Sie sind auf Seite 1von 11

HISTORIA MATHEMATICA 23 (1996), 246–256

ARTICLE NO. 0026

An Example of the Secant Method of Iterative Approximation in


a Fifteenth-Century Sanskrit Text

KIM PLOFKER
Department of History of Mathematics, Box 1900, Brown University, Providence, Rhode Island 02912

Mathematical approximation by iterative algorithms is well attested in Sanskrit astronomical


texts, but its use has not been studied systematically. In his 14th-century supercommentary
on Govindasvāmin’s commentary on Bhāskara I’s Mahābhāskarı̄ya, Parameśvara, a student
of the renowned Kerala astronomer Mādhava, presents a one-point iterative technique for
calculating the Sine of a given angle, as well as a modification of this technique that involves
a two-point algorithm essentially identical to the modern secant method. This paper presents
a mathematical and historical interpretation of his remarks.  1996 Academic Press, Inc.
L’approximation mathématique par les procédés d’itération se manifeste fréquemment
dans les textes sanscrites astronomiques, mais elle n’est pas beaucoup étudiée. Dans son
surcommentaire sur le commentaire de Govindasvāmin sur le Mahābhāskarı̄ya de Bhāskara
I, Parameśvara, un élève de Mādhava, l’astronome renommé de Kerala, donne un calcul
itératif en virgule fixe pour le calcul du Sinus d’un angle donné, aussi bien qu’une modification
de cette méthode dans laquelle paraı̂t une technique virtuellement identique à la forme
discrète de la méthode de Newton–Raphson. L’article suivant présente une interpretation
mathématique et historique de ses rémarques.  1996 Academic Press, Inc.

Die mathematische Annäherung bei den Iterationsverfahren ist in den sanskritischen astro-
nomischen Texten gut bezeugt, aber darüber gibt es nicht viele Studien. In seiner Kommentar
über Govindasvāmins Kommentar über die Mahābhāskarı̄ya des ersten Bhāskaras gibt,
Parameśvara, ein Student des bekannten keralischen Astronomen Mādhavas, eine Festkom-
marechnung für die Berechnung des Sinus eines gegebenen Winkels, wie auch eine Modifika-
tion dieser Technik; diese Modifikation enthält einen Algorithmus der grundsätzlich der
diskreten Newton–Raphsonischen Methode gleichwertig ist. Dieser Artikel stellt eine ma-
thematische und historische Interpretation seiner Bemerkungen dar.  1996 Academic Press, Inc.

MSC subject classifications: 01A32, 4103, 6503.


KEY WORDS: Parameśvara, Kerala school, Sine computations, approximation, fixed-point iteration,
secant method.

INTRODUCTION
Iterative techniques for solving equations have been common in Indian astronomi-
cal calculations at least since the use in the Paitāmahasiddhānta (fifth century) of
such a technique for correcting planetary longitudes [10, 558]. Usually, these rules
involve some form of the method of successive approximations, in which the kth
approximation to a root x of g(x) 5 x 2 F(x) is given by
xk 5 F(xk21).
The Siddhāntadı̄pikā of Parameśvara (ca. 1380/1460), a supercommentary on the
commentary of Govindasvāmin on the Mahābhāskarı̄ya (henceforth MB), discusses
246
0315-0860/96 $18.00
Copyright  1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.
HM 23 THE SECANT METHOD IN INDIAN MATHEMATICS 247

such a method for evaluating the Sine of a given angle (in this notation Sin u 5
R sin u for a certain trigonometric radius R). In addition, Parameśvara gives a more
efficient approximation algorithm equivalent to what is now called the ‘‘secant
method,’’1 in which the root of some function f (x) is approximated by
f (xk )(xk 2 xk21)
xk11 5 xk 2 .
f (xk ) 2 f (xk21)
The following verses describing the prescribed procedure appear in the supercom-
mentary on MB IV 22 [6, 207–211] as part of an enumeration in 68 verses of various
methods of calculating and correcting sines. Some of the earlier verses have been
discussed elsewhere [2, 94–96; 4].

