CHENNAI MATHEMATICAL INSTITUTE

Undergraduate Entrance Examination, 7th May 2023

Time: 3.5 hours

Unless specified otherwise, in this exam all numbers are real, the domain of each function
is the set of real numbers (or an implied subset) and the codomain is also the set of real
numbers. You may use the following information wherever you find it relevant.

• 2023 = 172 ⇥ 7.

• One can use long division to find the gcd of two positive integers a, b (defined to be a
common divisor d of a and b such that d is divisible by any other such common divisor).

• The same procedure stays valid for finding gcd of polynomials in one variable with
rational/real/complex coefficients.

• Any nonzero polynomial of degree n has n complex roots counting multiplicity.

Part A instructions

• Part A has 10 questions, each worth 4 points, for a total of 40. Points for
part A will be given based only on the answers you enter into the computer.

• 7 out of the 10 questions are objective, each with a group of four statements.
(These statements are numbered 1 to 28 for technical reasons.) For each statement,
independently choose one of the three options True / False / No Attempt. In particular
there is no guarantee that at least one of the four statements in a given question is
true. If you do not choose an option for a statement, it will be treated as No Attempt.

• Grading scheme for the 7 objective questions is as follows.

All 4 answers correct 4 points Note that getting even one of the four
3 correct and 1 No Attempt 2 points answers wrong will result in zero points
for that question. So if you are not sure,
2 correct and 2 No Attempt 1 point you are advised to choose No Attempt
instead of guessing.
Anything else 0 points

• The remaining three questions have two parts each. For each part, enter only the final
answer into the computer in the precise format specified in the question. There is no
negative marking for these three questions.

• Part A will be used for screening. Part B is assured to be graded if you meet
any one of the following two conditions. (i) You score at least 24 in part A. (ii) You
are among the top 400 students in part A. Thus part B will be graded for at least 400
students, more if enough students score at least 24 in part A.

Instrutions for part B !

 Part B instructions

• Part B has 6 problems worth a total of 80 points. See each question for the
break-up. You are advised to spend at least 2 hours on part B.

• Clearly explain your entire reasoning. No credit will be given without correct
reasoning. Partial solutions may get partial credit. You may solve a later part of a
problem by assuming a previous part, even if you could not do the earlier part.

• Solve each part B problem on the designated pages in the answer booklet. Use
the blank pages at the end for rough work OR if you need extra space for any problem.
Clearly label any such solution overflowing to last pages. For problems with multiple
parts, clearly label your solution to each part separately.
 Part A Questions for CMI BSc entrance exam on May 7, 2023

A1. Define the right derivative of a function f at x = a to be the following limit if it exists.
f (a + h) f (a)
lim+ , where h ! 0+ means h approaches 0 only through positive values.
h!0 h
Statements
(1) If f is di↵erentiable at x = a then f has a right derivative at x = a.
(2) f (x) = |x| has a right derivative at x = 0.
(3) If f has a right derivative at x = a then f is continuous at x = a.
(4) If f is continuous at x = a then f has a right derivative at x = a.

A2. Suppose a rectangle EBF D is given and a rhombus ABCD is inscribed in it so that
the point A is on side ED of the rectangle. The diagonals of ABCD intersect at point G.
See the indicative figure below.

Statements
(5) Triangles CGD and DF B must be similar.
AC EB
(6) It must be true that BD = ED .
(7) Triangle CGD cannot be similar to triangle AEB.
(8) For any given rectangle EBF D, a rhombus ABCD as described above can be con-
structed.

A3. This question is about complex numbers.

Statements
(9) The complex number (e3 )i lies in the third quadrant.
(10) If |z1 | |z2 | = |z1 + z2 | for some complex numbers z1 and z2 , then z2 must be 0.
(11) For distinct complex numbers z1 and z2 , the equation |(z z1 )2 | = |(z z2 )2 | has at
most 4 solutions.
(12) For each nonzero complex number z, there are more than 100 numbers w such that
w2023 = z.

1
A4. Statements
1
(13) lim e x = +1.
x!0

(14) The following inequality is true.

ln x ln x
lim 100
< lim 1 .
x!1 x x!1 x 100

(15) For any positive integer n,

Z n
n
x2023 cos(nx)dx < .
n 2023

(16) There is no polynomial p(x) for which there is a single line that is tangent to the graph
of p(x) at exactly 100 points.

A5. Statements
p p
(17) 4 < 5 + 5 5.
1+log2 61
(18) log2 11 < 2
.
(19) (2023)2023 < (2023!)2 .
(20) 92100 + 93100 < 94100 .
P P
n
A6. For a sequence ai of real numbers, we say that ai converges if lim ai is finite.
n!1 i=1
In this question all ai > 0.
Statements
P
(21) If ai converges, then ai ! 0 as i ! 1.
P
(22) If ai < 1i for all i, then ai converges.
P P
(23) If ai converges, then ( 1)i ai also converges.
P P
(24) If ai does not converge, then i tan(ai ) cannot converge.

A7. Statements
(25) To divide an integer b by a nonzero integer d, define a quotient q and a remainder r to
be integers such that b = qd + r and |r| < |d|. Such integers q and r always exist and are
both unique for given b and d.
(26) To divide a polynomial b(x) by a nonzero polynomial d(x), define a quotient q(x) and
a remainder r(x) to be polynomials such that b = qd + r and degree(r) < degree(d). (Here
b(x) and d(x) have real coefficients and the 0 polynomial is taken to have negative degree
by convention.) Such polynomials q(x) and r(x) always exist and are both unique for given
b(x) and d(x).

2
(27) Suppose that in the preceding question b(x) and d(x) have rational coefficients. Then
q(x) and r(x), if they exist, must also have rational coefficients.
(28) The least positive number in the set

{(a ⇥ 20232020 ) + (b ⇥ 20202023 )}

as a and b range over all integers is 3.

A8. You play the following game with three fair dice. (When each one is rolled, any one of
the outcomes 1, 2, 3, 4, 5, 6 is equally likely.) In the first round, you roll all three dice. You
remove every die that shows 6. If any dice remain, you roll all the remaining dice again in
the second round. Again you remove all dice showing 6 and continue.
Questions
(29) Let the probability that you are able to play the second round be ab , where a and b are
integers with gcd 1. Write the numbers a and b separated by a comma. E.g., for probability
10
36
you would type 5,18 with no quotations, space, full stop or any other punctuation.
(30) Let the probability that you are able to play the second round but not the third round
be dc where c and d are integers with gcd 1. Write only the integer c as your answer.
E.g., for probability 34
36
you would type 17 with no quotations, space, full stop or any other
punctuation.

A9. Two lines `1 and `2 in 3-dimensional space are given by

`1 = {(t 9, t + 7, 6) | t 2 R} and `2 = {(7, s + 3, 3s + 4) | s 2 R}.
Questions
(31) The plane passing through the origin and not intersecting either of `1 and `2 has equation
ax + by + cz = d. Write the value of |a + b + c + d| where a, b, c, d are integers with gcd = 1.
(32) Let r be the smallest possible RADIUS of a circle that has a point on `1 as well as a
point on `2 . It is given that r2 (i.e, the SQUARE of the smallest radius) is an integer. Write
the value of r2 .

A10. Consider the part of the graph of y 2 + x3 = 15xy that is strictly to the right of the
Y-axis, i.e., take only the points on the graph with x > 0.
Questions
(33) Write the least possible value of y among considered points. If there is no such real
number, write NONE (without any spaces or quotation marks or any other punctuation).
(34) Write the largest possible value of y among considered points. If there is no such real
number, write NONE (without any spaces or quotation marks or any other punctuation).

3
 Part B Problems for CMI BSc entrance exam on May 7, 2023

B1. [11 points] We want to find odd integers n > 1 for which n is a factor of 2023n 1.

