Preliminary Exam 2019

Solutions to Morning Exam

Part I.
Solve four of the following five problems.
Problem 1. Consider the function f on R such that f (x) = x2 ln |x| if x 6= 0
and f (0) = 0. Prove that f 0 (0) exists but f 00 (0) does not.
Solution: The limit of the difference quotient for f at 0 is
x2 ln |x|
lim = lim x ln |x| = 0
x→0 x x→0

by L’Hôpital’s Rule. So f (0) exists and equals 0. Now if x 6= 0 then f 0 (x) =

0

2x ln |x| + x, so the limit of the difference quotient for f 0 at 0 is

2x ln |x| + x
lim = 1 + 2 lim ln |x|,
x→0 x x→0

which does not exist as a finite quantity (the limit is −∞).

Problem 2. Suppose that y : R → R is twice-differentiable √ and satisfies the
00 0
differential
√ equation y + y + y = 0. If y(0) = 1 and y(π/ 3) = 0 then what is
y(2π/ 3)? Simplify your answer to the extent possible.
√
Solution: Since r2 + r + 1 = (r − eλ )(r − eλ ) with λ = e2πi/3 = −1/2 + i 3/2, we
√
can write y = aeλt + beλt with constants a, b ∈ C. But y(0) = 1 and y(π/ 3) = 0,
or in other words. (
a + b√= 1
e−π/ 3 (a − b)i = 0,
so a = b = 1/2. Therefore
√ √ √
y = (e(−1/2+i 3/2)t
+ e(−1/2−i 3/2)t
)/2 = e−t/2 cos( 3t/2),
√ √
so y(2π/ 3) = −e−π/ 3 .
2 R tan x
Problem 3. Let f (x) = ex and g(x) = 0 f (t) dt. Find g 0 (x), and then find
a constant c 6= 0 such that cg 0 (x) = h(x)eh(x) for some function h(x).
Solution: By the Fundamental Theorem of Calculus and the Chain Rule,
2
g 0 (x) = f (tan x) sec2 x = etan x
sec2 x.
Recalling that tan2 x + 1 = sec2 x, we see that we may take c = e and h(x) = sec2 x.
Problem 4. Decide whether each series converges, justifying your answer:
(a) n>2 (cos(1/ log n) − 1)n−1
P

Solution: From the Taylor series of the cosine function (or simply by Taylor’s
theorem) we have | cos x − 1| 6 M x2 for x near 0, so in particular
| cos(1/ log n) − 1|n−1 6 M n−1 (log n)−2 .
n−1 (log n)−2 converges by the Integral Test, because
P
Now n>2
Z T
dx
= −(log x)−1 |T2 = (log 2)−1 − (log T )−1 → (log 2)−1
2 x(log x)2
1
2

as T goes to infinity. Therefore n>2 (cos(1/ log n)−1)n−1 is absolutely convergent

P

(and hence
P convergent) by the Comparison Test.
(b) n>2 sin(1/ log n)n−1
Solution: Since limx→0 sin x/x = 1, we see that for some M > 0 we have sin x >
M x if x is close to 0 and positive. In particular,
(M/ log n)n−1 < sin(1/ log n)n−1 .
n−1 (log n)−1 diverges by the Integral Test, because
P
But n>2
Z T
dx
= log(log x)|T2 = log log T − log log 2 → ∞
2 x log x
as T goes to infinity. So n>2 sin(1/ log n)n−1 diverges by the Comparison Test.
P

Problem 5. Let I be an open interval in R and f a differentiable function on

I, and suppose that f 0 is identically 0. Prove that f is a constant function. You
may quote theorems from calculus that are logically prior to the present assertion.
Solution: Fix a ∈ I. By the Mean Value Theorem, for any b ∈ I there exists a
number c between a and b such that f (b) − f (a) = f 0 (c)(b − 1). Since f 0 = 0, we
get f (b) = f (a), and since b ∈ I was arbitrary we conclude that f is constant.

Part II.
Solve three of the following six problems.
Problem 6. If n 7→ an is a surjective or “onto” map from the set of positive
integers to the set Q ∩ [0, 1] of rational
P numbers between 0 and 1, what is the radius
of convergence of the power series n>1 an xn ? Prove your answer.
1/n
Solution: Let ρ be the radius of convergence. Then 1/ρ = lim supn→∞ an . Put
1/n
L = lim supn→∞ an . We claim that L = 1, whence ρ = 1. Certainly L 6 1,
1/n
because an 6 1 for all n, whence an 6 1 for all n also. On the other hand,
given ε > 0, there are infinitely many rational numbers between 1 − ε and 1, hence
infinitely many n such that 1 − ε 6 an 6 1. Therefore
1 − ε < (1 − ε)1/n 6 a1/n
n

for infinitely many n. Therefore L > 1 − ε, and since ε > 0 is arbitrary, we conclude
that L > 1, whence L = 1.
Problem 7. Show that
Z 1
X (−1)n 1 x
= (√ − ) dx
(2n + 1)(2n + 2) 0 2−x 2 1 + x2
n>0

by computing both sides explicitly.