TEXT
cāpāj jı̄vāpi sādhyā syād aviśes.ākhyakarman.ā i40i
ardhı̄kuryād is.t.acāpam . tadardham ca tathā punah. u
. bhavet i41i
2
yāvad ardham . vr.ttapādatriśatām. śonakam
tasyāpy ardhāc ca vā tāvad ardhı̄karan.am is.yate u
tadārdhitasya cāpasya samastajyāpi tatsamā i42i
alpatvād utkramasyātra prāyo ’rdhajyāpi tatsamā u
ardhajı̄vāvaśāt kot.ih. kot.itaś cotkramah. punah. i43i
utkramajyāvargahı̄nasamastajyākr.teh. padam u
sphut.ārdhajyā tadvaśāc ca kot.ih. kot.yās tathotkramah. i44i
samastajı̄votkramayor vargabhedapadam . bhavet u
sphut.ārdhajyā punaś caivam . kr.tvā tām aviśes.ayet i45i
dvighnāviśes.t.ā tv ardhajyā samastajyā sphut.ā bhavet u
dvighnasya tasya dhanus.as tato ’rdhajyā ca sādhyate i46i
kiñcidūnām
. samastajyām ardhajyām . parikalpya vā u
samastajyātaddhanus.or bhedāt trim . śaddhatāc chivaih. i47i
. prakalpya vā u
3
labdhenonām . samastajyām ardhajı̄vām
sādhyā kot.iś cotkramaś ca samastajyotkramajyayoh. i48i
kr.tibhedapadam . cāviśes.ayet u
. cārdhajı̄vā tām
dvighnā sā syāt samastajyā dhanus.o dvigun.asya ca i49i
evam . dvigun.acāpasya kramāt sādhyā samastajā u
jı̄vāviśis.t.ā cārdhajyā yāvat syād is.t.acāpajā i50i
is.t.acāpodbhavā hy ardhajı̄vābhı̄s.t.abhujā bhavet u
kot.icāpāt tathā sādhyā kot.ijyā coktavartmanā i51i
bahvāvr.ttyaivāviśes.ah. kadācit sam . bhavaty atah. u
aviśes.e lāghuvāya karma kiñcit pradarśyate i52i
ardhajyotkramavargaikyapadam . jyā syāt samastajā u
tattadardhajyotkramābhyām
. samastajyām . ca sādhayet i53i

1
Other names for this technique currently include ‘‘Lin’s method,’’ ‘‘iterated linear interpolation,’’
and ‘‘regula falsi with successive adjacent points.’’
2
The printed text has ‘‘triśatām
. śenakam
. ,’’ corrected in [6, 440].
3
Text has ‘‘jyārdhajı̄vām
. .’’
248 KIM PLOFKER HM 23

tattatkālāt pūrvalabdhasamastajyādvayāntaram u
hāro gun.aś cotkramayoh. pūrvayor antaram . bhavet i54i
yā tv abhı̄s.t.ā samastajyā pūrvakālabhavā ca yā u
tayor bhedo gun.yarāśis tatrāvāptam . phalam . punah. i55i
pūrvalabdhotkrame deyam . śodhyam . vā syād yathā vidhi u
pūrvalabdhasamastākhyajı̄vāto mahatı̄ yadā i56i
samastajyes.t.ajā deyā tadā śodhyamato ’nyathā u
utkramah. sa sphut.as tena sādhyārdhajyā ca pūrvavat i57i
. cotkramādi punar apy uktavartmanā u
4
samastajyām
kr.tvāviśes.ayec chı̄ghram aviśes.o ’tra sidhyati i58i

