Problems of the Week

These problems were downloaded from

https://www.physics.harvard.edu/academics/undergrad/problems
Introduction
For those of you looking for some problems and puzzles to brood over, Ill post a new problem here
each week, the solution to which Ill post the following week. !ome are new, and some are classics. I
wont limit them to physics, so many will be of the mathematical sort. "The pattern I seem to have
settled into is physics problems on the odd weeks, and math problems on the even weeks.# In most
cases, theyre $uite difficult. %fter all, I call them &'roblems of the (eek,& and not &'roblems of the
)our&*
Feel free to stop by my office "+yman ,--# if you get an answer or want any hints. )ave fun*
.any of the physics problems can be found scattered throughout the te/tbook for the 'hysics 01 course
here at )arvard.
David Morin
List of Problems
Week Date Problem Name
(eek 0 ,22,324301 'roblem 0 5asketball and tennis ball
(eek , ,22,3243,- 'roblem , 6reen3eyed dragons
(eek - ,22,3243-2 'roblem - 5alancing a pencil
(eek 7 ,22,302328 'roblem 7 'assing the spaghetti
(eek 9 ,22,302307 'roblem 9 The raindrop
(eek 1 ,22,3023,0 'roblem 1 Flipping a coin
(eek 8 ,22,3023,: 'roblem 8 .ountain climber
(eek : ,22,300327 'roblem : !ub3rectangles
(eek 4 ,22,300300 'roblem 4 Fractal moment
(eek 02 ,22,30030: 'roblem 02 'roduct of lengths
(eek 00 ,22,3003,9 'roblem 00 5reak or not break;
(eek 0, ,22,30,32, 'roblem 0, <ecreasing numbers
(eek 0- ,22,30,324 'roblem 0- =nchanged velocity
(eek 07 ,22,30,301 'roblem 07 Find the angles
(eek 09 ,22,30,3,- 'roblem 09 .a/imal gravity
(eek 01 ,22,30,3-2 'roblem 01 +etters in envelopes
(eek 08 ,22-320321 'roblem 08 Icosahedron of resistors
(eek 0: ,22-32030- 'roblem 0: <istribution of primes
(eek 04 ,22-3203,2 'roblem 04 5lock and bouncing ball
(eek ,2 ,22-3203,8 'roblem ,2 >ollinear points
(eek ,0 ,22-32,32- 'roblem ,0 5all on turntable
(eek ,, ,22-32,302 'roblem ,, Trading envelopes
(eek ,- ,22-32,308 'roblem ,- V"x# versus a hill
(eek ,7 ,22-32,3,7 'roblem ,7 ?erifying weights
(eek ,9 ,22-32-32- 'roblem ,9 .a/imum deflection angle
(eek ,1 ,22-32-302 'roblem ,1 <runken walk
(eek ,8 ,22-32-308 'roblem ,8 @elativistic cookies
(eek ,: ,22-32-3,7 'roblem ,: @ectangle in a circle
(eek ,4 ,22-32-3-0 'roblem ,4 5alls in a semicircle
(eek -2 ,22-327328 'roblem -2 <ifference of powers
(eek -0 ,22-327307 'roblem -0 !imultaneous claps
(eek -, ,22-3273,0 'roblem -, The game show
(eek -- ,22-3273,: 'roblem -- 5all rolling in a cone
(eek -7 ,22-329329 'roblem -7 >ounterfeit coin
(eek -9 ,22-32930, 'roblem -9 @ising hoop
(eek -1 ,22-329304 'roblem -1 .onochromatic triangle
(eek -8 ,22-3293,1 'roblem -8 5ouncing down a plane
(eek -: ,22-32132, 'roblem -: !um over 0
(eek -4 ,22-321324 'roblem -4 ?iewing the spokes
(eek 72 ,22-321301 'roblem 72 !taying ahead
(eek 70 ,22-3213,- 'roblem 70 !peedy travel
(eek 7, ,22-3213-2 'roblem 7, )ow much change;
(eek 7- ,22-328328 'roblem 7- Infinite %twoods machine
(eek 77 ,22-328307 'roblem 77 @elatively prime numbers
(eek 79 ,22-3283,0 'roblem 79 !liding along a plane
(eek 71 ,22-3283,: 'roblem 71 The birthday problem
(eek 78 ,22-32:327 'roblem 78 !liding ladder
(eek 7: ,22-32:300 'roblem 7: The hotel problem
(eek 74 ,22-32:30: 'roblem 74 Falling chimney
(eek 92 ,22-32:3,9 'roblem 92 A$ual segments
(eek 90 ,22-324320 'roblem 90 %ccelerating spaceship
(eek 9, ,22-32432: 'roblem 9, >onstruct the center
(eek 9- ,22-324309 'roblem 9- >ircles on the ice
(eek 97 ,22-3243,, 'roblem 97 @olling the die
(eek 99 ,22-3243,4 'roblem 99 Fi/ed highest point
(eek 91 ,22-302321 'roblem 91 !tirlings formula
(eek 98 ,22-30230- 'roblem 98 Throwing a beach ball
(eek 9: ,22-3023,2 'roblem 9: >oins and 6aussians
(eek 94 ,22-3023,8 'roblem 94 6etting way ahead
(eek 12 ,22-30032- 'roblem 12 >ereal bo/ prizes
(eek 10 ,22-300302 'roblem 10 Falling rope
(eek 1, ,22-300308 'roblem 1, +eftover dental floss
(eek 1- ,22-3003,7 'roblem 1- .inimal surface
(eek 17 ,22-30,320 'roblem 17 %ttracting bugs
(eek 19 ,22-30,32: 'roblem 19 @elativistic cart
(eek 11 ,22-30,309 'roblem 11 5owl of spaghetti
(eek 18 ,22-30,3,, 'roblem 18 Inverted pendulum
(eek 1: ,22-30,3,4 'roblem 1: Tower of circles
(eek 14 ,227320329 'roblem 14 >ompton scattering
(eek 82 ,22732030, 'roblem 82 'ainting a funnel
(eek 80 ,227320304 'roblem 80 .a/imum traBectory length
(eek 8, ,2273203,1 'roblem 8, Find the foci
(eek 8- ,22732,32, 'roblem 8- >hain on a scale
(eek 87 ,22732,324 'roblem 87 >omparing the numbers
(eek 89 ,22732,301 'roblem 89 )anging chain
(eek 81 ,22732,3,- 'roblem 81 >rawling ant
(eek 88 ,22732-320 'roblem 88 @elativistic momentum parado/
(eek 8: ,22732-32: 'roblem 8: Infinite s$uare roots
(eek 84 ,22732-309 'roblem 84 'ropelling a car
(eek :2 ,22732-3,, 'roblem :2 Cine divisible by 4
(eek :0 ,22732-3,4 'roblem :0 @ainbows
(eek :, ,227327329 'roblem :, !tanding in a line
(eek :- ,22732730, 'roblem :- The brachistochrone
(eek :7 ,227327304 'roblem :7 'oisson and 6aussian
(eek :9 ,2273273,1 'roblem :9 Tower of cylinders
(eek :1 ,22732932- 'roblem :1 !hifted intervals
(eek :8 ,227329302 'roblem :8 +eaving the hemisphere
(eek :: ,227329308 'roblem :: +adder envelope
(eek :4 ,2273293,7 'roblem :4 @ope between inclines
(eek 42 ,2273293-0 'roblem 42 The game of CI.
Week 1 (9/16/02)
Basketball and tennis ball
(a) A tennis ball with (small) mass m
2
sits on top of a basketball with (large)
mass m
1
. The bottom of the basketball is a height h above the ground, and
the bottom of the tennis ball is a height h + d above the ground. The balls
are dropped. To what height does the tennis ball bounce?
h
Note: Work in the approximation where m
1
is much larger than m
2
, and
assume that the balls bounce elastically.
(b) Now consider n balls, B
1
, . . . , B
n
, having masses m
1
, m
2
, . . . , m
n
(with
m
1
m
2
m
n
), sitting in a vertical stack. The bottom of B
1
is a