(a) Find the two smallest such integers.

(b) Prove that there are infinitely many such integers.

B2. [12 points] Let Z+ denote the set of positive integers. We want to find all functions
g : Z+ ! Z+ such that the following equation holds for any m, n in Z+ .

g(n + m) = g(n) + nm(n + m) + g(m).

P
Prove that g(n) must be of the form di=0 ci ni and find the precise necessary and sufficient
condition(s) on d and on the coefficients c0 , . . . , cd for g to satisfy the required equation.

B3. [13 points] Suppose that for a given polynomial p(x) = x4 + ax3 + bx2 + cx + d, there
is exactly one real number r such that p(r) = 0.

(a) If a, b, c, d are rational, show that r must be rational.

(b) If a, b, c, d are integers, show that r must be an integer.

Possible hint: Also consider the roots of the derivative p0 (x).

B4. [14 points] There are n students in a class and no two of them have the same height.
The students stand in a line, one behind another, in no particular order of their heights.

(a) How many di↵erent orders are there in which the shortest student is not in the first
position and the tallest student is not in the last position?

(b) The badness of an ordering is the largest number k with the following property. There
is at least one student X such that there are k students taller than X standing ahead
of X. Find a formula for gk (n) = number of orderings of n students with badness k.

Example: The ordering 64 61 67 63 62 66 65 (the numbers denote heights) has badness

3 as the student with height 62 has three taller students (with heights 64, 67 and 63)
standing ahead in the line and nobody has more than 3 taller students standing ahead.

Possible hints for (b): It may be useful to first count orderings of badness 1 and/or to find
fk (n) = the number of orderings of n students with badness less than or equal to k.

4
B5. [15 points] Throughout this question every mentioned function is required to be a
di↵erentiable function from R to R. The symbol denotes composition of functions.

(a) Suppose f f = f . Then for each x, one must have f 0 (x) = or f 0 (f (x)) = .
Complete the sentence and justify.

(b) For a non-constant f satisfying f f = f , it is known and you may assume that the
range of f must have one of the following forms: R, ( 1, b], [a, 1) or [a, b]. Show that
in fact the range must be all of R and deduce that there is a unique such function f .
(Possible hints: For each y in the range of f , what can you say about f (y)? If the
range has a maximum element b what can you say about the derivative of f ?)

(c) Suppose that g g g = g and that g g is a non-constant function. Show that g must
be onto, g must be strictly increasing or strictly decreasing and that there is a unique
such increasing g.

B6. [15 points] Starting with any given positive integer a > 1 the following game is played.
If a is a perfect square, take its square root.
p Otherwise take a + 3. Repeat the procedure
with the new positive integer (i.e., with a or a + 3 depending on the case). The resulting
set of numbers is called the trajectory of a. For example the set {3, 6, 9} is a trajectory: it
is the trajectory of each of its members.
Which numbers have a finite trajectory? Possible hint: Find the set

{n | n is the smallest number in some trajectory S}.

If you wish, you can get partial credit by solving the following simpler questions.
(a) Show that there is no trajectory of cardinality 1 or 2.
(b) Show that {3, 6, 9} is the only trajectory of cardinality 3.
(c) Show that for any integer k 3, there is a trajectory of cardinality k.
(d) Find an infinite trajectory.

5
 2023 CMI BSc entrance exam solutions

Part A

A1. Define the right derivative of a function f at x = a to be the following limit if it exists.

f (a + h) f (a)
lim+ , where h ! 0+ means
h!0 h
h approaches 0 only through positive values.

Statements
(1) If f is di↵erentiable at x = a then f has a right derivative at x = a.
(2) f (x) = |x| has a right derivative at x = 0.
(3) If f has a right derivative at x = a then f is continuous at x = a.
(4) If f is continuous at x = a then f has a right derivative at x = a.

(1) True. Obvious from the definition of the derivative.

(2) True. Right derivative is 1.
(3) False. Consider the floor function at integer values.
(4) False. Take x sin x1 made continuous at 0.

A2. Suppose a rectangle EBF D is given and a rhombus ABCD is inscribed in it so that
the point A is on side ED of the rectangle. The diagonals of ABCD intersect at point G.
See the indicative figure below.

Statements
(5) Triangles CGD and DF B must be similar.
AC EB
(6) It must be true that BD = ED .
(7) Triangle CGD cannot be similar to triangle AEB.
(8) For any given rectangle EBF D, a rhombus ABCD
as described above can be constructed.

(5) True. Both are right angled and \F BD = \BDA = \GDC.

(6) True. G bisects AC and BD. Use the similarity of CGD with DEB ' DF B.
(7) False. They are similar when ABC is equilateral, which is possible.
(8) False. ED AD = AB EB (hypotenuse), so need ED EB.
A3. This question is about complex numbers.

Statements
(9) The complex number (e3 )i lies in the third quadrant.
(10) If |z1 | |z2 | = |z1 + z2 | for some complex numbers z1 and z2 , then z2 must be 0.
(11) For distinct complex numbers z1 and z2 , the equation |(z z1 )2 | = |(z z2 )2 | has at
most 4 solutions.
(12) For each nonzero complex number z, there are more than 100 numbers w such that
w2023 = z.

(9) False. Second quadrant. The argument of e3i is 3 radian, which is just under 172 .
(10) False. Take z2 = rz1 with r real and 1  r < 0.
(11) False. |(z z1 )2 | = |z z1 |2 , so z is equidistant from z1 and z2 . Solutions form a line.
1 i(2⇡k+✓)
z
(12) True. There are 2023 such w. Letting |z|
= ei✓ , w = |z| 2023 e 2023 , k = 0, 1, . . . , 2022.

A4. Statements

(13)
1
lim e x = +1.
x!0

(14)
ln x ln x
lim < lim 1 .
x!1 x100 x!1 x 100

(15) For any positive integer n,

Z n
n
x2023 cos(nx)dx < .
n 2023

(16) There is no polynomial p(x) for which there is a single line that is tangent to the graph
of p(x) at exactly 100 points.

(13) False. Limit from the left is 0.

(14) False. Both limits are 0. The numerator ln x is dominated by any positive power of x.
(15) True. The function x2023 cos(nx) is odd so by symmetry the integral is 0.
(16) False. For p(x) with exactly 100 multiple roots the X-axis is such a line. (This is
essentially the only way: if y = ax + b is such a line for a polynomial q(x), then the X-axis
is such a line for the polynomial q(x) ax b, which must have exactly 100 multiple roots.)
A5. Statements
p p
(17) 4 < 5 + 5 5.
1+log2 61
(18) log2 11 < 2
.
(19) (2023)2023 < (2023!)2 .
(20) 92100 + 93100 < 94100 .

p p
(17) True. 16 < 5 + 5 5 as 11 < 5 5 as 121 < 125. Taking square roots preserves order.
(18) True. 22 log2 11 = 121 < 21+log2 61 = 122. Taking log2 preserves order.
(19) True. Pair numbers on the RHS symmetrically. n(2024 n) > 2023 for n = 1, . . . , 2023.
1 2
(20) True. Divide by 92100 and use binomial theorem for (1 + x)100 with x = 92
and x = 92
.

P P
n
A6. For a sequence ai of real numbers, we say that ai converges if lim ai is finite.
n!1 i=1
In this question all ai > 0.

Statements
P
(21) If ai converges, then ai ! 0 as i ! 1.
P
(22) If ai < 1i for all i, then ai converges.
P P
(23) If ai converges, then ( 1)i ai also converges.
P P
(24) If ai does not converge, then i tan(ai ) cannot converge.

(21) True.
1
(22) False. Take ai = 2i
P
n
(23) True. Given ai are all positive, so ( 1)i ai remains a Cauchy sequence.
i=1
(24) False. Take all ai = ⇡.