TRANSLATION
The Sine is to be derived from the arc by a method called ‘‘aviśes.a’’ [literally,
‘‘without difference’’; i.e., an iterative approximation technique]. i40i
One should halve the given arc, and the half of that, and so on in this way.
[Successive] halving from the half of that [arc] is required until the half is less than
1/300 of the quadrant [5 189]. Then the Chord of the halved arc is equal to that
[arc]. i41–42i
Because of the smallness of the Versine, the Sine too is nearly equal to that [arc].
The Cosine [is calculated] by means of the Sine, and the Versine from the Cosine.
Repeatedly, the square root of the square of the Chord diminished by the square
of the Versine is the corrected Sine, and from that, the Cosine [is calculated], and
thus from the Cosine, the Versine. i43–44i
The square root of the difference of the squares of the Chord and the Versine
is the corrected Sine; and having done thus repeatedly, one may make that indistin-
guishable [from the previous value, by iterative correction]. i45i
The [final] iterated Sine, multiplied by 2, is the accurate Chord of twice that arc.
And the Sine is derived from that [as follows:] i46i
When one has assumed the Sine [to be] the Chord diminished by something [i.e.,
an arbitrary small quantity, «] (or when one has chosen the Sine [to be] the Chord
diminished by the difference of the Chord and its arc, multiplied by 30 [and] divided
by 11), the Cosine and Versine are produced [from that]. And the [new] Sine is
the square root of the difference of the squares of the Chord and the Versine. One
should correct that iteratively [as in the above method]. That [corrected Sine],
multiplied by 2, is the Chord of twice the arc. i47–49i
In this way should be obtained the Chord of the successively doubled arc and
[its] iteratively corrected Sine, until there results the desired Sine produced from
the given arc. In the same way, the Cosine is to be derived from the complementary
arc in the said manner. i50–51i
This iterative correction sometimes involves much repetition. Hence, a certain
method for speed in correction is explained: i52i
The square root of the sum of the squares of the Sine and the Versine is the
Chord. One should derive the Chord from each Sine and Versine. i53i
4
Text has ‘‘samastajyā cotkramādi.’’
HM 23 THE SECANT METHOD IN INDIAN MATHEMATICS 249

At each time [of iteration], the difference between the two Chords previously
obtained is the ‘‘divisor,’’ and the ‘‘multiplier’’ is the difference between the two
previous Versines. i54i
The difference between the given Chord and [the Chord] produced at the previous
time [of iteration] is the ‘‘multiplicand.’’ The result obtained is repeatedly added
to or subtracted from the previously obtained Versine. The rule [for determining
the sign] is as follows: when the Chord resulting from the given [Sine of half the
arc] is greater than the previously obtained Chord, then it is added; otherwise,
subtracted. This is the corrected Versine, and the Sine is to be derived by means
of that, as before. i55–57i
When one has computed the [subsequent] Chord and the Versine and so on
again by the said method, one should correct iteratively. Then the correction is
quickly accomplished. i58i
After the final verse appears a worked example of the rule for ‘‘speed in correc-
tion,’’ using Āryabhat.a’s Sine values with R 5 3438, in which Sin 52;30 is calculated
from Crd 52;30 5 2 3 Sin 26;15 5 3040 using the initial approximation
Sin 52;30 P 3000.

A METHOD OF SUCCESSIVE APPROXIMATIONS


Parameśvara begins by explaining how to find the Sine of a given angle using a
method of successive approximations, and how to calculate the initial approximate
value. He then modifies this approach by introducing his version of the secant
method.
The first part of the procedure makes use of the small-angle approximation by
requiring the given angle u (in minutes of arc) to be halved n times so that u /2 n P
Crd(u /2 n). The rather stringent requirement that u /2 n must be less than 189 is
necessitated by the fact that in this method the initial approximation Crd (u /2 n) P
u /2 n is never corrected, so for accuracy’s sake the arc must be quite small. (The
reason for specifying exactly 189 as the upper bound, however, is not clear.) The
first approximation to Sin(u /2 n) is also taken to be u /2 n, and the subsequent approxi-
mations are found by manipulating the well-known trigonometric identities

