equal temperament,using only the well-known
Euclidean method for insertinga single mean between two lines [2], [1]. By
this method he firstobtained a major thirdwith the ratio of (3- /2) :2. Since
the major thirdcontains foursemitones,it was then easy to insertthree.mean
proportionalswithinit, and similarlyto insertseven proportionalsbetween the
major thirdand the octave. In decimals his ratio forthe major thirdwas .7929,
which is .1% shorterthan the correctvalue forthe temperedthird,.7939, and
this,of course,was the maximumerrorin the entireconstruction.Table 1 gives
the string-lengths forequal temperamentand Table 2 is Mersenne's approxima-
tion. The error is expressedas the logarithmof the quotient of the approxima-
tionby the correct value.
Galilei's approximation.Vincenzo Galilei, father of the astronomer,gave
an approximationthat was less accurate than Mersenne's,but much simplerto
construct.He took 18:17 as the ratio of the temperedsemitoneon the lute, and
* partofthisarticlewas presented,
The first withthetitle"Aneighteenthcentury approxima-
tionto theequallytempered scale,"to the MichiganSectionofthe MathematicalAssociationof
America,April24, 1951.The authorwishesto thankProfessor V. G. Groveforhis assistancein
boththeinitialand finalstagesofthisstudy.

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
2 A GEOMETRICALAPPROXIMATIONTO THE ROOTS OF NUMBERS [January
TABLE1 TABLE2 TABLE3
MERSENNE'S GALILEI'S
EQUALTEMPERAMENT APPROXIMATION APPROXIMATION

Notes String-
lengths * String-
lengths 05 Xlog
error String-
lengths 105Xlog
error

C 100000 2 100000 0 100000 0

94387 211/12 94365 -11 94444 26
D 89090 25/16 89045 -22 89197 52
Eb 84090 23A 84028 -33 84242 79
E 79370 22/3 79290 -44 79562 105
F 74915 27/12 74850 -38 75142 131
F# 70711 21/2 70658 -33 70967 157
G 66742 26/12 66700 -27 67024 183
G# 62996 21/3 62966 -22 63301 210
A 59460 21/4 59439 -16 59784 236
B13 56123 21/6 56110 -11 56463 262
B 52973 21/12 52968 -5 53326 288
C 50000 1 50000 0 50000 0

* In Table 1,thestring-lengths in thethirdcolumn.

are 50000timesthemeanproportionals

then foundthe remainingsemitonesby proportion[3], [1]. However, (17/18)12

=.50363, whichis .7% too long. An easy way to reduce thiserroris to take twice
the excess, or .00726, fromthe opposite end of the string,thus making Galilei's
12th fretthe exact middle. With this correction,the maximum error,for the
augmented fourthof six semitones,is .07%. Even without the correction,this
is a veryeffectivepractical method,and, as such, it was referredto in the litera-
ture fortwo and a half centuries.

Str'ahle's approximation.In 1743 Daniel P. Striahlepresenteda curious and

easy geometricalapproximationto equal temperament[4]. This, as the present
author has describedit, was his method (Fig. 1): "Upon the line QR, 12 units in
length,erect an isosceles triangle,QOR, its equal legs being 24 units in length.
Join 0 to each of the 11 points of division in the base. On QO locate P, 7 units
fromQ, and draw RP, extendingit its own lengthto M. Then, ifRM represents
the fundamentalpitch and PM its octave, the points of intersectionof RP with
the 11 rays from0 will be the 11 semitoneswithinthe octave" [1].
Striihle had apparently arrived at his approximation by intuition,for its
testingwas leftto Jacob Faggot in the second part of Striihle'sarticle. Faggot
(1699-1777), geometer and pioneer political economist, had been a charter
member of the Swedish Academy at its foundingin 1739, was for three years
its secretary,had 18 articles published in its Proceedings,the last in 1770, and
in 1776 was ranked No. 4 in the Academy, with Carl Linneaus as No. 2. Faggot,
the distinguishedsavant, proceeded to work out by trigonometrythe string-
lengthsforStriihle'sapproximation.