Continued !
A7. Statements
(25) To divide an integer b by a nonzero integer d, define a quotient q and a remainder r to
be integers such that b = qd + r and |r| < |d|. Such integers q and r always exist and are
both unique for given b and d.
(26) To divide a polynomial b(x) by a nonzero polynomial d(x), define a quotient q(x) and
a remainder r(x) to be polynomials such that b = qd + r and degree(r) < degree(d). (Here
b(x) and d(x) have real coefficients and the 0 polynomial is taken to have negative degree
by convention.) Such polynomials q(x) and r(x) always exist and are both unique for given
b(x) and d(x).
(27) Suppose that in the preceding question b(x) and d(x) have rational coefficients. Then
q(x) and r(x), if they exist, must also have rational coefficients.
(28) The least positive number in the set {(a ⇥ 20232020 ) + (b ⇥ 20202023 )} as a and b range
over all integers is 3.
(25) False. For r > 0, we can increase quotient by 1 and make remainder negative.
(26) True. q1 d + r1 = q2 d + r2 gives (q1 q2 )d = r2 r1 . Compare degrees.
(27) True. Uniqueness and the long division procedure ensure this.
(28) False. It is the gcd of 20232020 and 20202023 , which is 1.

A8. You play the following game with three fair dice. (When each one is rolled, any one of
the outcomes 1, 2, 3, 4, 5, 6 is equally likely.) In the first round, you roll all three dice. You
remove every die that shows 6. If any dice remain, you roll all the remaining dice again in
the second round. Again you remove all dice showing 6 and continue.
Questions
(29) Let the probability that you are able to play the second round be ab , where a and b are
integers with gcd 1. Write the numbers a and b separated by a comma. E.g., for probability
10
36
you would type 5,18 with no quotations, space, full stop or any other punctuation.
(30) Let the probability that you are able to play the second round but not the third round
be dc where c and d are integers with gcd 1. Write only the integer c as your answer. E.g.,
for probability 34
36
you would type 17 with no quotations, space or any other punctuation.
(29) The probability is 1 ( 16 )3 = 215
216
, so 215,216 is the answer.
1115
(30) The probability is 66
by the calculation below, so 1115 is the answer.
P(3 dice left after first round) ⇥ P(all 3 remaining dice show 6 in round 2) +
P(2 dice left after first round) ⇥ P(both remaining dice show 6 in round 2) +
P(1 die left after first round) ⇥ P(the remaining die shows 6 in round 2)
⇣ 5 ⌘3 ⇣ 1 ⌘3 ⇣ 5 ⌘2 ⇣ 1 ⌘ ⇣ 1 ⌘2 ⇣ 5 ⌘⇣ 1 ⌘2 ⇣ 1 ⌘
= ⇥ + 3 ⇥ + 3 ⇥ .
6 6 6 6 6 6 6 6
A9. Two lines `1 and `2 in 3-dimensional space are given by
`1 = {(t 9, t + 7, 6) | t 2 R} and `2 = {(7, s + 3, 3s + 4) | s 2 R}.

Questions
(31) The plane passing through the origin and not intersecting either of `1 and `2 has equation
ax + by + cz = d. Write the value of |a + b + c + d| where a, b, c, d are integers with gcd = 1.
(32) Let r be the smallest possible RADIUS of a circle that has a point on `1 as well as a
point on `2 . It is given that r2 (i.e, the SQUARE of the smallest radius) is an integer. Write
the value of r2 .

(31) (1, 1, 0) and (0, 1, 3) are the direction vectors of the given lines. (3, 3, 1) is a common
perpendicular to both direction vectors. So 3x + 3y z = 0 is an equation for the desired
plane. Thus the answer is |3 + 3 1 + 0| = 5.

(32) Each of the two mentioned points must be the only intersection of such a circle with
the respective line. The segment joining these points must be perpendicular to both `1 and
`2 and is a diameter of any specified circle. Taking a general point on each line, a vector
representing the segment joining the two points is (16 t, s + t 4, 3s 2). Solving

(16 t, s + t 4, 3s 2) · (1, 1, 0) = 0 and (16 t, s + t 4, 3s 2) · (0, 1, 3) = 0

gives s = 0, t = 10. So (2r)2 = (16 10)2 + (10 4)2 + ( 2)2 = 76. Thus the answer is 19.

A10. Consider the part of the graph of y 2 + x3 = 15xy that is strictly to the right of the
Y-axis, i.e., take only the points on the graph with x > 0.

Questions
(33) Write the least possible value of y among considered points. If there is no such real
number, write NONE (without any spaces or quotation marks or any other punctuation).
(34) Write the largest possible value of y among considered points. If there is no such real
number, write NONE (without any spaces or quotation marks or any other punctuation).
p
(33) Regarding the equation as a quadratic in y gives y = 12 15x ± 225x2 4x3 . So
there is a y-value for every x  225 . We also have y > 0 for x > 0 since in that case
p 4
15x > 225x 2 3
4x . As x ! 0, y also ! 0, so there is no minimum y-value.

(34) On the closed interval [0, 225

4
], each sign in the formula for y gives a continuous function.
So y must have a maximum, which cannot occur at x = 0 as y > 0 for x > 0. So we inspect
y-values for the endpoint x = 225 4
and for any critical points. Di↵erentiating implicitly,
dy 2
dx
= 0 precisely when 3x = 15y for (x, y) on the graph. This gives only the point (50, 500).
As 500 > (y-value at x = 225 4
), the answer is 500. One can also argue without considering
225
x = 4 at all. Plot the graph to see that there is no “endpoint”.
 Part B solutions

B1. [11 points] We want to find odd integers n > 1 for which n is a factor of 2023n 1.
(a) Find the two smallest such integers.
(b) Prove that there are infinitely many such integers.
Solution: (a) 2023 is 1 mod 3, so n = 3 works. Similarly using modular arithmetic one
checks that 5 and 7 do not work but 9 does. (b) If n = k works so does n = 3k by induction:
20233k 1 = (2023k 1)(20232k +2023k +1) = (multiple of k)(multiple of 3) as each summand
in the second factor is 1 mod 3. Thus all powers of 3 satisfy the required condition.

B2. [12 points] Let Z+ denote the set of positive integers. We want to find all functions
g : Z+ ! Z+ such that the following equation holds for any m, n in Z+ .
g(n + m) = g(n) + nm(n + m) + g(m).
P
Prove that g(n) must be of the form di=0 ci ni and find the precise necessary and sufficient
condition(s) on d and on the coefficients c0 , . . . , cd for g to satisfy the required equation.

Solution: Setting m = 1 gives g(n + 1) Pg(n) = n(n + 1) + g(1). Apply this repeatedly
starting with n = 1 to get g(n) = ng(1) + ni=11 i2 + i, which works out to ng(1) + 13 (n3 n).
Set g(1) to be an arbitrary positive integer k and verify that the resulting formula satisfies
the given condition for all m, n. Thus g(n) = 13 n3 + kn with k = (any positive integer) 13 .

B3. [13 points] Suppose that for a given polynomial p(x) = x4 + ax3 + bx2 + cx + d, there
is exactly one real number r such that p(r) = 0.
(a) If a, b, c, d are rational, show that r must be rational.
(b) If a, b, c, d are integers, show that r must be an integer.
Possible hint: Also consider the roots of the derivative p0 (x).

