1

EDUCATIVE COMMENTARY ON

ISI 2023 MATHEMATICS PAPERS

(Last Revisedd 07/06/2023)

Introduction 1

Paper 1 2

Paper 2 29

Concluding Remarks 43

Introduction
The original motivation for writing the annual commentaries on ISI B.Stat.
Entrance Tests is now waning. They began in 2016 because it was the year
in which some exciting questions came to the notice of the author. Later on
commentaries from 2014 were added. Unfortunately, after 2017, there was a
noticeable decline in the quality of questions. Some readers also brought to the
notice of the author that a few questions which were hailed in the commentary
were picked from some reputed international contests. This did not detract
from the intrinsic quality of the questions. Nor is it likely that some candidates
had seen the solutions already. But apart from loss of originality, it is an insult
to a good problem (for which originally ample time was allowed) to expect the
candidates to crack it in a couple of minutes.
Still, the annual commentaries on ISI Entrance Test continued more as a
mission, because some students pointed out that unlike the JEE, there are no
sources to get the answers of ISI Entrance tests. But things are changing here
too. In addition to the official key to the multiple choice questions in Paper
1, videos giving worked out solutions are available. As for the second paper,
solutions for some of the problems are presented on the Artofsolvingproblems.
But even with both the justifications for the commentaries losing their
strengths, I believe there is still some room to bring out the educative aspects
of some of the problems and solutions. So, here is the commentary on ISI 2022.
As in the other commentaries, unless otherwise stated, all the references
are to the author’s Educative JEE Mathematics, Third Edition, published by
Universities Press, Hyderabad.
I am thankful to the persons who made the question papers available to
me. The pure geometry solution to Q.3 in Paper 2 was contributed by Pankaj
Meena while the alternate solution to Q.6 in it was contributed by Saad Hassan.
Readers are invited to send their comments and point out errors, if any, in the
commentary. They may be sent either by e-mail (kdjoshi314@gmail.com) or by
an SMS or a WhatsApp message on mobile (9819961036).
2

ISI BStat-BMath-UGA-2023 Paper 1 with Comments

N.B. Each question has four options of which ONLY ONE is correct. There are
4 marks for a correctly answered, 0 marks for an incorrectly answered question
and 1 mark for each unattempted question.

Q.1 For a real number x,

x3 − 7x + 6 > 0
if and only if

(A) x > 2 (B) −3 < x < 1

(C) x < −3 or 1 < x < 2 (D) −3 < x < 1 or x > 2

Answer and Comments: (D). There is a formula to give the roots of

a cubic. But it it is far more complicated than the quadratic formula
and is seldom used. But when one root of a cubic can be found easily,
it can be factored as the product of a linear polynomial and a quadratic
polynomial. The roots of the latter are the missing two roots of the cubic.
In the present problem, x3 − 7x + 6 has 1 as an obvious root and it factors
as (x − 1)(x2 + x − 6) and further as (x − 1)(x + 3)(x − 2). Thus there
are three distinct real roots, viz. −3, 1 and 2 in ascending order. The
cubic is positive when all three factors are positive, which is the case for
x > 2, or when two of them are negative and the third positive, for which
to happen, x must lie between −3 and 1. Put together, (D) is the answer.
An absolutely simple problem since all three roots are easy to identify
and distinct. The problem would have been more interesting if there were
only one real root, or a double root.

Q.2 The number of consecutive zeroes adjacent to the digit in the units place
of 40150 is

(A) 3 (B) 4 (C) 49 (D) 50

Answer and Comments: (A). Since 401 = 400+1, we can expand 40150
by the binomial theorem. The last term would be 1. The question asks to
identify the terms before this and to count the number of trailing zeros in
their sum.
It is a little easier to rewrite (401)50 as (1 + 400)50 and find the numbers of
trailing zeros from the first terms onwards. The first term, viz. 1 has no
trailing zeros. The second term is 50 1 49


1 (400) 1 , i.e. 20000. When added
to 1, we get 20001 which has three zeros adjacent to 1. The remaining
terms will have higher powers of 100 dividing them. So when added to
20001, the tail end of the sum will remain 20001. So the answer is 3.
A simple but good problem on binomial expansion. Although the
reasoning is time consuming to write out, it does not take much time to
 3

conceive it and then the calculations are minimal. So the problem suits
well as an MCQ.
Q.3 Consider a right angled triangle △ABC whose hypotenuse AC is of length
1. The bisector of 6 ACB intersects AB at D. If BC is of length x, then
what is the length of CD?

B x C

r
2x2 1
(A) (B) √
r1 + x 2 + 2x
x 1
(C) (B) √
1+x 1 − x2

Answer and Comments: (A). Since CD2 = BC 2 + BD2 = x2 + BD2 ,

we shall be through if we can find BD. By the angle bisector
√ property
the ratio BD : DA equals x : 1. But BD + DA = BA = 1 − x 2 . Hence
√ x x2
BD = 1 − x2 × x+1 . So CD2 = x2 + BD2 = x2 + (1 − x2 ) × (x+1) 2.

2x2
which simplifies to 1+x . Hence (A) is the answer.
A simple trigonometric problem based on the Pythagoras theorem
and the angle bisection property. The calculations are somewhat prone to
mistakes of hand.
Q.4 Define a polynomial f (x) by

1 x x
f (x) = x 1 x
x x 1

for all x ∈ IR, where the right hand side is the determinant. Then the
roots of f (x) are of the form
(A) α, β ± iγ where α, β, γ ∈ IR, γ 6= 0 and i is a square root of −1.
(B) α, α, β where α, β ∈ IR are distinct.
(C) α, β, γ where α, β, γ ∈ IR are all distinct.
(D) α, α, α for some α ∈ IR.
4

Answer and Comments: (B). The notation for the determinant is so

standard that it was hardly necessary to specify what it meant. Perhaps
it is an implied hint that some properties of the determinant are to be
used. If we expand the determinant it will come out to be a cubic in x.
The problem then will be similar in spirit to Q.1 which also dealt with the
roots of a cubic. In the present problem, if we look at the determinant, we
see easily that it vanishes for x = 1 as two rows become identical. Hence
x − 1 is a factor of the determinant, or equivalently, 1 is a root of the cubic
f (x). Also, if you add the second and the third rows to the first, you get
2x + 1 as a factor of f (x). Hence x = − 12 is also a root.
What about the third root? We can expand f (x) and find it after factoring
(x − 1)(2x + 1) out. But there is a more intelligent way. If we are extra
sharp we will notice that (x − 1) will be a double factor, because putting
x = 1 makes the first two rows equal and this is enough for the determinant
to vanish. But putting x = 1 also makes the second and the third rows
equal, independently of the equality of the first two rows and hence x − 1 is
a factor again. (We cannot apply this reasoning to the first and the third
row now because their equality follows from that of the first two rows and
of the last two rows. So it cannot give us anything more. In particular, we
cannot conclude that (x − 1)3 is a factor of f (x).) In fact, once we know
that (x − 1)2 and 2x + 1 are factors of the determinant, there is no need
to expand it. Since it is a polynomial of degree 3 in x, the other factor, if
any, must be a constant and its presence does not affect the roots. As 1
is a double root and − 21 is the third root, the correct answer is (B).
The problem is a good way to reward those who work it without expansion.
But the MCQ nature of the question mars its beauty because as it happens,
in the present case a brute force expansion of the determinant and further
factorisation is not prohibitively lengthy. The problem is actually a part
of the JEE 2001 question discussed at the end of Comment No. 6 of
Chapter 24. That problem asked to find the number of distinct roots of
sin x cos x cos x
the equation cos x sin x cos x = 0 lying in the interval [−π/4, π/4].
cos x cos x sin x
Q.5 Let S be the set of those real numbers x for which the identity
∞
X
cosn x = (1 + cos x) cot2 x
n=2

is valid, and the quantities on both the sides are finite. Then

(A) S is the empty set.

(B) S = {x ∈ IR : x 6= nπ for all n ∈ ZZ}.
(C) S = {x ∈ IR : x 6= 2nπ for all n ∈ ZZ}.
(D) {S = {x ∈ IR : x 6= (2n + 1)π for all n ∈ ZZ}.
 5

Answer and Comments: (B). The L.H.S. is an infinite geometric series

with starting term cos2 x and common ratio cos x. So it will be finite if
and only if −1 < cos x < 1. So real numbers of the form nπ, for n ∈ ZZ are
to be excluded from S. To see if there are any additional restrictions, we
solve the equation for the values of x other than those of the form nπ. For
cos2 x
such values, the L.H.S. sums to . Hence the equation becomes
1 − cos x

cos2 x
= (1 + cos x) cot2 x (1)
1 − cos x
which simplifies to

cos2 x
cos2 x = (1 − cos2 x)
sin2 x
= cos2 x (2)

which is true for all x for which sin x 6= 0. But that happens only when
x is an integral nultiple of π, which is already excluded. Hence no new
restrictions are imposed by the equation. Therefore the solution se S is
{x ∈ IR : x 6= nπ, n ∈ ZZ}.
An absolutely simple problem.

Q.6 Let   
πθ 1 πθ
S= θ sin , cos : θ ∈ IR, θ > 0
1+θ θ 1+θ
1
and T = {(x, y) : x, y ∈ IR, xy = 2 }. How many elements does S ∩ T
have?