height h above the ground, and the bottom of B
n
is a height h + above the
ground. The balls are dropped. In terms of n, to what height does the top
ball bounce?
B
1
B
2
B
3
B
4
h
n= 4
Note: Work in the approximation where m
1
is much larger than m
2
, which is
much larger than m
3
, etc., and assume that the balls bounce elastically.
If h = 1 meter, what is the minimum number of balls needed for the top one to
bounce to a height of at least 1 kilometer? To reach escape velocity? Assume
that the balls still bounce elastically (which is a bit absurd here). Ignore wind
resistance, etc., and assume that is negligible.
Week 2 (9/23/02)
Green-eyed dragons
You visit a remote desert island inhabited by one hundred very friendly dragons,
all of whom have green eyes. They havent seen a human for many centuries and
are very excited about your visit. They show you around their island and tell you
all about their dragon way of life (dragons can talk, of course).
They seem to be quite normal, as far as dragons go, but then you nd out
something rather odd. They have a rule on the island which states that if a dragon
ever nds out that he/she has green eyes, then at precisely midnight on the day of
this discovery, he/she must relinquish all dragon powers and transform into a long-
tailed sparrow. However, there are no mirrors on the island, and they never talk
about eye color, so the dragons have been living in blissful ignorance throughout
the ages.
Upon your departure, all the dragons get together to see you o, and in a tearful
farewell you thank them for being such hospitable dragons. Then you decide to tell
them something that they all already know (for each can see the colors of the eyes of
the other dragons). You tell them all that at least one of them has green eyes. Then
you leave, not thinking of the consequences (if any). Assuming that the dragons are
(of course) infallibly logical, what happens?
If something interesting does happen, what exactly is the new information that
you gave the dragons?
Week 3 (9/30/02)
Balancing a pencil
Consider a pencil that stands upright on its tip and then falls over. Lets idealize
the pencil as a mass m sitting at the end of a massless rod of length .
1
(a) Assume that the pencil makes an initial (small) angle
0
with the vertical, and
that its initial angular speed is
0
. The angle will eventually become large,
but while it is small (so that sin ), what is as a function of time?
(b) You might think that it would be possible (theoretically, at least) to make the
pencil balance for an arbitrarily long time, by making the initial
0
and
0
suciently small.
It turns out that due to Heisenbergs uncertainty principle (which puts a con-
straint on how well we can know the position and momentum of a particle),
it is impossible to balance the pencil for more than a certain amount of time.
The point is that you cant be sure that the pencil is initially both at the top
and at rest. The goal of this problem is to be quantitative about this. The
time limit is sure to surprise you.
Without getting into quantum mechanics, lets just say that the uncertainty
principle says (up to factors of order 1) that xp h (where h = 1.06
10
34
Js is Plancks constant). The precise implications of this are some-
what vague, but well just take it to mean that the initial conditions satisfy
(
0
)(m
0
) h.
With this condition, nd the maximum time it can take your solution in part
(a) to become of order 1. In other words, determine (roughly) the maximum
time the pencil can balance. (Assume m = 0.01 kg, and = 0.1 m.)
1
It actually involves only a trivial modication to do the problem correctly using the moment of
inertia and the torque. But the point-mass version will be quite sucient for the present purposes.
Week 4 (10/7/02)
Passing the spaghetti
At a dinner party, there are N people seated around a table. A plate of spaghetti
starts at the head of the table. The person sitting there takes some spaghetti and
then passes the (very large) plate at random to his/her right or left. Henceforth each
person receiving the plate takes some spaghetti and then passes the plate at random
to his/her right or left. (Diners who have already received the plate can simply pass
it on, without taking any more.) When all the diners have nally received their
spaghetti, the plate stops being passed, and the eating begins.
(a) What are the chances of being the last to be served, as a function of position
(relative to the head) at the table of N people?
(b) If this procedure is repeated over the course of many dinners, what is the
average number of times the plate is passed?
Week 5 (10/14/02)
The raindrop
Assume that a cloud consists of tiny water droplets suspended (uniformly dis-
tributed, and at rest) in air, and consider a raindrop falling through them. What
is the acceleration of the raindrop? (Assume that when the raindrop hits a wa-
ter droplet, the droplets water gets added to the raindrop. Also, assume that the
raindrop is spherical at all times.)
Week 6 (10/21/02)
Flipping a coin
(a) Consider the following game. You ip a coin until you get a tails. The number
of dollars you win equals the number of coins you end up ipping. (So if you
immediately get a tails, you win one dollar; if you get one heads before a tails,
you win two dollars, etc.) What is the expectation value of your winnings?
(b) Play the same game, except now let the number of dollars you win be equal
to 2
n1
, where n is the number of coins you end up ipping. How much do
you expect to win now? Does your answer make sense?
Week 7 (10/28/02)
Mountain climber
A mountain climber wishes to climb up a frictionless conical mountain. He wants
to do this by throwing a lasso (a rope with a loop) over the top and climbing up
along the rope. Assume that the mountain climber is of negligible height, so that
the rope lies along the mountain, as shown.