Solution: (a) The multiplicity of the root r must be either 2 or 4. In the latter case
p(x) = (x r)4 = x4 4rx3 + 6r2 x2 4r3 x + r4 . As 4r is a rational, so is r. If the multiplicity
is 2 then the two non-real roots are complex conjugates and r is the only repeated root
of p(x). So r is the only common root (whether real or not) of p(x) and p0 (x) AND r
is a simple root of the polynomial p0 (x). So gcd(p, p0 ) = x r by looking at complete
factorization of p(x) and p0 (x) into linear terms (including complex roots). As p(x) and p0 (x)
have rational coefficients, so does their gcd by looking at each step of the division algorithm.
(The preceding three sentences deserve careful consideration.) Therefore r must be rational.
(b) It is standard that a rational root of a monic polynomial with integer coefficients must
be an integer. (Proof: write a rational root r = mn
with gcd(m, n) = 1, substitute into the
polynomial, clear powers of n in the denominators and deduce that n cannot be divisible by
any prime because that prime would then need to divide m as well. So n = ±1.)
B4. [14 points] There are n students in a class and no two of them have the same height.
The students stand in a line, one behind another, in no particular order of their heights.

(a) How many di↵erent orders are there in which the shortest student is not in the first
position and the tallest student is not in the last position?

(b) The badness of an ordering is the largest number k with the following property. There
is at least one student X such that there are k students taller than X standing ahead
of X. Find a formula for gk (n) = number of orderings of n students with badness k.

Example: The ordering 64 61 67 63 62 66 65 (the numbers denote heights) has badness

3 as the student with height 62 has three taller students (with heights 64, 67 and 63)
standing ahead in the line and nobody has more than 3 taller students standing ahead.

Possible hints for (b): It may be useful to first count orderings of badness 1 and/or to find
fk (n) = the number of orderings of n students with badness less than or equal to k.

Solution: (a) There are (n 1)! + (n 1)! (n 2)! orderings with the shortest student
first or the tallest student last or both. So the desired number = n! 2(n 1)! + (n 2)!,
i.e., (n 2)!(n2 3n + 3). Alternatively, first order all but the shortest and the tallest
students in (n 2)! ways. The number of ways to insert the shortest and then the tallest is
(n 2)(n 1) + 1. (What is the extra 1 for?)
(b) Following both the hints, first consider badness 1 and use induction. Leave out the
shortest student and order the remaining n 1 students with badness 1. The shortest student
can now go in place 1 or 2. There is one more possibility where the n 1 students have 0
badness (i.e., are in increasing order) and the shortest student goes in place 2. Inductively
one gets the formula g1 (n) = 2n 1 1 (valid even for n = 1, giving g1 (1) = 0).
Induction to find fk (n) is easier. Leave out the shortest student and order the remaining
n 1 students with badness at most k. To maintain badness at most k, out of the n
available slots for the shortest student, the allowed ones are precisely 1, 2, . . . , min(k + 1, n).
So fk (n + 1) = fk (n) min(k + 1, n). Answer: fk (n) = n! if n  k + 1 and fk (n) = k!(k + 1)n k
if n k + 1. (The formulas agree for n = k + 1.)
Now gk (n) = fk (n) fk 1 (n). This works out to 0 if n  k (as expected) and for n k, one
gets gk (n) = k! (k + 1)n k k n k .

B5. [15 points] Throughout this question every mentioned function is required to be a
di↵erentiable function from R to R. The symbol denotes composition of functions.
 (a) Suppose f f = f . Then for each x, one must have f 0 (x) = or f 0 (f (x)) = .
Complete the sentence and justify.
(b) For a non-constant f satisfying f f = f , it is known and you may assume that the
range of f must have one of the following forms: R, ( 1, b], [a, 1) or [a, b]. Show that
in fact the range must be all of R and deduce that there is a unique such function f .
(Possible hints: For each y in the range of f , what can you say about f (y)? If the
range has a maximum element b what can you say about the derivative of f ?)
(c) Suppose that g g g = g and that g g is a non-constant function. Show that g must
be onto, g must be strictly increasing or strictly decreasing and that there is a unique
such increasing g.
Solution: (a) f 0 (f (x))f 0 (x) = f 0 (x) for each x by chain rule, so f 0 (x) = 0 or f 0 (f (x)) = 1.
(b) (Argument taken from the answer by Dan Shved to question 365363 on stackexchange.)
For each y = f (x) in the range, f (y) = f (f (x)) = f (x) = y, so f is the identity function
on the range. Therefore it is enough to show that the range is all of R. As the range is
given to be an interval (a proof is given below), at each y in the range, f 0 (y) = 1 by direct
calculation. Note that if the range has endpoint(s), this derivative calculation is one sided
at such a point. We will show that the range does not have an endpoint on either side. If the
range interval has a left/right endpoint f (c) = c, then f has a minimum/maximum value at
c, so it must be true that f 0 (c) = 0. (Recall that the domain is all of R, so Fermat’s theorem
applies at x = c.) This contradicts the earlier calculation of a one-sided derivative at c being
1. So the range cannot be of the form ( 1, b], [a, 1) or [a, b] and must be all of R.
(c) By applying g to the given equation, g g g g = g g. So if g g is non-constant, it
has to be the identity by part (b), i.e., g is its own inverse. In particular, being invertible,
g is onto and one-to-one. Due to continuity, being one-to-one implies that g is monotonic.
(This is standard. If g is one-to-one and not monotonic, we have some a, b, c for which
a < b < c and WLOG g(a) < g(c) < g(b) by replacing g with its reflection in one/both axes
if necessary. By the intermediate value theorem, we have d 2 (a, b) with g(c) = g(d), giving a
contradiction to g being one-to-one.) If g is increasing, g(x) < x implies x = g(g(x)) < g(x)
and vice versa, so the only possibility is g(x) = x.
For completeness, here is a proof of the fact that you were asked to assume in part (b), namely
that the range of f must be of the form R, ( 1, b], [a, 1) or [a, b]: As f is continuous, if
f (p) < r < f (q) then r is also in the range by intermediate value theorem. So the range
must be one of the intervals (m, M ) or (m, M ] or [m, M ) or [m, M ], where m is the greatest
lower bound of the range (possibly m = 1), M is the least upper bound of the range
(possibly M = 1), and it is understood that if m and/or M is not finite then only the open
interval makes sense on the corresponding side. It remains to show that if either of m and
M is finite, it must belong to the range. Now f (m + h) = m + h for small enough h > 0.
(Recall that the function is non-constant so m 6= M .) So lim+ f (m + h) = lim+ m + h = m.
h!0 h!0
At the same time, because f is continuous, the same limit must⇣ be f (m), so f (m) =
⌘ m and
thus m is in the range of f . For M , take limit from the left lim f (M + h), etc.
h!0
B6. [15 points] Starting with any given positive integer a > 1 the following game is played.
If a is a perfect square, take its square root.
p Otherwise take a + 3. Repeat the procedure
with the new positive integer (i.e., with a or a + 3 depending on the case). The resulting
set of numbers is called the trajectory of a. For example the set {3, 6, 9} is a trajectory: it
is the trajectory of each of its members.
Which numbers have a finite trajectory? Possible hint: Find the set

{n | n is the smallest number in some trajectory S}.