(A) 0 (B) 1 (C) 2 (D) 3

Answer and Comments: (B). The set T is a familiar hyperbola with

asymptotes along the axes. This is of little help here since the question
merely asks how many points (x, y) on this hyperbola lie in the set S.
πθ
For this to happen, x and y must be, respectively, of the form θ sin 1+θ
1 πθ πθ πθ
and θ cos 1+θ for the same θ. But then xy would equal sin 1+θ cos 1+θ
which is precisely 12 sin 1+θ
2πθ
. Therefore the condition xy = 21 reduces to
2πθ
sin 1+θ = 1. If we solve this indiscriminately, we would get
π
2πθ = (1 + θ)(2πk + ) (1)
2
for some integer k. For each integer k, there will be a solution for θ which
can be written down, after some simplification, as

4k + 1
θ= (2)
3 − 4k
6

It is thus tempting to think that S ∩T is an infinite set. As this is not given

as an alternative, we need to check further. Note that θ is required to be
positive. This means the numerator and the denominator of the ratio on
the R.H.S. of (2) must be of the same sign. As k takes only integral values,
if k > 0 then the numerator is positive and the denominator negative. The
opposite holds if k is a negative integer. So, it is only for k = 0, that θ > 0.
Hence there is only one point in S ∩ T . (Although not asked, this common
point occurs for θ = 31 . Hence it is the point ( 3√ 1
2
, √32 ).)
An alternate method to rule out the multiplicity of common points of
θ
S ∩ T is to observe that for θ > 0, the ratio 1+θ lies between 0 and 1. (It
2πθ
increases monotonically from 0 to 1 as θ varies over (0, ∞).) Hence 1+θ
lies in the interval (0, 2π) for all θ > 0 and there is only one point in this
interval, viz. π2 at which the sine function takes the value 1.
The problem requires some care in eliminating other candidates of
S ∩ T . By not giving ‘infinitely many’ as a possible answer, the paper-
setters have pointed out the need for such care.
Q.7 How many numbers formed by rearranging the digits of 234578 are divis-
ible by 55?

(A) 0 (B) 12 (C) 36 (D) 72

Answer and Comments: (B). Since 11 × 5 is the prime factorisation of

55, divisibility by 55 is equivalent to simultaneous divisibility by 5 and by
11. The first is easy to check, since a number is divisible by 5 if and only
if its last digit is 5 or 0. In the present case, this means that the number
is of the form abcde5 where abcde is a permutation of 23478. There are
5! = 120 such permutations and we have to find those which, after putting
a 5 at the end give a number divisible by 11.
The criterion for divisibility by 11 is that the sums of the alternate digits
should differ by a multiple of 11. In the present problem, this means that
the difference (a + c + e) − (b + d + 5) should be a multiple of 11. Since
a+c+e+b+d+5 equals the sum of the original digits 2, 3, 4, 5, 7, 8, viz. 29,
and is odd, the difference has to be odd too. Moreover it cannot exceed 29.
So divisibility by 11 is possible only in the two cases (a+c+e)−(b+d+5) =
±11. The first possibility gives a + c + e = 20 and b + d + 5 = 9, the second
gives a + c + e = 9 and b + d + 5 = 20. The first possibility cannot occur
because no three digits out of 2, 3, 4, 7, 8 can make a total of 20. In the
second possibility we have b + d + f = 20, i.e. b + d = 15 which is possible
only when b = 7, d = 8 or vice versa. In each case, the remaining three
digirs, viz. 2, 3, 4 have to add up to 9. They can be put as a, c, e in 3! = 6
ways. Hence in all there are 6 × 2 numbers of the given type that are
divisible by 55.
A simple but good problem requiring very elementary number theory
and combinatorics. The reasoning appears long when written down fully.
 7

But it does not take much time to do it as a rough work. So the problem
is well suited as a MCQ.
Q.8 The limit
3
  12
lim n− 2 (n + 1)(n+1) (n + 2)(n+2) . . . (2n)(2n)
n

n→∞

equals
1 3
(A) 0 (B) 1 (C) e− 4 (D) 4e− 4

Answer and Comments: (D). Call the given expression as an . It is a

complicated expression involving, besides powers, the product of n factors.
As the number of factors changes with n, theorems such as the limit of a
product is the product of limits are not applicable.
The standard practice in such cases (after ensuring that it is positive for
all n) is to take the logarithm, say bn , of the given expression, find its
limit, say L, and then take eL as the limit of the original sequence. So we
set bn = ln an to get, using properties of logarithms,
n
3 1 X
bn = ln an = − ln n + 2 (n + k) ln(n + k)
2 n
k=1
n
3 1X k
= − ln n + (1 + ) ln(n + k) (1)
2 n n
k=1

If the factor ln(n + k) were absent in the k-th term of the summation,
the second term on the R.H.S. of (1) would be a Riemann sum of the
Z 2
integral xdx, specifically a Riemann sum arising from a partition of
1
the interval [1, 2] into n equal parts. As n tends to infinity, the mesh n1 of
the partition tends to 0 and all such Riemann sums tend to the respective
definite integrals.
Here the factor ln(n + k) creates a problem. To tackle it, we rewrite n + k
as (1 + nk )n. Then ln(n + k) = ln(1 + nk ) + ln n and (1) becomes
" n # n
3 1 X k k 1X k
ln an = − ln n + (1 + ) ln(1 + ) + [(1 + ) ln n] (2)
2 n n n n n
k=1 k=1

which simplifies to
" n #  
3 1 X k k ln n n(n + 1)
ln an = − ln n + (1 + ) ln(1 + ) + n+
2 n n n n 2n
k=1
" n #  
3 1 X k k (3n + 1)
= − ln n + (1 + ) ln(1 + ) + ln n (3)
2 n n n 2n
k=1
8

Z 2
As n → ∞, the middle term tends to the definite integral x ln xdx,
1
which we shall evaluate a little later. The first term tends to −∞ while
the last one tends to ∞. But their sum, say cn , may tend to some finite
limit. To find it we note that
3n + 1 3
cn = ln n( − )
2n 2
ln n
= (4)
2n
Thus both the numerator and the denominator of cn tend to ∞ as n → ∞.
But it is quite well known that ln n tends to ∞ much slower than n and
hence the ratio tends to 0. (This can also be seen by replacing n by some
continous variable u which tends to ∞ and applying the l’Hôpital rule.)
Z 2
Summing up, lim an is the definite integral x ln xdx. If we integrate
n→∞ 1
by parts, this comes out to be

2 2
x2 2 x
Z Z
x ln xdx = ln x − dx
1 2 1 1 2
3 3
= 2 ln 2 − = ln 4 − (5)
4 4
3 4
Therefore the original limit lim an equals eln 4− 4 = 3 .
n→∞ e4
A complicated and lengthy question involing properties of logarithms,
their rates of growth, and finally, the evaluation of a certain definite inte-
gral involving logarithms. A better question would have been to do away
with the last part, give the expression
  12
nr (n + 1)(n+1) (n + 2)(n+2) . . . (2n)(2n)
n

and ask to find for which value of r it has a finite, non-zero limit as n → ∞.
Or, ask which power of n has comparable rate of growth as the expression
  12
(n + 1)(n+1) (n + 2)(n+2) . . . (2n)(2n)
n
.

Q.9 Consider a triangle with vertices (0, 0), (1, 2) and (4, 2). Let A be the area
of the triangle and B be the area of the circumcircle of the triangle. Then
B
equals
A
π 5π 3
(A) (B) (C) √ π (D) 2π
2 4 2

Answer and Comments: (B).

 9

(−4,2) P (1,2)
Q

O
(0,0)

Call the vertices as O, P, Q respectively. From the area formula in terms

of coordinates of the vertices,
0 0 1
1
A = ± 1 2 1
2
−4 2 1
= ±5 (1)

As the area is positive, we take A = 5 sq. units.

There is no similar formula for B, the area of the circumcircle. We would
be through if we can identify its radius. But, again, there is no general
formula for it either. But we can make use of special features of △OP Q.
As the slopes of the sides OP and OQ are 2 and − 21 respectively, we
get that 6 P OQ is a right angle. Hence P Q is a diameter. Its length is
1 − (−4) = 5 units. Hence the radius of the circumcircle is 25 . Therefore,
B, its area, is 25π B 5π
4 sq. units. Therefore the ratio A equals 4 .
An absolutely simple problem once it is realised that the triangle is
right angled.
Q.10 The value of
202  
X 202
k kπ
(−1) cos
k 3
k=0
equals
202 202 202
(D) cos202 ( π3 )




(A) sin 3 π (B) − sin 3 π (C) cos 3 π

Answer and Comments: (C). Call the given sum as S. If we can find
some number t such that cos kπ k
3 = t for all integers k = 0, 1, 2, . . . , 202,
then by the binomial theorem we would have

S = (1 − t)202 (1)

Unfortunately, there is no real number (not even a complex number) t

with this property, for if there were then {cos kπ
3 } would be a geometric
progression, which it surely is not since cos2 2π 1 π 3π
3 = 4 , but cos 3 cos 3 =
1
−4.
10

Nevertheless we do have a complex number t such that cos kπ

3 is the real
part of the k-th power of t. Indeed, if we take
π π
t = cos + i sin (2)
3 3
then by DeMoivre’s theorem we get

kπ kπ
tk = cos + i sin (3)
3 3

A compact notation for t is eiπ/3 . With this notation, we get that S is

the real part of

(1 − eiπ/3 )202 (4)

This is not of much use by itself, because if we expand it using the binomial
theorem and take its real part, we shall get back to the sum we started
with. So let us see if the number 1 − eiπ/3 can be recast so that its 202-th
power can be calculated efficiently. This turns out to be easy. Indeed

1 − eiπ/3 = 1 − cos(π/3) − i sin(π/3)

√
1 3
= 1− −i
2√ 2
1 3
= −i
2 2
= cos(π/3) − i sin(π/3)
= e−iπ/3 (5)

Combining this with (4), we now get that S is the real part of

202π 202π
(e−iπ/3 )202 = e−i202π/3 = cos − i sin (6)
3 3

Hence the given sum S equals cos 202π

3 .
A good problem based on the elementary poperties of complex roots
of unity.