At the bottom of the mountain are two stores. One sells cheap lassos (made
of a segment of rope tied to loop of rope of xed length). The other sells deluxe
lassos (made of one piece of rope with a loop of variable length; the loops length
may change without any friction of the rope with itself).
cheap
deluxe
When viewed from the side, this conical mountain has an angle at its peak.
For what angles can the climber climb up along the mountain if he uses:
(a) a cheap lasso?
(b) a deluxe lasso?
Week 8 (11/4/02)
Sub-rectangles
A rectangle is divided into many smaller rectangles, each of which has the prop-
erty that at least one pair of sides has integer length. Prove that the original
rectangle also has this property.
Week 9 (11/11/02)
Fractal moment
Take an equilateral triangle of side , and remove the middle triangle (1/4
of the area). Then remove the middle triangle from each of the remaining three
triangles (as shown), and so on, forever. Let the nal object have mass m. Find the
moment of inertia of this object, around an axis through its center and perpendicular
to its plane.
l
Week 10 (11/18/02)
Product of lengths
Inscribe a regular N-gon in a circle of radius 1. Draw the N 1 segments
connecting a given vertex to the N 1 other vertices. Show that the product of the
lengths of these N 1 segments equals N. The gure below shows the case where
N = 10; the product of the lengths of the 9 segments is 10.
R=1
N=10
Week 11 (11/25/02)
Break or not break?
Two spaceships oat in space and are at rest relative to each other. They are
connected by a string. The string is strong, but it cannot withstand an arbitrary
amount of stretching. At a given instant, the spaceships simultaneously (with re-
spect to their initial inertial frame) start accelerating (along the direction of the line
between them) with the same acceleration. (Assume they bought identical engines
from the same store, and they put them on the same setting.)
Will the string eventually break?
Week 12 (12/2/02)
Decreasing numbers
Pick a random number (evenly distributed) between 0 and 1. Continue picking
random numbers as long as they keep decreasing; stop picking when you obtain a
number that is greater than the previous one you picked. What is the expected
number of numbers you pick?
Week 13 (12/9/02)
Unchanged velocity
A ball rolls without slipping on a table. It rolls onto a piece of paper. You slide
the paper around in an arbitrary (horizontal) manner. (Its ne if there are abrupt,
jerky motions, so that the ball slips with respect to the paper.) After you allow the
ball to come o the paper, it will eventually resume rolling without slipping on the
table. Show that the nal velocity equals the initial velocity.
Week 14 (12/16/02)
Find the angles
Quadrilateral ABCD has angles

