Académique Documents
Professionnel Documents
Culture Documents
Section 1 3
Section 2 25
Concluding Remarks 46
Introduction
K. D. Joshi
2
Section 1
" #
a b
Q.1 Suppose A = is a real matrix with non-zero entries, ad−bc 6= 0
c d
and A2 = A. Then a + d equals
A. 1 B. 2 C. 3 D. 4
3
Incidentally, a matrix A which satisfies A2 = A is called an
idempotent matrix because all its powers, A, A2 , A3 , . . . , are the same.
Q.2 On any given interval of positive length on the unit circle |z| = 1 in the
complex plane,
Answer and Comments: (D). It would have been better had the
question included a definition of a root of unity, not that it is hard
to guess, once you know that ‘unity’ is an old name for the number
1 (‘naught’ being an old name for the number 0). So, logically, a
complex number z is a root of unity if there is some positive integer
n such that z n = 1. For every positive integer n, there are precisely
n distinct n-th roots of unity. They are the complex numbers of the
form zk = e2πki/n where k = 0, 1, 2, . . . , n − 1. (Other values of k do
not give new roots because e2πri/n = e2πsi/n whenever r and s differ by
an integral multiple of n. See Chapter 7 for more on complex roots of
unity and their applications in trigonometry.)
p
Put differently, if r = is a rational number where p, q are positive
q
integers, then z = e2πri is a root of unity, specifically, a q-th root
of unity. An arc of the unit circle will be a set of the form C =
{z ∈ C | : z = eiθ , α ≤ θ ≤ β} where α, β are some real numbers
with 0 ≤ α < β < 2π. It will contain the root e2πpi/q if and only if
α ≤ 2πp q
α
≤ β, which is equivalent to 2π ≤ pq ≤ 2π
β
. So there are as many
roots of unity in the arc C as there are rational numbers in the interval
α β
[ 2π , 2π ].
Every open interval contains infinitely many rationals. Moreover,
since β−α
2π
α β
< 1, no two distinct rationals in the interval [ 2π , 2π ] differ by
an integer. So the corresponding roots of unity are distinct. Hence (D)
is true.
A simple problem once you get the correct thought.
4
Q.3 For 0 < θ < π2 , four tangents are drawn at four points (±3 cos θ, ±2 sin θ)
2 2
to the ellipse x9 + y4 = 1. If A(θ) denotes the area of the quadrilateral
formed by these four tangents, the minimum value of A(θ) is
A. 21 B. 24 C. 27 D. 30
A. S = ∅
B. S is a non-empty finite set
C. S is an infinite proper subset of IR\{0}
D. S = IR\{0}
5
Q.5 On a rectangular hyperbola x2 − y 2 = a2 , a > 0, three points A, B, C
are taken as follows: A = (−a, 0); B and C are placed symmetrically
w.r.t. the x-axis on the branch of the hyperbola not containing A.
Suppose that the triangle ABC is equilateral. If the side length of the
triangle ABC is ka, then k lies in the interval
A. 0 B. 1 C. 2 D. infinite
6
2 2
Q.7 Let xa2 + yb2 = 1, a > b, be an ellipse with foci F1 , F2 . Let AO be
its semi-minor axis, where O is the centre of the ellipse. The lines
AF1 and AF2 , when extended, cut the ellipse again at points B and C
respectively. Suppose that the triangle ABC is equilateral. Then the
eccentricity of the ellipse is
A. √1 B. √1 C. 1
D. 1
2 3 3 2
Q.8 Let a = cos 1◦ and b = sin 1◦ . We say that a real number is algebraic if
it is a root of a polynomial with integer coefficients. Then
7
Expressions for the sines and cosines of angles of the form k degrees
are available when k = 180, 90, 60, 45, 36, 30, 15 and a few other values,
but not when k = 1. But if we recall how some of these values are
obtained, we see that they crucially use the formulas for the sines and
cosines of an angle kα in terms of the sines and cosines of α. For
example, cos 30◦ = 2 cos2 15◦ − 1. So, if we know cos 30◦ , we get cos 15◦
as a root of a quadratic.
In general for every integer n ≥ 0, cos(2n + 1)θ and sin(2n + 1)θ can
be expressed as polynomials with integer coefficients of degree 2n + 1
in cos θ and sin θ respectively. The cases n = 0, 1 are very familiar, viz.
cos 3θ = 4 cos3 θ − 3 cos θ and sin 3θ = 3 sin θ − 4 sin3 θ. The general
case can be proved by induction on n using the formulas
but the proof gets rather complicated because in the inductive step we
also need that cos(2n − 1)θ and sin(2n − 1)θ contain only odd degree
terms in cos θ and sin θ respectively. (So, this is an example where it
is easier to prove by induction, a more specific statement.)
Anyway, if we assume this result, then we can express cos 45◦ as a
polynomial of degree 45 in a (i.e. cos 1◦ ), say
1
√ = c45 a45 + c43 a43 + . . . + c3 a3 + c1 a (3)
2
where c45 , c43 , . . . , c3 and c1 are some integers. Squaring and multiply-
ing by 2, we get
8
Interestingly, if we consider the complex number z = a + ib =
cos 1 +i sin 1◦ , then it is much easier to show that z is algebraic because
◦
by DeMoivre’s rule, z 360 = cos 360◦ +i sin 360◦ = 1. More generally, this
argument shows that any root of unity is an algebraic complex number.