Q.11 For real numbers a, b, c, d, a′ , b′ , c′ , d′ , consider the system of equations

ax2 + ay 2 + bx + cy + d = 0
a′ x2 + a′ y 2 + b′ x + c′ y + d′ = 0

If S denotes the set of all real solutions (x, y) of the above system of
equations, then the number of elements in S can never be

(A) 0 (B) 1 (C) 2 (D) 3

 11

Answer and Comments: (D). Clearly S = S1 ∩ S2 where S1 is the

solution set of the first equation and S2 that of the second. Assuming
a and a′ are non-zero, both S1 and S2 are circles since neither equation
has a xy-term. Now, two circles may have no point in common (i.e. be
disjoint), or just one point in common (when they touch each other) or
two points in common. But if they have 3 points in common, then they
are identical and hence have infinitely many points in common. So the
number of points of intersection can never be 3.
The degenerate cases when a and/or a′ vanishes are easy to deal with.
For then S1 and/or S2 is a straight line. A circle cannot meet a straight
line in 3 (distinct) points, nor can two straight lines. (Those who are very
fussy, will also consider the cases where these lines are degenerate. But
even then S1 ∩ S2 cannot have exactly 3 points.
The underlying theme of the problem is that a circle is uniquely
determined by three points. Once this is realised, there is no computation
involved. This problem, is therefore, a good test of the ability to get down
to the underlying theme. Unfortunately, the data does not specify that
the equations are non-degenerate. Those who consider the degeneracies
are wasting their time. Yet another illustration where scruples do not pay
in an MCQ.
Q.12 Let f, g be continuous functions from [0, ∞) to itself,
Z 3x
h(x) = f (t)dt, x > 0,
2x

And Z h(x)
F (x) = g(t)dt, x > 0
0

If F ′ is the derivative of F , then for x > 0,

(A) F ′ (x) = g(h(x))
(B) F ′ (x) = g(h(x))[f (3x ) − f (2x )]
(C) F ′ (x) = g(h(x))[x3x−1 f (3x ) − x2x−1 f (2x )]
(D) F ′ (x) = g(h(x))[3x f (3x ) ln 3 − 2x f (2x ) ln 2]

Answer and Comments: (D). The theme of the problem is the deriva-
tive of a function defined by an integral. The first equation defines h(x)
as the integral of the function f (x) whose upper and lower limits are both
functions of x. Using a generalisation of the Fundamental Theorem of
Calculus, we have
d x d
h′ (x) = f (3x ) (3 ) − f (2x ) (2x )
dx dx
= 3x ln 3f (3x ) − 2x ln 2f (2x ) (1)
12

The second equaion defines a new function F (x) by integrating yet another
continuous function g(x) (about which nothing is given) in the limits 0 to
h(x). As the lower limit is constant, we have

F ′ (x) = g(h(x))h′ (x) (2)

If we substitute (1) into (2), we get (D).

An absolutely simple, in fact dull, problem based on differentiation of
a function defined by an integral.
Q.13 Suppose x and y are positive integers. If 4x + 3y and 2x + 4y are divided
by 7, then the respective remainders are 2 and 5. If 11x + 5y is divided
by 7, then the remainder equals

(A) 0 (B) 1 (C) 2 (D) 3

Answer and Comments: (B). The given statements can be written

down as

4x + 3y = 7p + 2 (1)
and 2x + 4y = 7q + 5 (2)

where p, q are some integers. As their values are not given, these equations
cannot be solved for x and y. Nor is it needed because the question merely
asks the remainder when 11x + 5y is divided by 7. We could change x and
y by adding any multiples of 7 and that will not affect the remainders.
It is therefore convenient to work in terms of the residue classes modulo
7. (See Comment No. 15 of Chapter 4. Regrettably, the Index gives a
wrong location.) The residue class modulo 7 of an integer a is generally
denoted by [a]7 or simply by [a] when the modulus 7 is understood. The
basic rules of residue arithmetic are [a ± b] = [a] ± [b] and [ab] = [a][b]. As
a result, the two equations above reduce to

[4][x] + [3][y] = [2] (3)

and [2][x] + [4][y] = [5] (4)

respectively. If we add these two equations, the coefficient of [y] is [3] +

[4] = [7] which is the same as [0] modulo 7. Similarly, the right hand sides
[2] and [5] add to [0]. So we are left with

[6][x] = [0] (5)

This means that the integer 6x is divisible by 7. But as 7 is relatively

prime to 6, we must have that 7 divides x, or equivalently, [x] = [0].
The equation (3) now reduces to [3y] = [2]. That means that 3y − 2 is
divisible by 7. Again, this does not define y uniquely, but only modulo 7.
With trial, we can take y as 3 since 3 × 3 = 9 has the same residuce class
 13

as 2 modulo 7. (This also tallies with (4) because 4 × 3 = 12 is indeed

congruent to 5 modulo 7.)
Summing up, we now have that [x] = [0] and [y] = [3]. So [11x + 5y] =
[11][0] + [5][y] = [15] = [1]. Hence when 11x + 5y is divided by 7 the
remainder will be 1.
A very good problem based on residue arithmetic. Those totally
unfamiliar with it are likely to be confused that prima facie the data
is insufficient. It is only after realising that what is asked is only the
remainder that the confusion is resolved.

Q.14 Suppose F : IR −→ IR is a continuous function which has exactly one local

maximum. Then which of the following is true?

(A) F cannot have a local minimum.

(B) F must have exactly one local minimum
(C) F must have at least two local minima
(D) F must have either a global maximum or a local minimum

Answer and Comments: (D). As the domain IR is the entire real line,
there is no room for controversies that may arise because of end points.
Easy counterexamples can be given to dismiss the first two options. For
example, take the cosine function. It has infinitely many local maxima
and local minima. The value at any local maximum is 1. Take any two
consecutive maxima, say 0 and 2π. Modify the the graph of the function
on [0, 2π] by joining (0, 1) and (2π, 1) to the point (π, 2). Leave intact the
other part of the graph. Then there is a unique local maximum at π. But
(A) and (B) fail as there still are infinitely many local minima.
1
For (C), define F (x) = |x| for |x| ≥ 1. But on [−1, 1] join (−1, 1) and
(1, 1) to (0, 2) to get the graph of a function which has a unique local
maximum at 0, but no local minimum.
As only one alternative is correct, it has to be (D). For an honest answer,
here is a proof. Assume that the unique local maximum of F is at x0 . If
it is not the global maximum, there is some x1 such that F (x1 ) > F (x0 ).
Without loss of generality assume that x1 > x0 . Then the point, say x2
where F attains its minimum on the interval [x0 , x1 ] will be an interior
point of [x0 , x1 ] and hence a point of local minimum of F on IR.
A purely thought oriented problem. Regrettably, being an MCQ, it is
impossible to tell whether those who select (D) can give a proof.

Q.15 Suppose z ∈ C
| is such that the imaginary part of z is non-zero and
25
z = 1. Then
2023
X
zk
k=0
14

equals

(A) 0 (B) 1 (C) −1 − z 24 (D) −z 24

Answer and Comments: (D). Since z 25 = 1, z is one of the twentyfive

25-th complex roots of unity. As 25 is odd, all of them except 1 have
non-zero imaginary parts. Moreover, for any such root (other than 1), its
powers recur in cyles of length 25 (or sometimes a of length 5, which is
the only divisor of of 25). In either case, the sum of powers occurring in
any cycle is 0.
In this problem, the given sum, say S, consists of 2024 terms
z 0 (= 1), z, z 2, . . . , z 2023 . The first 2000 terms from 1 to z 1999 fall into 80
cycles of 25 consecutive powers each. As each cycle adds to 0, we have

S = z 2000 + z 2001 + . . . + z 2022 + z 2023 (1)

There 24 terms equal, respectively, 1, z, z 2, . . . , z 23 . If we add z 24 we get

a cycle of 25 consecutive powers. As their sum is 0, we get

S = 0 − z 24 = −z 24 (2)

An easy problem based on the peiodicity of the complex roots of unity of

a given order.

Q.16 The limit

     
1 1
lim cos(x) + cos − cos(x) cos −1
x→0 x x

(A) equals 0 (B) equals 21

(C) equals 1 (D) does not exist

Answer and Comments: (A). Neither the second nor the third term
has a limit as x → 0 because cos( x1 ) tends to no definite limit. But as
in Q.8, sometimes the sum of two functions neither of which has a limit
can have a limit. To see if that is so here, we note the function, say f (x)
whose limit is to be taken factors as
1
f (x) = (cos x − 1)(1 − cos ) (0)
x
The first factor tends to 0 as x → 0. Although the second factor does not
tend to any limit, it is bounded. That is, there is some number B > 0
such that
1
|1 − cos | ≤ B (1)
x
 15

for all x 6= 0. Hence we have

|f (x)| ≤ B| cos x − 1| (2)

As the second factor of the R.H.S. can be made arbitrarily small for x
sufficiently small, so can |f (x)|. Therefore |f (x)| → 0 and hence finally
f (x) → 0 as x → 0.
A simple problem illustrating that the product of two functions of which
one is bounded and the other tends to 0 also tends to 0. The problem
would have been a little more interesting if f (x) had an additional factor
1
x . This factor is not bounded near x = 0. But cos x − 1 is of the order
of x2 and so cos xx−1 tends to 0 too. (Of course, this can also be seen from
cos x − 1
the fact that lim is the derivative of cos x at x = 0.)
x→0 x
We pointed to Q.8 where the sum of two sequences had a limit when
neither did. So, in a sense their ‘bad’ parts canceled each other, rather
like two jagged pieces of a jigsaw puzzle fitting together to form something
smooth. In the present problem, the sum of the two terms not having a
limit did not have a limit either. But between two factors, if the one
without a limit is bounded and the other tends to 0, then the product
tends to 0. The vanishing factor by its strength ‘cures’ the anomaly. The
reasoning will not apply if the other factor had a limit different from 0.
Q.17 Let n be a positive integer having 27 divisors including 1 and n, which are
denoted by d1 , ..., d27 . Then the product of d1 , d2 , ..., d27 equals
27
(A) n13 (B) n14 (C) n 2 (D) 27n

Answer and Comments: (C). In general, whenever d is a divisor of an

integer d so is nd . They occur in pairs√and their product equals n. So the
smaller one among them is less than n and the other is bigger. This also
means that the number of factors of n is even. The only exception is when
n is a perfect square, say n = k 2 . In this case the factor k is counted only
once. All others pair off. So the number of factors is odd.
The present problem falls in this case. As the number of factors, viz. 27, is
odd, n is a perfect square and n1/2 is an integer. The remaining 26 factors
pair off into 13 pairs and each pair multiplies to n. Hence the product of
1 27
all 27 factors of n is n13 n 2 = n 2 .
An absolutely simple problem once the key idea, viz. pairing off the
factors, strikes.
Q.18 If f : [0, ∞) −→ IR is a continuous function such that
Z x
f (x) + ln 2 f (t)dt = 1, x ≥ 0
0

then for all x ≥ 0,

16

(A) f (x) = ex ln 2 (B) f (x) = e−x
ln 2

x
(C) f (x) = 2x (D) f (x) = 12

Answer and Comments: (D). The function f is not given to be differ-

entiable.Z But it is continuous and so the function obtained by integrating
x
it, viz. f (t)dt is differentiable. So, if we rewrite the given condition
0
as
Z x
f (x) = 1 − ln 2 f (t)dt (1)
0

then the R.H.S. and hence f (x) is differentiable. Taking derivatives of

both the sides of (1), we have

f ′ (x) = − ln 2f (x) (2)

This is a linear differential equation with general solution

1
f (x) = ce−x ln 2 = c(e− ln 2 )x = c( )x (3)
2
where c is an arbitrary constant. Putting x = 0 in the given equation, we
get f (0) = 1 which determines c as 1. Hence the solution is (D).
Conversion to a diffrential equation and finding its solution are fairly
routine. The less trivial part is to show that f (x) is differentiable, even
though it is given to be merely continuous. But once again, the MCQ
format fails to test it.

