2021 CMI BSc entrance examination

Part A Answers with explanation

1. Consider the two equations numbered [1] and [2]:

log2021 a = 2022 − a [1]

2021b = 2022 − b [2]

(a) Equation [1] has a unique solution.

(b) Equation [2] has a unique solution.
(c) There exists a solution a for [1] and a solution b for [2] such that a = b.
(d) There exists a solution a for [1] and a solution b for [2] such that a+b is an integer.

Correct options: a,b,d Wrong: c

a = 2021, b = 1 is the unique solution. (This can also be solved qualitatively, e.g., the
graph of y = 2022 − x is decreasing, with range the set of real numbers. It is easy
to see that it must intersect each of the graphs y = 2021x and y = log2021 x (both
of which are increasing) exactly once. Substituting c = 2022 − a, the first equation
is equivalent to 2021c = 2022 − c, which is the same as the second equation, so the
(unique) solutions c and b are equal, i.e., b = c = 2022 − a, so a + b = 2022. If a = b,
both would need to be 1011, which is manifestly not a solution to either equation.)

2. A prime p is an integer ≥ 2 whose only positive integer factors are 1 and p.

(a) For any prime p the number p2 − p is always divisible by 3.

(b) For any prime p > 3 exactly one of the numbers p − 1 and p + 1 is divisible by 6.
(c) For any prime p > 3 the number p2 − 1 is divisible by 24.
(d) For any prime p > 3 one of the three numbers p + 1, p + 3 and p + 5 is divisible
by 8.

Correct options: b,c Wrong: a,d

(a) is false for any number that is 2 modulo 3, in particular for 2. To see that (b) and
(c) are true, note that any prime p greater than 3 is not divisible by 3, so p is either
1 mod 3 (which makes p − 1 divisible by 3) or 2 mod 3 (which makes p + 1 divisible
by 3). Note also that both p − 1 and p + 1 are even, which gives (b). In fact they
are consecutive even numbers, so one of them is a multiple of 4, making their product
p2 −1 a multiple of 8, giving (c). Finally, note that any prime p that is 1 mod 8 violates
(d), e.g., p = 17.

1
3. We want to construct a triangle ABC such that angle A is 20.21◦ , side AB has length
1 and side BC has length x where x is a positive real number. Let N (x) = the number
of pairwise noncongruent triangles with the required properties.

(a) There exists a value of x such that N (x) = 0.

(b) There exists a value of x such that N (x) = 1.
(c) There exists a value of x such that N (x) = 2.
(d) There exists a value of x such that N (x) = 3.

Correct options: a,b,c Wrong: d

Draw a ray with endpoint A. The point C will be chosen on this ray later on. Draw a
segment AB of length 1 making an angle 20.21◦ with this ray. Now, to fulfil the required
properties, a necessary and sufficient condition for C is that it is on the original ray as
well as on the circle with center B and radius x. As x increases from 0, the number
of intersections of the expanding circle with the ray goes from 0 to 1 (when ABC is a
right angled triangle, i.e., when x = sin 20.21◦ ) to 2 and finally back to 1.

4. Consider polynomials of the form f (x) = x3 + ax2 + bx + c where a, b, c are integers.

Name the three (possibly non-real) roots of f (x) to be p, q, r.

(a) If f (1) = 2021, then f (x) = (x − 1)(x2 + sx + t) + 2021 where s, t must be integers.
(b) There is such a polynomial f (x) with c = 2021 and p = 2.
(c) There is such a polynomial f (x) with r = 21 .
(d) The value of p2 + q 2 + r2 does not depend on the value of c.

Correct options: a,d Wrong: b,c

(a) is true by the remainder theorem. Long division automatically gives integers s, t.
Uniqueness of quotient and remainder for polynomial long division means those are the
only values of s, t that work. (b) is false by substituting x = 2 into f (x) and noting that
c = 2021 forces f (2) to be odd, in particular nonzero. To see that (c) is false, substitute
x = 21 into f (x), multiply by 8 to clear denominators and see that the leading term
makes the integer 8f ( 12 ) odd. So f ( 21 ) is nonzero. (General version of (b) and (c) that
one gets by the similar reasoning: suppose a polynomial p(x) with integer coefficients
has a rational root rs written in lowest form. Then the leading coefficient of p(x) is
divisible by s and the constant term of p(x) is divisible by r. Often used special case: for
a polynomial p(x) = xn + lower terms with integer coefficients, any rational root must
be an integer.) For (d) note that p2 +q 2 +r2 = (p+q +r)2 −2(pq +pr +qr) = (−a)2 −2b
does not depend on c.