If you wish, you can get partial credit by solving the following simpler questions.
(a) Show that there is no trajectory of cardinality 1 or 2.
(b) Show that {3, 6, 9} is the only trajectory of cardinality 3.
(c) Show that for any integer k 3, there is a trajectory of cardinality k.
(d) Find an infinite trajectory.
p
Solution: Let S = a trajectory, n = the smallest number in S. Note that 1 2
/ S, so n > n.
p p
(a) |S|= 1 implies n = n, so n = 1, which is impossible. As n > n, n cannot be a perfect
square. So |S| = 2 implies S = {n, n + 3} and n + 3 = n2 , which cannot happen for n > 1.
(b) Similarly |S| = 3 implies S = {n, n + 3, n + 6} and n + 6 = n2 , which gives n = 3.
(c) To get any finite cardinality repeatedly square 6 (or 9) and add these numbers to {3, 6, 9}.
(d) (3k)2 = 9k 2 is 0 mod 3. Next, (3k + 1)2 = 9k 2 + 6k + 1 and (3k + 2)2 = 9k 2 + 12k + 4 are
1 mod 3. As all squares are 0 or 1 mod 3, any S containing a 2 mod 3 number is infinite.
Claim: For a trajectory S with smallest number n, exactly one of the following two happens.
1. No square occurs after n in the trajectory. Hence n is 2 mod 3 and S is infinite.
2. A square does occur after n and n = 3. Hence S is finite.
Proof of the claim: The smallest number in S cannot be a square, so let k 2 < n < (k + 1)2 .
Assuming a square occurs after n, we will show that n = 3. The first encountered square
after n is at most (k + 3)2 (e.g., make cases depending on what k and n are mod 3.) So
k 2 < n  k + 3, but k 2 < k + 3 only for k = 1, 2. Hence n < (k + 1)2  9. Now n cannot be
2, 5, 8 because adding 3 repeatedly to these will never give a square. And n cannot be 4, 6, 7
because in each case one gets a smaller number by playing the game (respectively 2, 3, 2). So
n = 3 is the only possibility. For the second sentence in case 1, note that repeatedly adding
3 to n will eventually give a square if and only if n is 0 or 1 mod 3.
Main answer: If a number in S is / is not divisible by 3, then the same is true for all numbers
in S (check this). If the initial number a is a multiple of 3, then so is n, and hence we must
be in case 2 of the claim. If a is not a multiple of 3, then nor is n, so n 6= 3 and we must be
in case 1. Thus multiples of 3 are precisely the numbers with finite trajectories.
Notes: (1) The above pattern was discovered earlier by Stephan Wagner. See problem 1 in
IMO 2017 for a slightly di↵erent formulation. (2) The analysis in the solution generalizes
naturally if 3 is replaced in the game by any prime p. (Why prime?) What happens for
p = 2? For p = 5? For p = 7? In general?
 Draft solutions for CMI BSc entrance exam on May 19, 2024

Send any comments to ugadmit2024@cmi.ac.in by May 25, 2024.

In the online exam the instruction page counted as question 1, so the 21 part A questions
were numbered from 2 to 22. For any correspondence refer to the numbering below.

A test developed to detect Covid gives the correct diagnosis for 99% of people with Covid.
1
It also gives the correct diagnosis for 99% of people without Covid. In a city 1000 of the
population has Covid. Answer questions (1) and (2) as per the instruction below.

Questions

(1) What is the probability that a randomly selected person tests positive? (We assume that
in our random selection every person is equally likely to be chosen.) [2 points]

(2) Suppose that a randomly selected person tested positive. What is the probability that
this person has Covid? [2 points]

Instruction for (1) and (2)

If the probability is x%, then your answer should be the integer closest to x. E.g., for
probability 13 = 33.33 . . . %, you should type 33 as your answer. For probability 23 you should
type 67 as your answer.

Solution: Let C = event that the person has Covid, T = event that the test is positive.
Question (1) asks for P (T ) and question (2) asks for P [C|T ].
1 999 1 10.98 549
P [T ] = ⇥ 0.99 + ⇥ 0.01 = (0.99 + 9.99) = = = 1.098% ⇡ 1%.
1000 1000 1000 1000 50000
1
P [C&T ] 1000
⇥ 0.99 0.99 11
P [C|T ] = = 10.98 = = ⇡ 9.02% ⇡ 9%.
P [T ] 1000
10.98 122

Note: Most of the positive tests are false positives coming from Covid-free people, which
“explains” why the answer to (2) is so low. Because only 0.1% of the population has Covid,
positive tests coming from this group are an order of magnitude fewer than false positives
coming at 1% rate from Covid-free people who make up 99.9% of the population. This also
explains the answer to (1): when rounded down, it is the “same” 1% rate.
Consider the polynomial

p(x) = x6 + 10x5 + 11x4 + 12x3 + 13x2 12x 11.

Questions

(3) Find the remainder when p(x) is divided by x + 1. [1 point]

P6 2
(4) Let z1 , z2 , z3 , z4 , z5 , z6 be the six complex roots of p(x). Evaluate i=1 zi . [2 points]

(5) Find an integer n with the least possible absolute value such that p(x) has a real root
between n and n + 1. Write this number along with your reason as per the instruction below.
[2 points]

Instruction for (5)

Write two numbers separated by a comma: value of n, number of the theorem below that
justifies this answer. E.g., if you think that n = 5 because of the factor theorem, then type
5,1 as your answer with no space, full stop or any other punctuation.

1. Factor theorem

2. Mean value theorem

3. Intermediate value theorem

4. Fundamental theorem of algebra

5. Fundamental theorem of calculus

Solution: (3) p( 1) = 4
P P P
(4) 6i=1 zi2 = ( i zi )2 2 i<j zi zj = ( 10)2 2(11) = 78.

(5) p(0) = 11 and p(1) > 0. By intermediate value theorem p has a root between 0 and 1.
Two mighty frogs jump once per unit time on the number line as described below.

Questions

(6) The first frog is at x = 2i at time t = i. How many numbers of the form 7n + 1 (with n
an integer) does the frog visit from t = 0 to t = 99 (both endpoints included)? [3 points]

(7) The second frog starts at x = 0 and jumps i + 1 steps to the right just after t = i, so that
at times t = 0, 1, 2, 3, . . . this frog is at positions x = 0, 1, 3, 6, . . . respectively. How many
numbers of the form 7n + 1 (with n an integer) does the frog visit from t = 0 to t = 99 (both
endpoints included)? [3 points]

Solution: (6) For t = 0, 1, 2, 3, 4, 5, . . . the positions mod 7 are 1,2,4,1,2,4,. . . by repeatedly

multiplying by 2 modulo 7. So positions 1 mod 7 are occupied at t = every multiple of 3,
starting at 0. There are 34 multiples of 3 from 0 to 99 including both endpoints.

(7) Here we repeatedly add the sequence 1, 2, 3, 4, 5, 6, 7 = 0 mod 7 to previous position. For
t = 0, 1, 2, 3, 4, 5, 6, . . . we get the repeating pattern 0, 1, 3, 6, 3, 1, 0 of positions mod 7. Thus
a 1 mod 7 position occurs precisely for t = 1 mod 7 and t = 5 mod 7. So till 98 = 14 ⇥ 7
we get 14 ⇥ 2 = 28 such occurrences. We also get an occurrence at 99, which is 1 mod 7. So
the answer is 29.
Let O = (0, 0, 0), P = (19, 5, 2024) and Q = (x, y, z) be points in 3-dimensional space where
Q is an unknown point.
! !
Consider vector u = OP = 19 î + 5 ĵ + 2024 k̂ and unknown vector v = OQ = x î + y ĵ + z k̂.
Instruction: for each of the sets below choose the correct option describing it and enter the
number of that option. E.g., if you think a given set is a line, enter 3 as your answer with
no full stop or any other punctuation.
Questions
(8) {Q | u · v = 2024}. [1 point]
p
(9) {Q | u · v = 2024 v · v }. [2 points]
(10) {Q | u · v = 2024 (v · v)}. [2 points]

Options
1. The empty set
2. A singleton set
3. A line
4. A pair of lines
5. A circle
6. A plane perpendicular to u
7. A plane parallel to u
8. An infinite cone
9. A finite cone
10. A sphere
11. None of the above

Solution: (8) A plane perpendicular to u.

p p
(9) u · v = u · u v · v cos(✓) where ✓ is the angle between u and v (between 0 and 180 ).
So the given condition for nonzero v is cos(✓) = p2024
u·u
, which is a constant between 1 and 0
p
as u · u > 2024. So the given set consists of endpoints of all vectors making a fixed obtuse
angle with u, along with the origin. This is an infinite cone.
(10) This is the set of points satisfying 19x + 5y + 24z = 2024(x2 + y 2 + z 2 ). Taking LHS
to RHS and completing three squares, we get the equation of a sphere passing through the
origin.
An integer d is called a factor of an integer n if there is an integer q such that n = dq.
In particular the set of factors of n contains n and contains 1. You are given that 2024 =
8 ⇥ 11 ⇥ 23.

