SERIES ARTICLE

Dawn of Science
20. Calculus is Developed in Kerala

T Padmanabhan

Foundations of calculus were developed by a sc-

hool of mathematicians in Kerala during 1400{
1600, years before similar developments took
place in the west.
An ancient text from Kerala has a verse, giving the cir-
T Padmanabhan works at cumference of a circle, the translation of which goes as:
IUCAA, Pune and is
interested in all areas \Multiply the diameter by four. Subtract from it and add
of theoretical physics, to it alternatively the quotients obtained by dividing four
especially those which times the diameter to the odd numbers 3,5 etc. "
have something to do with
gravity. In modern notation this reduces to a remarkable innite
series expansion for =4 given by
1 1 1
= 1 + + : (1)
4 3 5 7
Previous parts:
Res onance, Vol.15: p.498, This result { usually attributed to Gregory (1638{1675)
p.590, p.684, p.774, p.870, and others who were to be born more than a century
p.1009, p.1062; Vol.16: p.6,
after the text in question was authored { is one of the
p.110, p.274, p.304, p.446,
p.582, p.663, p.770, p.854, many gems to be found in the ancient Kerala texts (espe-
p.950, p.103; Vol.17: p.6. cially the one called Yuktibhasha) which laid the foun-
dation for a branch of mathematics that we now call
calculus.
The discovery of the contributions from the Kerala school
of mathematicians is comparatively recent and many de-
tails are still being probed extensively. This is inspite
of the fact that some of these ancient Kerala texts have
been referred to explicitly in an article by C M Whish
who { having learnt Malayalam and collected palm leaf
Keywords manuscripts from Kerala { found to his astonishment a
Yuktibhasha, Madhava, tantra- \complete system of uxions" in them. He published a
samgrha, Kerala calculus.
paper in 1834 in the Transactions of the Royal Asiatic

106 RESONANCE February 2012

 SERIES ARTICLE

)WHEN

Figure 1.

Figure 2.

WHERE

RESONANCE February 2012 107

 SERIES ARTICLE

Virtually all these Society of Great Britian and Ireland with a fairly ex-
results were obtained plicit title `On the Hindu quadrature of the circle and
by using techniques the innite series of the proportion of the circumfer-
which can be thought ence to the diameter exhibited in the four sastras, the
of as belonging to Tantrasangraham, Yukti-Bhasha, Carana Padhati and
calculus Sadratnamala'.
In addition to the result quoted above, these texts con-
tain the innite series expansion for tan1 (x) (from which
the above result is but one step; you just put x = 1),
the series expansion for sin and cos, and the indenite
integral of xn { just to list a few. Virtually all these
results were obtained by using techniques which can be
thought of as belonging to calculus.
Who are these mathematicians and how did they de-
velop these techniques? Though many of the details are
still somewhat sketchy, historians of Indian mathemat-
ics have now put together the following picture. Most
of these developments took place in villages around a
river called Nila in the ancient days (and currently called
river Bharatha, the second longest river in Kerala) dur-
ing 1300{1600 or so. One of the key villages was called
Sangama-grama in the ancient times and is thought of
as referring to the village Irinhalakkuta (about 50 km to
the south of Nila) in present day Kerala. (There are a
few other candidates, like Kudalur and Tirunavaya, for
which one could raise arguments in favour of them be-
ing the Sangama-grama; so this issue is not completely
settled.) What is more certain was the existence of a re-
markable lineage of mathematicians in Sangama-grama
of which Madhava ( 1350{1420) seems to be the one
The infinite series for who discovered many of the basic ideas of calculus. The
sine, cosine, arctan innite series for sine, cosine, arctan as well as rudiments
as well as rudiments of integration are attributed to Madhava by many later
of integration are sources. He was strongly inuenced, like many other
attributed to Indian scholars of that period, by Aryabhata { which
Madhava by many is no surprise since Aryabhateeyam (circa 500 AD) was
later sources. a very inuential text in India as well as (through its