Q.19 If [x] denotes the largest integer less than or equal to x, then
√ 20
 
9 + 80

equals
√
20 √
20
(A) 9 + 80 − 9 − 80
√
20 √
20
(B) 9 + 80 − 9 − 80 − 20
√
20 √
20
(C) 9 + 80 − 9 − 80 −1
√
20
(D) 9 − 80

Answer and Comments: (C). When expanded by the binomial theo-

rem,
20  
√ 20 X 20 20−k k/2
(9 + 80) = 9 80 (1)
k
k=0
 17
√
√ 20 because 80 and hence all its odd powers
Terms for which k are irrational
are irrational. So (9 + 80) is not an integer. Our task is to find the
largest integer, say I less than it.
√
However, if we consider the conjugate surd (9 − 80)20 , then we get
20
√ 20 X
 
k 20
(9 − 80) = (−1) 920−k 80k/2 (2)
k
k=0

If we add (1) and (2), the terms for which k is odd get cancelled and we
are left with
10  
√ 20 √ X 20 20−2r r
(9 + 80) + (9 − 80)20 = 2 9 80 (3)
r=0
2r

which
√ 20is clearly an integer. Of course, it is an integer greater than (9 +
80) because the second term on the L.H.S. is positive. So we have
√ √
I < (9 + 80)20 + (9 − 80)20 (4)
√ 20
But at the same time, the new term we are adding, √ viz. (9 − 80) is
very small because
√ 9 is only slightly bigger than 80 and therefore the
difference 9 − 80 is small√and its 20-th power would be negligibly small.
1 1
(To be more precise, 9 − 80 equals 9+√ 80
and hence is less than 9+8
−1
√ 20
which itself is less than 10 . Hence the term (9 − 8) will be less than
10−20 .)
So, (4) can be tightened by
√ √ √ √
(9 + 80)20 + (9 − 80)20 − 1 < I < (9 + 80)20 + (9 − 80)20 (5)
√ √
which shows that the integral part of (9 + 80)20 is (9 + 80)20 + (9 −
√
80)20 − 1.
A good problem based on the application of conjugate surds. Similar
problems are given in Exercise (6.27).

Q.20 Let f : IR −→ IR be a twice differentiable one-to-one function. If f (2) = 2,

f (3) = −8 and
Z 3
f (x)dx = −3
2

then Z 2
f −1 (x)dx
−8

equals

(A) 25 (B) −25 (C) −31 (D) 31

18

Answer and Comments: (A). The function f (x) is not given explicitly.
Only its values at the two points 2 and 3 are given. The integral of
f (x) over the interval [2, 3] is also given. But this does not determine
the function uniquely. In fact, there are infinitely many choices for f (x)
satisfying these three constraints. Accordingly, there are infinitely many
possibilities for the inverse function f −1 too. So, on the face of it, the
problem will not have a unique answer. The only saving grace is that the
end points of the interval of integration of the integral to be evaluated are
precisely the values of the integrand at the end points of the interval of
integration of the first integral whose value is given.
This sort of reciprocity relationship suggests that the only way to evaluate
the second integral is by relating it with the first one in some general way.
Such a relationship indeed exists and is given by the following equation
where instead of [2, 3] we have a more general interval [a, b]. (The other
hypotheses remaining the same.)

Z b Z f (b)
f (x)dx + f −1 (x)dx = bf (b) − af (a) (1)
a f (a)

b
To prove this, we note that the R.H.S. is the expression xf (x) a , which
would occur if we were evaluating the integral over [a, b] of the derivative
of xf (x). The derivative of xf (x) is simply f (x) + xf ′ (x). In other words,
Z b Z b
f (x)dx + xf ′ (x)dx = bf (b) − af (a) (2)
a a

So the proof of (1) is reduced to shwoing that

Z f (b) Z b
f −1 (x)dx = xf ′ (x)dx (3)
f (a) a

To prove this, we substitute x = f (t) in the integral on the L.H.S. This

is a valid substitution since the function f is one-to-one and hence by
continuity, a strictly monotonic bijection. Then the integral on the L.H.S.
Z b
becomes tf ′ (t)dt. Here t is a dummy variable of integration and we
a
can change it at will. Replacing it by x gives (3) and hence (1).
It is now trivial to finish the solution. Apply (1) with a = 2, b = 3, f (a) = 2
and f (3) = −8. Then f −1 (2) = 2 and f −1 (−8) = 3. Then we get
Z 3 Z −8
f (x)dx + f −1 (x)dx = 3f (3) − 2f (2) = −24 − 4 = −28 (4)
2 2

As the first term on the L.H.S. is given to be −3, the second one is −28 +
3 = −25.
 19

Z 2
But there is something more. The integral asked is f −1 (x)dx and not
−8
Z −8
−1
f (x)dx. So the correct answer is not −25 but 25.
2
The result (1) is not among the well known results about integrals such
as the fundamental theorem of calculus (two forms) or the mean value
theorem for integrals. It is more suited for problems. As a result, the
problem is rather tricky. But such tricks are worn out if used frequently.
See the solution to Q.7 of Paper 2 of ISI 2016 where a similar trick was
used.

Q.21 Three left brackets and three right brackets have to be arranged in such a
way that if the brackets are serially counted from the left, then the number
of right brackets counted is always less than or equal to the number of left
brackets counted. In how many ways can this be done?

(A) 3 (B) 4 (C) 5 (D) 6

Answer and Comments: (C). The description given is precisely that

of what is called a balanced arrangement of n pairs of brackets. It is a
well known result (see Comment No.6 of Chapter 1)that the number of
1 2n
balanced arrangements of n pairs of brackets is . This is called
n+1 n
the n-th Catalan number and pops up in many combinatorial problems.
1
20
For n = 3, this comes out to be × 63 = = 5.
4 4
The result used here is non-trivial and proved by first counting the
number of unbalanced pairs. Those who are not familiar with this result
can do a hand count. For n = 1, there is only one arrangement, viz. [ ].
For n = 2, there are 2, viz. [ ] [ ] and [ [ ] ]. For n = 3, the first bracket
has to be a left bracket. Classify the balanced arrangements according to
which of the three right brackets cancels it. If it is the first, there are two
possible arrangements after it. If it is the second, there is only one before
and only one after it. If the canceling right bracket is the third one, then
it is at the end and before it there are two possible arrangements. So, in
all, there are 2 + 1 + 2 = 5 arrangements.
The papersetters have probably deliberately avoided the term bal-
anced arrangements as this will give an automatic clue to those who know
Catalan numbers. That would be unfair in a competitive examination.

Q.22 The limit

  −n
n−1 2
2−2 + 2−2
n+1
lim
n→∞

equals

(A) =1 (B) √1 (C) 0 (D) 1

2 4
20

Answer and Comments: (B). Another complicated limit involving pow-

ers. As a first measure of simplification, we note that the variable n oc-
curs only in the powers 2n and 2−n . Secondly, its integrality is nowhere
involved. So we may as well put t = 2n , treat it as a continuous variable
and take limit as t → ∞. So our problem is now reduced to finding L
where
 1/t
L = lim 2−2t + 2−t/2 (1)
t→∞
 1/t
1 1
= lim + (2)
t→∞ 22t 2t/2
 1/t 
1 3t/2
1/t
= lim 1 + 2 (3)
t→∞ 22t

The first factor is a constant 41 . But the second factor is in the indetermi-
nate form ∞0 . To find its limit, we take its log.

 1/t ln(1 + 23t/2 )

ln 1 + 23t/2 = (4)
t

To find the limit of the R.H.S. as t → ∞, we apply L’ôpital’s rule. For

this,

d   1 3
ln(1 + 23t/2 ) = × ln 2 × 23t/2 (5)
dt 1 + 23t/2 2
3
2 ln 2
= (6)
1 + 2−3t/2

The numerator and the first term of the denominator are constant. The
second term tends to 0 as t → ∞. Therefore by Lôpital’s rule, (4) tends
to 23 ln 2 as t → ∞. Hence the second factor of the R.H.S. of (3) tends to
3 √ √
e 2 ln 2 = 23/2 = 8 = 2 2. Therefore, we get, finally

1 √
L = ×2 2
4
√
2
=
2
1
= √ (7)
2

Q.23 In the following figure, OAB is a quarter-circle. The unshaded region is

a circle to which OA and CD are tangents.
 21

10 D
C

2r 2r
R
O A

If CD is of length 10 and is parallel to OA, then the area of the shaded

region in the above figure equals

(A) 25π (B) 50π (C) 75π (D) 100π

Answer and Comments: (A). Ler R and r be the radii of the bigger
and the smaller circle respectively. Then the area, say A of the shaded
region is
π 2
A= R − πr2 (1)
4
We could find A if we can find the values of R and r. But the data is not
sufficient for this. But if we look at (1) carefully, we can recast it as
π 2
A= (R − 4r2 ) (2)
4
And, we can get the value of R2 − 4r2 from the right angled triangle OCD
if we note that the hypotenuse OD is also a radius of the bigger circle
and hence equals R. OA and CD are parallel being the tangents to the
smaller circle at the ends of a diameter. Hence OC equals the perpendiculr
distance between them. The Pythagoras theorem gives R2 − 4r2 = 100
and so A = 25π.
The problem is more suited as a puzzle.
Q.24 Consider the function f : C
| −→ C
| defined by

f (a + ib) = ea (cos b + i sin b), a, b ∈ IR

where i is a square root of −1. Then

(A) f is one-to-one and onto.
(B) f is one-to-one but not onto
(C) f is onto but not one-to-one
(D) f is neither one-to-one nor onto.
22

Answer and Comments: (D).

