Luc TARTAR to John MACKEY, September 16, 2009

Remark 1: Your last exercise yesterday was about 1+sin x 22

cos x = a (with a = 7 ), and one must assume that the
0
indetermination 0 is excluded, so that one does not have cos x = 0 and sin x = −1; multiplying by cos x
gives 1 + sin x = a cos x, and then multiplying by 1 − sin x gives cos2 x = a cos x (1 − sin x), so that

after
dividing by cos x (which is 6= 0) one obtains cos x = a (1 − sin x); eliminating sin x gives 2 = a + a1 cos x,
2a2 a2 −1
so that cos x = a22a
+1 , and then 1 + sin x = a2 +1 , so that sin x = a2 +1 .
If one does not see the trick of multiplying by 1 − sin x, one may put cos x = z (which √ is better than
using sin x as the unknown, because a particular solution is cos x = 0), and one obtains 1 ± 1 − z 2 = a z,
2a
so that by taking the squares 1 − z 2 = 1 − 2a z + a2 z 2 , and dividing by z gives z = 1+a 2.

Remark 2: For the problem of the cube with vertices on the sphere S2 , one may say that a particular cube
is Q0 having vertices A0 , B0 , . . . , F0 , and that one finds all the other configurations by choosing a rotation
P ∈ SO(3), and considering the cube Q having vertices A = P A0 , B = P B0 , . . . , F = P F0 .
Implicitly, you assumed that there is a probability µ on SO(3) such that, for every point A ∈ S2 , the
image νA of µ by the mapping P 7→ P A is the uniform probability on S2 .
Since SO(3) is a compact (nonAbelian) Lie group (of dimension 3, as a manifold), it has a Haar measure
π (nonzero and unique up to multiplication by a positive constant) invariant by left action of the group, so
that the image of π by the mapping P 7→ R P with R ∈ SO(3) is π; by multiplying π by a suitable positive
constant one obtains a probability µ, and then νA is also the image of µ by the mapping P 7→ R P A, i.e.
νA is invariant by multiplication on the left by R, so that it is the uniform probability of S2 .

For your problem set 4 on September 15,

Your problem 1: The only strategy that I see is to go up and drop the first ball at floors a1 , a2 , . . . , an = 100,
until the first ball is shattered; if it is shattered at level am , then one goes down to the precding level am−1
(or the ground floor if a1 > 1, and one lets a0 = 0), and one then tries levels am−1 + 1, am−1 + 2, . . . , am − 1
until one finds the level k where it is shattered. In that way, one may have used m + am − am−1 − 1 drops. If
the optimal number expected is `, the best choice for the first level is `; if one continues, the next best floor
is ` + (` − 1), and so on, so that if there were enough floors one could go to level ` + (` − 1) + . . . + 1 = ` (`+1)
2 ;
` (`+1)
one must then have 2 ≥ 100, i.e. ` ≥ 14. The optimal number is then 14, and the successive levels are
14, 27, 39, 50, 60, 69, 77, 84, 90, 95, 99, 100 (but if there were 105 floors, the end of the list would be 99,
102, 104, 105).
Your problem 2 (Putnam 1978-B1): Find the area of a convex octagon that is inscribed in a circle and has
four consecutive
√ sides of length 3 units and the remaining four sides of length 2 units. Give the answer in
the form r + s t with r, s, and t positive integers.
Hint: If 2α and 2β are the angles under which one sees sides of length 2 and 3 from the center of the circle of
radius R, i.e. 2 = 2R sin α and 3 = 2R sin β, then the hypothesis is α + β = π4 . One deduces that sin α = R1
√
and sin β = sin π4 − α = cos α−sin α 3 2+3 2


√
2
= 2R ; using the value of sin α, one finds that cos α = 2R , so that
√ 2 √
R2 = (R sin α)2 + (R cos α)2 = 1 + (2+34 2) = 26+12 4
2
. One deduces that R2 sin 2α = 2R sin α R cos α =
√ √ 2 √ √
2 + 3 2 and R2 sin 2β = R2 cos 2α = (R cos α)2 − (R sin α)2 = (2+34 2) − 1 = 18+12 2
= 9+6 2
. Since the
√ √ 4 √2
area of the octagon is 4R2 sin α cos α + 4R2 sin β cos β, it is 2(2 + 3 2) + 9 + 6 2 = 13 + 12 2.
Second method (used in class by Richard BIGGS, I think): One reorders the 8 triangles to alternate the sides
of length 3 and the√sides of length 2, and this does not change √ the area. One obtains a Latin cross of width
3 and length 3 + 2 2, because the corners should have size 2 so that the hypotenuse√ has length 2. For the
area, one
√ finds an inner square of side 3, and
√ area 9, four rectangles of width 2 and length 3, for a√total
area 12 2, and four half squares of side 2, for a total area 4; adding these gives the area 13 + 12 2 for
the octagon.
Your problem 3 (Putnam 1980-A3): Evaluate