BDA =

CDB = 50

,

DAC = 20

, and

CAB = 80

. Find angles

BCA and

DBC.
20
80
50
50
A
B
D
C
Week 15 (12/23/02)
Maximal gravity
Given a point P in space, and given a piece of malleable material of constant
density, how should you shape and place the material in order to create the largest
possible gravitational eld at P?
Week 16 (12/30/02)
Letters in envelopes
You are given N addressed letters and N addressed envelopes. If you randomly put
one letter in each envelope, what is the probability that no letter ends up in the
correct envelope?
Week 17 (1/6/03)
Icosahedron of resistors
Each edge of an icosahedron is a 1 resistor. Find the eective resistance between
two adjacent vertices. (An icosahedron consists of 20 equilateral triangles. It has
12 vertices and 30 edges, with 5 edges meeting at each vertex.)
Week 18 (1/13/03)
Distribution of primes
Let P(N) be the probability that a randomly chosen integer, N, is prime. Show
that
P(N) =
1
lnN
.
Note: Assume that N is very large, and ignore terms in your answer that are of
subleading order in N. Also, make the assumption that the probability that N is
divisible by a prime p is exactly 1/p (which is essentially true, for a large enough
sample size of numbers).
Week 19 (1/20/03)
Block and bouncing ball
A block with large mass M slides with speed V
0
on a frictionless table towards a
wall. It collides elastically with a ball with small mass m, which is initially at rest
at a distance L from the wall. The ball slides towards the wall, bounces elastically,
and then proceeds to bounce back and forth between the block and the wall.
M
L
m
V
0
(a) How close does the block come to the wall?
(b) How many times does the ball bounce o the block, by the time the block
makes its closest approach to the wall?
Assume that M m, and give your answers to leading order in m/M.
Week 20 (1/27/03)
Collinear points
You are given a nite number of points in space with the property that any line
that contains two of these points contains three of them. Prove that all the points
must lie on a straight line.
Week 21 (2/3/03)
Ball on turntable
A ball with uniform mass density rolls without slipping on a turntable. Show that
the ball moves in a circle (as viewed from the inertial lab frame), with a frequency
equal to 2/7 times the frequency of the turntable.
Week 22 (2/10/03)
Trading envelopes
(a) I give you an envelope containing a certain amount of money, and you open
it. I then put into a second envelope either twice this amount or half this
amount, with a fty-fty chance of each. You are given the opportunity to
trade envelopes. Should you?
(b) I put two sealed envelopes on a table. One contains twice as much money
as the other. You pick an envelope and open it. You are then given the
opportunity to trade envelopes. Should you?
(c) If your answers to (a) and (b) are the same, explain why. If they are dierent,
explain why.
Week 23 (2/17/03)
V (x) versus a hill
A bead, under the inuence of gravity, slides along a frictionless wire whose height
is given by the function V (x). Find an expression for the beads horizontal acceler-
ation. (It can depend on whatever quantities you need it to depend on.)
You should nd that the result is not the same as the x for a particle moving in
one dimension in the potential mgV (x), in which case x = gV