108 RESONANCE February 2012

 SERIES ARTICLE

translations) in the Arab world and medieval Europe. Another student of

Dhamodara was
Madhava founded a lineage which lasted possibly till Jyeshtadeva
the late seventeenth century. His student Paramesh-
(~ 15001610), the
wara ( 1360{1455) was an astronomer-mathematician
author of Yuktibhasha
who had authored more than two dozen works and made
which probably has
astronomical observations for a very extended period of
the clearest exposition
nearly ve decades. It appears that his son Dhamodara
of some of the topics
was the teacher of another key gure, Neelakanta So-
in calculus.
mayaji ( 1450{1550) who is the author of Tantrasan-
graha which contains extensive discussions on astron-
omy and related mathematics. Another student of
Dhamodara was Jyeshtadeva ( 1500{1610), the author
of Yuktibhasha which probably has the clearest exposi-
tion of some of the topics in calculus.
The quotation in the beginning of this article is quoted
in Yuktibhasha attributing it to Tantrasangraha which,
in turn, attributes the result to Madhava. Similarly
there is another detailed verse in Yuktibhasha which
attributes to Madhava the innite series expansion for
arctan. The rst sentence of Yuktibhasha states that
it is going to explain all the mathematics \... useful
in the motion of heavenly bodies following Tantrasan-
graha" but actually does much more as an independent
treatise than work merely as a bhasya. (It is also re-
markable for being written in Malayalam { not Sanskrit
{ and is in prose, not poetry.) Unlike Tantrasangraha,
this work provides detailed argumentation for the re-
sults and is written in a style which is straightforward
and unpretentious. (An English translation with de- Unlike
tailed commentary is now available [1]). Tantrasangraha,
Yuktibhasha
Here, I will concentrate on two specic results in Yuk- provides detailed
tibhasha and highlight the role of calculus, as we call it argumentation for the
today, in it. Let us begin with the series expansion for results and is written
=4. To understand what is involved in obtaining the
in a style which is
innite series expansion for =4 given above, it is useful
straightforward and
to recall the procedure in modern notation. If t = tan
unpretentious.

RESONANCE February 2012 109

 SERIES ARTICLE

It is clear that the with 0 (=4), then (dt=d) = 1 + t2 . Rewriting

derivation involves the this equation by taking the reciprocals (which is actu-
concepts of ally a rather insightful step!) and using the geometric
differentiation and series expansion, one gets (d=dt) = 1 t2 + t4 + .
integration in some Integrating both sides between 0 and 1 will give you
form as well as the tan1 (1) = =4 as an innite series. Of course, the in-
knowledge of denite integral also gives the innite series expansion
indefinite integrals of of in terms of tan . It is clear that the derivation
powers. involves the concepts of dierentiation and integration
in some form as well as the knowledge of indenite in-
tegrals of powers. How did the ancient mathematicians
do this?
The basic idea was to consider the geometry of the circle
and give meaning to all the relevant quantities in terms
of suitably dened geometrical constructions . In Fig-
ure 1, AB is a unit tangent at A, and P is an arbitrary
point on AB so that the length AP gives t = tan . Our
problem then reduces to nding the length of the arc
AP0 corresponding to the length AP which will give us
as a function of t. Yuktibhasha begins by dividing the
line AB into a large number n of equal segments mark-
ing out A0 = A; A1 ; A2; :::; An = B with the line OAi
intersecting the arc at Qi . The key insight is to real-
ize that as the half chord of a segment gets smaller and
smaller, its length approaches the arclength. In modern
language

Figure 3. Construction used

in Yuktibhasha.

110 RESONANCE February 2012

 SERIES ARTICLE

language this corresponds to the realization sin(=n) ! If the segments of

=n as n ! 1. There is an explicit statement in Yuk- the side ... are
tibhasha to this eect: \If the segments of the side ... very very small,
are very very small, these half chords will be almost the these half chords
same as arc segments." With this realization, as well as will be almost the
some very innovative geometrical reasoning, one obtains same as arc
the dierential relationship = [t=(1 + t2)] between segments.
t = (1=n) (given by Ai Ai+1) and (given by Qi Qi+1 )
which is the relevant arc length in modern notation.
To get we now need to sum the series (1 + i2 =n2 )1
from i = 1 to n and nally take the limit n ! 1.
Yuktibhasha, of course, cannot do this directly and hence
resorts to expanding the right-hand side into a series:

1 i2 i4
i = 3 + 5 : (2)
n n n
(Interestingly enough, this is done by a powerful tech-
nique of recursive rening { called samskaram, in Yuk-
tibhasha.) This reduces the problem to one of nding
the sums of powers of integers. While the sum of ik was
known already to Aryabhata for k = 1; 2; 3, one did not
know the result for higher values of k. Once again Mad-
hava uses the fact that he only needs the result for large
n and any sub-dominant term can be discarded. This
limiting process is done with great care in Yuktibhasha,
nally leading to the result (called samkalitam, or `dis-
crete integration') which in modern notation gives
n
1 X 1
lim ik = : (3)
n!1 nk+1 i=1
k+1
This limiting process
k is done with great
This result (equivalent to integrating x and obtaining
xk+1 =(k + 1)) was obtained using a procedure which re- care in Yuktibhasha,
duces the sum of k-th powers to the sum of (k 1)-th finally leading to the
power, a technique we now call mathematical induction. result (called
From reading the relevant section of Yuktibhasha it is samkalitam, or
clear that its author was fully aware of the dierence discrete integration)

RESONANCE February 2012 111

 SERIES ARTICLE

The clarity of the between this method of proof and the more traditional
exposition regarding approaches usually based on geometrical reasoning; in
ideas like limiting fact, this is probably the rst known instance of proof by
process and mathematical induction in Indian mathematics. With
infinitesimals shows this step completed, the result for the series can be ob-
how adept Madhava tained fairly easily. The clarity of the exposition regard-
was at manipulating ing ideas like limiting process and innitesimals shows
these to achieve his how adept Madhava was at manipulating these to achieve
goal. his goal.
Equally { if not more { innovative is the approach used
to obtain the series for sines and cosines. The rst part
of the analysis is again geometrical and closely parallels
the discussion given above. By similar reasoning one
can obtain the dierential (again in modern notation)
d(sin ) = cos d. In fact this is done with a clever ge-
ometrical construction leading to a discretized dierence
formula for sine and cosine of the form

si si+1 si1 = 2s1 ci ; ci ci+1 ci1 = 2s1 si ;

(4)
where si sin(i=2n) and ci cos(i=2n) for an angle
divided into 2n equal parts. Unlike the case of the arctan
series, one cannot now integrate this equation because
the derivative of sin itself is not a simple function to
handle. The way Yuktibhasha handles this situation is
absolutely ingenious.
It converts the two coupled rst order dierential equa-
tions for sine and cosine into a second order dierence
equation by introducing the quantity 2 fi fi+1
fi1 , where fi is si or ci and obtaining the result 2 fi =
4s21 fi . This is, of course, nothing but the dierential
This is a discretized equation f 00 + f = 0 in discretized form. The next step
version of the is to solve the discretized equation by adding up the
fundamental theorem second dierences to get rst dierences and adding up
of calculus and again the rst dierences to get the function itself. This is a
requires careful discretized version of the fundamental theorem of cal-
handling of limits. culus and again requires careful handling of limits. The

112 RESONANCE February 2012

 SERIES ARTICLE

relevant sections of Yuktibhasha dwell on this important This leads to a

issue showing clearly that the author was aware of these samkalitam which, in
subtleties. the first instance,
produces what we
This leads to a samkalitam which, in the rst instance,
would now call an
produces what we would now call an integral equation
for sin in the discretised form. This equation is solved integral edsquation
by another nite samkalitam by repeated substitution for sin in the discre-
of the equation into itself. which requires calculating a tised form.
set of sums of sums (samkalitasamkalitam, as you might
have guessed P the name should be!) dened recursively
i
by Sk (i) j=1 Sk1 (j) with appropriate boundary
conditions. The result for these sums of sums is given in
Yuktibhasha but this is one instant in which the proof is
not provided.
All this is done keeping a nite n; in fact Yuktibhasha
generally uses the technique of delaying the step of tak-
ing the limit as much as possible. After the relevant
sums are done, the time is ripe to make n tend to inn-
ity and the k-th term in the expansion of the sine series
pops out to have the right value (1)k 2k+1=(2k + 1)! {
and one obtains the innite series expansion for sine. In
modern langauage what has been achieved is equivalent
to converting the dierential equation for sin into an
integral equation of the form
After the relevant
Z Z sums are done, the
sin = d d sin (5) time is ripe to make
0 0
n tend to infinity and
and iterating it repeatedly to give the k-th term in the
3 expansion of the
sin = + + sine series pops out
3!
Z Z 2k1 to have the right
k
(1) d1 d2k sin 2k : (6) value of the sine serie
ansion
0 0
(1)k2k+1=(2k + 1)!
During the entire process, Yuktibhasha does not lose and one obtains
sight of the need for taking n ! 1 limit in the end. the infinite series
The word used in this context is sunya-prayam which expansion for sine.