However, deducing from this that its real and imaginary parts are both
algebraic is not so easy. Clearly, z = a − ib is also algebraic because
(z)n = z n . It takes some work to show that the sum of two algebraic
numbers is algebraic. Once we grant that, we get that 2a = z + z and
2ib = z − z are algebraic. From there, the algebraicity of a and b is but
one step away.
There is, in fact, a way using DeMoivre’s rule, to express cos nθ
and sin nθ as certain polynomials in cos θ and sin θ. The resulting
polynomials are called Chebyshev or Tchebyshev polynomials. See
Exercise (7.21), parts (d) and (e).
Q.9 A rectangle with its sides parallel to the x-axis and the y-axis is in-
scribed in the region bounded by the curves y = x2 −4 and 2y = 4 −x2 .
The maximum possible area of such a rectangle is closest to the integer
A. 10 B. 9 C. 8 D. 7
Answer and Comments: (B). The region bounded by the two parabo-
2
las y = x2 −4 and y = 4−x
2
and an inscribed rectangle P QRS are shown
in the figure below.
y
2 4
y=x−
(0,2)
R Q
x
(−2,0) O (2,0)
S P
Clearly the rectangle is symmetric about the y-axis. Assume that its
2
base is 2a. Then P = (a, a2 − 4) and Q = (a, 4−a
2
). Hence the area of
9
2
the rectangle is 2a × ( 4−a
2
− (a2 − 4)) which comes out as a(12 − 3a2 ).
Call this as f (a). We have to maximise this as a varies over [0, 2].
Setting f ′ (a) = 0 gives a = √23 . As this is the only critical point and
f (0) = f (2) = 0, f attains its maximum at a = √23 . The maximum
q √
area is √23 × (12 − 4) = √163 = 256 3
= 85.33... Among the given
options the one whose square is closest to 85.33.. is 9.
Answer and Comments: (B). The functions x and sin x are dif-
ferentiable everywhere. But |x| is not differentiable at x = 0. So,
| sin x| is not differentiable when sin x = 0, i.e. when x = nπ for
n = 0, ±1, ±2, ±3, . . .. For n 6= 0, the factor x is non-zero and hence
the product x| sin x| is non-differentiable at x = nπ. However, when
n = 0, we have to check the differentiability of f (x) from the first prin-
ciples. In a small neighbourhood N of 0, we have f (x) = x sin x for
x > 0 and f (x) = −x sin x for x < 0. Then the right handed derivative
x sin x − 0
f+′ (0) equals lim+ = lim+ sin x = 0. Also the left handed
x→0 x x→0
′ −x sin x − 0
derivative f− (0) equals lim− = lim− − sin x = 0. As both
x→0 x x→0
these are equal, f is differentiable at 0.
A. {x ∈ [−1, 1] : x 6= 0}
2
B. {x ∈ [−1, 1] : x = 0 or x = n+1
,n ∈ ZZ}
10
2
C. {x ∈ [−1, 1] : x = n+1
,n ∈ ZZ}
D. [−1, 1]
Answer and Comments: (C). This is very similar to the last ques-
tion. But the first factor is x2 and we have to check if it is strong
enough
to cure the non-differentiability of the second factor, viz. cos( πx ). It
second factor at 0 was possible because the second factor was bounded
in a neighbourhood of 0. The moral is that when you are dealing
with a product of two functions, even if one of the factors is bad, the
other factor is sometimes strong enough to cure this defect. This does
not apply to sums. If h(x) = f (x) + g(x) and g is non-differentiable
at some point a, then differentiability of f at a always means non-
differentiability of h at a because otherwise we would have g = h − f
differentiable at a. So, the goodness of one term does not cure the
badness of the other. Interestingly, sometimes the sum of two bad
functions can be good. An extreme example is when g = −f . Here
the defects of the two terms cancel each other and so together they
live happily like a homely looking girl married to a hardly working boy.
(The descriptions of the qualifications of both the parties are borrowed
from popular matrimonial advertisements in cheap newspapers!)
Problems dealing with such good-bad pairs, especially their products,
have been asked many times in the JEE. It is important to note that if
11
we are dealing with some h(x) = f (x)g(x) where f (x) is good in some
sense, say differentiability at some point a, and g(x) is bad, then for the
product h(x) to be good despite the badness of g(x) at a, f (a) must
be 0. For otherwise, by continuity, f (x) would be non-zero in some
neighbourhood N of a and in this neighbourhood we can write g(x) as
h(x)
f (x)
which will make it good, being the ratio of two good functions.
