Solutions to Introductory Exam for January, 2005

√
1. The function f : [0, 1] → R defined by f (x) := 1/ x for 0 < x ≤ 1 and f (0) := 0
is unbounded and henceR not Riemann integrable on [0, 1]. Show that the improper
integral of f on [0, 1] is 01 f (x) dx = 2.
Solution:
√ √ 1 √
For any c > 0 we have c1 f (x) dx = c1 (1/ x) dx = 2 x |c = 2 (1 − R c). Hence,
R R
1
letting c → 0+,
√ we obtain that the improper integral of f on [0, 1] is 0 f (x) dx =
limc→0+ 2 (1 − c) = 2.

2. With A an n × n matrix, the formula y = Ax can be regarded as instructions to

substitute, for each of the n variables yi , the corresponding linear combination of the
variables xj . Suppose this is combined with another such substitution, x = Bz, so
that ultimately each variable yi is substituted-for by an expression in the variables
zk . What matrix represents this last substitution?
Solution:
For this very simple problem, one could proceed component-wise, but it is simpler
to observe, from y = Ax = A(Bz) = (AB)z, that the desired matrix is just AB.
[Historically, this is the reason why matrix multiplication is defined the way it is; it
also exhibits how the operation of composition (for linear transformations) is in a
technical sense “represented” by the operation of multiplication (for matrices).] Note
that the restriction to square matrices in this problem was unnecessary.

3. Let λ ∈ (0, ∞). Consider the joint probability density function fX,Y given by

fX,Y (x, y) := λ2 e−λy for 0 ≤ x ≤ y < ∞

and fX,Y (x, y) := 0 elsewhere.

(a) Show that the marginal distribution of X is exponential with (inverse-scale)

parameter λ.
(b) Show that the marginal distribution of Y is gamma with index parameter 2 and
inverse-scale parameter λ.

Solution:
R∞
(a) fX (x) = x λ2 e−λy dy = λe−λx , x ≥ 0.

1
 Ry
(b) fY (y) = 0 λ2 e−λy dx = λ2 ye−λy , y ≥ 0.

4. Let n ≥ 2 be an integer and let N := {1, 2, . . . , n}. How many different trees T
can we form having V (T ) = N and vertex 1 as a leaf? Justify your answer.
Solution:
The answer is (n − 1)n−2 . Because vertex 1 is a leaf, T 0 = T − {1} is a tree and, by
Cayley’s formula, there are (n − 1)n−3 different trees T 0 . Then, for each choice of T 0 ,
there are n−1 choices for a neighbor for vertex 1. Hence, there are (n−1)·(n−1)n−3 =
(n − 1)n−2 such trees.

5. Prove that if an n × n matrix A can be partitioned in the form

" #
E Z
A= ,
G H

where E is a square matrix and Z consists entirely of zeros, then det A = (det E)(det H).
Solution:
Recall that det A is a signed sum of all products of n factors which can be formed by
selecting exactly one factor from each row and each column of A. Suppose that E is
m × m. Then because Z = 0, the m factors from the first m columns of A for such
a product to be nonzero must come from E; this also uses up the first m rows of A.
The remaining n − m factors must come appropriately from the entries of H. When
signs are taken into account (using the fact that sign is a multiplicative function on
the symmetric group), we obtain the desired result.
Solutions can also be based upon induction on the size of E or on the size of H.

6. Let V be the linear subspace of R3 spanned by the (column) vectors v = [1, 1, 0]T
and w = [0, 1, 1]T .

(a) Find an orthonormal basis {e, f } for V (with respect to the usual inner product).
h√ √ √ iT
(b) Let g = 3/3, − 3/3, 3/3 . Verify that {e, f, g} forms an orthonormal
basis of R3 .
(c) Let T : R3 → R3 denote the projection map onto V , and suppose A is the matrix
for T with respect to the usual basis of R3 . What is the rank of A?