The given definition of f is an elaborate way of saying that the function
is simply the exponentiation function, f (z) = ez . It is not onto since it
never vanishes. Nor is it one-to-one since e2πin = 1 for all n ∈ IN .
The purpose of the question is not clear. Once it is realised that
the function is the exponential function, all that is needed is recalling
some well-known properties of it. Those who can do only computational
problems sometimes miss such obvious interpretations. (An anecdote is
that a candidate for a lecturership in mathematics was asked to give the
Fourier expansion of sin x. The reply was that it will take a long time to
calculate it as many integrals will have to be evaluated!)
Q.25 Suppose f : ZZ −→ ZZ is a non-decreasing function. Consider the following
two cases:

Case 1: f (0) = 2, f (10) = 8,

Case 2: f (0) = −2, f (10) = 12
In which of the above cases it is necessarily true that there exists an n
with f (n) = n?
(A) In both cases
(B) In neither case
(C) In Case 1, but not necessarily in Case 2.
(D) In Case 2, but not necessarily in Case 1.

Answer and Comments: (C). It is easy to give a counterexample to

disprove the assertion in Case 2. Simply put f (x) = −2 for all x =
1, 2, . . . , 9. (For points x < 0, put f (x) = x − 100 and for x > 10 put
f (x) = x + 100. (These are essentially arbitrary choices.) But we claim
that the assertion is actually true in Case 1. For, if not then for every
integer x either f (x) > x or f (x) < x. Colour x red in the first case and
blue in the other. Then 0 is red and 10 is blue. So as we scan from 0
to 10, the colour will change from red to blue at some consecutive pair.
That is, there will be some x such that x < f (x) and x + 1 > f (x + 1).
But f (x) ≤ f (x + 1) as f is given to be non-decreasing. Put together, we
have

x < f (x) ≤ f (x + 1) < x + 1 (1)

So, the integers f (x) and f (x + 1) (which may be equal) lie in the open
interval (x, x + 1). But as x itself is an integer, there are no integers in
(x, x + 1), a contradiction. So the assertion is true in Case 1 but not
necessarily true in Case 2.
An excellent problem. It is the discrete analogue of the result that
if f (x) is continuous on [a, b] and f (a) > a but f (b) < b, then there
 23

exists some c ∈ (a, b) for which f (c) = c. (Geometrically, if one end of

the graph of a continuous function is below and the other above the line
y = x, then somewhere it must cross the line y = x.) A simple proof
is by applying the Intermediate Value Property (IVP) to the function
g(x) = f (x) − x. Note that no condition like f is monotonically increasing
is needed. This is because continity is a strong condition. For a function of
a discrete variable, continuity does not mean much sicnce every function
of a discrete variable is continuous. But the hypothesis of monotonicity
gives the strength needed.

Interestingly, the technique of colouring the points of the domain with

two colours and looking for a change of colours can be applied to prove
the continuous analogue of the result. For, suppose f : [a, b] −→ IR is
continuous with f (a) > a and f (b) < b. If there is no c ∈ (a, b) for which
f (c) = c, colour a point x of [a, b] red or blue according as f (x) > x or
f (x) < x. Then a ia red but b is blue. For a positive integer n, divide
the interval [a, b] into n equal parts. The nodes of this partition will be
coloured red or blue with the first node (viz. a) red and the last one b
blue. So, there are two consecutive nodes, say xn and yn = xn + b−a n such
that xn is red and yn blue. (In fact, the number of such pairs must be
odd. But that is not needed here. All we need is that there is at least
one such pair.) By choosing n sufficiently large, we get that there are
arbitrarily close pairs of points with one red and the other blue. This
is as far as discrete mathematics takes you. After this the proof can be
completed using completeness of real numbers, the most basic result of
continous mathematics.

Q.26 Suppose that f (x) = ax3 + bx2 + cx + d where a, b, c, d are real numbers
with a 6= 0. The equation f (x) = 0 has exactly two distinct real solutions.
If f ′ (x) is the derivative of f (x), then which of the following is a possible
graph of f ′ (x)?

(A) (B)
y y

x
24

(C) (D)
y y

Answer and Comments: (B). Luckily the four possibilities for the real
roots of a cubic are listed as the options in Q.4. In the present case, we
are given that f (x) = 0 has exactly two distinct real solutions. For this
to happen, one root, say, α must be a double root and the other, say β
distinct from it. But then f (x) is of the form

f (x) = a(x − α)2 (x − β) (1)

Hence

f ′ (x) = 2a(x − α)(x − β) + a(x − α)2 (2)

= a(x − α)(2(x − β) + x − α) (3)
= a(x − α)(3x − 2β − α) (4)

Thus f ′ (x) has two roots, one α and the other different from α (since
2β + α 6= 3α). Among the four possible graphs of f ′ (x), in (A) and (C)
there is a double root, while in (D), there is no root of f ′ (x). In (B),
clearly there are two distinct roots.
The essence of the solution os to show that f ′ (x) has only two distinct
roots. This need not have been done so elaborately as above. We could
use the general result that a multiple root of any polynomial is also a root
of its derivative. That gives one root, say α of f ′ . But there is also a root
β 6= α of the the cubic. Hence by Rolle’s there is a root of f ′ between α
and β.

Q.27 Suppose a, b, c ∈ IR and

f (x) = ax2 + bx + c, x ∈ IR

If 0 ≤ f (x) ≤ (x − 1)2 for all x and f (3) = 2, then

1 1
(A) a = , b = −1, c =
2 2
1
(B) a = , b = −2, c = 0
3
2 5
(C) a = , b = − , c = 1
3 3
 25

3 5
(D) a = , b = −2, c =
4 4

Answer and Comments: (A). Since 0 ≤ f (1) ≤ (1 − 1)2 we get that

f (1) = 0, i.e. 1 is a root of f (x). So the other root, say α, has to be real.
If it is different from 1, then on the interval between 1 and α will assume
negative values if a > 0. But if a < 0, then f (x) will assume negative
values outside the closed interval with end points 1 and α. In either case,
they hypothesis f (x) ≥ 0 for all x is violated. So we conclude that α = 1,
i.e. 1 is a double root of f (x), or equivalently f (x) = k(x − 1)2 for some
constant k. Comparing the leading coefficients, we have k = a.
All put together

f (x) = a(x − 1)2 = ax2 − 2ax + a (1)

Hence

b = −2a, c = a (2)

So, b, c are determined as soon as we know. To find a, we use f (3) = 2

which gives a(3 − 1)2 = 2 whence a = 21 . From (2), we get b = −1 and
c = 12 .
A simple, but unusual problem on quadratics. The key idea is that
a quadratic with different real roots assumes both positive and negative
values somewhere.
Q.28 The polynomial x10 + x2 + 1 is divisible by

(A) x2 + x + 1 (B) x2 − x + 1
(C) x2 + 1 (D) x5 − 1

Answer and Comments: (A). We can, of course, apply the long division
in each case to find the remainder and get the answer. But obviously, we
look for short cuts wherever poddible. (C) and (D) can be dismissed
easily. Call x10 + x2 + 1 as f (x). If f (x) were divisible by x2 + 1, so would
f (x) − (x2 + 1) which equals x10 . But a polynomial with no constant term
can never be divisible by one whose constant term is non-zero. For (D),
note that f (x) = x10 − 1 + x2 + 2. But x10 is divisible by x5 − 1. So, if
f (x) is divisible by x5 − 1 so would be x2 + 2 which is impossible since
x5 − 1 has a higher degree.
A short cut of this type is not available for (A) and (B) because no easy
multiples of these polynomials can be subtracted from f (x) so that the
polynomial left is readily seen to be divisible or not. But fortunately the
polynomials in (A) and (B) are quadratics and hence their roots can be
easily obtained. Once they are, we can apply the result that whenever
a polynomial f (x) is divisible by another polynomial g(x), every root of
26

g(x) is also a root of f (x). Moreover, the converse is also true if g(x) has
no multiple roots. (Actually, even multiple roots are all right, provided
the multiplicity of a root of g(x) is at least its multiplicity as a root of
f (x).)

We begin
√
with (A). The roots of x2 + x + 1 are ω and ω 2 , where ω =
1 3
− 2 + i 2 = e2πi/3 is a complex cube root of unity. As a result, powers of
ω recur in cycles of length 3 with ω 3 = 1. This gives

f (ω) = ω 10 + ω 2 + 1 = ω 2 + ω + 1 = 0 (1)

Similarly,

f (ω 2 ) = ω 20 + ω 4 + 1 = ω 2 + ω + 1 (2)

Thus both the roots of x2 +x+1 are also roots of f (x). Therefore it divides
f (x). As only one oprion is correct, it is now pointless to investigate (B).
√
Still, for the sake of completeness, the roots of x2 − x + 1 = 0 are 21 ± i 23
which can be written compactly as e±iπ/3 . Their powers recur in cylces
of length 6 and their cubes equal −1 and not 1. So,

f (eiπ/3 ) = e10iπ/3 + ei2π/3 + 1

= −eiπ/3 + ei2π/3 + 1

whose imaginary part − sin π3 + sin 2π3 is non-zero. So without any further
checking, eiπ/3 is not a root of f (x). Hence f (x) is not divisible by x2 −
x + 1. (We could also have applied this reasoning to (C) because the roots
of x2 + 1 are ±i and f (i) = i10 + i2 + 1 = −1 − 1 + 1 = −1 6= 0. But there
a short cut was available anyway.)

In JEE, there are some questions where one or more answer is correct.
If this were the case here, each option would have to be dealt with. But
as it is given that only one option is correct, and the very first one turns
out to be correct, it is a waste of time to check other options. This is
one place where the JEE practice is superiror. The normal tendence is to
try the options in the given order. it would have been better to keep the
correct option as (C) or (D).