Solution: The left-hand side can be rewritten as
X 1 1 X (−1)n 1 X (−1)n
(−1)n ( − )= − .
2n + 1 2n + 2 2n + 1 2 n+1
n>0 n>0 n>0

The first and second series on the right-hand side converge by the Alternating Series
Test, and they are the values at x = 1 of the Taylor series
X x2n+1
tan−1 x = (−1)n
2n + 1
n>0
 3

and
X xn+1
log(x + 1) = (−1)n
n+1
n>0
√
respectively. Thus the left-hand side is tan−1 (1) − (log 2)/2 = π/4 − log 2. (The
Taylor series are obtained by integrating the geometric series for (1 + t2 )−1 and
(1 + t)−1 respectively for 0 6 t 6 x. Strictly speaking, since the geometric series
converges only for |x| < 1, one should also quote Abel’s theorem that a power series
that converges on a closed interval is continuous there.)
The right-hand side can be written as
√ log(1 + x2 ) 1
sin−1 (x/ 2)|10 − |0 = π/4 − (log 2)/2,
2
which coincides with the left-hand side.
R
Problem 8. Find the value of the line integral C x dy − y dx, where C is the
curve r = cos 2θ for −π/4 6 θ 6 π/4, oriented counterclockwise.
Solution: By Green’s Theorem, the line integral coincides with
Z Z Z Z
∂ ∂
( x− (−y)) dx dy = 2 dx dy,
R ∂x ∂y R
where R is the region enclosed by C. Thus the line integral coincides with
Z π/4 Z cos 2θ Z π/4
2 r dr = cos2 (2θ) dθ.
−π/4 0 π/4
2
Since cos (2θ) is an even function, the right-hand side coincides with
Z π/4 Z π/4
2
2 cos (2θ) dθ = (1 + cos 4θ) dθ,
0 0
which is π/4.
p
Problem 9. Let D be the solid p region inside the p cone z = x2 + y 2 and
2 2 2 2
between the two hemispheres z = 4 − x − y and z = 1 − x − y . Given that
D has uniform density, find the “center of mass” or “centroid” of D. You may use
symmetry considerations to reduce the amount of computation.
Solution: Let (x, y, z) be the center of mass. The rotational symmetry of D
about the z-axis ensures that x = y = 0. We must compute the quantity z = I/J,
where Z Z Z
I= z dx dy dz
D
and Z Z Z
J= dx dy dz.
D
We compute I and J in spherical coordinates: Thus
Z 2π Z π/4 Z 2 √
J= ρ2 sin ϕ dρ dϕ dθ = 2π(1 − 2/2)(7/3)
0 0 1
and Z 2π Z π/4 Z 2
I= ρ3 cos ϕ sin ϕ dρ dϕ dθ = 15π/8.
0 0 1
√
So z = I/J = 45/(56(2 − 2)).
4

Problem 10. Let S be the portion of the sphere x2 + y 2 + z 2 = 4 defined by

z > 1, and let F(x, y, z) = (−y + z, x + z, z 2 ). Find the value of the surface integral
Z Z Z Z
(∇ × F) · dS = (∇ × F) · n dσ,
S S
(to use two common notations), where n is the unit outward normal vector, dσ is
an infinitesimal unit of surface area, and dS = n dσ.
Solution: By Stokes’ Theorem, the surface integral equals a line integral:
Z Z Z
(∇ × F) · dS = F · dr,
S ∂S
where ∂S is the boundary of S oriented counterclockwise. Now ∂S is a circle, and
we can parametrize it by
√ √
r(t) = ( 3 cos t, 3 sin t, 1) (0 6 t 6 2π).
√ √ √ √
Also F(r(t)) = (− 3 sin t + 1, 3 cos t + 1, 1) and r0 (t) = (− 3 sin t, 3 cos t, 0), so
√ √
F(r(t)) · dr0 (t) = 3 − 3 sin t + 3 cos t.
Hence Z Z 2π √ √
F · dr = (3 − 3 sin t + 3 cos t) dt = 6π.
∂S 0

Problem 11. Show that there are open neighborhoods D and D0 of (0, 0) ∈ R2
such that if (a, b) ∈ D0 then the system of equations
(
2ex − e2y − e4x−7y = a
e3x + 4ey − 5ex+y = b
has a unique solution (x, y) ∈ D.
Solution: Define f : R2 → R2 by f (x, y) = (2ex −e2y −e4x−7y , e3x +4ey −5ex+y ).
The Jacobian matrix of f is
 x
2e − 4e4x−7y −2e2y + 7e4x−7y

[f 0 (x, y)] = ,
3e3x − 5ex+y 4ey − 5ex+y
so  
0 −2 5
f (0, 0) = ,
−2 −1
and consequently det f 0 (0, 0) = 12 6= 0. Thus f 0 (0, 0) is invertible Note also that
f (0, 0) = (0, 0). Applying the Inverse Function Theorem, we deduce that there are
open neighborhoods D and D0 of (0, 0) such that f is a C ∞ diffeomorphism of D
onto D0 . In particular, f is a bijection of D onto D0 , so given (a, b) ∈ D0 there
exists a unique point (x, y) ∈ D such that f (x, y) = (a, b).
Solution:

Part III.
Solve one of the following three problems.
Problem 12. Let X be a metric space with metric d, and let {xn }n>1 and
{yn }n>1 be two Cauchy sequences in X. Show that {d(xn , yn )}n>1 is a Cauchy
sequences of real numbers. Do not use the fact that X can be embedded in a
complete metric space.
 5

Solution: By the triangle inequality we have

d(xn , yn ) 6 d(xn , xm ) + d(xm , ym ) + d(ym , yn ),
whence
d(xn , yn ) − d(xm , ym ) 6 d(xn , xm ) + d(ym , yn ).
Similarly
d(xm , ym ) − d(xn , yn ) 6 d(xn , xm ) + d(ym , yn ),
so we get
|d(xm , ym ) − d(xn , yn )| 6 d(xn , xm ) + d(ym , yn ).
Now if ε > 0 is given, there exists N such that if n, m > N then d(xn , xm ) < ε/2
and d(ym , yn ) < ε/2. Hence if n, m > N then |d(xm , ym )−d(xn , yn )| < ε, so indeed,
{d(xn , yn )}n>1 is a Cauchy sequences of real numbers.
Problem 13. Let I be an interval in R and {fn }n>1 a sequence of continuous
real-valued functions on I which is uniformly convergent to a real-valued function
f on I. In the following questions, “prove” means “justify by quoting general
theorems,” and “give a counterexample” includes proving that your counterexample
does what you claim.
(a) If I = [0, 1] then is f uniformly continuous? Prove or give a counterexample.
Solution: Yes, because (i) f , being the limit of a uniformly convergent sequence
of continuous functions, is continuous, and (ii) a continuous function on a compact
set is uniformly continuous.
(b) If I = R then is f uniformly continuous? Prove or give a counterexample.
Solution: No, f need not be uniformly continuous. For example, let fn (x) =
ex + 1/n and f (x) = ex . Then {fn } is uniformly convergent to f , because
|fn (x) − f (x)| = 1/n
for all x ∈ R, whence |fn (x) − f (x)| < ε for n > 1/ε. But f is not uniformly
continuous. Indeed let ε = 1, and suppose that there exists δ > 0 such that
|ex − ey | < 1 whenever |x − y| < δ. Then |ex − ex+δ/2 | < 1 for all x ∈ R. So
|1 − eδ/2 | < e−x for all x ∈ R. Taking the limit as x → ∞, we obtain |1 − eδ/2 | 6 0,
contradicting our assumption that δ > 0.
Problem 14. Define f, g : Rn → R by f (x1 , x2 , . . . , xn ) = x1 + x2 + · · · + xn
and g(x1 , x2 , . . . , xn ) = x1 x2 · · · xn . Given c > 0, consider the surface
S = {(x1 , x2 , . . . , xn ) ∈ Rn : g(x1 , x2 , . . . , xn ) = c and xi > 0 for 1 6 i 6 n}.
(a) Show that f attains a minimum value on S even though S is not compact.
Solution: Choose a number L > nc1/n . If (x1 , x2 , . . . , xn ) ∈ S and xi > L
for some i then f (x1 , x2 , . . . xn ) > L, whereas f (c1/n , c1/n , . . . , c1/n ) < L. Hence
writing x = (x1 , x2 , . . . , xn ), we have
inf{f (x) : x ∈ S} = inf{f (x) : x ∈ S and xi 6 L for 1 6 i 6 n}.
Let X be the set of x on the right-hand side, in other words, the set of x ∈ S
such that xi 6 L for all i. Then X is nonempty because (c1/n , c1/n , . . . , c1/n ) ∈ X.
Hence {f (x) : x ∈ X} is nonempty, and since the latter set is bounded below by
0, we see that inf{f (x) : x ∈ X} does exist. But X is compact (note that the
condition xi 6 L amounts to 0 6 xi 6 L) so inf{f (x) : x ∈ X} is actually the
minimum of f on X and hence on S.
6

(b) Show that the minimum value of f on S occurs at (c1/n , c1/n , . . . , c1/n ) and
at no other point.
Solution: If the minimum value of f on S occurs at the point x ∈ S then by
the method of Lagrange multipliers, we have ∇f (x) = λ∇g(x). In other words, for
1 6 i 6 n we have 1 = λc/xi . Taking the product of these equations for 1 6 i 6 n,
we obtain 1 = λn cn /c, whence λ = c−(n−1)/n . Substituting this value in 1 = λc/xi ,
we find that xi = c1/n for all i.
(c) Deduce that (x1 x2 · · · xn )1/n 6 (x1 +x2 +· · ·+xn )/n for all x1 , x2 , . . . , xn > 0,
with equality if and only if x1 = x2 = · · · = xn .
Solution: Let c = x1 x2 · · · xn . Then x ∈ S, so we know from (b) that f (x) >
nc1/n , the minimum value of f on S. Furthermore, this minimum is attained if and
only if x = (c1/n , c1/n , . . . , c1/n ). Thus f (x)/n > c1/n , with equality if and only if
all the xi are equal.