Cos 5 ÏR2 2 Sin2,


Vers 5 R 2 Cos, (1)
Sin 5 ÏCrd 2 Vers .
2 2

When there is no difference between two successive approximations, the value of


Sin is ‘‘aviśes.a’’ and the iteration ceases. Somewhat similar techniques involving
the repeated solution of right triangles are very common in jyotih.śāstra texts,
including the Mahābhāskarı̄ya itself [6, 188–191, 196] and, prior to that, the Paitāma-
hasiddhānta [9, 494] and the Pañcasiddhāntikā [7, 1: 96]. (The Pañcasiddhāntikā
seems to be the first to use the term ‘‘aviśes.a’’ for such an iterative method.)
Another procedure that bears some resemblance to Parameśvara’s is described by
250 KIM PLOFKER HM 23

Śaṅkara in his discussion of the power series for Sine and Cosine derived by
Parameśvara’s guru, Mādhava [12, 112–117]. Here, too, the initial assumption that
an arc approximately equals its Sine is used to produce a first approximation to
the Versine, which is then used to correct the Sine, and so on (although in this case,
the desired quantities are expressed in terms of series rather than by geometrical
formulas). If, as seems likely, Śaṅkara’s rationale originated with Mādhava, this
method may have partly inspired the rules presented by Parameśvara.
Once the corrected Sin(u /2 n) has been found, it is doubled to yield an accurate
value of the Chord of the next arc in the sequence, Crd(u /2 n21), and the iteration
process is repeated using the initial approximation

Sin0 S D
u
2 n21
u
S D
5 Crd n21 2 «,
2
(2)

or alternatively,

Sin0 S D
2
u
n21 5 Crd S D S
2
u
n21 2
2
u
n21 2 Crd S DD
2
u
n21 ?
30
11
. (3)

In this manner, it is possible (though laborious) to compute the Sines of all the
successively halved arcs, finally arriving at the desired quantity Sin u.
From a modern mathematical standpoint, we may consider the trigonometric
identities in Eq. (1) as equivalent to the iterative equation

Sink11 5 F(Sink ) 5 ÏCrd2 2 (R 2 ÏR2 2 Sin2k )2. (4)


Since this function is continuous and differentiable, it can be shown from the mean-
value theorem (see, e.g., [8, 38–40]) that the iteration will always converge when
uF9(Sin)u , 1, that is, when
(Sin)(R 2 ÏR2 2 Sin2 )
, 1,
Ï(Crd2 2 (R 2 ÏR2 2 Sin2 )2)(R2 2 Sin2)
or equivalently

Sin , ! R2 2 S R2
Crd 1 R
2
. D
Further application of trigonometric identities reduces this condition to
0 , Crd , R,
which requires that
0 , u , 608.
As Crd approaches R, the rate of convergence becomes extremely slow, and when
Crd . R, the iteration diverges. The constraint on the initial approximation Sin0
is given by inspection of the iterative expression in Eq. (4):
HM 23 THE SECANT METHOD IN INDIAN MATHEMATICS 251

FIGURE 1

Crd2 $ (R 2 ÏR2 2 Sin20 )2 or Sin0 # Ï2RCrd 2 Crd2.


Thus, with the given initial conditions R 5 3438 and Sin0 5 Crd where, say, 1 #
Crd , 18, the method will converge quite rapidly. Since the other two expressions
for Sin0 also satisfy this constraint, convergence is guaranteed—always assuming
that u , 608. Parameśvara makes no mention of such a restriction, although a little
experimentation with the Sines of angles greater than 608 would suffice to show
that in such cases the method fails.