2
Solution:

(a) Using the Gram–Schmidt process, we obtain an orthonormal basis {e, f }, where
"√ √ #T
1 2 2
e = √ [1, 1, 0]T = , ,0
2 2 2

and √ "√ √ #T
2 2 2 1 1 T
 
T
w − hw, eie = [0, 1, 1] − , ,0 = − , ,1 ,
2 2 2 2 2
and  s

1 1 T 3
− , ,1 = ,

2 2 2
so " √ √ √ #T
T ,
 1 1 T
1 1 6 6 6

f = − , ,1 − , ,1 = − , , .

2 2 2 2 6 6 3
(b) The vector g has unit length and is proportional to [1, −1, 1]T which is easily
seen to be orthogonal to v and w.
(c) The matrix has rank 2 since its columns span V .

7. Consider the following linear program:

Minimize cT x
subject to Ax ≥ b, x ≥ 0.

If it is known that the objective function of the problem is unbounded below in the
feasible region, can you change the vector b to make the problem solvable? Justify
your answer.
Solution:
By the duality theorem, the unboundedness of the linear program implies the infea-
sibility of its dual, which is

Maximize bT w
subject to AT w ≤ c, w ≥ 0.

The dual feasible region {w : AT w ≤ c, w ≥ 0} is independent of the vector b; it will

remain infeasible for any vector b. By the duality theorem again, no matter how the
vector b is chosen, the primal linear program is either infeasible or unbounded.

3
8. Let X be a Poisson random variable with parameter λ. Show that P(X = i)
increases monotonically and then decreases monotonically as i increases, reaching its
maximum value when i is the largest integer not exceeding λ.
Solution:
Consider the ratio P(X = i)/P(X = i − 1), which, after cancellation, is λ/i. The
frequency function is increasing when the ratio is greater than unity, i.e., when i < λ,
and decreasing when the ratio is less than unity, i.e., when i > λ. Thus, the largest
value occurs when i is the integer part of λ. (Note: If λ is an integer, then i = λ − 1
and i = λ both give the maximum.)

9. Let f (x) be continuously differentiable, strictly increasing, and concave for x ≥ 0,

with f (0) = 0. Show that d(x, y) := f (|x − y|) defines a metric on the nonnegative
reals.
Solution:
Clearly d(x, y) = 0 if and only if x = y, and d(x, y) = d(y, x). The triangle inequality
follows from showing that f (a + b) ≤ f (a) + f (b) for any a, b ≥ 0. From continuous
differentiability,
Z a+b
f (a + b) = f 0 (x)dx
0
Z a Z a+b
0
= f (x)dx + f 0 (x)dx
0 a
Z a+b
= f (a) + f 0 (x)dx
a
Z b
≤ f (a) + f 0 (x)dx
0
= f (a) + f (b),

where the inequality is due to the fact that f 0 (x) is decreasing.

10. When a current I (measured in amperes) flows through a resistance R (measured

in ohms), the power generated is given by by W = I 2 R (measured in watts). Suppose
that I and R are independent random variables with densities

fI (x) = 6x(1 − x) I{0 ≤ x ≤ 1}

and
fR (y) = 2y I{0 ≤ y ≤ 1}.
Determine the distribution function of W .

4
Solution #1:
For 0 < w < 1,
ZZ
FW (w) := P(W ≤ w) = 6x(1 − x) 2y dy dx,

where the integration is over the region x2 y ≤ w, 0 ≤ x ≤ 1, 0 ≤ y ≤ 1.

Continuing (note: for some students it may be useful to graph the region to help
determine limits of integration),
√
Z wZ 1 Z 1 Z w/x2
FW (w) = 12x(1 − x)y dy dx + √ 12x(1 − x)y dy dx
0 0 w 0
   
= 3w − 2w3/2 + 3w − 6w3/2 + 3w2
= 6w − 8w3/2 + 3w2 .

Solution #2:
By standard change-of-variables calculations, the joint density of R and W is
w
 r 
fR,W (r, w) = 6 1 − I{0 ≤ w ≤ r ≤ 1}.
r
Integrating over r gives, for 0 ≤ w ≤ 1,
Z 1
w
r  h  i  
fW (w) = 6 1− dr = 6 (1 − w) − 2w1/2 1 − w1/2 = 6 1 − 2w1/2 + w .
w r
Then for 0 ≤ w ≤ 1 we find FW (w) = 6w − 8w3/2 + 3w2 .