2
5. For any complex number z define P (z) = the cardinality of {z k |k is a positive integer},
i.e., the number of distinct positive integer powers of z. It may be useful to remember
that π is an irrational number.

(a) For each positive integer n there is a complex number z such that P (z) = n.
(b) There is a unique complex number z such that P (z) = 3.
(c) If |z| =
6 1, then P (z) is infinite.
(d) P (ei ) is infinite.

Correct options: a,c,d Wrong: b

2πi
(a) is true by z = e n or any primitive nth root of unity. (b) is false: there are 2
2πi
primitive third roots of 1, namely ω = e 3 and ω 2 . (c) is true because then each |z i |
is a distinct positive real number. (d) is true because P (z) is finite only if powers of
z cycle back to 1, which happens for z = reiθ only if (r = 1 and) the argument θ is a
rational multiple of π. But for z = ei , we have θ = 1.

6. A stationary point of a function f is a real number r such that f 0 (r) = 0. A polynomial

need not have a stationary point (e.g. x3 + x has none). Consider a polynomial p(x).

(a) If p(x) is of degree 2022, then p(x) must have at least one stationary point.
(b) If the number of distinct real roots of p(x) is 2021, then p(x) must have at least
2020 stationary points.
(c) If the number of distinct real roots of p(x) is 2021, then p(x) can have at most
2020 stationary points.
(d) If r is a stationary point of p(x) AND p00 (r) = 0, then the point (r, p(r)) is neither
a local maximum nor a local minimum point on the graph of p(x).

Correct options: a,b Wrong: c,d

(a) p0 (x) is a polynomial of degree 2021, which is odd, so it has a root by intermediate
value theorem by looking at behaviour as x → ±∞. (b) The graph of p(x) has to turn
between any two consecutive zeros, giving a stationary point, in fact a local max/min
(c) The graph of p(x) can turn more than once between zeros, or turn outside extreme
zeros or have stationary points that are not maxima or minima. (d) is false, e.g.,
p(x) = x4 .

3
7. Given three distinct positive constants a, b, c we want to solve the simultaneous equa-
tions
√
ax + by = 2
√
bx + cy = 3

(a) There exists a combination of values for a, b, c such that the above system has
infinitely many solutions (x, y).
(b) There exists a combination of values for a, b, c such that the above system has
exactly one solution (x, y).
(c) Suppose that for a combination of values for a, b, c, the above system has NO
solution. Then 2b < a + c.
(d) Suppose 2b < a + c. Then the above system has NO solution.

Correct options: a,b,c Wrong: d

Each of the given equations defines a line in the XY plane. (a) One can arrange
both lines to√ be identical by having each equation a scalar multiple of the other, e.g.,
a = 1, b = √32 , c = 32 . (b) There is a unique solution when the two lines are distinct
and not parallel. (c) The two lines are given to be parallel. So slopes are equal, i.e.,
b2 = ac. Thus b is the geometric mean of a and c, so b < the arithmetic mean a+c 2
.
2
(Recall that a, b, c are distinct and positive.) (d) is absurd. Just ensure b 6= ac.

8. Given two distinct nonzero vectors v1 and v2 in 3 dimensions, define a sequence of

vectors by

vn+2 = vn × vn+1 (so v3 = v1 × v2 , v4 = v2 × v3 and so on).

Let S = {vn |n = 1, 2, . . .} and U = { |vvnn | |n = 1, 2, . . .}. (Note: Here × denotes the

cross product of vectors and |v| denotes the magnitude of the vector v. The vector 0
with 0 magnitude, if it occurs in S, is counted. But in that case of course the 0 vector
is not considered while listing elements of U .)

(a) There exist vectors v1 and v2 for which the cardinality of S is 2.

(b) There exist vectors v1 and v2 for which the cardinality of S is 3.
(c) There exist vectors v1 and v2 for which the cardinality of S is 4.
(d) Suppose that for some v1 and v2 , the set S is infinite. Then the set U is also
infinite.

Correct options: b,c Wrong: a,d

4
 It is easiest to do this geometrically, remembering that the cross product p×q of vectors
p and q is perpendicular to both of them and |p × q| = |p| |q| sin(angle between p
and q) = |p| |q| if p and q are perpendicular. The cross product of nonzero vectors is
zero if and only if the vectors are collinear. It is easy to see that the only way the zero
vector is in S is if v3 is zero, which will happen only when the nonzero vectors v1 and
v2 are collinear, and in that case the sequence is zero all the way from v3 onwards.

