GENERAL  ARTICLE

Madhava, Gregory, Leibnitz, and

Sums of Two Squares
Shailesh A Shirali

This article describes a connection between the

Gregory{Leibnitz series for pi and the represen-
tations of positive integers as sums of two squares.
It emerges from a study of the geometry of lat-
tice points in the plane.
Shailesh Shirali heads the
Community Math Centre 1. The M¹
adhav¹
a{Gregory{Leibnitz Series for ¼
in Rishi Valley School and
works in the field of The identity generally called the Gregory{Leibnitz se-
teacher education. He is
ries,
the author of many books
and articles in mathemat- ¼ 1 1 1 1
ics, written for interested = 1¡ + ¡ + ¡ ¢¢¢ ; (1)
4 3 5 7 9
students in the age range
13–19 years. He also has a was ¯rst proved by the Kerala mathematician M¹adhav¹a
close involvement with the in the fourteenth century (see [3] for details). Here we
mathematical olympiad
movement in the country.
describe a curious connection between the M¹adhav¹a se-
ries and the representations of positive integers as sums
of two squares; it emerges from a study of the geometry
of lattice points in the plane. Like so many other things
The connection
in mathematics, it was ¯rst pointed out by Gauss. (See
between the Madhava [2] for another account of this connection.)
series and
representation of There are many ways of proving (1), the simplest being
integers as sums of term by term integration of both sides of the following
two squares was first identity from x = 0 to x = 1,
noted by Gauss. 1
= 1 ¡ x2 + x4 ¡ x6 + ¢ ¢ ¢ + (¡1)n¡1 x2n¡2
1 + x2
x2n
+ (¡1)n ;
1 + x2
Keywords followed by showing that
Gregory–Leibnitz series, lattice Z 1 2n
points, sums of two squares, x
2
dx ! 0; as n ! 1:
Gauss circle problem. 0 1+x

116 RESONANCE  February 2010

 GENERAL  ARTICLE

R ¼=4
Another proof comes from evaluating 0
tann x dx, us- f(n) is the number of
ing integration by parts. lattice points within
2. Lattice Points Within a Circle the circle C (n) which
p is centered at the
Let C(n) denote the circle with radius n, centered at origin and has radius
the origin, (0; 0), and let f (n) denote the number of p
n.
lattice points within this circle. An equivalent de¯nition
for f(n), more convenient for computational purposes,
is: f (n) is the number of pairs (x; y) of integers for which
x2 + y 2 does not exceed n.
It seems plausible that f (n) is approximately equal to
the area of C(n), i.e., that [f (n)=(¼n)] is close to 1, when
n is large. The ¯gures bear this out convincingly:

n ¼n f(n) f(n)=¼n
101 31:4159 37 1:17775
102 314:159 317 1:00904
3
10 3141:59 3149 1:00236
104 31415:9 31417 1:00003
105 314159:3 314197 1:00012
6
10 3141592:6 3141549 0:99999

The ¯gure f(10) = 37 may be checked by hand calcula-

tion.
To formally prove that [f (n)=(¼n)] ! 1 as n ! 1, we
associate a 1 £ 1 square with each lattice point in C(n),
centered at that lattice point, with sides parallel to the
coordinate axes. Figure 1 shows this for the case n = 11,
with the squares shown shaded. The total area of the
shaded region is clearly f(n).
Within each of these unit squares, the point furthest
from the center is a corner, its distance from the center
being p12 . Therefore, the circle with center (0; 0) and
p
radius n + p12 completely encloses the shaded region,

RESONANCE  February 2010 117

 GENERAL  ARTICLE

Figure 1. Lattice points con-

tained within C (n), for n =
11; the three circles have
p p
radii, n ¡ p12 ; n , and
p
n +p p1 , respectively;
2
the smallest circle lies com-
pletely within the shaded
region, and the largest
circle completely encloses
this region.

p
and the circle with center (0; 0) and radius n ¡ p12 lies
completely within the shaded region. Hence:
µ ¶2 µ ¶2
p 1 p 1
¼ n¡ p  f(n)  ¼ n+ p : (2)
2 2
It follows that
r r
1 2  f(n)  1 2
1+ ¡ 1+ + ; (3)
2n n ¼n 2n n
and therefore that [f(n)=(¼n)] ! 1 as n ! 1.
Relation (2) yields simple bounds for the error term,
f(n) ¡ ¼n:
¯ ¼ ¯¯ p
¯
The problem of ¯f (n) ¡ ¼n ¡ ¯ < ¼ 2n: (4)
2
representing integers
as sums of two 3. Which Numbers are Representable as Sums
squares was first of Two Squares?
studied by the Greek
mathematician The problem of representing integers as sums of two
Diophantus. squares is a very well studied topic, the ¯rst known re-
sults being due to the Greek mathematician Diophantus

