Week7 Inequalities Sols 2
Week7 Inequalities Sols 2
Week7 Inequalities Sols 2
Solution: Use the geometric mean - arithmetic mean inequality, with (a1 , . . . , an ) = (1, . . . , n).
p
2. 3(a + b + c) a + b + c for positive a, b, c.
Solution: Use the power means inequality, with (a1 , a2 , a3 ) = (a, b, c) and r = 1/2, s = 1.
Solution: Guess: the minimum is n, achieved when all x1 = 1. Then use geometric mean -
arithmetic mean inequality to show
x1 + . . . + xn
n x1 . . . xn = 1
n
for positive xi satisfying x1 . . . xn = 1.
4. Minimize
x2 y2 z2
+ +
y+z z+x x+y
subject to x, y, z 0 and xyz = 1.
x y z
Solution: Apply Cauchy-Schwartz with the vectors y + z, z + x, x + y and y+z , z+x , x+y
to get 2
y2 z2
2 x
(x + y + z) + + 2 (x + y + z) ,
y+z z+x x+y
leading to
x2 y2 z2 x+y+z
+ + .
y+z z+x x+y 2
By the AM-GM inequality,
x+y+z
3 xyz = 1,
3
so
x2 y2 z2 3
+ + .
y+z z+x x+y 2
This lower bound can be achieved by taking x = y = z = 1, so the minimum is 3/2.
1
5. If triangle has side lengths a, b, c and opposite angles (measured in radians) A, B, C, then
aA + bB + cC
.
a+b+c 3
Solution: Consider f (x) = (1999 x) ln(1999 + x). We have ef (0) = 19991999 and ef (1) =
20001998 , so we want to see what f does on the interval [0, 1]: increase or decrease? The
derivative is
1999 x
f 0 (x) = ln(1999 + x) + ,
1999 + x
which is negative on [0, 1] (since, for example,
1999 x
1 = ln e < ln(1999 + x)
1999 + x
on that interval). So
20001998 < 19991999 .
8. Minimize
sin3 x cos3 x
+
cos x sin x
on the interval 0 < x < /2.
Solution: We can use the rearrangement inequality on the pairs sin3 x, cos3 x (which sat-
isfies sin3 x cos3 x on [0, /4], and sin3 x cos3 x on [/4, /2]), and (1/ cos x, 1/ sin x)
(which also satisfies 1/ cos x 1/ sin x on [0, /4], and 1/ cos x 1/ sin x on [/4, /2]), to
get
sin3 x cos3 x sin3 x cos3 x
+ + = sin2 x + cos2 x = 1
cos x sin x sin x cos x
on the whole interval. Since 1 can be achieved (at x = /4) the minimum is 1.
Source: These problems were all taken from a Northwestern Putnam prep problem set.
2
The Putnam problems
1. Show that for non-negative reals a1 , . . . , an and b1 , . . . , bn ,
Solution: If any ai is 0, the result is trivial, so we may assume all ai > 0. Dividing through
by (a1 . . . an )1/n , the inequality becomes
where ek is the sum of the products of the ci s, taken k at a time. So it is enough to show
that for each k,
n X Y
(c1 . . . cn )k/n ci .
k
A{1,...,n}, |A|=k iA
Q
We apply the AM-GM inequality to the numbers iA ci as A ranges over all subsets of size
k of {1, . . . , n}. Note that each ai appears exactly n1
k1 times in all these numbers. So we
we get P Q
n1 n |A|=k iA ci
(c . . . c )(k1) (k )
/ A{1,...,n},
1 n . n
k
n1 n
Since k1 / k = k/n, this is the same as
P Q
k/n A{1,...,n}, |A|=k iA ci
(c1 . . . cn ) n
,
k
2. Minimize
9 2
p
2 2
(u v) + 2u
v
in the range 0 < u < 2, v > 0.
3
completely above its tangent line, while the circle (in the first quadrant) lies completely below
its tangent line; so the distance between ant other two points is at least the distance between
the two tangent lines.
(m + n)! m! n!
< m n.
(m + n)m+n m n
Since all terms are strictly positive, we get the required inequality.
Solution: We estimate the integral of ln x, which is convex and hence easy to estimate. Take
the integral from 1 to 2n 1. This is less than 2(ln 3 + ln 5 + . . . + ln(2n 1)). But the
antiderivative of ln x is x ln x x, so the integral evaluates to (2n 1) ln(2n 1) 2n + 2.
Hence (2n1) ln(2n1)(2n1) < (2n1) ln(2n1)2n+2 < 2(ln 3+ln 5+. . .+ln(2n1)).
Exponentiating gives the right-hand inequality.
4
Similarly, the integral from e to 2n + 1 is greater than 2(ln 3 + ln 5 + . . . + ln(2n 1)), and an
explicit evaluation of the antiderivative here leads to the right-hand side of the inequality. The
choice of lower bound e for the integral here is just the righ thing to make the computations
work out nicely.
6. Let f be a real function with a continuous third derivative such that f (x), f 0 (x), f 00 (x) and
f 000 (x) are positive for all x. Suppose that f 000 (x) f (x) for all x. Show that f 0 (x) < 2f (x)
for all x.
Solution: See Kedlaya, Poonen & Vakil, The William Lowell Putnam Mathematical Com-
petition 19852000; Problems, Solutions and Commentary, page 272. This was one of the
hardest Putnam Competition problems ever of the top 205 performers in the 1999 Putnam,
only one contestant received a score of more than 0 for this problem, and that score was 2!
7. Maximize Z y p
x4 + (y y 2 )2 dx
0
on the interval [0, 1].
Ryp
Solution: Let f (y) = 0 x4 + (y y 2 )2 dx. At y = 0, f (y) = 0, and at y = 1, f (y) =
R1 2
0 x dx = 1/3. Also, f is non-negative on [0, 1]. It looks like it will be impossible to evaluate
f (y) at any value other than y = 0, 1; this strongly leads to the suspicion that the maximum
is at y = 1. To prove this, we need to show that f 0 (y) 0 on the interval [0, 1]. A process by
which this can be done is outlined in Kedlaya, Poonen & Vakil, The William Lowell Putnam
Mathematical Competition 19852000; Problems, Solutions and Commentary, page 138.
Heres a (really) slick solution from the same source: Since x2 and y y 2 are positive in the
range 0 x 1, 0 y 1, we have
and so Z y Z y
p 2
x2 + (y y 2 ) dx = y 2 y 3 .
x4 + (y y 2 )2 dx
0 0 3
The derivative of + y2 (2/3)y 3
is 2y 2y 2
= 2y(1 y) which is non-negative on [0, 1], so the
maximum is at y = 1, where it is 1/3. So
Z yp
x4 + (y y 2 )2 dx 1/3;
0