A RULE FOR THE INITIAL APPROXIMATION


Parameśvara’s third and most complicated rule for obtaining Sin0 , shown in Eq.
(3) above, may have been inspired by geometrical reasoning, but its given form
probably owes more to numerical intuition. Generalizing the formula to any angle
a, it may be rewritten as
30
Crd a 2 Sin0 a P (a 2 Crd a) ? .
11
A possible geometric motivation for such a formula is illustrated in Fig. 1.
It is suggested that the discoverer of the formula, perhaps inspired by similar
geometric demonstrations of other Sine approximations (to be discussed further
by the present author in a forthcoming paper), may have proceeded as follows:
taking a/2 P Crd(a/2), he sought a geometric relationship between (Crd a 2
252 KIM PLOFKER HM 23

Sin a) and (2 Crd(a/2) 2 Crd a), or equivalently (Crd(a/2) 2 Sin(a/2)). This


involves solving the two oblique triangles shown in the figure. When this is done,
the true relationship between the given quantities is represented by
Crd a 2 Sin a 2(Cos(a/4))(R 1 Cos(a/4))
5 .
2 Crd(a/2) 2 Crd a R2
This result is not particularly useful for creating a reliable and simple formula.
Under the circumstances, at best one might note that if these two small triangles
are considered to be right and all the quantities linearly proportional, the desired
relation might be roughly
Crd a 2 Sin a 1
P .
2 Crd(a/2) 2 Crd a 2
Perhaps, due to the difficulty of extracting a useful formula from these data, the
originator of the rule fell back upon numerical examples easily obtained from
Āryabhat.a’s Sine values, as shown in the table below:
a(9) Sin a Crd a (Crd a 2 Sin a)/(a 2 Crd a)

450 449 450 1/0


900 890 898 8/2
1350 1315 1342 27/8
1800 1719 1780 61/20
2250 2093 2210 117/40
2700 2431 2630 199/70
3150 2728 3040 312/110
3600 2978 3438 460/162
4050 3177 3820 643/230
4500 3321 4186 865/314
4950 3409 4534 1125/416
5400 3438 4862 1424/538

Looking at the sample values of this factor produced from the twelve angles
whose Chords are easily determined from the table, one sees that the ratio
312/110 lies approximately in the middle, both in position and quantity, of the set
of usable values. It is possible that the chosen factor 30/11 is merely a simplification
of this convenient value, derived more by empirical observation than by demonstra-
tive reasoning.
THE SECANT METHOD
The last part of this discussion of approximation techniques consists of a refine-
ment of the method of successive approximations which, as the author justly boasts,
significantly improves the speed of the correction. Again, to estimate the value of
the Sine of an angle whose true Chord (henceforth called C) is known, one uses
the initial approximation Sin0 5 C 2 «, from which the corresponding Versine is
computed as before by
Vers 5 R 2 Cos 5 R 2 ÏR2 2 Sin2.
HM 23 THE SECANT METHOD IN INDIAN MATHEMATICS 253

Now, however, the iterative technique is modified by requiring the user to calculate
from each Sink and Versk a kth approximation to C, to wit,
Crdk 5 ÏSin2k 1 Vers2k.

Once Vers0 , Vers1 and Crd0 , Crd1 have been obtained, the user is instructed to
apply the following formula:

Versk11 5 Versk 6 U (C 2 Crdk )(Versk21 2 Versk )


Crdk21 2 Crdk
UH 1 if C . Crdk ,
2 if C , Crdk .
(5)

This process of finding two successive approximations, and then computing the
third by means of this difference rule, is repeated until the result is ‘‘aviśes.a,’’ that
is, until Crdk 5 C. This ingenious technique does indeed accelerate the correction,
as we now show.
In terms of modern numerical analysis, we may think of the original procedure
as an iterative equation for successive approximations to the Versine, namely,
Versk11 5 F(Versk ) 5 R 2 ÏR2 2 Sin2 5 R 2 ÏR2 2 (C 2 2 Vers2k). (6)