Questions

(11) Write the number of even positive integers that are factors of 20242 . [2 points]

(12) Write the number of ordered pairs (a, b) of positive integers such that a2 b2 = 20242 .
If there are infinitely many such pairs, write the word infinite as your answer. [3 points]

Solution: (11) Answer: 54. As 20242 = 26 ⇥112 ⇥232 , the number of factors is 7⇥3⇥3 = 63.
Of these 6 ⇥ 3 ⇥ 3 = 54 are even as the power of 2 in an even factor cannot be 0.

(12) Answer: 22. Observe that 20242 = (a + b)(a b) is a factorization with unequal factors
of the same parity which must be even. Conversely, given a factorization dq of 20242 into
even unequal factors, we get a unique solution for a, b by solving a + b = the larger factor
and a b = the smaller factor. The number of factors d such that both d and 2024 d
are even
is 5 ⇥ 3 ⇥ 3 = 45 as the power of 2 in d cannot be 0 or 6. Of these the equal factors case
d = q = 2024 should be discarded (as that gives b = 0). The other 44 values of d break into
22 pairs, giving 22 solutions.
A good path is a sequence of points in the XY plane such that in each step exactly one of
the coordinates increases by 1 and the other stays the same. E.g.,

(0, 0), (1, 0), (2, 0), (2, 1), (3, 1), (3, 2), (3, 3)

is good path from the origin to (3,3). It is a fact that there are exactly 924 good paths from
the origin to (6,6).

Questions

(13) Find the number of good paths from (0, 0) to (6, 6) that pass through both the points
(1, 4) and (2, 3). [1 point]

(14) Find the number of good paths from (0, 0) to (6, 6) that pass through both the points
(1, 2) and (3, 4). [2 points]

(15) Find the number of good paths from (0, 0) to (6, 6) such that neither of the two points
(1, 2) and (3, 4) occurs on the path, i.e., the path must miss both of the points (1, 2) and
(3, 4). [3 points]

Solution: (13) 0. Going to one of (1, 4) and (2, 3) precludes going to the other as for that
to happen one coordinate would have to decrease and that is not possible in a good path.
2+1 3+4 (1+2) 6+6 (3+4) 3 4 5
(14) 1
⇥ 3 1
⇥ 6 3
= 1
⇥ 2
⇥ 3
= 3 ⇥ 6 ⇥ 10 = 180.

(15) We use basic set theory.

3 9
Number of paths through (1, 2) is 1
⇥ 5
= 3 ⇥ 126 = 378.
7 5
Number of paths through (3, 4) is 3
⇥ 3
= 35 ⇥ 10 = 350.
Number of paths through (1, 2) or (3, 4) is 350 + 378 180 = 548.
Number of paths missing both (1, 2) and (3, 4) is 924 548 = 376.
Suppose f is a function whose domain is X and codomain is Y . It is given that |X| > 1 and
|Y | > 1. No other information is known about X, Y and f . Instruction: for each question
below write the number of a single correct option for the given statement S.

Questions [1 point each]

(16) S = “For each x in X, there exists y in Y such that f (x) = y.”

(17) S = “For each y in Y , there exists x in X such that f (x) = y.”

(18) S = “There exists a unique x in X such that for each y in Y it is true that f (x) = y.”

(19) S = “There exists a unique y in Y such that for each x in X it is true that f (x) = y.”

Options (with each question’s number written next to its matching option)

1. S is always true. (16)

2. S is always false. (18)

3. S is true if and only if f is one-to-one.

4. If S is true then f is one-to-one but the converse is false.

5. If f is one-to-one then S is true but the converse is false.

6. S is true if and only if f is onto. (17)

7. If S is true then f is onto but the converse is false.

8. If f is onto then S is true but the converse is false.

9. S is true if and only if f is a constant function. (19)

10. If S is true then f is a constant function but the converse is false.

11. If f is a constant function then S is true but the converse is false.

12. None of the above.

Solution: This is a matter of careful reading and interpretation of precise language.

Suppose a di↵erentiable function f from R to R has a local maximum at (a, f (a)) (This
means there are numbers m and M such that (i) m < a < M and (ii) f (a) f (x) for any
x 2 [m, M ].) The proof of a standard result is sketched below. Complete it as instructed.
Proof: For sufficiently 1 h > 0, it is given that f (a + h) 2 3 .
Therefore for such h the quantity 4 must be 5 6 .
By taking the limit of this quantity as h ! 0 from the right, we get that 7 must be
8 9 .
A parallel argument for suitable negative values of h gives that 10 must be 11 12 .
Combining both conclusions gives the desired result: 13 14 15 . Note that the
mentioned limits exist because 16 .
Questions
(20) Write a sequence of 9 letters indicating the correct options to fill in the numbered blanks
1 to 9. Do not use any spaces, full stop or other punctuation. E.g., ABACDIJKB is in
the correct format. [3 points]
(21) Write a sequence of 7 letters indicating the correct options to fill in the numbered blanks
10 to 16. [2 points]
Options
A. small B. large

C. D. >

E.  F. <

G. = H. 6=

I. 0 J. f (a)

f (a+h) f (a)
K. h
L. f 0 (a)

M. f is di↵erentiable N. f is continuous

Solution: For sufficiently small h > 0, it is given that f (a + h)  f (a).

f (a+h) f (a)
Therefore for such h the quantity h
must be  0.
By taking the limit of this quantity as h ! 0 from the right, we get that f 0 (a) must be  0.
A parallel argument for suitable negative values of h gives that f 0 (a) must be 0.
Combining both conclusions gives the desired result: f 0 (a) = 0.
Note that the mentioned limits exist because f is di↵erentiable.
 Solutions to Part B Problems

B1. [10 points] (a) Draw a qualitatively accurate sketch of the unique bounded region
R in the first quadrant that has maximum possible finite area with boundary described as
follows. R is bounded below by the graph of y = x2 x3 , bounded above by the graph of
an equation of the form y = kx (where k is some constant), and R is entirely enclosed by
the two given graphs, i.e., the boundary of the region R must be a subset of the union of the
given two graphs (so R does not have any points on its boundary that are not on these two
graphs). Clearly mark the relevant point(s) on the boundary where the two given graphs
meet and write the coordinates of every such point.

(b) Consider the solid obtained by rotating the above region R around Y-axis. Show how to
find the volume of this solid by doing the following: Carefully set up the calculation with
justification. Do enough work with the resulting expression to reach a stage where the final
numerical answer can be found mechanically by using standard symbolic formulas of algebra
and/or calculus and substituting known values in them. Do not carry out the mechanical
work to get the final numerical answer.

Solution to B1: (a) Increasing k from 0 rotates the line counterclockwise and increases the
desired area until the line stops intersecting
the graph of y = x2 x3 in the first quadrant.
So bounded R with maximum possible area
is obtained when the line is tangent to the
graph of y = x2 x3 in the first quadrant.
See the picture. At the point of tangency the
slope is k = 2x 3x2 and the y coordinate
is kx = x2 x3 . Solving gives the point of
tangency = (0.5, 0.125) and slope k = 0.25.