Rπ
Q.12 The value of the integral (1 − | sin 8x|)dx is
0
A. 0 B. π − 1 C. π − 2 D. π − 3
Answer and Comments: (C). The term | sin 8x| changes its def-
inition whenever x crosses a point at which sin 8x vanishes. In the
interval [0, π], sin 8x = 0 when x = kπ 8
for k = 0, 1, . . . , 8. So we
π
consider 8 intervals of the form [(k − 1) 8 , k π8 ] for k = 1, 2, . . . , 8. Cor-
respondingly, the given integral splits into 8 integrals, one over each
one of these subintervals. But it is hardly necessary to evaluate each
of these 8 integrals separately. Because of the periodicity of the sine
function, sin(8( π4 + x)) = sin 8x and so the graphs of f (x) on alternate
intervals will be the same, while on two adjacent intervals of the form
[(k−1)π/8, kπ/8] and [kπ/8, (k+1)π/8], where k is odd, the graphs will
be reflections of each other in the line x = kπ/8. Hence the integral
over each subinterval will be the same. Therefore the given integral
π/8
equals 8 times the integral (1 − sin 8x)dx which comes out to be
R
0
π/8
π
+ 81 cos 8x = π
− 14 . Therefore the given integral equals π − 2.
8 0 8
Q.13 Let ln x denote the logarithm of x with respect to the base e. Let S ⊂ IR
be the set of all points where the function ln(x2 − 1) is well defined.
Then the number of functions f : S −→ IR that are differentiable,
satisfy f ′ (x) = ln(x2 − 1) for all x ∈ S and f (2) = 0 is
12
A. 0 B. 1 C. 2 D. infinite
Q.14 Let S be the set of all real numbers p such that there is no non-zero
Rx
continuous function f : IR −→ IR satisfying f (t)dt = pf (x) for all
0
x ∈ IR. Then S is
13
the set of those real numbers p for which at least one function f (x) sat-
isfying the conditions given exists. That set would be the complement
of the set S in the question and so the problem would be the same.
Instead of asking every man for whom there is no woman who can be
called his wife to raise his hand, it is far better to say ‘Those who are
married may raise their hands’ !
Anyway, coming to the question, in the equation
Z x
f (t)dt = pf (x) (1)
0
for all x ∈ IR. Note that p 6= 0 as otherwise f (x) would vanish identi-
cally. So we assume p 6= 0. Then (2) is a differential equation whose
general solution is
0 = ke0 = k (4)
14
Q.15 The probability of men getting a certain disease is 12 and that of women
getting the same disease is 51 . The blood test that identifies the disease
gives the correct result with probability 45 . Suppose a person is chosen
at random from a group of 30 males and 20 females, and the blood test
of that person is found to be positive. What is the probability that the
chosen person is a man?
75 3 15 3
A. 107
B. 5
C. 19
D. 10
15
of the problem means that |M| = 30, |W | = 20, |M ∩ D| = 12 |M| = 15
and |W ∩ D| = 51 |W | = 4. It is convenient to show this in a Venn
diagram, where the set D of diseased persons is shown in the middle
and the figures in the various regions represent their cardinalities. (Such
diagrams are only indicative and so it is not necessary that the figures
be proportional to the areas, as is the case in a pie cut diagram.)
4
15 16
D
W
15
Now, for E1 to occur, the person must be in D and the test correct.
The probabilities of these two events are 19
50
and 45 respectively. As
the correctness of the test does not depend on whether the person is
diseased or not, these two sub-events are independent of each other.
Therefore,
19 4 76
P (E1 ) = × = (2)
50 5 250
Similarly, for E2 to occur, the person must be healthy but the test
wrong. So
31 1 31
P (E2 ) = × = (3)
50 5 250
Putting these into (1), we get
76 31 107
P (E) = + = (4)
250 250 250
So we have found the probability that a randomly chosen person tests
positive. Now, given this, we have to find the probability that the
16
person tested is a man. For this we need to find P (E ∩ M), i.e. the
probability that the person chosen is a man and tests positive. Once we
do that, Bayes theorem will give that the ratio P (E∩M
P (E)
)
is the probability
that a person testing positive is a man.
Clearly,
where P (M) is the probability that the chosen person is a man and
P (F/M) is the probability that a man chosen at random tests positive.
As there are 30 men in a population of 50, P (M) is easy to find. It
comes as
30 3
P (M) = = (6)
50 5
But finding P (F/M), i.e. the probability that a randomaly chosen
man tests positive needs some work. The caclulations are very similar
to those for P (E), except that we replace the entire population set S
with the set M of all men. Then M ∩ D is the set of diseased men and
has 15 elements in it while M ∩ H is the set of healthy men and also
has 15 elements in it. So
15 4 15 1 75 1
P (F/M) = × + × = = (7)
30 5 30 5 150 2
From (5), (6) and (7),
1 3 3
P (E ∩ M) = × = (8)
2 5 10
So, finally by Bayes’ theorem, the desired probability p, i.e. the prob-
ability that a person testing positive is a man is
P (E ∩ M) 3/10 75
p = P (M/E) = = = (9)
P (E) 107/250 107
Probability problems often have a real life setting which makes them
very appealing. But such problems are also often prone to controversies
arising from different interpretations of the data. For example, in this
17
problem, let us take the first interpretation that if a person x tests pos-
itive then he/she is in D with probability 45 and in H with probability
1
5
. If x ∈ D, then x is a man with probability 15 19
while if x ∈ H, then
15
x is a man with probability 31 . Then p can be obtained by drawing
an appropriate tree diagram (similar to those in the solutions of the
problems in Comment No. 11 of Chapter 22). It comes out as
4 15 1 15
p = × + ×
5 19 5 31
12 3 429
= + = (10)
19 31 589
As this is not one of the options, a candidate is alerted that the paper-
setters have in mind some other meaning of the probability of correct-
ness of the blood test. But that is likely to cause confusion. Instead
of the statement about the blood test as given, it would have been far
better had the paper-setters spelt it out by saying that a person having
the disease tests positive with probability 54 and a healthy person tests
positive with probability 15 .