11. Let B(m, n) denote the complete bipartite graph with m vertices in one part
and n vertices in the other. Consider the graph G = B(4, 4). With justification,
determine
(a) how many cycles of length 5 are contained in G, and
(b) how many cycles of length 4.
Solution:
(a) Since one characterization of bipartite graphs is as graphs with no odd-length
cycles, part (a) has the immediate answer “zero.”
(b) Let X and Y be the two “parts” of G, so that any cycle must alternate between X
and Y . The critical fact, not true for (say) 6-cycles, is that a bipartite 4-cycle
C is (why?) uniquely determined by its set of   vertices. By completeness, the
4
two X-vertices of C can be chosen in any of 2 = 6 ways and, independently
of that, the two Y -vertices of C can also be chosen in any of 6 ways. So the
number of cycles of length 4 is 36.

5
12. Find all the values of α and β such that the following problem has an optimal
solution with basic variables x3 and x4 :
Minimize αx1 + 13x2 − αx3 + 3x4
subject to x1 + x2 − 2x3 + 3x4 = −1
2x1 − x2 − 3x3 + 4x4 = β
x1 , x2 , x3 , x4 ≥ 0.

Solution:
To move the variables x3 and x4 to the basis,
" we consider
# their corresponding columns
−2 3
in the system to form the matrix B = . Premultiply the linear system by
−3 4
" #
4 −3
B −1 = and compute the reduced costs to get the following tableau:
3 −2

3−α 7α − 2 0 0
−4 − 3β −2 7 1 0
−3 − 2β −1 5 0 1

If this is a tableau for an optimal solution then we have

−4 − 3β ≥ 0, −3 − 2β ≥ 0, 3 − α ≥ 0, 7α − 2 ≥ 0.
Therefore α and β should have values in the following ranges:
2
7
≤ α ≤ 3 and β ≤ − 32 .

13. Let X be a real-valued random variable with density f , and let Y1 , . . . , Yn be

Bernoulli variables which are conditionally independent given X. We assume that
p(x) := P(Yk = 1 | X = x) is independent of k.
Compute the conditional density of X given Y1 = y1 , . . . , Yn = yn as a function of
f , p, and ny , where ny is defined as the number of values k such that yk = 1.
Solution:
Denote this conditional density by x 7→ f (x|y1 , . . . , yn ). It is defined by requiring the
following for any interval A:
Z
P (X ∈ A | Y1 = y1 , . . . , Yn = yn ) = f (x|y1 , . . . , yn ) dx.
A

6
The computation
Z
P(X ∈ A, Y1 = y1 , . . . , Yn = yn ) = P(Y1 = y1 | X = x) · · · P(Yn = yn | X = x)f (x) dx
ZA
= p(x)ny (1 − p(x))n−ny f (x) dx
A

leads directly to

p(x)ny (1 − p(x))n−ny f (x)

f (x|y1 , . . . , yn ) = R ∞ ny n−ny f (u) du
.
−∞ p(u) (1 − p(u))

14. Estimate the probability that in 60 independent tosses of a pair of fair dice the
sum is never equal to 4. Use the Poisson approximation to the binomial distribution.
Solution:
Let Xn := 1 if the sum obtained on the nth toss is 4 and let Xn := 0 otherwise,
n = 1, 2, . . . , 60. Then X1 , . . . , X60 are independent Bernoulli random variables with
P(Xn = 1) = 3/36 = 1/12 for each n, and therefore S := 60
P
n=1 Xn is binomial
with sample size 60 and success probability 1/12. By the Poisson approximation, the
distribution of S is approximately Poisson with mean λ = 60/12 = 5. Consequently,

P(S = 0) ≈ e−5 .

(Note: The exact value is P(S = 0) = (11/12)60 = 0.0054 . . . , whereas e−5 =

0.0067 . . . . Of course, students taking the exam will not have calculators.)