As the starting vectors v1 and v2 are distinct and nonzero, the third vector v3 = v1 ×v2 ,
being perpendicular to both v1 and v2 , is distinct from them. This is true even if v3
is 0 due to v1 and v2 being collinear. So (a) is false.

Basic calculation. Taking v1 = i and v2 = j, the sequence cycles: i, j, k, i, j, k, . . ., so

(b) is true. (c) is also true because we can arrange the sequence to be the following:
a vector not in {i, j, k}, j, k, i, j, k, . . . (e.g., take v1 = i + j and v2 = j. The basic
calculation repeats after v3 .) Note that any vector in the sequence depends only on
the previous two vectors.

(d) is false. The set S can easily be infinite (e.g., if you start with i and 2j, the mag-
nitudes of subsequent vectors will keep increasing), but U is always finite. First note
that the cycle of three vectors occurs whenever one starts with any two perpendicular
vectors of unit length. Now U consists of unit vectors in the direction of each nonzero
vector in S. So depending on the angle θ between v1 and v2 , the cardinality of U is
either 1 (when θ = 0), 2 (when θ = π), 3 (when θ = π/2) or 4 (in all other cases,
because v2 and v3 are still perpendicular).

9.
x x 4 + x6
f (x) = and g(x) = .
x + sin x e x − 1 − x2
(a) Limit as x → 0 of f (x) is 21 .
(b) Limit as x → ∞ of f (x) does not exist.
(c) Limit as x → ∞ of g(x) is finite.
(d) Limit as x → 0 of g(x) is 720.

Correct options: a,c Wrong: b,d

Calculate (a) and (c) using L’Hôpital’s rule. (Or in (a) use that sin x behaves like x
near 0 and in (c) the limit is 0 because ex dominates any polynomial for large x.) In
x
(b) the limit is 1 as f (x) is sandwiched between x±1 , both of which → 1. L’Hôpital’s
rule is not applicable as the expression one gets after attempting it does not have a
limit as x → ∞, so L’Hôpital’s rule does not tell us anything. In (d) the limit is 0 by
L’Hôpital’s rule used correctly. Only one application is enough.

5
 10. Let f (u) = tan−1 (u), a function
R v whose domain in the set of all real numbers and whose
π π
range is (− 2 , 2 ). Let g(v) = 0 f (t)dt.

(a) f (1) = π4 .
(b) f (1) + f (2) + f (3) = π.
(c) g is an increasing function on the entire real line.
(d) g is an odd function, i.e., g(−x) = −g(x) for all real x.

Correct options: a,b Wrong: c,d

(a) is direct and (b) is a straightforward calculation using the formula for tan(A + B)
keeping in mind the range of tan−1 . By the fundamental theorem of calculus, g 0 (x) =
f (x), so g is increasing when f is positive, which is true only in (0, ∞). g is an
even
R q function Rasp its derivative f is odd. Note that g(x) is defined for all real x as
p
f (t)dt = − q f (t)dt.

Part B Solutions

B1. Solve the following two independent problems on pages 2–3 of the answer booklet.

(i) Let f be a function from domain S to codomain T . Let g be another function from
domain T to codomain U . For each of the blanks below choose a single letter corre-
sponding to one of the four options listed underneath. (It is not necessary that each
choice is used exactly once.) Write your answers on page 2 as a sequence of four letters
in correct order. Do NOT explain your answers.

If g ◦ f is one-to-one then f B and g D .

If g ◦ f is onto then f D and g C .

Option A: must be one-to-one and must be onto.

Option B: must be one-to-one but need not be onto.
Option C: need not be one-to-one but must be onto.
Option D: need not be one-to-one and need not be onto.

Recall: g ◦ f is the function defined by g ◦ f (a) = g(f (a)). The function f is said
to be one-to-one if, for any a1 and any a2 in S, f (a1 ) = f (a2 ) implies a1 = a2 . The
function f is said to be onto if, for any b in T , there is an a in S such that f (a) = b.

6
 (ii) In the given figure ABCD is a square. Points X and Y , respectively on sides BC and
CD, are such that X lies on the circle with diameter AY . What is the area of the
square ABCD if AX = 4 and AY = 5? (Figure is schematic and not to scale.)