It is instructive to look at the difference in the two interpretations
of the probability of correctness of a test in terms of conditional prob-
ability. In this problem, if a person is known to be diseased (through
some other means perhaps), then 54 is the probability that the test is
positive. In other words, it is the conditional probability that a person
tests positive given that he/she is diseased. Effectively, this means that
you are testing the test and not the patient! This is usually done when
the test is in its experimental stages, for example, when it is claimed
as a cheap, convenient alternative to some other well established, reli-
able but invasive and expensive test. After conducting a large number
of such trials (both on diseased as well as healthy persons), a picture
emerges about the diagnostic value of the test, that is, the conditional
probability, say p∗ , that a person has a disease given that he/she tests
positive. In real life, it is this interpretation that is relevant because
now you are testing an actual patient (whose state vis-a-vis the disease
is unknown).
So, in a sense the two interpretations of the probability of the
correctness of the test are poles apart. That makes it all the more im-
perative that the question clearly tells which interpretation is intended.
18
Q.16 The number of functions f : [0, 1] −→ [0, 1] satisfying |f (x) − f (y)| =
|x − y| for all x, y in [0, 1] is
A. exactly 1
B. exactly 2
C. more than 2, but finite
D. infinite
19
Q.17 Suppose A is a 3 ×3 matrix consisting of integer entries that are chosen
at random from the set {−1000, −999, . . . , 999, 1000}. Let P be the
probability that either A2 = −I or A is diagonal, where I is the 3 × 3
identity matrix. Then
A. P < 10118 B. P = 1
1018
2 52 54
C. 10518 ≤ P ≤ 1018
D. P ≥ 1018
2018
!2 2018
!
kx2k
X X
I. kxk ≤N
k=1 k=1
2018
!2 2018
!
k 2 x2k
X X
II. kxk ≤N
k=1 k=1
Then
20
Answer and Comments: (A). The sums on the L.H.S. in both the
inequalities are the same, while the sum on the R.H.S. of I is clearly
smaller than that of II. So, if I is true, then automatically so is II.
To check if I is true it is very tempting to apply the Cauchy-Schwarz
inequality to the L.H.S. In its classic form it says that for any positive
integer n and any real numbers u1 , u2 , . . . , un , v1 , v2 , . . . , vn we have
n
!2 n
! n
!
u2k vk2
X X X
uk vk ≤ (1)
k=1 k=1 k=1
But this may not be true because the coefficient of x2k in the L.H.S. is
a fixed number 2037
3
while that on the R.H.S. is k for k = 1, 2, . . . , 2018.
So nearly two third terms in the sum in the L.H.S. are greater than the
corresponding terms in the sum on R.H.S. Moreover, no upper bounds
are given on any xk . So even a single huge x1 can render (3) false.
The failure of this approach can be analysed if we recall when
equality holds in the Cauchy-Schwarz inequality. In (1) it holds if
and only if the vk ’s are proportional to the uk ’s. So, in (2) equality
holds, if and only if x1 , x2 , . . . , x2018 are proportional to 1, 2, . . . , 2018
respectively. This precludes the possibility of choosing x1 very huge as
compared to the other xk ’s. So, in (2) the R.H.S. is likely to be a lot
bigger than the L.H.S. Hence it will not be of any help in proving (3).
Nothing wrong in aiming at a higher target, because if you can shoot it
then you can comfortably shoot the lower one. But the target you are
21
aiming at should not be so high that it is beyond your capacity. So,
it was a mistake to apply the Cauchy Schwarz inequality in the most
obvious way.
Let us see if we can apply the same inequality a little more efficiently.
2018 2018
k 2 . Had the
X X
A hint is provided by giving N as k rather than as
k=1 k=1
latter been the case, then, the constant on the L.H.S. of (3) would
go away and then (3) would be indeed be true. If the first sum on the
2018
X √
R.H.S. of (2) were k then it would exactly equal N. But k = ( k)2 .
k=1 √ √
So if we rewrite kxk as k × kxk and then apply the Cauchy-Schwarz
inequality, then, instead of (2) we would get
2018
!2 2018
! 2018
! 2018
!
kx2k kx2k
X X X X
kxk ≤ k =N (4)
k=1 k=1 k=1 k=1
and now we are home because the R.H.S. of (4) is the same as that of
I. Thus we have proved I and as already observed II as well.
A somewhat tricky problem. Also the lower bounds given on xk ’s
serve no purpose. Such redundant stipulations only serve to confuse
the candidates.
Q.19 Let x2 = 4ky, k > 0, be the parabola with vertex A. Let BC be its latus
rectum. An ellipse with centre on BC touches the parabola at A, and
cuts BC at points D and E such that BD = DE = EC (B, D, E, C
in that order). The eccentricity of the ellipse is
√ √
(A.) √1 (B.) √1 (C.) 5
(D). 3
2 3 3 2
22
y
2
x = 4y
x
A
The ellipse has its semi-minor axis 23 and semi-major axis 1. (Note that
the major axis is along the y-axis and not parallel to the
√ x-axis. Hence
4
its eccentricity is given by 9 = 1 − e which gives e = 35 .