15. Show that a unitary n × n matrix U which is also upper triangular must be
diagonal.
Solution:
Writing out the unitarity condition U ∗ U = I, with I the n × n identity, in terms of
components gives the conditions

|u11 |2 = 1; u11 u1k = 0, k > 1;

|u12 |2 + |u22 |2 = 1; u12 u1k + u22 u2k = 0, k > 2;
|u13 | + |u23 |2 + |u33 |2 = 1;
2
u13 u1k + u23 u2k + u33 u3k = 0, k > 3;

and so forth. These equations may be solved successively to give

|ukk | = 1, k = 1, ..., n; u`,k = 0, ` < k.

7
A solution can also be based upon induction on the size of U .

16. Let f be a nonnegative continuous function on a compact interval J = [a, b] with

a < b, and let M := sup{f (x) : x ∈ J}. Prove that if Mp is defined by
(Z )1/p
b
p
Mp := [f (x)] dx
a

for 0 < p < ∞, then M = limp→∞ Mp . Do not make any appeal to theory of
Lp -spaces.
Solution:
First note that 0 ≤ M < ∞ and that M is achieved by f . Direct estimation shows
that Mp ≤ M (b − a)1/p , so lim supp→∞ Mp ≤ M . Conversely, given  > 0 we have
f (x) ≥ M −  on some subinterval [c, d] of J with c < d; then
(Z )1/p (Z )1/p
d d
p p
Mp ≥ [f (x)] dx ≥ (M − ) dx = (M − )(d − c)1/p ,
c c

so lim inf p→∞ Mp ≥ M − . Letting  drop to 0, we find

M ≤ lim inf Mp ≤ lim sup Mp ≤ M,

p→∞ p→∞

and the desired result follows.

17. Let S = {1, 4, 9, 16, 25, . . . } be the set of squared positive integers, and define
(
0 if i ∈
/S
xi :=
1 if i ∈ S.

(a) Compute lim inf i→∞ xi and lim supi→∞ xi .

n
1X
(b) Evaluate lim xi .
n→∞ n
i=1
n √
1 X
(c) Evaluate lim 2 i.
n→∞ n
i=1

Solution:

(a) The sequence (xi ) is a 0/1-valued sequence with a subsequence of 0s and a

subsequence of 1s, so the lim inf is 0 and the lim sup is 1.

8
(b) The nonnegative sum ni=1 xi , which counts the
P
√ number of perfect squares less
than or equal to n, clearly does not exceed n. Thus the limit in question
vanishes.
(c) We have
n √ n √
n = n3/2 ,
X X
0≤ i≤
i=1 i=1

so again the desired limit is 0.

18. Let n, a, b be nonnegative integers with n ≥ a + b. Give a combinatorial proof of

the following identity:
! ! ! !
n n−a n a+b
= .
a b a+b a

Note: It is not acceptable to convert the binomial coefficients into expressions in-
volving factorials and proceed by algebraic methods.
Solution #1:
Let N be an n-element set. We ask: In how many ways can we choose an ordered
pair (A, B) of disjoint subsets of N with |A| = a and |B| = b?
 
n
On the one hand, we can select the elements of A in a
ways. For each such choice,
    
n−a n n−a
there are b
ways to choose the elements of B, giving a b
pairs in total.
 
n
On the other hand, we can begin by choosing the elements for A ∪ B in ways
 a+b
a+b
and for each such choice, select which elements of this union are in A in a
ways,
  
n a+b
giving a+b a
choices.
     