. But if you grab

hold of the wire, is there any way you can move it so that the beads x is equal to
the x = gV

result due to the one-dimensional potential, mgV (x)?

Week 24 (2/24/03)
Verifying weights
(a) You have a balance scale and wish to verify the weights of items that come in
weights from 1 to 121 (all integral). What is the minimum number of xed
weights you need? What are the weights?
(b) Using n wisely-chosen xed weights, what is the largest integer W, for which
you can verify all the integral weights less than or equal to W? What xed
weights should you choose?
Week 25 (3/3/03)
Maximum deection angle
A mass M collides elastically with a stationary mass m. If M < m, then it is
possible for M to bounce directly backwards. However, if M > m, then there is a
maximum angle of deection of M. Find this angle.
Week 26 (3/10/03)
Drunken walk
A drunk performs a random walk along a street. At one end of the street is a river,
and at the other end is a police station. If he gets to either of these ends, he remains
there. He starts n steps from the river, and there are N total steps between the
river and the police station.
(a) What is the probability that he ends up at the river? At the police station?
(b) What is the expected total number of steps he takes?
Week 27 (3/17/03)
Relativistic cookies
Cookie dough (chocolate chip, of course) lies on a conveyor belt which moves along
at speed v. A circular stamp stamps out cookies as the dough rushed by beneath
it. When you buy these cookies in a store, what shape are they? That is, are they
squashed in the direction of the belt, stretched in that direction, or circular?
Week 28 (3/24/03)
Rectangle in a circle
Given a cyclic quadrilateral ABCD, draw the diagonals AC and BD. Prove that
the centers of the inscribed circles of triangles ABC, BCD, CDA, and DAB are
the vertices of a rectangle.
A
B
C
D
Week 29 (3/31/03)
Balls in a semicircle
N identical balls lie equally spaced in a semicircle on a frictionless horizontal table,
as shown. The total mass of these balls is M. Another ball of mass m approaches
the semicircle from the left, with the proper initial conditions so that it bounces
(elastically) o all N balls and nally leaves the semicircle, heading directly to the
left.
m
M Total mass
(a) In the limit N (so the mass of each ball in the semicircle, M/N, goes to
zero), nd the minimum value of M/m that allows the incoming ball to come
out heading directly to the left.
(b) In the minimum M/m case found in part (a), show that the ratio of ms nal
speed to initial speed equals e

.
Week 30 (4/7/03)
Dierence of Powers
Find the smallest positive value of 33
m
7
n
, where m and n are positive integers.
Week 31 (4/14/03)
Simultaneous claps
With respect to the ground, A moves to the right at speed c/

3, and B moves to
the left, also at speed c/

3. At the instant they are a distance L apart (as measured

in the ground frame), A claps his hands. B then claps his hands simultaneously (as
measured by B) with As clap. A then claps his hands simultaneously (as measured
by A) with Bs clap. B then claps his hands simultaneously (as measured by B)
with As second clap, and so on. As measured in the ground frame, how far apart
are A and B when A makes his nth clap? What is the answer if c/