2
Q.20 Let f : [0, 1] −→ [−1, 1] and g : [−1, 1] −→ [0, 2] be two functions such
that g is injective and g ◦ f : [0, 1] −→ [0, 2] is surjective. Then
A. f must be injective but need not be surjective.
B. f must be surjective but need not be injective.
C. f must be bijective
D. f must be a constant function
23
g(y) = y + 1. For f to be surjective, it is enough if it maps some part,
say [0, 1/2] of [0, 1] onto [−1, 1]. The rest half, viz. (1/2, 1] can be
mapped arbitrarily to any points in [−1, 1]. The data of the problem
will be satisfied because g is a bijection. For an actual counter-example,
we construct f by taking it of the form f (x) = Ax + B for some real
numbers, chosen so that f (0) = −1 and f ( 21 ) = 1. This gives B = −1
and A2 + B = 1. So, A = 4. Hence f (x) = 4x − 1 for x ∈ [0, 21 ] and for
x ∈ ( 21 , 1] choose f (x) to be any element of [−1, 1].
24
Section 2
A. 3 B. 4 C. 5 D. 6
R T
25
is sheer common sense. But it has a formal name, the pigeon hole
principle. It says that if a set with more than rs elements (with r and
s some positive integers) is expressed as a union of r subsets, then at
least one of these subsets contains more than s elements.
In many applications of the pigeon hole principle, little more is
needed. The classic example of this is the Ramsey problem. Suppose
that there are six persons at a party. Prove that either there are three
among them every two of whom know each other or that there are three
no two of whom know each other. (There is a more colourful version of
this. Suppose that there are six points in plane, and every two of them
are joined by an arc. We ignore the intersections of these arcs, except
when the intersection is at one of these six points. Ramsey problem asks
to show that no matter how we colour these 15 arcs with two colours,
say red and blue, there will be a monochromatic triangle, i.e. a triangle
all three of whose edges are of the same colour.) The problem is simple
but not trivial. For a solution see Comment No. 16 of Chapter 6. But
Ramsey’s problem is just the beginning. Using generalised versions of
the pigeon hole principle, highly non-trivial results can be obtained.
There is a whole branch of mathematics, called Ramsey theory.
Of course, just because there cannot be more than 6 points common
to the perimeters of R, C and T , does not, by itself, mean that there
can be a situation where there are six points. Perhaps by a more clever
argument than the pigeon hole principle, one can get a better upper
bound. This happens frequently in mathematics. For example, for an
acute angle θ, we have sin θ cos θ ≤ 1 because both the factors are at
most 1. But if we rewrite sin θ cos θ as 21 sin 2θ, we see that it can be at
most 12 . This is the best upper bound because it is actually attained
when θ = π4 and so we cannot improve on it.
Returning to the problem, the deceptive word in it is ‘rectangle’.
Normally, we take a rectangle so that its length is considerably bigger
than its breadth. It is clear that if R is such a rectangle, then a circle
can cut it only on its longer sides. But if the rectangle is more like
a square, then it is possible to draw a circle C which cuts all its four
sides in two points each. This will happen if the circle√ has the same
centre as the square and its radius lies between a and 2a where 2a is
the side of the square.
26
Once this idea strikes, the solution is easy. We draw such a square
R and the circle C as shown in the figure. We take both the points of
their intersection on one pair of parallel sides. But from the other two
sides we pick only one each so that they lie on a line parallel to the
other two sides. (This can be done in several ways, But we need only
one.) Then we draw a triangle T in such a way that each side contains
two of these points. One of the sides contains an edge of the square
too. So, the points chosen on it qualify to lie both on R and T .
Some persons hesitate to accept a square as a rectangle. (It all
depends on how inclusive you are about degeneracies. A square is a
degenerate case of a rectangle. But then a circle is a degenerate case
of an ellipse too.) Anyway, if instead of a square we want a ‘genuine’
rectangle, i.e. a rectangle which is not a square, that can be arranged
by elongating the two vertical sides of the square downwards slightly
and pushing the base of T downwards accordingly.
Q.82 The number of different possible values for the sum x + y + z, where
x, y, z are real numbers such that x4 + 4y 4 + 16z 4 + 64 = 32xyz is
A. 1 B. 2 C. 4 D. 8
27
If x, y, z are all non-negative,
√ then by √the A.M.-G.M. inequality the
L.H.S. is at least 4 × x × 2y × 2z × 2 2. But this product is simply
32xyz which is the R.H.S. of (1). So, the only possible solution is
√ √
x = 2y = 2z = 2 2 (2)
i.e.
√ √
x = 2 2, y = 2, z = 2 (3)
So the given equation has only one solution if x, y, z are all positive.
But if we change the signs of any two values in (3), the equation still
holds. This can be done in three different ways. So, in all the given
equation has four distinct solutions.
In commenting on Q.6 we already remarked about the duplication
of ideas. We now have a case of triplication!