(b) This can be done by either the “washer”

method or the “shell” method. The washer
method divides R into horizontal slices of
“tiny height dy” and integrates along Y-axis
the volumes of the resulting thin washers of
revolution around Y-axis. This gives
Z 0.125
desired volume = ⇡x2 dy (volume of a cone of radius 0.5 and height 0.125),
0

2
where x and y in the integral are related R 0.5 2 by y = 2x x3 . So dy = (2x 3x2 )dx. After
substitution the integral becomes ⇡ 0 x (2x 3x )dx. Cone volume = ⇡3 (0.5)2 (0.125) (or
R 0.125 2
R 0.5 ⇡
0
⇡(4y) dy or 0
(0.25)⇡x2 dx). Overall answer (not asked for) is 480 .
The shell method divides R into vertical slices of “tiny width dx” and integrates along X-axis
the volumes of resulting shells of revolution around Y-axis. This gives the following integral
and the same numerical answer.
Z 0.5 Z 0.5
desired volume = 2⇡x(y1 y2 )dx = 2⇡ x(0.25x (x2 x3 ))dx.
0 0

B2. [15 points] (a) Find the domain of the function g(x) defined by the following formula.
Z x
g(x) = log10 log10 (t2 1000t + 101000 ) dt.
10

Calculate the quantities below. You may give an approximate answer where necessary, but
clearly state which answers are exact and which are approximations.
(b) g(1000).
(c) x in [10,1000] where g(x) has the maximum possible slope.
(d) x in [10,1000] where g(x) has the least possible slope.
ln(x)
(e) limx!1 g(x)
if it exists.

Solution to B2: (a) The parabola t2 1000t takes lowest value 250000 (at t = 500), which
is absolutely dwarfed by 101000 = (1 followed by 1000 zeros). So log10 (t2 1000t + 101000 )
is defined for any t and is always greater than 999. So log10 log10 (t2 1000t + 101000 ) is
defined for any t. The integrand is a continuous function, so g(x) is defined for all real x.

(b) The log function increases very slowly, so log log increases extremely slowly. In the
interval [10,1000], the values of t2 1000t are in [ 250000, 0] and have no practical e↵ect
on the order of magnitude of the far larger 101000 . So throughout the interval of integration
[10,1000], log10 (t2 1000t + 101000 ) is extremely close to 1000 and log10 of that is extremely
close to 3. So we are essentially integrating the constant function 3. The answer is ⇡
3 ⇥ (1000 10) = 2970. This answer is correct to a very high degree of accuracy and a
rigorous calculation bounding the error can be given, but for this exam the above simple
qualitative answer was enough. (Idea taken from “Lucky numbers” in Surely You’re Joking,
Mr. Feynman! “People started giving me problems they thought were difficult such as
integrating a function ... which hardly changed over the range they gave me”. Also worth
reading are Feynman’s thoughts on education near the end of “O Americana, Outra Vez!”,
starting with “In regard to education in Brazil, I had a very interesting experience ...”)

(c,d) Slope of g is g 0 (x), which is log10 log10 (x2 1000x + 101000 ) by the fundamental
theorem of calculus. As log10 is an increasing function, we may simply discard the log10 log10
in the front and find extrema of x2 1000x+101000 over the interval [10,1000]. This parabola
takes minimum value at its vertex which is at x = 500. By symmetry of the parabola, the
maximum value occurs at the endpoint farther from the vertex of the parabola, namely at
x = 1000. (Note that while these are theoretically exact answers, g 0 (x) sees extremely tiny
variation thanks to flattening due to log log.)

(e) Both ln(x) and g(x) go to 1 as x ! 1 but the denominator dominates. (Why?) Proof:
By L'Hôpital’s rule the required limit can be checked via limx!1 g1/x
0 (x) , which is 0 as 1/x ! 0
x ln(x)
and g 0 (x) = log10 log10 (x2 1000x + 101000 ) ! 1. (Note: calculating limx!1 g(x)
is a
bit more interesting. What is the answer?)

B3. [15 points] (a) For non-negative numbers a, b, c and any positive real number r prove
the following inequality and state precisely when equality is achieved.

ar (a b)(a c) + br (b a)(b c) + cr (c a)(c b) 0

Hint: Assuming a b c do algebra with just the first two terms. What about the third
term? What if the assumption is not true?
(b) As a special case obtain an inequality with a4 + b4 + c4 + abc(a + b + c) on one side.
(c) Show that if abc = 1 for positive numbers a, b, c, then
a2 + b 2 b 2 + c 2 c 2 + a2
a4 + b 4 + c 4 + a3 + b 3 + c 3 + a + b + c + + + 3.
c a b

Solution to B3: (a) This is known as Schur’s inequality. Here is the standard argument.

We may assume a b c because the inequality is completely symmetric in a, b, c, i.e.,

any interchange of two letters gives the same inequality (check this), and therefore so does
any permutation of a, b, c. (Alternatively, if a b c is not true, then arrange a, b, c in
a decreasing order and use similar reasoning as below by separating the term with three
occurrences a least one among of a, b, c. Check this in two cases: b a c and b c a.)

So let a b c. First note ar (a b)(a c) + br (b a)(b c) = (a b) ar (a c) br (b c) .

Now ar br 0 and a c b c 0. Multiplying the left sides and the middle terms we
get ar (a c) br (b c), so ar (a c) br (b c) 0. Now multiplying by (a b) gives

(a b) ar (a c) br (b c) = ar (a b)(a c) + br (b a)(b c) 0.

Adding cr (c a)(c b) 0 (which is true when a b c), we get the desired inequality.

To have equality, trace all used inequalities to deduce the necessary and sufficient condition
⇣ ⌘
(a = b) OR (ar = br ) and (a c = b c) AND cr (c a)(c b) = 0.
This happens exactly when (a = b) AND (c = 0 or c = a or c = b). Using symmetry we get
equality precisely when a = b = c or when one of a, b, c is zero and the other two are equal.
(b) Substitute r = 2 and do algebra to get

a4 + b4 + c4 + abc(a + b + c) ab(a2 + b2 ) + bc(b2 + c2 ) + ca(c2 + a2 ).

(c) (Problem by Dan Sitaru, this solution by Imad Zak, source www.cut-the-knot.org) Use
abc = 1 in part (b). Cancel abc from the left and divide by abc on the right to get

a2 + b 2 b 2 + c 2 c 2 + a2
a4 + b 4 + c 4 + a + b + c + + .
c a b
We will be done by showing a3 + b3 + c3 3 = 3abc, which is true by AM GM .

B4. [10 points] Find all solutions of the following equation where it is required that
x, k, y, n are positive integers with the exponents k and n both > 1.

20xk + 24y n = 2024.

Solution to B4: First solve 20a + 24b = 2024 in integers. For any solution (a, b), the pair
(a 6k, b + 5k) is also a solution for any integer k.

Moreover any solution (a0 , b0 ) is obtainable this way from a single known solution (a, b).
Proof: we have 20(a a0 ) + 24(b b0 ) = 0, so 5(a a0 ) + 6(b b0 ) = 0. As gcd(5, 6) = 1, we
see that 6 must be a factor of a a0 (and 5 must be a factor of b b0 ). Letting a a0 = 6k,
the equation 5(a a0 ) + 6(b b0 ) = 0 implies b b0 = 5k, so (a0 , b0 ) = (a 6k, b + 5k) as
claimed.

Observe that a = 100, b = 1 is a solution. As subtracting 5 from b = 1 gives a negative

number, we can take only positive values of k to generate the remaining positive integer
solutions. One gets 17 solutions in positive integers, the last one being a = 100 16 ⇥ 6 = 4
paired with b = 1 + 16 ⇥ 5 = 81. Listing all 17 solutions shows that (100,1) and (4,81) are the
only ones in which a, b are both perfect powers: 100 = 102 , 1 = 1any n , 4 = 22 , 81 = 34 = 92 .

B5. [15 points] (a) Find all complex solutions of z 6 = z + z.

(b) For an integer n > 1, how many complex solutions does z n = z + z have?