Q.29 As in the following figure, the straight line OA lies in the second quadrant
of the (x, y) plane and makes an angle θ with the negative half of the
x-axis, where 0 < θ < π2 . The line segment CD of length 1 slides on the
(x, y) plane in such a way that C is always on OA and D on the positive
side of the x-axis. The locus of the mid- point of CD is
 27

P. (h,k)
θ
x
O D

1
(A) x2 + 4xy cot θ + y 2 (1 + 4 cot2 θ) = 4
1
(B) x2 + y 2 = 4 + cot2 θ
1
(C) x2 + 4xy cot θ + y 2 = 4
1
(D) x2 + y 2 (1 + 4 cot2 θ) = 4

Answer and Comments: (A). Finding the locus of a given point (usually
the midpoint) on a ladder moving with one end on the horizontal flooe and
the other on a vertical wall is a standard problem in coordinate geometry.
The present problem is a generalisation of this, in that the wall is not
vertical but a tilted one, making an angle θ with the negative axis.
As usual, we let P = (h, k) be midpoint of the segment CD. We need
an equation in h and k. That would be easy if the wall were the y-axis.
But here we need to introduce an auxiliary variable. Let α > 0 be the
perpendicular distance of C from the y-axis. Then

C = (−α, α tan θ) (1)

As (h, k) is the midpoint of CD we get

D = (2h + α, 2k − α tan θ) (2)

But D lies on the x-axis. So, 2k − α tan θ = 0, i.e.

α = 2k cot θ (3)

Equating the distance CD with 1 we get another equation

(2h + 2α)2 + α2 tan2 θ = 1 (4)

The locus of P = (h, k) can be obtained by eliminating α between (3) and

(4). If we substitute (3) into (4), we get

4(h + 2k cot θ)2 + 4k 2 = 1 (5)

28

On expansion,
1
h2 + 4hk cot θ + 4k 2 cot2 θ + k 2 = (6)
4
which gives (A) if we replace h, k by x, y respectively.
A straightforward problem on finding a locus. Art lies only in intro-
ducing an auxiliary variable.
Q.30 How many functions f : {1, 2, . . . , 10} −→ {1, 2, . . . , 2000}, which satisfy
f (i + 1) − f (i) ≥ 20, for all 1 ≤ i ≤ 9, are there?

1829 1830



(A) 10! 10 (B) 11! 11

1829 1830



(C) 10 (D) 11

Answer and Comments: (C). The requirement implies, in particular,

that the function f is strictly monotonically increasing. So its range,
say R, is a set with 10 elements, say x1 , x2 , . . . , x10 where xi = f (i).
Conversely, every subset R with 10 elements x1 , x2 , . . . , x10 in ascending
order determines a unique strictly increasing function from {1, 2, . . . , 10}.
Let Ai be the set of integers strictly between xi and xi+1 for i = 1, 2, . . . , 9.
Also let A0 = {1, 2, . . . , x1 − 1} and A10 = {x10 + 1, . . . , 2000}. If the
restriction xi+1 − xi ≥ 20 was merely that xi+1 > xi , then this would be a
familiar problem of putting 2000−10 = 1990 identical balls into 11 distinct
boxes A0 to A10 . Each such placement amounts to an arrangement of 1990
identical balls and 10
identical interbox partitions and the number of such
arrangements is 200010 .
The given restriction on f amounts to saying that each of the boxes A1
to A9 has at least 19 balls. So we modify the argument above. First we
put 19 balls into each of these 9 boxes. We are now left with 1990 −
171 = 1819 balls and
10 identical interbox partitions. The number of such
arrangements is 1829
10 .
The problem is a good test of paraphrasing the data so that the problem
transforms into some more familiar problem. By giving the upper index
in the binomial coefficients in all options as either 1829 or 1830, the paper
setters have given an implicit hint that the requirement f (i+1)−f (i) ≥ 20
amounts to saying that the i-th box has at least 19 balls.
 29

ISI BStat-BMath-UGA-2023 Paper 2 with Comments

Q.1 Determine all integers n > 1 such that every power of n has an odd number
of digits.

Solution and Comments: Trivially 1 has the desired property. But

that has been excluded. Excluded are also double digit numbers and,
more generally, all numbers which have an even number of digits. If we
try the single digit numbers 2, 3, 4, . . . , 9 we see that some power of them
runs into two digits. By the binomial theorem, some powers of 102, 103, . . .
etc. will have an even number of digits.
But what if we take powers of 100? They are 100, 10000, 100000, . . . and
more generally (10)2k which has 2k + 1 digits. Of course, since powers of
10000 are included in those of 100, 10000 is also a successful candidate.
More generally, all powers of 100 will work.
We claim that these are the only numbers all whose powers have an odd
number of digits. For this we need to find a convenient expression for the
number of digits in a number which is such that the number of digits in
any power can be calculated easily. As we are using the decimal system,
logarithms with base 10 provide such an expression. If k is a fixed positive
integer, then note that the smallest and the largest numbers with k digits
are 10k and 10k+1 − 1 respectively. Put differently, an integer n has k
digits if and only if k ≤ log10 n < k + 1, or equivalently, if and only if the
integral part of log n is k. (From now on we suppress the base 10 from the
notation.)
It is now easy to come up with a solution. Suppose all powers of n have
an odd number of digits. Let
log n = I + f (1)
(where I and f are, respectively the integral and the fractional parts of
log n.) Then I is even and 0 ≤ f < 1. Taking the k-th power,
log nk = k log n = kI + kf (2)
Note that kI is an integer. But the integral part of log nk may be bigger
than kI if kf equals or exceeds 1. Let us see when this happens for the
first time as k increases. f ∈ [0, 1). Unless f = 0, {f, 2f, 3f, . . . , kf, (k +
1)f, . . .} is a strictly increasing sequence, in fact an aithmetic progression.
Since f < 1, the first time kf equals or exceeds 1, kf cannot exceed 2,
because (k − 1)f < 1 implies kf < 1 + f < 2. (This is also obvious
intuitively because when the common difference of an A.P. (Here {kf }) is
less than 1, it cannot jump across an interval of length 1 (here the interval
[1, 2]) in one step.)
So, the firt time kf equals or exceeds 1, the integral part of nk is I + 1
which is odd (since I is even). But that means that nk has an even number
of digits.
30

The only exception is when f = 0. But that means n is an even power of

10, viz. 10I .
Hence integers of the form 10I where I is even are the only numbers which
have an odd number of digits in every power.
A simple but imaginative problem. The solution comes easily once
the idea of taking logs w.r.t. base 10 and their integral parts strikes. This
idea comes after some experimentation. So the problem is a good test of
this ability on the part of a candidate.
1
Q.2 Let a0 = 2 and an be defined inductively by
r
1 + an−1
an = , n ≥ 1.
2

(a) Show that for n = 0, 1, 2, . . . ,

π
an = cos θn for some 0 < θn < ,
2
and determine θn .
(b) Using (a) or otherwise, calculate

lim 4n (1 − an )
n→∞

Solution and Comments: (a) amounts to showing that 0 < an < 1

for every n. This follows easily by induction on n. a0 is given to be 12 .
Assume 0 < an−1 < 1. Squaring the given relation,

1 + an−1
a2n = (1)
2

By induction hypothesis, the R.H.S. lies between 12 to 1 and hence its

(positive) squareroot lies between √12 to 1 and hence certainly between 0
and 1.
Now we define θn = cos−1 (an ). To determine θn explicitly in terms of n,
we replace an−1 and an by cos θn−1 and θn respectively in (1) and get

2 cos2 θn − 1 = cos θn−1 (2)

The L.H.S. equals cos 2θn . Since both θn−1 and 2θn lie in the interval
[0, π] on which the cosine function is injective, we get

θn−1
θn = (3)
2
 31

Thus {θn } is a geometric progression with initial term θ0 and common

ratio 12 . Since a0 = 21 , θ0 = π3 . Therefore

π/3
θn = (4)
2n
which completes the solution to (a).
For (b) we use (a). Although in (a) n was a discrete variable, in finding
the limit in (b), we may treat 2n as a continuous variable t and take the
limit as t → ∞ (as was done in Q.22 of Paper 1). Then the problem
reduces to evaluating L where
π
L = lim t2 (1 − cos ) (5)
t→∞ 3t

We make one more substitution, u = 1t . Then

1 − cos(uπ/3)
L = lim (6)
u→0+ u2
This can be evaluated using L’ôpital’s rule. Taking first derivatives, we
π
sin(uπ/3)
have to find lim 3 . We can apply L’ôpital’s rule again or use
u→0+ 2u
2
the more basic fact that sinα α → 1 as α → 0. Either way the limit is π18 .
n
So this is also the value of lim 4 (1 − an ).
n→∞

Q.3 In a triangle ABC, consider points D and E on AC and AB, respectively,

and assume that they do not coincide with any of the vertices A, B, C.
If the segments BD and CE intersect at F , consider the areas w, x, y, z
of the quadrilateral AEF D and the triangles BEF , BF C, CDF , respec-
tively.

E
w1 w
2
F D
x
z
y
B C

(a) Prove that y 2 > xz.