Z π/2
dx
√ .
1 + (tan x) 2
0

1
 √
Hint: One replaces 2 by any real number r, and one notices that the integral is finite. One makes the
R π/2 R π/2 dy
R π/2 (tan y)r
change of variable x = π2 − y, and one finds Ir = 0 1+(tan dx
x)r = 0 1+(cot y)r = 0 1+(tan y)r dy =
π
R π/2 dy π π
2 − 0 1+(tan y)r = 2 − Ir , so that Ir = 4 .
More generally, for any continuous function f on an interval [a, b], such that f (x)+f (a+b−x) does not vanish
Rb
on [a, b] (or vanishes in such a way that the integral considered exists), one has a f (x)+ff (x) b−a
(a+b−x) dx = 2 ,
by using the change of variable x = a + b − y. The same trick is used for
Putnam 1987-B1: Evaluate Z 4 p
ln(9 − x) dx
p p .
2 ln(9 − x) + ln(x + 3)
p
Hint:
p Defining f by f (x) = log(9 − x) on [2, 4], the symmetry change of variable x = 6 − y gives f (6 − y) =
log(y + 3), so that this is an example analogous to 1980-A3: as f is positive on [2, 4], the integral exists
and is equal to 1, i.e. half the length of the interval.
Your problem 4 (Putnam 1986-A1): Find, with explanation, the maximum value of f (x) = x3 − 3x on the
set of all real numbers x satisfying x4 + 36 ≤ 13x2 .
I had not written a solution before: Because f 0 (x) = 3(x2 − 1), f increases from −∞ to −1 with f (−1) = 2,
decreases from −1 to +1 with f (+1) = −2, and then increases from +1 to +∞. The inequality x4 +36 ≤ 13x2

2  2 13 
is equivalent to x2 − 132 ≤ 25 5 13 5 2 13 5 2
4 , i.e. x − 2 ≤ 2 , i.e. 2 − 2 ≤ x ≤ 2 + 2 , or 4 ≤ x ≤ 9, so that the
 
set on which one wants to maximize f is made up of two intervals, −3 ≤ x ≤ −2 and 2 ≤ x ≤ 3. On the
interval −3 ≤ x ≤ −2, f increases from −18 to −2, while on the interval 2 ≤ x ≤ 3, f increases from +2 to
+18, so that the maximum value is 18.
Your problem 5 (Putnam 1996-A1): Find the least number A such that for any two squares of combined
area 1, a rectangle of area A exists such that the two squares can be packed into that rectangle (without the
interiors of the squares overlapping). You may assume that the sides of the squares will be parallel to the
sides of the rectangle.
Hint: One must have A ≥ a (a + b) whenever a ≥ b > 0 with a2 + b2 = 1. Using a = cos θ and b = sin θ with
2
0 ≤ θ ≤ π/4, A must be the maximum of cos θ (cos θ+sin θ), i.e. 2A is the maximum
√ of 2 cos θ+2 cos θ sin θ =
1 + cos √
2θ + sin 2θ; the maximum is attained for 2θ = π/4, giving 2A = 1 + 2, so that the least value is
A = 1+2 2 .
Your problem 6 (Putnam 1993-A1): The horizontal line y = c intersects the curve y = 2x − x3 in the first
quadrant as in the figure. Find c so that the areas of the two shaded regions are equal.
I had not written q a solution qbefore: If f (x) = 2x − x3 , one has f 0 (x) = 2 − 3x2 , so that f increases from

4√2 √ √
x = 0 until x = 23 , with f 2
3 = 3 3 , and decreases then until x =
√ 2, with f ( 2) = 0. One then takes
√ √
0 < c < 43√23 , so that there are two values a, b with f (a) = f (b) = c and 0 < a < b < 2.
Ra Rb
If I guess correctly what the picture was, one wants to have a c − 0 f (x) dx = a f (x) dx − (b − a) c, i.e.
Rb 4

0
f (x) dx = b c, or b2 − b4 = b (2b − b3 ), and since b 6= 0, one obtains b2 = 34 , so that c = b (2 − b2 ) = 3√
4
3
.