Upon substituting R 2 ÏR2 2 Sin2 for Vers, this is immediately seen to be identical
to Eq. (4), the iterative equation for Sink11 . Similarly, inspection of F(Vers) and
its derivative reveals that the convergence conditions are also equivalent to those
given for the Sine expression, namely, 0 , C , R and Vers0 # C.
Then what is the significance of Parameśvara’s difference rule? If we express
Crd as a function of Vers alone, namely,
Crdk 5 Ï2R Versk ,
and define
f (Vers) 5 C 2 Crd(Vers) 5 C 2 Ï2R Vers, (7)
we see that the difference rule is simply a form of the secant method, here used
for finding a root of f (Vers) 5 0. For that method gives
f (Versk )(Versk 2 Versk21)
Versk11 5 Versk 2
f (Versk ) 2 f (Versk21)
(C 2 Crdk )(Versk 2 Versk21)
5 Versk 2 (8)
(C 2 Crdk ) 2 (C 2 Crdk21)
(C 2 Crdk )(Versk 2 Versk21)
5 Versk 2 ,
Crdk21 2 Crdk

which is identical to the formula from the Siddhāntadı̄pikā shown in Eq. (5). (The
correctness of the given rule for the sign of the correction term is obvious when
one considers that (Versk 2 Versk21) and (Crdk21 2 Crdk ) will always have opposite
254 KIM PLOFKER HM 23

sign: Versk . Versk21 implies Cosk , Cosk21 and therefore Sink . Sink21 and
Crdk . Crdk21).
Since the rate of convergence of the method of successive approximations can
be very slow and that of the secant method is superlinear [8, 40], Parameśvara is
quite justified in his claim that by this method, ‘‘the correction is quickly accom-
plished.’’ For example, when applied to Parameśvara’s sample problem, the method
of successive approximations alone requires 15 iterations to converge to
Vers 52;30 5 1344, but when alternated with the secant rule, it converges in 6.
(Actually, Vers 52;30 5 1345, but since Āryabhat.a’s Sine table establishes Crd
52;30 5 3040 instead of the more accurate 3041, no greater accuracy can be obtained
in this calculation.5) If, after obtaining the first two approximations, the iterative
equation were abandoned and the secant method alone used, only four iterations
would be required for convergence; but as it is, the improvement is significant.
Moreover, the technique using the secant method has the great advantage that it
may converge even when C is greater than R; in fact, since the first and second
derivatives of f (Vers) with respect to Vers are nonzero everywhere, the secant
method converges for all values of C [5, 150–152]. Parameśvara gives no explicit
indication that he is aware of this advantage; yet in his example he chooses a value
of C quite close to R (in fact, it is the closest of all the values of C less than R that
can be easily picked out of Āryabhat.a’s table), where the improvement in the rate
of convergence is especially striking. It seems unlikely that he would have stumbled
upon such a neat example without understanding the first technique’s dependence
upon the value of C, but it seems no less peculiar that he should have understood
it yet neglected to discuss it.

DISCUSSION OF ATTRIBUTION
Although Parameśvara does not claim to have discovered this rule himself, it is
not, like other rules in this collection, explicitly attributed to the mysterious ‘‘some
others’’ [6, 204–205; 2, 95]. A similar interest in the question of the speed of
the convergence of the method of successive approximations, as well as a similar
willingness to modify the method for the sake of faster convergence, is displayed
in Parameśvara’s later work, the Goladı̄pikā of 1443 (IV 20–23 [11, 48–52]). Al-
though the secant rule itself is not repeated there, it seems reasonable to suppose
that it was in fact Parameśvara (probably motivated by a desire to reduce the
immense labor of computation that would naturally fall to the lot of such a conscien-
tious observational astronomer) who was responsible for its formulation.
It is difficult to determine exactly where in the general history of this method
Parameśvara’s discussion should be placed. Although the closely related iterative
technique known as regula falsi appears in Indian texts as early as the fifth century
(cf. [9, 498–499]), the secant method proper is somewhat more elusive. It may be