3 is replaced
by a general speed v?
Week 32 (4/21/03)
The game show
A game show host oers you the choice of three doors. Behind one of these is the
grand prize, and behind the other two are goats. The host announces that after
you select a door (without opening it), he will open one of the other two doors and
reveal a goat. You select a door. The host then opens one of the other doors and
reveals a goat, and oers you the chance to switch your choice to the remaining
door. Should you switch?
Week 33 (4/28/03)
Ball rolling in a cone
A ball (with moment of inertia I = (2/5)MR
2
) rolls without slipping on the inside
surface of a xed cone, whose tip points downward. The half-angle at the vertex of
the cone is . Initial conditions have been set up so that the ball travels around the
cone in a horizontal circle of radius , with the contact points (the points on the
ball that touch the cone) tracing out a circle (not necessarily a great circle) on the
ball.
What should the radius of the circle of these contact points be, if you want
the sphere to travel around the cone as fast as possible? (You may work in the
approximation where R is much less than . Also, assume that the coecient of
friction between the ball and the cone is arbitrarily large.)
Week 34 (5/5/03)
Counterfeit coin
(a) You are given twelve coins, eleven of which have the same weight, and one
of which has a weight dierent from the others (either heavier or lighter, you
do not know). You have a balance scale. What is the minimum number of
weighings it takes to determine which coin has the dierent weight, and also
whether it is heavier or lighter than the rest?
(b) You are given N coins, N 1 of which have the same weight, and one of
which has a weight dierent from the others (either heavier or lighter, you do
not know). You are allowed W weighings on a balance scale. What is the
maximum value for N, as a function of W, for which you can determine which
coin has the dierent weight, and also whether it is heavy or light?
Week 35 (5/12/03)
Rising hoop
Two beads of mass m are positioned at the top of a frictionless hoop of mass M and
radius R, which stands vertically on the ground. The beads are given tiny kicks,
and they slide down the hoop, one to the right and one to the left, as shown. What
is the smallest value of m/M for which the hoop will rise up o the ground at some
time during the motion?
M
m m
Week 36 (5/19/03)
Monochromatic Triangle
(a) Seventeen points, no three of which are collinear, are connected by all the
possible lines between them (136, in fact). Each line is colored either red,
green, or blue. Prove that within the resulting network of lines, there is at
least one triangle all of whose sides are the same color.
(b) Let a denote the smallest integer greater than or equal to a. Let n!e points,
no three of which are collinear, be connected by all the possible lines between
them. Each line is colored one of n colors. Prove that within the resulting
network of lines, there is at least one triangle all of whose sides are the same
color.
Week 37 (5/26/03)
Bouncing down a plane
Consider a ball (with moment of inertia I = (2/5)MR
2
) which bounces elastically
o a surface. Assume that the balls speed in the direction perpendicular to the
surface is the same before and after a bounce. Also, assume that the ball is made
of a type of rubber which allows it to not slip on the surface (which has friction)
during the bounce. (This implies that the angular and linear motions may aect
each other.)
The ball is projected from the surface of a plane which is inclined at angle .
The initial velocity of the ball is perpendicular to the plane and has magnitude V .
The initial angular velocity is zero. Find the component of the balls velocity along
the plane, immediately after the nth bounce.
Week 38 (6/2/03)
Sum over 1
(a) You are given a random number (evenly distributed) between 0 and 1. To
this, you add a second such random number. Keep adding numbers until the
sum exceeds 1, and then stop. How many numbers, on average, will you need?
(b) When the sum nally exceeds 1 and the game stops, what is the average result
for the sum?
Week 39 (6/9/03)
Viewing the spokes
A wheel with spokes rolls on the ground. A stationary camera takes a picture of the
wheel. Due to the nonzero exposure time of the camera, the spokes will generally
appear blurred. At what location(s) in the picture does (do) the spoke(s) not appear
blurred?
Week 40 (6/16/03)
Staying ahead
In a two-way election, candidate A receives a votes and candidate B receives b
votes, with a > b. If the ballots are removed one at a time from the ballot box, and
a running total of the score is kept, what is the probability that at all times As
sub-total is greater than or equal to Bs sub-total?
Week 41 (6/23/03)
Speedy travel
A straight tube is drilled between two points (not necessarily diametrically opposite)
on the earth. An object is dropped into the tube. How much time does it take to
reach the other end? Ignore friction, and assume (erroneously) that the density of
the earth is constant.
Week 42 (6/30/03)
How much change?