Solution to B5: (b) z = 0 is a solution. From now on let z 6= 0. Suppose z = re✓i . So

z n = rn en✓i = z + z = 2r cos ✓, which is real. So the equation is true if and only if
⇣ ⌘ ⇣ ⌘
en✓i = 1 and rn = 2r cos ✓ OR en✓i = 1 and rn = 2r cos ✓ .
Case 1. In the first case e✓i must be an nth root of 1 and its real part cos ✓ must be positive
n 1
because cos ✓ = r 2 is positive. Conversely for any one of such values of ✓, the preceding
equation uniquely determines r, thus giving a solution of the given equation. The number
of such ✓ can be calculated by cases mod 4.
• For n = 4k we get 2k 1 valid values of ✓ as two of the nth roots of 1 are imaginary
and half of the remaining 4k 2 have positive real part.
• For n = 4k + 1, 4k + 2, 4k + 3, we get 2k + 1 valid values of ✓, namely ✓ = 0 and k
values each in the first and fourth quadrant because 2k⇡
n
< ⇡2 < 2(k+1)⇡
n
.
Case 2. Otherwise e✓i must be an nth root of 1 and its real part cos ✓ must be negative
n 1
because cos ✓ = r 2 is negative. Conversely for any one of such values of ✓, the preceding
equation uniquely determines r, giving a solution of the given equation. The number of
valid ✓ can again be calculated by cases mod 4. Let S = (set of angles of nth roots of 1) =
{ (2j+1)⇡
n
| j = 1, . . . , n}, i.e., the set of angles obtained by rotating each nth root of 1 halfway
to the next root. We need to find angles ✓ in S with ⇡2 < ✓ < 3⇡ 2
.
• For n = 4k we get 2k valid values of ✓ as exactly half the angles in S are on each side
of the Y axis.
• For n = 4k + 2 we get 2k valid values of ✓ as two of the angles in S are on the Y axis.
For odd n, note that if z is a root, then so is z. So for odd n, (number of roots in case 2)
= (number of roots in case 1), giving 2k + 1 case 2 roots for n = 4k + 1, 4k + 3. We can also
get this by continuing brute force analysis.
(2j+1)⇡
• For n = 4k + 1 we get 2k + 1 valid values of ✓, namely ✓ = 4k+1
for j = k, . . . , 3k.
(2j+1)⇡
• For n = 4k+3, we get 2k+1 valid values of ✓, namely ✓ = 4k+3
for j = k+1, . . . , 3k+1.
Adding up the number of solutions of two types and counting z = 0, we get the following
result. The total number of solutions of the equation z n = z + z (for integer n > 1) is n + 2
when n = 1 mod 4 and n otherwise. One can check this for several n on a numerical solver,
e.g., at https://www.wolframalpha.com.

Aside: if z is a root, then so is z. (Why?) This symmetry does not seem helpful to reduce
work, but it does explain some of the numerology, e.g., case 1 always gives an odd number
of roots (because ✓ = 0 always gives a root) while case 2 gives odd or even number of roots
depending on whether ✓ = ⇡ gives a root or not. But ✓ = ⇡ means z is a negative real
number and such a root exists exactly when n is odd.

(a) Applying the general analysis above gives five nonzero solutions: three with positive real
part (case 1 above), and two with negative real part (case 2 above). Working out the details
gives the following list of six solutions.
p ✓ p ◆
p5 ± ⇡i 1 ± 3i p ± 5⇡i
10
p
10 3±i
0 2 e 3 = 3e 6 = 3
2 2
B6. [15 points] A list of k elements, possibly with repeats, is given. The goal is to find
if there is a majority element. This is defined to be an element x such that the number of
times x occurs in the list is strictly greater than k2 . (Note that there need not be such an
element, but if it is there, it must be unique.) A celebrated efficient way to do this task uses
two functions f and m with domain {1, 2, . . . , k}. The functions are defined inductively as
follows.

Define f (1) = first element of the list, m(1) = 1.

Assuming f and m are defined for all inputs from 1 to i, define
(
f (i) if m(i) > 0
f (i + 1) = th
(i + 1) element of the list if m(i) = 0
(
m(i) 1 if m(i) > 0 and (i + 1)th element of the list is other than f (i)
m(i + 1) =
m(i) + 1 otherwise

(a) For the example of length 15 given below, write a sequence of 15 letters showing the
values of f (i) and a sequence of 15 numbers directly underneath showing the values of m(i)
for i = 1, 2, . . . , 15.
aababccbbbabbcb

(b) Prove that in general the list can be divided into two disjoint parts A and B such that
• Part A contains m(k) elements of the list each of which is f (k).

• Part B contains the remaining k m(k) elements of the list and B can be written as
disjoint union of pairs such that the two elements in each pair are distinct.

(c) If there is a majority element, show that it must be f (k). You may assume part (b) even
if you did not do it.

(d) Assuming f (k) is the majority element, answer the following two questions. Show by
examples that the number of occurrences of f (k) in the list does not determine the value
of m(k). Can the value of m(k) be anything in {0, . . . , k}? Find constraints if any on the
possible values of m(k).
k
(e) Now assume instead that an element occurs exactly 2
times in the list. Is it necessary
that f (k) is such an element?

Solution to B6: This is the well known Boyer-Moore majority vote algorithm. Note some
features of this clever procedure. To calculate f (i + 1) and m(i + 1) one needs only f (i) and
m(i). No need to “remember” previous values! The inductive nature means a single pass of
the list gives a unique candidate f (k) for the majority element. (To verify that f (k) indeed
has majority, one pass won’t do. One can count occurrences of f (k) in a second pass.)
(a) One follows the procedure to get the following table of values of f and m for i = 1, . . . , 15.

i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
list a a b a b c c b b b a b b c b
f (i) a a a a a a c c b b b b b b b
m(i) 1 2 1 2 1 0 1 0 1 2 1 2 3 2 3

(b) Use induction. Initially part A contains f (1) and part B is empty, so the claim is true.
Assuming it is true at stage i we make three cases at stage i + 1.
• If m(i) = 0, by induction part A is empty. We just put the (i + 1)th element in part A
and leave part B untouched.
• If m(i) > 0 and the incoming element is the same as f (i), again put the (i+1)th element
in A and leave B untouched. This case can also be combined with the previous one.
• If m(i) > 0 and the incoming element is di↵erent from f (i), we remove one copy of
f (i) from part A (which is there by induction as m(i) > 0), pair it with the incoming
element and place this pair of distinct elements in part B.
In all cases the claim stays true by using induction and the definition of f (i+1) and m(i+1).
(c) By the claim proved in (b), the number of occurrences of any element in part B can be
at most half of the size of B. So the majority element has to occur in part A (otherwise it
cannot occur more that k2 times). But all elements in part A are f (k), so f (k) must be the
majority element.
(e) No. The list cabc of length 4 has c occurring twice but f (4) = b.
(d) The lists abccc and aaccc both have the majority element c occurring thrice but di↵erent
m(5), namely 3 and 1 respectively.
Some constraints on possible values of m(k) are as follows. First, the answer to (b) shows
that if there is a majority element, it has to occur in part A, so m(k) = size of part A cannot
be 0. Second, by part (b) we also have that m(k)  number of occurrences of f (k). Finally
as m(1) = 1 and since the value of m changes by 1 at each step, m(k) must have the same
parity as k. The second and third properties are true for all lists regardless of whether there
is a majority element.
Conversely, given the list length k, the number of occurrences n of the majority element (say
x) and any nonzero number p  n with p of the same parity as k, one can easily construct
a list with m(k) = p as follows. Put p occurrences of x at the end of the list and ensure
that every element in an even numbered slot among the first k p elements di↵ers from the
preceding element. This means that values of m(i) follow the pattern 1, 0, 1, 0, . . . , 1, 0 until
index i = k p, which is even. (This simpleminded scheme can realize a desired value p of
m(k) of the correct parity regardless of existence of the majority element if one is given an
element x with pre-specified frequency p. If there are more constraints, e.g., if frequencies
of all list elements are pre-specified, then possible values of m can have more restrictions.)