B X C

A D

Solution: AXY is a right angle, being an angle in a semicircle. Therefore by Pythagoras,

XY = 3. Triangles ABX and XCY are similar, because both are right angled triangles and
at the point X the three angles add to 180◦ , with the middle angle AXY being a right angle.
We have the following three equations in three unknowns.
AB 4
= by similarity BX + CX = AB AB 2 + BX 2 = 16.
CX 3
Solving these gives the answer, e.g., CX = 43 AB by the first equation, so BX = 14 AB by
1 2
the second equation, so AB 2 + 16 AB 2 = 16 by the third equation, so area = AB 2 = 16
17
.

B2. Solve the following two independent problems on pages 4–5 of the answer booklet.

(i) A mother and her two daughters participate in a game show. At first, the mother
tosses a fair coin.

Case 1: If the result is heads, then all three win individual prizes and the game ends.

Case 2: If the result is tails, then each daughter separately throws a fair die and wins
a prize if the result of her die is 5 or 6. (Note that in case 2 there are two independent
throws involved and whether each daughter gets a prize or not is unaffected by the
other daughter’s throw.)

(a) Suppose the first daughter did not win a prize. What is the probability that the
second daughter also did not win a prize?
Solution: Since the first daughter did not win a prize, the coin toss must have
shown tails. Now the second daughter does not win precisely when she throws 1,
2, 3 or 4. The probability of this is unaffected by the first daughter’s throw. So
the desired probability is 46 = 23 . One can also do this more pedantically in a way
similar to part (b), see below.

7
 (b) Suppose the first daughter won a prize. What is the probability that the second
daughter also won a prize?
Solution: Let T = the event that the coin toss gives tails. Similarly H = heads,
F = first daughter wins, S = second daughter wins. We want P(S | F). Note that
the outcome of the throw of each die is independent of that of the other die and
is unaffected by the coin toss that preceded it.
P(S | F) = P(S & F)/P(F) = ( 59 )/( 23 ) = 5
6
because
1
P(F) = P(H) + P(T) P(first die = 5 or 6 | T) = 2
+ 21 26 = 23 , and
1
P(S & F) = P(H) + P(T) P(each die = 5 or 6 | T) = 2
+ 12 ( 26 )2 = 59 .
We can also solve part (a) similarly to find the desired P(not S | not F).
P(not S | not F) = P((not S) & (not F))/ P(not F) = 64 because
P(not F) = P(T) P(1 ≤ first die ≤ 4 | T) = 21 46 , and
P((not S) & (not F)) = P(T) P(1 ≤ both dice ≤ 4 | T) = 12 ( 64 )2 .

(ii) Prove or disprove each of the following statements.

(a) 240 > 20!
Solution: False. In fact 262 > 20! > 261 , so even crude estimation is enough
to solve this. For example LHS = 240 = 420 = 4 × 4 × · · · × 4 (20 times).
RHS = 1 × 2 × · · · × 20. Consider the ratio RHS/LHS and pair the 20 numbers
in each product. The initial three fractions less than 1, namely 41 , 24 , 34 are easily
overpowered by the remaining ones, e.g., they are individually matched by 16 , 8, 6.
4 4 4
(OR, using 2 × 3 > 22 , 4 × · · · × 7 > 44 = 28 , 8 × · · · × 15 > 88 = 224 and
16 × · · · × 20 > 165 = 220 , one gets 20! > 254 .)
1
(b) 1 − x
≤ ln x ≤ x − 1 for all x > 0.
Solution: True. Let f (x) = x − 1 − ln x. By analyzing the sign of f 0 (x) = 1 − x1
(or by looking at the sign of f 00 (x)), see that f (x) has a global minimum at x = 1
and that this minimum value is 0, giving ln x ≤ x − 1. For the other inequality,
substitute x = 1t in ln x ≤ x − 1 to get ln 1t = − ln t ≤ 1t − 1, i.e., 1 − 1t ≤ ln t
for all t such that 1t > 0, which is equivalent to having the same inequality for all
t > 0. So we may replace t by x, giving the desired result. (Of course, it is also
possible to repeat the earlier logic by analyzing ln x − 1 + x1 .)

B3. You are supposed to create a 7-character long password for your mobile device.
(i) How many 7-character passwords can be formed from the 10 digits and 26 letters?
(Only lowercase letters are taken throughout the problem.) Repeats are allowed, e.g.,
0001a1a is a valid password.
Solution: For each character there are 36 choices. So number of passwords = 367 .

8
 (ii) How many of the passwords contain at least one of the 26 letters and at least one of
the 10 digits? Write your answer in the form: (Answer to part i) − (something).