RESONANCE February 2012 113

 SERIES ARTICLE

Box 1. A Piece of Pi
Estimating the circumference of a circle of given radius has intrigued all the ancient
geometers and astronomers. Most of them used the procedure of inscribing and cir-
cumscribing the circle by a polygon of a large number of sides and approximating the
circumference between the perimeters of the two polygons. This, of course, requires one
to nd the side of the polygon as a function of the radius and consequent development
of trigonometry.
While the approximate value of ' (22=7) was available since antiquity, better results
were known to many ancient mathematicians. Archimedes (287{212 BC) used a polygon
of 91 sides to get 3 10
71
< < 3 10
70
. Aryabhatiiya (500 AD) gives the \... circumference of
a circle of diameter 20000 is 62832" getting 3:1416 while Bhaskara ( 1114{1185)
gives the circumference to be 3927 for a circle of diameter 1250 obtaining 3:14155.
What is remarkable about Yuktibhasha is that it does not try to give yet another ap-
proximate value for but an exact innite series expansion { which is a giant conceptual
leap forward. In fact, in addition to the result described in the text, Yuktibhasha also
gives an algebraic recursive method for nding
p . In modern notation this corresponds
1
to using the recursion relation xn+1 = xn 2
1 + xn 1 with x0 = 1 and obtaining
as the limit

= 4 lim 2n xn : (a)
n!1

This method, however, requires evaluation of square roots which is not an easy procedure
while the series expansion given in the text has no square roots.

has a literal translation of `being similar to zero'; today,

we would have called it innitesimal! There is also the
usage of the word yathestam meaning `as one wishes'
while, for example, talking about dividing lines into ar-
bitrarily large subdivisions.
Yuktibhasha also obtained the surface area and volume
of the sphere by integration of the innitesimal elements.
For example, surface area of a sphere of radius R is
computed by considering a small strip between circles of
latitudes and + d in, say, the upper hemisphere of
the sphere. The area of this strip is shown to be equal to
dA = (2R sin )(Rd) by very careful and meticulous
reasoning. In the discretised version equals =2n and
= i=2n with i = 0; 1; 2; :::; 2n and the total area is
obtained by summing over all strips labelled by i and

114 RESONANCE February 2012

 SERIES ARTICLE

then taking the limit of n ! 1. This is essentially the Historians are still
same as nding the integral of sin as cos . Though unsure whether the
it was known that rst dierence of cosine is sine, this developments on
summation is done by eectively going through the orig- the banks of Nila
inal steps once again. influenced
corresponding later
Historians are still unsure whether the developments on
developments in the
the banks of Nila inuenced corresponding later devel-
west and as far as
opments in the west and { as far as we know { there is
we know there is
no direct evidence that it did. (Kerala, of course, had
direct linkage with the west, for example through trad- no direct evidence
ing, from very ancient days.) It is also somewhat unclear that it did.
how the Nila tradition died down after the seventeenth
century though several obvious historical factors can be
identied as causes for its demise.
Acknowledgement
I have benetted signicantly from several discussions
with P P Divakaran on this subject.

Suggested Reading
Address for Correspondence
[1] K V Sarma, Ganita-Yukti-Bhasa of Jyesthadeva, Hindustan Book Agency, T Padmanabhan
2008. IUCAA, Post Bag 4
[2] P P Divakaran, Notes on Yuktibhasha: The birth of calculus, Indian Pune University Campus
Journal of History of Science, 2011 [in press]; Calculus in India: The Ganeshkhind
historical and mathematical context, Current Science, Vol.99, p.8, 2010. Pune 411 007, India.
[3] S G Rajeev, Neither Newton nor Leibnitz: The pre-history of calculus in Email: [email protected]
medieval Kerala, Lectures at Canissius College, Buffalo, New York, [email protected]
2005, available at http://www.river-valley.com/tug

RESONANCE February 2012 115