118 RESONANCE  February 2010

 GENERAL  ARTICLE

(third century AD). Let n be a positive integer. It is Fermat used the

known that: principle of infinite
descent to show that a
1. If n ´ 3 (mod 4), then n cannot be written as a sum prime number of the
of two squares. Example: 7 is such a number. More form 4k+1 is
generally, if n contains a prime factor p ´ 3 (mod 4) expressible as a sum
raised to an odd power, then n cannot be written as of two squares.
a sum of two squares. Example: 78 = 2 £ 3 £ 13 is
such a number.

2. If n is a prime number of the type 1 (mod 4), then

n can be written as a sum of two squares in precisely
one way. Example: 13 = 22 +32 . (We do not consider
13 = 32 + 22 as distinct from 13 = 22 + 32 .) In this
case the total number of pairs (x; y) of integers such
that x2 + y 2 = n is 8. For example, for n = 13 we get
the pairs (§2; §3) and (§3; §2). This result is due to
Fermat, who proved it using the principle of in¯nite
descent (see [3]). It is one of the gems of elementary
number theory, and quite a challenge to prove.

3. For the general case, when n is an arbitrary positive

integer, a formula for the number of pairs (x; y) of
integers such that x2 + y 2 = n was ¯rst found by
Jacobi. It is expressed in terms of the number of
divisors of n, thus: Let the number of divisors of n
of the type 1 (mod 4) be d1(n), and let the number of
divisors of n of the type 3 (mod 4) be d3 (n). Then the
number of pairs (x; y) of integers such that x2 +y 2 = n
is equal to 4 [d1 (n) ¡ d3 (n)].
Example: Take n = 65. The divisors of 65 are
1; 5; 13; 65. These are all of the form 1 (mod 4), so
d1 (n) = 4, d3 (n) = 0, 4 [d1 (n) ¡ d3 (n)] = 16. The in-
teger pairs (x; y) for which x2 +y2 = 65 are (§1; §8),
(§8; §1), (§4; §7), (§7; §4), and these are indeed
16 in number.
Or take n = 39. The divisors of 39 are 1; 3; 13; 39, so
that d1 (n) = 2, d3 (n) = 2, d1 (n) ¡ d3(n) = 0. And

RESONANCE  February 2010 119

 GENERAL  ARTICLE

Jacobi’s theorem
indeed there are no pairs (x; y) of integers for which
for f(n) allows us to
x2 + y 2 = 39.
write it as a Jacobi used elliptic functions and `q-series' to prove
summation. his identity (see [4] for details), so his was certainly
not an elementary proof.

4. A Formula for f (n)

Jacobi's result allows us to write f(n) as a summation:
n
X
f (n) = 1 + 4 [d1 (k) ¡ d3 (k)] ;
k=1

the `1' coming from the representation of 0 as 02 + 02.

Let us write this as
n
X Xn
f (n) = 1 + 4 d1(k) ¡ 4 d3 (k) ; (5)
k=1 k=1

and see what each sum on the right might signify. Con-
sider ¯rst the quantity
n
X
d1(k) = d1 (1) + d1 (2) + d1 (3) + ¢ ¢ ¢ + d1 (n):
k=1

For convenience we introduce a symbol ai;j to codify

divisibility:
(
1; if i divides j;
ai;j = (6)
0; if i does not divide j:
P
Then d1 (k) = a1;k +a5;k +a9;k +a13;k +¢ ¢ ¢ = i´1 (mod 4) ai;k ,
and so:
Xn X n X
d1 (k) = ai;k :
k=1 k=1 i´1 (mod 4)

As only ¯nite summations are involved, we may freely

interchange the order of summation:
n
X X n
X
d1 (k) = ai;k :
k=1 i´1 (mod 4) k=1

120 RESONANCE  February 2010

 GENERAL  ARTICLE

P
The quantity nk=1 ai;k is easily computed: it is simply
the number of multiples of i not exceeding n; hence
n
X jnk
ai;k = ; (7)
i
k=1

where bzc denotes the largest integer not exceeding z.