(b) Determine w in terms of x, y, z.
32

Solution and Comments: Both the parts can be solved with a variety
of methods. We prefer to prove (a) using vectors and (b) by pure geometry
using areas.
−→ −→
Take A as the origin and call AB and AC (i.e. the position vectors of B
and C) as ~b and ~c respectively. Similarly, let d,
~ ~e and f~ be the position
vectors of D, E, F respectively. Then

d~ = λ~c (1)
and ~e = µ~b (2)
α
for some λ, µ ∈ (0, 1). Assume F divides BD in the ratio 1−α and CE in
β
the ratio 1−β . Then

f~ = (1 − α)~b + αd~ = (1 − α)~b + αλ~c (3)

But at the same time,

f~ = (1 − β)~c + β~e = (1 − β)~c + βµ~b (4)

As the vectors ~b, ~c are linearly independent, (3) and (4) together give

1−α = βµ (5)
and 1 − β = αλ (6)

which can be solved easily to express α and β in terms of λ and µ

1−µ
α = (7)
1 − λµ
1−λ
and β = (8)
1 − λµ
With this spadework, we are ready to prove (a), i.e. y 2 > xz which is
equivalent to proving that yz > xy . Now,

y BF α
= = (9)
z FD 1−α
and similarly,
x EF 1−β
= = (10)
y FC β
So, (a) is equivalent to proving that

αβ > (1 − α)(1 − β) (11)

On expansion, this reduces to proving that

α+β >1 (12)

 33

For this we apply (7) and (8) which together imply

2−λ−µ
α+β = (13)
1 − λµ

Hence (a) finally reduces to showing that 2 − λ − µ > 1 − λµ which can be

written as (1 − λ)(1 − µ) > 0. But that is true because λ and µ are both
less than 1 each.
For (b), we draw the line AF . It divides the area w into w1 and w2 which
are, respectively, the areas of the triangles AEF and AF D. We shall get
a system of two equations in w1 , w2 , x, y, z. It can be solved to express
each of w1 and w2 in terms of x, y, z. Adding, we shall get w in terms of
x, y, z.
We note that
AD △ABD x + w1 + w2
= = (14)
DC △CBD y+z
But at the same time
AD △AF D w2
= = (15)
DC △CF D z

By a property of ratios,
AD x + w1 + w2 − w2 x + w1
= = (16)
DC y+z−z y
Hence,
x + w1 w2
= (17)
y z

and therefore

zw1 − yw2 + xz = 0 (18)

Interchanging the roles of w1 and w2 , and also of x and z, we get a similar

equation

−yw1 + xw2 + zx = 0 (19)

(18) and (19) can together be solved as a system of two linear equations
in the two unknowns w1 and w2 . Skipping the details,

xz(y + x)
w1 = (20)
y 2 − xz
xz(y + z)
and w2 = (21)
y 2 − xz
34

Adding,
zx(x + 2y + z)
w = w1 + w2 = (22)
y 2 − xz
which gives an expression for w in terms of x, y and z. It also proves (a)
because as the areas are positive, the denominator y 2 − xz is positive, i.e.
y 2 > xz.
The second solution is shorter and more elegant because it makes use of the
powerful concept of areas. The arguments are very similar to those used
in the proof of a well-known result in geometry, called Ceva’s theorem.
It is very simple to state. Suppose the line AF when produced meets the
side BC at G as shown below.

E
w1 w
2
F D
x
z
y
B G C

Then Ceva’s theorem says that

AD CG BE
× × =1 (23)
DC GB EA
Although this is a result purely about the ratios of lengths of line segments,
the standard proof consists of expressing each ratio as a ratio of areas. The
△ABD
AD
ratio DC equals the ratio △CBD and also the ratio △AF D
△CF D . Hence by the
same property of ratios used above, we get
AD △ABD − △AF D △AF B
= = (24)
DC △CBD − △CF D △BF C
Similarly,
CG △AF C
= (25)
GB △BF A
BE △BF C
and = (26)
EA △AF C
Ceva’s theorem follows by multiplying these three equations.
 35

A good problem on areas in geometry. In (b), it would have been

better to ask to prove (22) instead of leaving the expression as open ended,
because there could be other expressions for w and some of them very
complicated.

Q.4 Let n1 , n2 , . . . , n51 be distinct natural numbers each of which has exactly
2023 positive integer factors. For instance, 22022 has exactly 2023 positive
integer factors 1, 2, 22 , . . . , 22021 , 22022 . Assume that no prime larger than
11 divides any of the ni ’s. Show that there must be some perfect cube
among the ni ’s. You may use the fact that 2023 = 7 × 17 × 17.

Solution and Comments: Let S = {n1 , n2 , . . . , n51 }. Each n ∈ S has

an odd number of divisors. So from the reasoning in Q.17 of Paper 1, each
n is a perfect square. Moreover, its only prime divisors are 2, 3, 5, 7 and
11. Hence it is of the form

n = 2a 3b 5c 7d 11e (1)

where a, b, c, d, e are all non-negative even integers (which will change as

n varies over S). Every divisor of n is of the form 2p 3q 5r 7s 11t where p
can take values from 0 to a, q can take values from 0 to b and so on. So
the number of possible divisors of n are (a + 1)(b + 1)(c + 1)(d + 1)(e + 1).
This is given to be 2023 whose prime factorisation is 7 × 17 × 17. So we
get

(a + 1)(b + 1)(c + 1)(d + 1)(e + 1) = 7 × 17 × 17 (2)

In order that n is a perfect cube, all exponents a, b, c, d, e must be multiples

of 3. But we already know that they are all even. So the assertion that
S contains a perfect cube is equivalent to saying that there is some n ∈ S
for which a, b, c, d, e are all divisible by 6, or equivalently, each of a + 1, b +
1, c + 1, d + 1 and e + 1 takes values which are congruent to 1 modulo 6.
But (2) puts a severe restriction on the possible values of these factors. The
possible factorisations of the R.H.S. of (2) lead to the following possible
factorisations of the L.H.S.

(i) (7,17,17,1,1)-type (That means one of the five factors (a + 1), (b +

1), (c + 1), (d + 1) and (e + 1) is 7, two are 17 and the remaining two
are 1).
(ii) (7, 289, 1,1,1)-type
(iii) (17,119, 1,1,1)-type
(iv) (2023, 1, 1, 1,1)-type

In (i), the factor 17 is not congruent to 1 modulo 6, i.e. is not of the form
6k + 1 for any integer k. So if n falls into this type, it cannot be a perfect
cube. The same is the case with (iii). However, in (ii) and (iv), all entries
36

are congruent to 1 modulo 7 and hence any n for which the factors (a + 1)
to (e + 1) are of this type will be a perfect cube, in fact a perfect sixth
power.
We are given that |S| = 51. Let us see how many of these 51 integers
are of the ‘bad’ types, i.e. of types (i) and (iii). This is an elementary
counting problem.
In (i), the factor 7 can be picked from the five factors in 5 ways. Once
done, from the remaining four factors, the two 17’s can be picked in 42 ,

i.e. 6 ways. Hence there are at most 30 elements of S of the type (i).
Similarly there are at most 5 × 4 = 20 elements of type (iii). As S has 51
elements, there is at least one n ∈ S which falls in type (ii) or (iv) and
hence is a perfect cube (in fact, a perfect sixth power).
An excellent problem requiring only elementary number theory and
combinatorics. The criterion for having an odd number of divisors was
already needed in Q.17 of Paper 1. This duplication could have been
avoided.

Q.5 There is a rectangular plot of size 1 × n. This has to be covered by three

types of tiles - red, blue and black. The red tiles are of size 1 × 1, the blue
tiles are of size 1 × 1 and the black tiles are of size 1 × 2. Let tn denote
the number of ways this can be done. For example, clearly t1 = 2 because
we can have either a red or a blue tile. Also, t2 = 5 since we could have
tiled the plot as: two red tiles, two blue tiles, a red tile on the left and a
blue tile on the right, a blue tile on the left and a red tile on the right, or
a single black tile.

(a) Prove that t2n+1 = tn (tn−1 + tn+1 ) for all n > 1.

P n−d
n−2d
(b) Prove that tn = d 2 for all n > 0. Here
d≥0

   m!
m r!(m−r)! , if 0 ≤ r ≤ m
=
r 0, otherwise

for integers m, r.

Solution and Comments: (a) Consider a plot of size 1 × (2n + 1) with

2n + 1 squares marked from 1 to 2n + 1 as shown in the diagram below
where there is a thick line between the squares marked n and n + 1.

1 2 n n+1 2n 2n+1

This thick line divides the plot into two (unequal) subplots, the left one
of size 1 × n and the right one of size 1 × (n + 1).
 37

There are two types of tilings of the entire plot of size 1 × (2n + 1). We
can tile the left subplot in any of the tn ways and simply juxtapose it with
any of the tn+1 tilings of the right subplot. So, in all, there are tn tn+1
tilings of this type.
But there are also some tilings of the entire plot which are not obtained
by freely juxtaposing tilings of the two subplots. These are the tilings in
which the n-th and the (n + 1)-th squares are together tiled with a black
tile. In this case we can tile the plot with the first n − 1 squares in any
of tn−1 ways and also the 1 × n plot from square n + 2 to square 2n + 1
in any of tn ways. We then put a 1 × 2 black tile in between. There are
tn−1 tn tilings of this type.
Put together, we get

t2n+1 = tn tn+1 + tn−1 tn = tn (tn−1 + tn+1 ) (1)

For (b), we classify the tn valid tilings according to the number, say d, of
black tiles. In such a tiling, there are n − 2d non-black tiles. For the time
being, we do not distinguish the red tiles from the blue ones. We call both
as ‘non-black’.
In a tiling with d black tiles, there are n − 2d non-black tiles. Hence the
total number of tiles is n − 2d + d = n − d. The number of ways to align
d
black and n − 2d non-black tiles is simply the binomial coefficient n−dd .
But in every such arrangement, the n − 2d non-black tiles can be coloured
with red or blue in 2n−2d ways. Hence
X n − d
tn = 2n−2d (2)
d
n≥0

which proves (b).

A good combinatorial problem requiring careful classification. (b)
is very close to the answer to Exercise (7.4.2) in Foundations of Discrete
Mathematics.
Q.6 Let {un }n≥1 be a sequence of real numbers defined as u1 = 1 and
1
un+1 = un + for all n ≥ 1
un
√
3 n
Prove that un ≤ 2 for all n.
√
Solution and Comments: Call 3 2 n as vn . We have to prove that
un ≤ vn for all n. It is tempting to apply induction on n. But the trouble
is that in un+1 = un + u1n even if un ≤ vn , there is no guarantee that
1
un ≤ (vn+1 − vn ). In fact, an upper bound on un gives rise to a lower
bound and not an upper bound on the reciprocal u1n .
38