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
1957] A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS 3

Faggot's trigonometric solutionof Strahle's construction.The trigonometric

solution was not difficult:cos L OQR = 6/24, whence L OQR = 750 31'. Then A
PQR, with two sides and the included angle known,was solved by the law of
tangents. All of the other angles formedbetween QR and one of the rays from
0 could also be found directly;e.g., ctg L (O II R) =6V515/3, whence L 0 II R
=77? 52'. Therefore,each of the 11 triangleshaving its base along QR and a
leg along PR could be solved by the law of sines, since both its base and its
angles were known. However, there was one little difficulty.Faggot had com-
puted ZPRQ as 400 14', whereas it should have been 330 32'. If he had made a
mistake in any of the otherangles, he mighthave caught it. But this was fatal,
since Z PRQ was used in the solution of each of the othertriangles,and exerted
its baleful influenceimpartiallyupon them all. (This errorin the size of L PRQ
was equivalent to using 8.605 forPQ in place of 7.)

x mZ arNwm ~zrm}s

FIG. 1

Afterhaving computedall the lengthsfromR along MR, Faggot subtracted

each of them fromMR (= 2PR) and then divided by the lengthof MR. As can
be seen fromTable 4, Faggot did not put Strahle's methodin a very good light,
forthe maximumerroris about 1.7%, whichis fivetimesas greatas a musician's
ear would consideracceptable.
Algebraic solution of Strahle's construction. A corrected trigonometric
solution of Strahle's approximation is unnecessary,for this problem can be

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
4 A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS [January

solvedalso by algebraicmethods,and the lattersolutionlendsitselfbetterto

generalizationas wellas to numerical Usingthesamebasicfigure
computation.
as before(Fig. 2), let C be anypointon AR. Draw OC, cuttingPR at D. Draw
DJIAR. Then trianglesOAC and DJC are similarand so are trianglesBAR
and DJR. Fromthisconditionof similitudeit followsthatDJ/OA= JC/AC;
DJI/BA= JR/AR=DR/BR. WhenDJ is eliminatedfromtheabove equations,
JR/AR = JC X OA/BA X AC.
Substitute(JR-CR) forJC and (AR - CR) forA C:
JR/AR = OA(JR - CR)/BA(AR - CR).

24

Q A JC R
IA

FIG. 2

collectterms,and solve forJR:

Cross-multiply,
JR = CR X OA X AR/{AR(OA - BA) + BA X CR}.
Then
DR/BR(= JR/AR) = CR X OA/{AR(OA - BA) + BA X CR}.

If m is the fractionalpowerto be approximated,

CR =1-rm and AR= . Let
OA =a and BA =5.

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
1957] A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS 5

thedenomi-
thesevaluesin theabove equationand simplifying
Substituting
nator:
DR/BR = 2a(1 - m)/{a + b- 2bm}.
When m=0, PR/BR=2a/(a+b). When N (the number whose root is to be
approximated)= 2, MP =PR and MD = 2PR -DR. Hence
(M4a 2a(1/-Mm) \a + b) a + b + (a-3b)m
(1) 2" _ MDIMP - I I=
a + b a+ b- 2bm/i 2a / a b+-2bm
+
let us turnagainto theorigi-
To fitStriihle'svaluesintotheabove formula,
naldiagram,anddropa perpendicular PE fromP to QR (Fig. 3). ThenQE = 7/4;
ER=41/4; PR/BR=41/24. Fromthislast relation,2a/(a+b) =41/24, a 41,
b=7.
0

41L

Q E A R

FIG. 3

thesevalues fora and b in formula(1),

Substituting
24 + lOm
(2) 2 ___ .
24 - 7m
Then, for
408 398 388 288
1,
M-II/12, 10/12, 0, 2m 21 28
204 211 218 288

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
6 A GEOMETRlCAL APPROXIMATION TO THE ROOTS OF NUMBERS [January

In order to compare with Faggot's incorrecttable forStrahle's approximation,

Table 5 shows the above fractionsmultipliedby 5000.
It is unfortunatethat Faggot's erroneous figureswere copied, with no
explanation of Striihle'smethoditself,in Marpurg's excellenttreatiseon musical
temperament [5]. Since Marpurg had not troubled himselfto check Faggot's
computations,he was naturally superciliousregardingthe method. As can be
seen fromTable 5, Striahle'sgreatesterrorsare about -.11% and .15%. It will
be noted that the patternof the errorsresemblessomewhat a sine curve, rising
above the axis forsmall values of m, passing throughzero almost at the mid-
point, and then dipping similarlybelow the axis. Neither arch of the curve,
however,is completelysymmetrical.