Q.83 Let Γ be a circle with diameter AB and centre O. Let ℓ be the tangent
to Γ at B. For each point M on Γ different from A, consider the tangent
t at M and let it intersect ℓ at P . Draw a line parallel to AB through
P intersecting OM at Q. The locus of Q as M varies over C is
A. an arc of a circle
B. a parabola
C. an arc of an ellipse
D. a branch of a hyperbola
28
A
Q
M
C
S
O
T
L
B P
The triangles OSQ and P MQ are right angled and have one angle
common. Also OS = BP = P M. Hence ∆OSQ ≡ ∆P MQ. Therefore
OQ = QP . Hence the point Q is equidistant from the fixed point O
and the fixed line ℓ. Therefore its locus is a parabola with focus O and
directrix ℓ.
x2 − 2y = 1 (6)
which is a parabola.
Clearly, the pure geometry solution is shorter and gives more specific
information about the locus, viz. its focus and directrix. Of course, we
can get this from (6) too.
A. 0 B. 1 C. 2 D. 3
A. 0
30
B. 1
C. more than 1, but finite
D. infinite
31
Q.86 Suppose the limit
√ Z 1 1
L = lim n dx
n→∞ 0 (1 + x2 )n
exists and is larger than 12 . Then
1
A. 2
<L<2 B. 2 < L < 3 C. 3 < L < 4 D. L ≥ 4
Answer and Comments: (A). The problem does not ask us to iden-
tify L or even to prove that it exists. It is also given that L > 12 . Our
task is to decide which of the four possibilities hold for L. Call the
given integral as In . If we could evaluate In directly as√a function of
n, we can decide which possibility holds by considering nIn for large
values of In .
The substitution x = tan θ gives
π/4 π
Z
I1 = 1dθ = (1)
0 4
However, for n > 1, it is not easy to express In directly as a function of
n. The best we can do is to get a reduction formula for it, i.e. express
In in terms of In−1 . This can be done if we integrate by parts.
Z 1
In = (1 + x2 )−n dx
0
1 Z 1
= [x(1 + x2 )−n ] + 2n (1 + x2 )−n−1 x2 dx
0 0
1 Z 1
= n + 2n [(1 + x2 )−n−1 (1 + x2 − 1)dx
2 0
1
Z 1 Z 1
2 −n
= n + 2n (1 + x ) dx − 2n (1 + x2 )−n−1 dx
2 0 0
1
= n + 2nIn − 2nIn+1 (2)
2
Hence
1 2n − 1
In+1 = + In (3)
2n+1 n 2n
By induction on n we shall show that
√
nIn < 2 (4)
32
for all n. This does not quite prove that L < 2. But it does prove that
L ≤ 2 and eliminates options (B), (C) and (D). Surely (4) is true for
n = 1 since I1 = π4 < 2.
For the inductive step we use (3) and the inductive hypothesis to get
√ √
√ n + 1 2n − 1 n + 1 √
n + 1In+1 = + √ nIn
2n+1 n 2n n
√ s
n + 1 2n − 1 1
< + 1 + (5)
2n+1 n n n
√
So n + 1In+1 < 2 will follow if we can prove that
s √
2n − 1 1 n+1
1 + < 2 − n+1 (6)
n n 2 n
√
The substitution h = n1 reduces the L.H.S. to (2 − h) 1 + h. By the
Taylor approximation of order 2, we have
√ h h2
1+h<1+ − (7)
2 8
for all h > 0. Hence
√ h h2
(2 − h) 1 + h < (2 − h)(1 + − )
2 8
2 3
3h h
= 2− +
4 8
3 1
= 2− 2 + 3 (8)
4n 8n
Hence (6) will be proved if we can show that
√
3 1 n+1
2 − 2 + 3 < 2 − n+1 (9)
4n 8n 2 n
which can be rewritten as
√
n+1 3 1 6n − 1
n+1
< − 2 = (10)
2 4n 8n 8n2
33
or equivalently,
√
n + 1n2
< 2n−2 (11)
6n − 1
The L.H.S. is bounded by a rational function in n while the R.H.S. is
an an exponential function. Hence (11) holds for all sufficiently large n.
But
√ by a direct verification, it is true even for n = 2, 3, 4, . . . . Therefore
nIn < 2 for all n ≥ 2. (For the sake of completeness, an elementary
inductive proof of (11) can be given by observing that as n increases
by 1, the expression on the L.H.S. increases by a factor which is only
marginally bigger than 1, while the R.H.S. always increases by a factor
of 2.)
This proof is far too advanced for the KVPY standards. A much
shorter and more direct proof of (4) (which does not use the reduction
formula (3)) can be given simply by obtaining an upper bound on the
1
integrand on [0, 1]. All we need is to expand (1 + x2 )n by the
(1 + x2 )n
binomial theorem. As all the terms are positive, we get
(1 + x2 )n > 1 + nx2 (12)
for all n > 1. Hence
1 dx 1 dx
Z Z
In = 2 n
< (13)
0 (1 + x ) 0 1 + nx2
√
The substitution y = nx converts the integral on the R.H.S. to
1 1 dy
Z
√ π
√ which we already know to be n 4 . Hence,
n 0 1 + y2
√ π
nIn < (14)
4
for all n ∈ IN. This implies that L ≤ π4 and hence L < 2 since π4 < 2.
A reduction formula is a powerful tool. But when the question is
not so demanding, an elementary alternative can often be found as in
this problem.