Another line of attack is to hope that even though both un and vn increase
as n does, the increase in vn is higher than that in un . The trouble is that
we are not given un explicitly but only as a reurrence relation, which is far
from easy to solve. So there is no way to express the increase un+1 − un
directly in terms of n and compare it with the increase vn+1 − vn . In the
present case the increase, i.e. un+1 − un is u1n and this would be less than
the increase vn+1 − vn if and only if
1
< vn+1 − vn
un
3 √ √
= ( n + 1 − n)
2
3
= √ √ (1)
2( n + 1 + n)
which is the case if and only if
√ √
2( n + 1 + n)
≤ un (2)
3
So, insread of proving the given inequality we prove the double inequality
√ √ √
2( n + 1 + n) 3 n
≤ un ≤ (3)
3 2
by induction on n. Actually, the first inequality is miserably false for
n = 1, 2. But it is true for n = 3 because u2 = 2, u3 = 25 . On the other
√
hand, for n = 3 the expression on the left in (3) is 2(2+3 3) and so the first
√ √
inequality reduces to showing that 8 + 4 3 ≤ 15, i.e. to 4 3 ≤ 7 which
is true as we see by squaring both the sides.
So, we prove (3) only for n ≥ 3. The advantage of proving such a ‘double
engine inequality’ is that when dealing with u1n , we can use the other
inequality. We verified the truth of the first inequality just
√
now for n = 3.
The other inequality also holds for n = 3 since 25 ≤ 3 2 3 as we see by
squaring both the sides.
Now, for the inductive step, assume (3) holds for some n. To prove it for
n + 1, we must show that
√ √ √
2( n + 2 + n + 1) 1 3 n+1
≤ un + ≤ (4)
3 un 2
From the inductive hypothesis (3), we have
√ √
1 2( n + 1 + n) 2
un + ≥ + √ (5)
un 3 3 n
√ √ √
2 n( n + 1 + n) + 2
= √ (6)
3 n
√ √
2n + 2 + 2 n n + 1
= √ (7)
3 n
 39

Hence proving the first inequality in (4) reduces to showing that

√ √ √ √ √
n + 1 + n n + 1 ≥ n( n + 1 + n + 2) (8)
or, equivalently, to proving that
√ √
n+1≥ n n+2 (9)
which is true as we see by squaring.
We still have to prove the right half of (4). For this we apply the inductive
hypothesis (3) with its left half to get an upper bound for u1n , so that
√
1 3 n 3
un + ≤ + √ √ (10)
un 2 2( n + 1 + n)
3 √ √ √
= ( n + ( n + 1 − n)) (11)
2√
3 n+1
= (12)
2
We have thus proved (4) fully. Hence by induction (3) holds for all n ≥ 3.
Our interest is, of course, only in the right half of (3) which holds true for
n = 1, 2 by special verification. Hence the given inequality holds for all n.
Admittedly, this is a long proof which was necessitated because un
was neither given nor could be obtained as an explicit function of n. It
turns out that there is a shorter solution which begins by realising that the
inequality to be proved is equivalent to showing that u2n < 9n 4 . Squaring
the given relation we get
1
u2n+1 = u2n + 2 + (13)
u2n
and recasting this as
1
u2n+1 − u2n = 2 + (14)
u2n
or (for n ≥ 2,
1
u2n − u2n−1 = 2 + (15)
u2n−1

This can be telescoped to get

n n
X X 1
u2k − u2k−1 = 2(n − 1) + (16)
u2
k=2 k=2 k−1

i.e.
n−1
X 1
u2n = 2n − 1 + (17)
r=1
u2r
40

The terms of the series on the R.H.S. are strictly decreasing (since un is
strictly increasing) and the largest among them is the first one, viz. 1. This
gives 3n − 2 as an upper bound on u2n , But a more parsimonious estimate
is to note that since u1 = 1 and u2 = 2, all terms of the series from the
second onwards are at most 14 and decreasing. Hence 2n − 1 + 1 + (n−2) 4
is a strict upper bound for u2n . But that is less than 9n
4 and we are done.

Q.7 (a) Let n ≥ 1 be an integer. Prove that X n + Y n + Z n can be written as

a polynomial with integer coefficients in the variables α = X + Y +
Z, β = XY + Y Z + ZX and γ = XY Z.
(b) Let Gn = xn sin(nA)+ y n sin(nB)+ z n sin(nC), where x, y, z, A, B, C
are real numbers such that A + B + C is an integral multiple of π.
Using (a) or otherwise, show that if G1 = G2 = 0, then Gn = 0 for
all positive integers n.

Solution and Comments: We first construct a cubic equation whose

roots are X, Y and Z. The simplest such cubic is

(u − X)(u − Y )(u − Z) = 0 (1)

which when expanded becomes

u3 − αu2 + βu − γ = 0 (2)

The functions X + Y + Z, XY + Y Z + ZX and XY Z are called ele-

mentary symmetric functions of the three variables X, Y, Z. There is
a non-trivial theorem of algebra that any symmetric polynomial function
of X, Y, Z (with integer coefficients) can be expressed as a polynomial in
these three elementary symmetric functions. Actually, the number of vari-
ables can be more than 3 but then so will be the number of elementary
symmetric functions.
The present problem asks to do this for the special case where the function
is of the form X n + Y n + Z n for a positive integer n. The standard
notation for this function is Sn . We shall prove the assertion directly for
n = 1, 2, 3 and then by induction for higher n. For notational uniformity
S0 = X 0 + Y 0 + Z 0 = 1 + 1 + 1 = 3 (even when one of X, Y, Z may be 0.)
The first case is trivial since S1 = X + Y + Z is α itself.
For n = 2 we write

S2 = X 2 + Y 2 + Z 2 = (X + Y + Z)2 − 2(XY + Y Z + ZX) = α2 − 2β (3)

which is also a polynomial in α, β and γ (which is actually missing) and

has integral coefficients.
The next case X 3 + Y 3 + Z 3 can be handled with the well-know identity

X 3 + Y 3 + Z 3 − 3XY Z = (X + Y + Z)(X 2 + Y 2 + Z 2 − XY − Y Z − ZX) (4)

 41

using (3). But we give a proof which is also indicative of the inductive
step when we shall prove the assertion for any n. Since X, Y, Z are roots
of (2), we have

X3 = αX 2 − βX + γ
Y3 = αY 2 − βY + γ
and Z 3 = αZ 2 − βZ + γ

Adding,

S3 = X 3 + Y 3 + Z 3 = α(X 2 + Y 2 + Z 2 ) − β(X + Y + Z) + 3γ (5)

= αS2 − βS1 + S0 γ (6)

We can now substitute for S2 , S1 and S0 . But that is hardly necessary.

The problem merely asks to show that S3 can be expressed as a polyno-
mial (with integer coefficients) in α, β and γ and not to give a particular
expression for S3 . Since S2 , S1 and S0 are already such polynomials, so is
S3 by (6).
Now, for the inductive step, exactly as above we get that for n ≥ 4

Sn = αSn−1 − βSn−2 + γSn−3 (7)

By the induction hypothesis, the R.H.S. is a polynomial in α, β and γ with

integral coefficients and hence so is Sn . This completes the proof of (a).
For (b), we apply (a) for a particular choice of X, Y, Z. Note that nowhere
is it needed in (a) that the roots be real. They could as well be complex.
We take

X = xeiA = x cos A + ix sin A (8)

Y = yeiB = y cos B + iy sin B (9)
and Z = zeiC = z cos C + iz sin C (10)

Since A + B + C is of the form kπ for some integer k,

XY Z = xyzei(A+B+C) = xyzeikπ = ±xyz (11)

depending upon whether k is even or odd. The point is that in either case,
XY Z is real, i.e. its imaginary part is 0.
We claim that the imaginary part of Gn also vanishes which will prove the
result because since x, y, z are all real, for every n ≥ 1,

Gn = xn sin nA + y n sin B + z n sin C (12)

is precisely the imaginary part of X n + Y n + Z n by DeMoivres theorem.

Suppose G1 and G2 vanish. Then the imaginary parts of X + Y + Z and
X 2 +Y 2 +Z 2 vanish. But then by (3), the imaginary part of XY +Y Z+ZX
also vanishes.
42

Summing up, XY Z is always real, while X +Y +Z and XY +Y Z +ZX are

real because of vanishing of G1 and G2 . By (a), for any n, X n + Y n + Z n
is a polynomial with integer coefficients in X + Y + Z, XY + Y Z + ZX
and XY Z. So X n + Y n + Z n is real and, as already shown, this means
that Gn = 0.
More generally, there are identities, called Newton’s identities which
express the sums of the n-th powers of the roots of a polynomial in terms
of its coefficients. Their use will give a shorter solution but the gain is not
substantial, since anyway, our interest is not in an explicit expression for
Sn , but only in showing that it can be expressed as a polynomial in the
symmetric functions.
A good but lengthy problem in algebra. The stipulation that A+B +C
is a multiple of π makes it tempting to think that A, B, C are the angles of
a triangle and then look for the millions of trigonometric identities about
triangles. But this is a trap and the only way to come out of it is to realise
that A + B + C may as well be 2π in which case many of these identities
fail. That makes the problem tricky, the correct trick being the choice of
the numbers X, Y, Z.
Q.8 Let f : [0, 1] −→ IR be a continuous function which is differentiable on
(0, 1). Prove that either f is a linear function f (x) = ax + b or there exists
t ∈ (0, 1) such that |f (1) − f (0)| < |f ′ (t)|.

Solution and Comments: Let A = (0, f (0)), B = (1, f (1)) and L be

the straight line segment joining A and B. Without loss of generality,
assume that f (1) ≥ f (0). If f is not linear, then the graph of f is not L.
Hence there is some point, say, C = (c, f (c)) which lies outside L. The
slope of L lies between that of AC and of CB. Hence either AC or BC
has slope greater than f (1) − f (0). Applying the Lagrange Mean Value
Theorem to (0, c) (or to (c, 1) as the case may be), we get a point t ∈ (0, 1)
for which f ′ (t) > f (1) − f (0).
A very simple problem based on LMVT. Note that f ′ is not given to
Z 1
be continuous. If it were so, then f (1) − f (0) equals f ′ (x)dx which is
0
the average rate of growth of f ′ . The function f is linear if and only if f ′
is constant. When it is not linear, we may take t as the point where f ′ is
maximum.
 43

CONCLUDING REMARKS

This year the problems in Paper 2 are much better than those in the recent
years. Except possibly for the last problem, every question demands some
innovative thinking. The last question could have been dropped to make room
for a problem requiring the maximality principle for the solution. (See Q.1 of
Paper 2 of ISI 2021.)
But excellent as the problems are, the time constraint may be unfair. It is
entirely possible that a candidate can barely solve two or three problems during
the test, but after the test solves nearly all in a day or two, while some other
candidate solves, say four during the test but afterwards hardly adds one more.
Paper 1 contains some good problems. These are indicated in the respective
comments. The best problem is probably Q.19 about an application of conjugate
surds. Problem 25 and the last problem (asking for numbers of functions f which
satisfy f (i + 1) − f (i) ≥ 20 are close seconds.