Solution: From 367 remove 267 passwords containing only letters and 107 passwords
containing only digits. Required number = 367 − (267 + 107 ).

(iii) How many of the passwords contain at least one of the 5 vowels, at least one of the 21
consonants and at least one of the 10 digits? Extend your method for part ii to write
a formula and explain your reasoning.

Solution: Apply the inclusion exclusion principle or use a Venn diagram. Out of 367
passwords, 317 contain no vowels (V), 157 contain no consonants (C) and 267 contain no
digits (D). As first step we take 367 −(317 missing V + 157 missing C + 267 missing D).
But this removes the passwords without two types of characters (i.e., those containing
only one type of character) twice. So we need to add these back so as to effectively
remove them only once. So we need to add (107 missing VC + 217 missing VD + 57
missing CD). So the final answer is

367 − (317 + 157 + 267 ) + (107 + 217 + 57 ).

(iv) Now suppose that in addition to the lowercase letters and digits, you can also use 12
special characters. How many 7-character passwords are there that contain at least
one of the 5 vowels, at least one of the 21 consonants, at least one of the 10 digits and
at least one of the 12 special characters? Write only the final formula analogous to
your answer to part iii. Do NOT explain.

Solution: The answer is

487 − (437 + 277 + 387 + 367 ) + (227 + 337 + 177 + 317 + 157 + 267 ) + (57 + 217 + 107 + 127 ).

A password missing one type of character is subtracted only in the second term. A
password missing two types of character is subtracted twice in the second term, so
added once in the third term. A password
missing three types of character is subtracted
thrice in the second term, added back 32 = 3 times in the third term and subtracted
once in the last term. (This is easier to understand as an application of the inclusion
exclusion principle. Venn diagram gets harder to keep track of as there are more
possibilities for overlaps.)

B4. Show that there is no polynomial p(x) for which cos(θ) = p(sin θ) for all angles θ in
some nonempty interval.
Hint: Note that x and |x| are different functions but their values are equal on an interval
(as x = |x| for all x ≥ 0). You may want to show as a first step that this cannot happen for
two polynomials, i.e., if polynomials f and g satisfy f (x) = g(x) for all x in some interval,

9
then f and g must be equal as polynomials, i.e., in each degree they must have the same
coefficient.
Solution: To prove the assertion in the hint, note that the polynomial f − g would have
infinitely many roots and hence must be the zero polynomial.
Suppose a polynomial p satisfies cos(θ) = p(sin θ) for θ in an interval. Let t = sin θ. Then
the following equality is true for the (infinitely many) values of t in some nonempty interval:

p(t)2 = cos2 (θ) = 1 − sin2 (θ) = 1 − t2 .

By the hint, this forces the polynomial 1 − x2 to be the square of the polynomial p(x). But
1 − x2 is not a square because, e.g., the square of the leading coefficient would need to be
−1, which cannot happen. (Or p would need to be a linear polynomial ax + b, etc.)
Note: The italicized part “in some nonempty interval” at the end of the problem statement
was missing in the actual exam due to oversight. As sin of two angles can be equal without
their cos being equal, a very easy solution now becomes possible, e.g., just plug in θ = 0 and
θ = π to get p(0) = 1 as well as p(0) = −1. So in fact there cannot be any function (not
just polynomial) p such that cos(θ) = p(sin θ) for ALL angles θ. In the exam, everyone who
gave any correct solution to the problem as stated there was given full credit.

B5. Define a function f as follows: f (0) = 0 and, for any x > 0,

Z L Z ∞
1  1 
f (x) = lim cos(t)dt or, in simpler notation, the improper integral cos(t)dt .
L→∞ 1 t2 1 t2
x x

(i) Show that the definition makes sense for any x > 0 by justifying why the limit in the
definition exists, i.e., why the improper integral converges.
Solution: As | cos(t)| ≤ 1, the integral defining f (x) is in fact absolutely convergent.
Z ∞ Z ∞
1
 1 1 ∞
cos(t) dt ≤ dt = −  1 = x.

 2 2
t t t

1 1
x x x

(ii) Find f 0 ( π1 ) if it exists. Clearly indicate the basic result(s) you are using.
df
Solution: For x 6= 0, let u = x1 . By the fundamental theorem of calculus1 , du = − u12 cos(u) =
df
−x2 cos( x1 ). Since du
dx
= − x12 , by chain rule dx = cos( x1 ), so f 0 ( π1 ) = cos(π) = −1.
(iii) Using the hint or otherwise, find limh→0+ f (h)−f
h
(0)
, i.e., the right hand derivative of f
at x = 0. We can take the limit only from the right hand side because f (x) is undefined for
negative values of x.
Hint: Break f (h) into two terms by using a standard technique with an appropriate choice.
Then separately analyze the resulting two terms in the derivative.
1
To use the standard version of the fundamental theorem where the lower endpoint is a fixed finite
R1
number, take some positive constant K. Then f (x) = − Kx t12 cos(t)dt + a constant (by part i).