Since na n−a
b
and n
a+b
a+b
a
are both correct answers to the same question, they
are equal.
Solution #2:
(This second solution is essentially the same as the first, but in a different guise.)
Let N be an n-element set.
Let X denote the set of ordered
  pairs
 (A, B) where |A| = a, |B| = b, A ∩ B = ∅, and
n n−a
A, B ⊆ N . Clearly |X | = a b
(as in Solution 1).
Let Y denote the set
 ofordered
 pairs (S, T ) where |S| = a, |T | = a + b, and S ⊆ T ⊆
n a+b
N . Clearly |Y| = a+b a (choose T and then choose S).
Let f : X → Y by f (A, B) = (A, A ∪ B). Note:

9
 • Since (A, B) ∈ X , it follows that |A ∪ B| = a + b and so f (A, B) is indeed an
element of Y.
• f is one-to-one because if f (A, B) = f (A0 , B 0 ) then A = A0 and A ∪ B = A0 ∪ B 0 .
It follows that B = B 0 by deleting the elements of A = A0 from both sides and
using the fact that A, B and A0 , B 0 are both disjoint pairs.
• f is onto because if (S, T ) ∈ Y then (S, T − S) ∈ X and f (S, T − S) = (S, T ).

Therefore f is a bijection between X and Y, and so |X | = |Y| and the result follows.

19. Suppose that f is a continuous function from R into R and that K is a compact
subset of R. Prove that f either has a zero in K or else is strictly bounded away from
zero on K, i.e., satisfies |f (x)| ≥  for some  > 0.
Solution #1:
If f is not bounded away from 0 on K, then there must exist a sequence xn ∈ K
such that |f (xn )| < 1/n for each positive integer n. Because K is compact, there is
a convergent subsequence xnk → x∗ ∈ K. Since |f (xnk )| < 1/nk and f is continuous,
taking k → ∞ gives f (x∗ ) = 0.
Solution #2:
The continuous image f (K) is also compact, hence closed, and its complement f (K)c
is open. Thus, if 0 ∈ f (K)c , then there exists some open ball B(0, ) = {x : |x| < }
with  > 0 such that B(0, ) ⊂ f (K)c . Thus, either 0 ∈ f (K) or else f (K) ⊂ B(0, )c .
Solution #3:
The continuous function |f | assumes a minimum value on K. Since |f | is nonnegative,
this minimum value is either 0 or else some  > 0. These two possibilities yield,
respectively, the alternative situations stated in the problem.

20. Consider the following matrix (with n ≥ 1):

 
0 1 2
... n
1 2 3
. . . n + 1
 
 
2
A= 3 4
. . . n + 2
.
. .. ..
.. .. 
 .. . .. . 
 
n n + 1 n + 2 ... 2n

(a) Find two vectors u and v such that for any x there exist a, b ∈ R such that
Ax = au + bv.
(b) Compute the eigenvalues of A, together with their (algebraic) multiplicities.

10
[Hint: A useful formula for solving Problem 20 is 1+4+· · ·+n2 = n(n+1)(2n+1)/6.
You may use this formula without proving it.]
Solution:

(a) Let u := (1, 1, . . . , 1)T and v := (0, 1, 2, . . . , n)T . The jth column in A is aj =
(j − 1)u + v. This implies that, for any x,
   
n+1
X n+1
X n+1
X
Ax = x j aj =  (j − 1)xj  u +  xj  v.
j=1 j=1 j=1

(b) The matrix A is symmetric and thus diagonalizable with orthogonal eigenspaces.
The eigenspace for the eigenvalue 0 is characterized by n+1
P Pn+1
j=1 (j−1)xj = j=1 xj =
0 and so is a space of codimension 2 (i.e., dimension n − 1), the vectors u and v
being shown linearly independent by computing the first 2-by-2 determinant.
To compute the last two eigenvalues, it suffices to compute the matrix of A with
respect to u and v, i.e., to express Au = au + cv, Av = bu + dv. Using the
formula in the hint,

a = 0 + 1 + · · · + n = n(n + 1)/2
c = n+1
b = 02 + 12 + · · · + n2 = n(n + 1)(2n + 1)/6
d = n(n + 1)/2

The problem thus reduces to solving for the eigenvalues of

" #
n+1 3n 6
M := .
6 n(2n + 1) 3n

Those are obtained by finding the roots of

n(n + 1)2 (n + 2)
t2 − n(n + 1)t −
12
and are thus
 s   s 
n n(2n + 1)  n n(2n + 1) 
λ1 = (n + 1)  + , λ2 = (n + 1)  − .
2 6 2 6

11