We indicate how a similar reduction formula for a closely related
integral, along with a well-known approximation formula for the fac-
torials (called the Stirling approximation) enables us to obtain the
value of L. (Of course, this is not asked in the present problem.)
34
Putting x = tan θ, we get
Z π/4
In = cos2n−2 θdθ (15)
0
= Kn − J n (16)
where
Z π/2
Kn = cos2n−2 θdθ (17)
0
Z π/2
and Jn = cos2n−2 θdθ (18)
π/4
π 1 √
Note that Jn ≤ 4 2n−1
, nJn → 0 as n → ∞. Hence
So
√ √
lim nIn = lim nKn (19)
n→∞ n→∞
which implies
2n − 3
Kn = Kn−1 (23)
2n − 2
We could have also derived (2) using the substitution x = tan θ. But
the advantage this time is that the first term in (21) vanishes and so
35
the reduction formula is much simpler. Indeed, repeated applications
of this formula give
2n − 3
Kn = Kn−1
2n − 2
(2n − 3)(2n − 5)
= Kn−2
(2n − 2)(2n − 4)
..
.
(2n − 3)(2n − 5) . . . 1
= . . . K1 (24)
(2n − 2)(2n − 4) . . . 2
Z π/2 π
By a direct calculation, K1 = 1dθ = . Multiplying and dividing
0 2
by (2n − 2)(2n − 4) . . . 2 we get
(2n − 2)! π
Kn = (25)
[2n−1 (n− 1)!]2 2
n! √
Theorem: As n → ∞, √ n n → 2π
n( e )
nn√
This theorem allows us to replace n! by for large n. Doing
2πn
en
so for the two factorials (2n − 2)! and (n − 1)! in (25), we get
√
√ π n(2n − 2)!
lim nKn = lim n−1
n→∞ 2 n→∞ [2 (n − 1)!]2
q √
π 2π(2n − 2)(2n − 2)2n−2 n e2n−2
= lim × 2n−2 ×
2 n→∞ e(2n−2) 2 2π(n − 1)2n−1
36
√ √
The bracketed expression simplifies to √1π × n−1
n−1
n
and tends to √1π
√ √ √
as n
√
→ ∞. Hence nKn → 2π as n → ∞, By (19) √nIn also tends
to 2π as n → ∞. In other words, L exists and equals 2π .
37
y
D
C
(0,n) E(n,n)
B Q R
P
A
L
x
(0,0) (n,0)
Then the region R lies in the first quadrant in between the concentric
circles C and D. So, if we denote the area of R by ∆n , we get an easy
upper bound on it, viz.,
πn2
" !#
1
∆n ≤ 2 − 1 + (1 − √ )2
4 n
2
!
πn 2 1
= √ −
4 n n
πn2
≤ √ (3)
2 n
38
π
we know that it is at most √ . As this tends to 0 when n → ∞, so
2 n
does Pn .
An excellent problem which tests a candidate’s ability to get to the
key idea, which in the present problem is that if ℓ touches the given
circle C then no point on it can be inside C. After that there is a jump
in concluding that the probability that a point be picked randomly from
some subset R of the square [0, n] × [0, n] is proportional to its area. It
will be easier to accept this if instead of the set An as given we consider
the set Bn of all ordered pairs of the form ( nx , ny ) where x, y are integers
ranging from 0 to n. In effect, Bn is obtained by compressing the square
[0, n] × [0, n] to the unit square S = [0, 1] × [0, 1] under the mapping
which sends a point (x, y) to the point ( nx , ny ). This compression does
not affect tangency and so the probability Pn remains the same. Note
that S is independent of n. But as n increases, not only the number
of points in Bn increases, their density also increases. As n → ∞, the
discrete set Bn of points tends to the continuous set of all points of the
unit square S. The concept of probability as the ratio of cardinalities
no longer applies now. Instead we have to take the ratios of their areas.
Problems of this type are not asked commonly. To help understand
the solution better, at the end of Section 9 of Chapter 23 on infinitistic
probability, there is a problem of a similar type. It asks to find lim pn
n→∞
where pn is the probability that two points picked at random from
the set { 2i
n
: 0 ≤ i ≤ n} are at a distance 1 or less apart from each
other. Here pn can be calculated explicitly and then its limit comes
out as 34 . But if our interest is only in the limit, it is the same as the
probability that two points taken at random from the interval [0, 2] are
at a distance 1 or less apart. And this is solved at the beginning of the
same section, by taking the ratio of certain areas.
A. 0
39
B. 1
C. more than 1, but finite
D. infinite
40
y
A ∆
x
a O h b
∆ = ∆1 − ∆2 (2)
where ∆1 is the area between the chord and the x-axis and ∆2 is the
area between the curve y = x2 and the x-axis, both lying between x = a
and x = b.
As ∆1 is the area of a trapezeum, a direct calulation gives
a2 + b2 Z b
∆ = (b − a) − x2 dx
2 a
a2 + b2 b3 − a3
= (b − a) − (3)
2 3
As the factor b − a is common to both the terms, it is conveninet to
call it h and write b as a + h. Then (3) becomes
41
maximum occurring when h = 1, i.e. when the chord is horizontal. So
the maximum value of ∆ is 61 .