TABLE 4 TABLE S TABLE 6

FAGGOT'S NUMERICAL CORRECTED NUMERlCAL APPROXIMATION FOR EQUAL

VALUES FOR STRXHLE'S VALUES FOR STRXHLE'S TEMPERAMENT, USING COR-
APPROXIMATION APPROXIMATION RECT VALUE FOR A2

String- 10Xlog String- 1O6Xlog String- 105Xlog

lengths error lengths error lengths error

10000 0 10000 0 100000 0

9379 -276 9432 -32 94305 -37
8811 -479 8899 -49 88978 -55
8290 -619 8400 -46 83981 -56
7809 -706 7931 -33 79290 -44
7365 -740 7490 -9 74874 -24
6953 -732 7073 11 70711 0
6570 -683 6680 38 66780 24
6213 -601 6308 58 63061 44
5881 -478 5955 65 59538 56
5568 -344 5620 60 56195 55
5274 -192 5302 38 53020 37
5000 0 5000 0 50000 0

Striih1e'sapproximationgeneralized. AlthoughStrahle's approximationwas

devised fora specificpurpose,i.e., to findthe successive powersof the 12th root
of 2, it should be possible to employa modificationof it to findthe rootsof num-
bers otherthan 2. To do thiswe shall need morethan the inspiredguess by which
Strahle made PQ/OQ = 7/24. The necessaryclue comes fromthe fact that his
constructionis very nearly perfectfor V/2,his fractionbeing 58/41= 1.41463,
which is only .03% too high. If Z MRQ can be adjusted so that MB is precisely
the mean proportionalbetweenMP and MR, thiswillbe the best approximation
by this method forthe roots of the numberMR/MP-at least when all (n-1)
mean proportionalsare to be found,as in Striihle'smusical problem,since the
negative and positive errorsare thenequalized.
Let us assume,then,thatMPXMR= MB2. Since MR=MP+PR and MB
=MR -BR,

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
1957] A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS 7

MP/MP 2a MP 2a \2
J=I + ~- Ii
BR BR a+bI\ BR a +b
This leads to a linear equation in the variable, which simplifiesto

MP a-b
BR 2b(a + b)
Then
MR a+b MB a-b
-= and BR-= 2b .
BR 2b BR 2b
From this result,
MR (a+b\2
MP a-b
It followsthat

(3) Ntm 4---

MD
MP
a+
a-
b (
b 'a-b + 2bm'
a + b-2bm
)
It may seem a pity to change so beautifullysymmetrica formula,but it will be
moreconvenientforgeneraluse ifwe let a - b-1, whencea+b = \/27.Substitut-
we obtain*
ing these values in formula(3) and simplifying,

Nm + V\N(1- m)
(4) NVm
.1
( +
mVN(l -m)
Table 6 shows the application of formula(4), to the solution of the musical
problem,withN= 2 and m= 1, 11/12,etc.If this table is compared withTable 5,
it will be seen how closelythe two approximationsresembleeach other (here the
maximumerroris less than .13%). One's admirationforStriahleincreases when
one contemplatesthe beautifulsimplicityof his method.
Construction.Here is the methodofconstructinga geometricalapproximation
to the roots of numbers,in which the correctmean proportionalis used;t
Problem.To construct,by approximation,(n -1) geometricmeans between
two given lines.
* Comment by the referee.This formulacan be obtainedby interpolating the exponential
y=Nx in the threepoints:x=O, x=j, x=1 by a linearfractional function(A+Bx)/(C-Dx).
Sincethelatterdependson exactlythreeparameters, thesolutionis unique.
t Comment bythereferee.Thisisa verybeautifulinterpretationofformula (4). The procedure
is quite natural,as the linearfractional
transformationclearlycalls fora perspectivity.Indeed,
theauthor'srequirement thatBA be perpendicularto RQ is irrelevant,sincea projectivecorre-
spondenceon linesis uniquelydetermined by threepairsofcorresponding points.

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
8 A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS [January

Hypothesis.Let MR be the longerof the two lines; NS the shorter(Fig. 4).