10
Solution: Integrate by parts using u = t12 and dv = cos(t)dt. Then du = − t23 dt and
v = sin(t). (The method below does not work with u = cos(t) and dv = t12 dt.) So

f (h) − f (0) 1 ∞ 1 1 sin(t) ∞ 1 ∞ 2 sin(t)

Z Z
= cos(t)dt =  + dt.
h h h1 t2 h t2 h1 h h1 t3
∞
1 sin(t) 
The first term in the sum = h t2  1
= −h sin( h1 ) → 0 as h → 0+ , where we have twice used
h
that | sin(t)| ≤ 1 for all t. For the the second term, by logic similar to part (i) we get
 1 Z ∞ 2 sin(t)  1 Z ∞ 2 1 −2 ∞
dt ≤ dt = − t  1 = h → 0 as h → 0+ .
 
h h1 t3 h h1 t3 h
 
h

So the desired limit is 0.

Rx
Note: Substituting s = 1t gives f (x) = the simpler looking improper integral 0 cos( 1s )ds.
This makesR x parts (i) and (ii) more transparent. Now x < 0 is also ok in the improper
1
integral 0 cos( s )ds. For the function g(x) defined by this new integral, the above analysis
gives one way to show that g 0 (0) exists (see math.stackexchange.com/questions/2127903). g
is differentiable everywhere (the only case requiring work being x = 0) but the derivative is
not continuous at x = 0, as g 0 does not have a limit at x = 0.

B6. n and k are positive integers, not necessarily distinct. You are given two stacks of cards
with a number written on each card, as follows.
Stack A has n cards. On each card a number in the set {1, . . . , k} is written.
Stack B has k cards. On each card a number in the set {1, . . . , n} is written.
Numbers may repeat in either stack. From this, you play a game by constructing a sequence
t0 , t1 , t2 , . . . of integers as follows. Set t0 = 0. For j > 0, there are two cases:
If tj ≤ 0, draw the top card of stack A. Set tj+1 = tj + the number written on this card.
If tj > 0, draw the top card of stack B. Set tj+1 = tj − the number written on this card.
In either case discard the taken card and continue. The game ends when you try to draw
from an empty stack. Example: Let n = 5, k = 3, stack A = 1, 3, 2, 3, 2 and stack B = 2, 5, 1.
You can check that the game ends with the sequence 0, 1, −1, 2, −3, −1, 2, 1 (and with one
card from stack A left unused).

(i) Prove that for every j we have −n + 1 ≤ tj ≤ k.

(ii) Prove that there are at least two distinct indices i and j such that ti = tj .

(iii) Using the previous parts or otherwise, prove that there is a nonempty subset of cards
in stack A and another subset of cards in stack B such that the sum of numbers in
both the subsets is same.

11
Solution: (i) Induction. Base case is true as t0 = 0. Assume the result up to tj . Now there
are two cases. If tj ∈ [−n + 1, 0] then tj+1 = tj + a number from stack A, which is between
1 and k, so tj+1 ∈ [(−n + 1) + lowest possibility 1, 0 + highest possibility k] ⊂ [−n + 1, k].
If tj ∈ [1, k] then tj+1 = tj − a number from stack B, which is between 1 and n, so tj+1 ∈
[1 − highest possibility n, k − lowest possibility 1] ⊂ [−n + 1, k].
(ii) Suppose the game ends when we try to draw from stack B. As there are k cards in stack
B, in all we must have tried to draw k + 1 times from stack B. At each one of these attempts,
the value of tj must have been positive and by part (i) each one of these k + 1 numbers is
between 1 and k (inclusive). So there must be a repeat among these k + 1 numbers.
If the game ends when we try to draw from stack A, the argument is parallel: there must
have been n + 1 attempts to draw from stack A, each one resulting from a value of tj that
lies among the n numbers from −(n − 1) to 0, so there must be a repeat.
(iii) Suppose ti = tj . Then, from the set of cards drawn at steps i + 1, . . . , j the sum of the
cards from stack A must equal the sum of the cards from stack B.

12