The problem is conceptually simple. But the introduction of h
makes it computationally simple too because the expression for ∆ in
terms of h is surprisingly simple. (It is also remarkable that ∆ depends
only on b − a and not on both a and b.) Had we not introduced h,
instead of (4) we would get
(b − a)3
∆= (5)
6
and the problem would have been to maximise (b − a)3 subject to the
constraint (b − a)2 + (b2 − a2 )2 = 1 (obtained by setting the length of
the chord AB equal to 1). If we let θ be the angle the chord AB makes
with the x-axis, then b2 − a2 = (b − a) tan θ and the constraint becomes
b − a = cos θ and we have to maximisie cos3 θ. Clearly, the maximum
occurs when cos θ = 1. This gives the same answer as before.
Parabolas, like other conics, have been extensively studied in classi-
cal geometry. But the formula (4) seems not to be well-known, despite
its simplicity. A possible explanation is that it deals with areas unlike
most other formulas which deal mostly with angles and lengths.
√
Q.90 Suppose z is any root of 11z 8 + 20iz 7 + 10iz − 22 = 0, where i = −1.
Then S = |z|2 + |z| + 1 satisfies
A. S ≤ 3 B. 3 < S < 7
C. 7 ≤ S < 13 D. S ≥ 13
42
impossible to answer this by actually identifying all the roots of (1).
But we can recast it as
The advantage of this recasting is not that we can solve the equation,
but that when we take the squares of the absolute values of both the
sides, some of the terms in the expansion of |11w 7 +10|2 will be common
to some of the terms in the expansion of |10w 7 + 11|2. Doing so, we get
from (3),
Both the sides have two factors both of which are positive. This does
not mean that their factors match. But it does imply that when one of
the factors on one side is greater than the corresponding factor on the
other side, the remaining two factors on the two sides must satisfy the
opposite inequality. That is,
i.e.
Summing up, |w| ≥ 2 holds if and only if |w| ≤ 1 holds. But these two
statements are mutually contradictory and can never hold together. So
both are false. This gives |w| < 2 and |w| > 1. Thus 1 < |w| < 2 and
as already noted, S = |w|2 + |w| + 1 lies between 3 and 7. So (B) holds.
Lest this solution appears too tricky, it is instructive to recast (3)
a little differently, viz.
w 10w 7 + 11
= (7)
2 11w 7 + 10
43
Let us put z = w 7 . (Earlier we let w = iz and hence z = −iw. But
that is all over now. So no harm in using z for something else now.
Those who are uncomfortable, may denote w 7 by the complex number
Z instead of z.) Then (7) can be written as
w
= T (z) (8)
2
where
az + b
T (z) = (9)
cz + d
with a = d = 10 and b = c = 11. T (z) is a ratio of two linear expressions
(i.e. two first degree polynomials) in z. For this reason, T (z) is an
example of what is called a fractional linear transformation or
FLT for short. It is also called a Mobius transformation. More
az+b
generally, an FLT is any transformation of the form T (z) = cz+d where
a, b, c, d are complex numbers with ad − bc 6= 0. The last stipulation is
added to exclude the degenerate case where T (z) is a constant mapping.
Note that T (z) is undefined when the denominator vanishes, i.e. when
z = − dc . It is customary to add one point ∞ to the complex plane C |
d a
and say that T (− c ) = ∞. To make up, we set T (∞) = c with the
understanding that if c = ∞, then T (∞) = ∞. (Note that a and c
cannot both vanish because of the condition ad − bc 6= 0.)
With this understanding, every FLT can be thought of as a transfor-
mation from the extended complex plane C | ∗ = C
| ∪ {∞} to itself.
44
Theorem: Suppose that a, b, c, d are real numbers with a = d, b = c
and |a| < |b|. Then the transformation (9) maps the unit circle onto
∗
itself, the region {z ∈ C
| : |z| < 1} onto the region {z ∈ C
| : |z| > 1}
and vice versa.
10z + 11
We now go back to (7). The R.H.S. is T (w 7) where T (z) = .
11z + 10
Since |w 7| = |w|7, the conditions |z| = 1, |z| < 1 and |z| > 1 are
equivalent, respectively, to the conditions |w| = 1, |w| < 1 and |w| > 1.
On the other hand | w2 | = |w| 2
. Hence |w| > 2 holds if and only if
|T (w)| > 1 holds and by the theorem above this happens if and only if
|w| < 1.
Thus the concept of a fractional linear transformation enables us to
cast the solution more systematically. It is possible that the problem
was inspired by the property of FLT’s given in the theorem above. If
so, the paper-setters must be commended for making an imaginative
use of FLT’s. But is not clear what is gained by introducing iz instead
of z. The problem could as well have given that a complex number
z satisfies the equation 11z 8 − 20z 7 + 10z − 22 = 0. Replacing z by
iz gives a nasty twist to the problem without enhancing its academic
quality.
45
Concluding Remarks
46
Shockingly absent are real life problems which can be reduced to simple
algebra or sometimes even to arithmetic. Such problems can even be tried by
laymen and serve to popularise the test from which they come. Two notable
problems of this type asked in KVPY in recent years come to mind. One
was about the loss of weight in watermelons and the other about the motion
of ants on a line.
47