It follows that
X n jnk jnk jnk j n k
d1 (k) = + + + + ¢¢¢ :
1 5 9 13
k=1

In the same way we get:

n
X jnk jnk jnk jnk
d3(k) = + + + + ¢¢¢ :
3 7 11 15
k=1

We thus get an unexpected formula for the number of

lattice points enclosed by C(n):
µj k j k j k
n n n
f (n) = 1 + 4 ¡ +
1 3 5
jnk jnk j n k ¶
¡ + ¡ + ¢¢¢ : (8)
7 9 11
Since [f(n)=n] ! ¼ as n ! 1, it follows that
µ
1 jnk jnk jnk jnk
limn!1 ¡ + ¡
n 1 3 5 7
jnk j n k ¶
¼
+ ¡ + ¢¢¢ = : (9)
9 11 4

This identity serves as a nice companion to the M¹adhav¹a{

Gregory{Leibnitz result quoted earlier!
1
The ‘big O’ notation used here
5. Closing Remarks has the following sense:
f(x) = O(g(x)) if there exists a
5.1 Gauss Circle Problem number M > 0 such that
f(x)<M g(x)for all
Relation (4) implies the following relation:1 sufficiently large values of x.
For example, x3 + 2x2 = O(x3),
1=2
f(n) ¡ ¼n = O(n ): (10) sin x = O(1), ; xlnx = O(x2).

RESONANCE  February 2010 121

 GENERAL  ARTICLE

The proof of (10) being quite easy, we may feel we can

The Gauss circle
get better results by using stronger methods. But this
problem is regarded
turns out to be much more challenging than anticipated!
as difficult, and
remains open. The determination of the least value µ of c such that
f(n)¡¼n = O(nc ) is known as the Gauss circle problem.
From the above we have µ  1=2. A literature search re-
veals that successively tighter upper bounds for µ have
been obtained over the decades: 1=3 (Voronoi, 1903;
Sierpi¶
nski, 1906), 17=53 (Vinogradov, 1935), and more
recently, 131=416 (Huxley, 2003). In the other direc-
tion it is known that µ ¸ 1=4 (Hardy; Landau). In [5],
the determination of µ is described as a \very di±cult
problem", and it remains open; see [6].
5.2 The Three Squares Problem
Much less is known about the three-dimensional version
of this problem. The function now under study is the
cardinality g(n) of the set S(n) = f(x; y; z) j x; y; z 2
ZZ; x2 + y 2 + z 2  ng. For example, g(5) = 57, and
g(10) = 147. Here we expect that g(n) is close to 43 ¼n3=2
for large n, and the ¯gures bear this out. De¯ne h(n) =
g(n) ¥ 43 ¼n3=2 ; then we have:

h(100) ¼ 0:99528; h(1000) ¼ 0:99992; h(10000) ¼ 0:99978

By associating a cube of side 1 with each point (x; y; z) 2

S(n) and arguing as earlier, we may show that
à p !3 à p !3
4 p 3 4 p 3
3
¼ n¡
2
 g(n)  ¼
3
n+
2
;

and deduce that g(n) ¥ 43 ¼n3=2 ! 1 as n ! 1.

Further analysis along these lines becomes di±cult, as
there is no simple formula for the number of integer
triples (x; y; z) for which x2 + y 2 + z 2 = n. As earlier,
we may ask for the least value ¿ of c such that g(n) ¡
4
3
¼n3=2 = O(nc ). Some upper bounds obtained for ¿ are:

122 RESONANCE  February 2010

 GENERAL  ARTICLE

2=3 (Vinogradov, 1963), 29=44 (Chamizo and Iwaniec,

1995), 21=32 (Heath{Brown, 1999). But this problem
too remains open.

Suggested Reading

[1] S A Shirali, Euler, Nilkantha and pi, Resonance, Vol.2, No.5, pp.29–43,
1997.
[2] D Hilbert, and S Cohn-Vossen, Geometry and the Imagination, New York,
Chelsea, pp.27–39,1999.
[3] S A Shirali, Fermat’s Two-Square Theorem, Resonance, Vol.2, No.3,
pp.69–83, 1997.
[4] Eric W Weisstein, Sum of Squares Function. From MathWorld–A Wol- Address for Correspondence
fram Web Resource. Shailesh A Shirali
http://mathworld.wolfram.com/SumofSquaresFunction.html Rishi Valley School
[5] G H Hardy and E M Wright, An Introduction to the Theory of Numbers, 5th Rishi Valley 517 352
ed. Oxford, England, pp.267–272, 1979. Andhra Pradesh, India
[6] Eric W Weisstein, Gauss’s Circle Problem, From MathWorld – A Wolfram Email:
Web Resource. [email protected]
http://mathworld.wolfram.com GausssCircleProblem.html

RESONANCE  February 2010 123