Construction.Construct,by the Euclidean method, the mean proportional
between MR and NS. On MR, take MP = NS and MB = the mean proportional.
Through R draw the line QR, formingan acute angle with MR, and drop a
perpendicularto it fromB, which intersectsit at A. Take AQ =AR, and draw
QP intersectingAB prolongedat 0. Draw OR. Divide QR into n equal parts
and drop rays from0 to each point of division.
Conclusion.The lines fromM to the points of intersectionwith the (n-1)
rays from0 will be approximationsto the geometricmeans between MR and
MP, i.e., between MR and NS.
0

A.- R

lV s

FIG. 4

Error for N< 2. In the range 1 <N <2, the errorsforthe above method of
approximationare indeed small, the maximumerrorforN=2 being .13%. This
error occurs when
1 ? V/.348
2
When N= 1, the maximumis at
1 + V-.333
2
which is imperceptiblysmaller than the previous values. The maximum for
N=1 is precisely that of the cubic: f(m)=m(1-m)(1-2m), in the range
0 <m < 1. Thus the errorcurve can be fittedremarkablywell to this cubic for

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions
1957] A GEOMETRICAL APPROXIMATION TO THE ROOTS OF NUMBERS 9

all values of N in the given range, and such a correctionreduces the original
very small errorby somethinglike 99%! However, a complicated third-degree
of conversionfora particularN, and
formulais needed to obtain the coefficient
this makes correctingthe errora somewhatlengthyprocess.
Error for N>2. As N becomes large, the maximum error increases very
rapidly,with max EN2 _8 max EN. Moreover,with N large, the errorcurve re-
tains the same generalshape as before,but the maximumis pushed nearer the
endpoints,0 and 1. Thus the curve resemblessomewhat a higher-order variant
of a cubic, namely, a curve having an equation of the form:f(x) =x(1 -X24),
wherea > 1 and is not necessarilyan integer.For most values of N it would be
impossibleto calculate the correctionfunctionby elementarymethods.Thus it is
not at all feasible to use this method of approximationwith large numbers.
Reduction by repeated mean proportionals.The roots of large numberscan
be approximatedwith a high degree of accuracy, however,by a simple adapta-
tion of our version of Strahle's geometricalapproximation.One of its essential
featureswas the findingof a singlemean proportional.Moreover,Mersennehas
used nothingbut mean proportionalsin his approximationfor equal tempera-
mentshownin Table 2. If mean proportionalsare taken repeatedly,the problem
of approximatingthe desired root may eventually be reduced to an intervalin
whichthe ratio is no greaterthan 2 :1-the range in whichthe erroris very small
-after which,ifdesired,the cubic-curvecorrectionmay be applied.
Conclusion. As a geometrical approximation, the method shown in this
article is simple and works exceedinglywell for small numbers.It is especially
advantageous forobtaining approximationsto the successive powers of a root,
as in its original application to the solution of a musical problem. For larger
numbers,the extendedmethodwhichuses repeated mean proportionalswill also
be easy to operate geometricallyuntil the mere length of the lines becomes a
burden. As a numericalapproximation,the method also gives good results for
small numbers,and is superfinewith the correctionwhich uses the cubic curve.
For large numbers,the labor of taking repeated square roots, multiplyingto-
getherthe numbersthus obtained, and applying a complicated correctionfor-
mula is not commensuratewith the resultsobtained.
References
1. J. M. Barbour,Tuningand Temperament, MichiganState CollegePress,1951; 2nded.,
1953,pp. 50-53; 54 f.; 57ff.;65-68.
2. MarinMersenne, Harmonieuniverselle, Paris,1636-37,p. 68.
3. VincenzoGalilei,Dialogodella musicaanticae moderna,Florence,1581,p. 49.
4. D. P. Straihle,
Nyttp&fund, tilat finnatemperatureni staLmningen
forthonerne p&claveret
ock dylikainstrumenter, Proceedings oftheSwedishAcademy,1743,IV, pp. 281-291.
5. F. W. Marpurg,Versuchfiberdie musikalische Temperatur, Breslau,1776,p. 167f.

This content downloaded from 194.27.18.18 on Mon, 15 Feb 2016 15:43:31 UTC
All use subject to JSTOR Terms and Conditions