5
Parameśvara certainly knew of more precise Sine values; in fact, these very verses are part of his
commentary on Govinda’s corrections to Āryabhat.a’s table (see also [3]). The cruder integer values,
however, are adequate to illustrate the given method, and much easier to manipulate in computation.
HM 23 THE SECANT METHOD IN INDIAN MATHEMATICS 255

considered either as an iterated version of the regula duarum falsarum positionum


or ‘‘al-khat. a‘ayn rule,’’ or as a variant of the Newton–Raphson method in which
a difference quotient is substituted for the derivative; I have been unable to locate
with certainty an early use of either form. The invention of the al-khat. a‘ayn rule
was ascribed by Arab mathematicians to the Indians [14, 45], but the iterated form
of the rule does not seem to be explicitly attested, although it has been suggested
[1] that it was probably used by Fibonacci to find a root of a cubic polynomial. The
secant modification of the Newton–Raphson method as a general technique for
approximating roots of both algebraic and transcendental functions was ‘‘discov-
ered’’ (repeatedly) in the 20th century, although it seems unlikely that it was
unknown earlier. In fact, it has been remarked that the secant method may claim
‘‘the distinction of being the most often rediscovered I.F. [iteration function] in the
literature’’ [13, 109]. It is now apparent that the somewhat sketchy history of this
useful technique must be extended to include the members of the Mādhava school
in the southernmost part of India at least as early as the 15th century.

ACKNOWLEDGMENTS
I am grateful to Professor David Pingree for suggesting the study of the Siddhāntadı̄pikā’s sine rules,
and for his subsequent comments on this paper. In addition, I am indebted to Dr. Davide P. Cervone
for several useful discussions on the subject of convergence conditions.

REFERENCES
1. Stanislaw Glushkov, On Approximation Methods of Leonardo Fibonacci, Historia Mathematica 3
(1976), 291–296.
2. R. C. Gupta, Second Order Interpolation in Indian Mathematics up to the Fifteenth Century, Indian
Journal of History of Science 4 (1969), 86–98.
3. R. C. Gupta, Fractional Parts of Āryabhat.a’s Sines and Certain Rules Found in Govindasvāmi’s
Bhās.ya on the Mahābhāskarı̄ya, Indian Journal of History of Science 6 (1971), 51–59.
4. R. C. Gupta, An Indian Form of Third Order Taylor Series Approximation of the Sine, Historia
Mathematica 1 (1974), 287–289.
5. Alston S. Householder, The Numerical Treatment of a Single Nonlinear Equation, New York:
McGraw–Hill, 1970.
6. T. S. Kuppanna Sastri, Mahābhāskarı̄ya of Bhāskarācārya with the Bhās.ya of Govindasvāmin and
the Supercommentary Siddhāntadı̄pikā of Parameśvara, Madras: Government Oriental Manuscripts
Library, 1957.
7. Otto Neugebauer and David Pingree, The Pañcasiddhāntikā of Varāhamihira, 2 vols., København:
Royal Danish Academy of Sciences and Letters, 1970–1971.
8. Alexander M. Ostrowski, Solution of Equations and Systems of Equations, New York: Academic
Press, 1966.
9. David Pingree, The Paitāmahasiddhānta of the Vis.n.udharmottarapurān.a, Brahmavidyā 31–32 (1967–
1968), 472–510.
10. David Pingree, History of Mathematical Astronomy in India, in Dictionary of Scientific Biography,
ed. Charles Gillispie, 16 vols., 2 supps. New York: Charles S. Scribner’s Sons, 1978, 15: 533–633.
11. K. V. Sarma, The Goladı̄pikā by Parameśvara, Adyar Library Pamphlet Series, No. 32, Madras:
Adyar Library and Research Centre, 1957.
256 KIM PLOFKER HM 23

12. K. V. Sarma, Tantrasaṅgraha of Nı̄lakan.t.ha Somayāji with Yuktidı̄pikā and Laghuvivr.ti of Śaṅkara,
Hoshiarpur: Vishveshvaranand Vishva Bandhu Institute of Sanskrit and Indological Studies, 1977.
13. Joseph F. Traub, Iterative Methods for the Solution of Equations, Englewood Cliffs, NJ: Prentice–
Hall, 1964.
14. A. P. Youschkevitch, Les mathématiques arabes, trans. M. Cazenave and K. Jaouiche, Paris: Librarie
Philosophique J. Vrin, 1976.

Das könnte Ihnen auch gefallen