You are out shopping one day with $N, and you nd an item whose price has a
random value between $0 and $N. You buy as many of these items as you can with
your $N. What is the expected value of the money you have left over? (You may
assume that $N is large compared to a penny, so that the distribution of prices is
essentially continuous.)
Week 43 (7/7/03)
Innite Atwoods machine
Consider the innite Atwoods machine shown below. A string passes over each
pulley, with one end attached to a mass and the other end attached to another
pulley. All the masses are equal to m, and all the pulleys and strings are massless.
The masses are held xed and then simultaneously released. What is the acceleration
of the top mass?
m
m
m
.
.
.
.
(Well dene this innite system as follows. Consider it to be made of N pulleys,
with a non-zero mass replacing what would have been the (N + 1)st pulley. Then
take the limit as N .)
Week 44 (7/14/03)
Relatively prime numbers
What is the probability that two randomly chosen positive integers are relatively
prime?
Week 45 (7/21/03)
Sliding along a plane
A block is placed on a plane inclined at angle . The coecient of friction between
the block and the plane is = tan. The block is given a kick so that it initially
moves with speed V horizontally along the plane (that is, in the direction perpen-
dicular to the direction pointing straight down the plane). What is the speed of the
block after a very long time?
Week 46 (7/28/03)
The birthday problem
(a) How many people must be in a room in order for the probability to be greater
than 1/2 that at least two of them have the same birthday? (By same birth-
day, we mean the same day of the year; the year may dier.) Ignore leap
years.
(b) Assume there is some large number, N, of days in a year. How many people
are now necessary for the odds to favor a common birthday? Equivalently,
assuming a normal 365-day year, how many people are required for the prob-
ability to be greater than 1/2 that at least two of them were born in the same
hour on the same date? Or in the same minute of the same hour on the same
date? Neglect terms in your answer that are of subleading order in N.
Week 47 (8/4/03)
Sliding ladder
A ladder of length and uniform mass density stands on a frictionless oor and
leans against a frictionless wall. It is initially held motionless, with its bottom end
an innitesimal distance from the wall. It is then released, whereupon the bottom
end slides away from the wall, and the top end slides down the wall. When it loses
contact with the wall, what is the horizontal component of the velocity of the center
of mass?
l
Week 48 (8/11/03)
The hotel problem
You are driving down a one-way road and pass a strip of a large number, N, of
hotels. These all have dierent rates, arranged randomly. You want to maximize
your chance of choosing the cheapest hotel, but you cant return to one youve passed
up. Assume that your only goal is to obtain the cheapest one (the second cheapest
is of no more value to you than the most expensive). If your strategy is to proceed
past a certain fraction, x, of them and then pick the next one that is cheaper than
all the ones youve seen so far, what should x be? What, then, is the probability of
success? Assume that N is very large, and ignore terms in your answer that are of
subleading order in N.
Week 49 (8/18/03)
Falling chimney
A chimney initially stands upright. It is given a tiny kick, and it topples over. At
what point along its length is it most likely to break?
In doing this problem, work with the following two-dimensional simplied model
of a chimney. Assume that the chimney consists of boards stacked on top of each
other, and that each board is attached to the two adjacent ones with tiny rods at
each end, as shown below. The goal is to determine which rod in the chimney has
the maximum tension. (Work in the approximation where the width of the chimney
is very small compared to the height.)
Week 50 (8/25/03)
Equal segments
Given a line segment, a line parallel to it, and a straightedge, divide the segment
into N equal segments. (With a straightedge, you are allowed only to draw straight
lines. You are not allowed to mark o distances on the straightedge.)
Week 51 (9/1/03)
Accelerating spaceship
A spaceship is initially at rest with respect to frame S. At a given instant, it starts
to accelerate with constant proper acceleration, a. (The proper acceleration is the
acceleration with respect to the instantaneous inertial frame the spaceship was just
in. Equivalently, if an astronaut has mass m and is standing on a scale, then the
scale reads a force of F = ma.) What is the relative speed of the spaceship and
frame S when the spaceships clock reads time t?
Week 52 (9/8/03)
Construct the center
Construct the center of a given circle, using only a compass. With a compass, you
are allowed to mark points with the needle, and to draw arcs of circles (which may
intersect at new points).
Week 53 (9/15/03)
Circles on the ice
A puck slides with speed v on frictionless ice. The surface is level, in the sense
that it is perpendicular to the direction of a hanging plumb bob at all points. Show
that the puck moves in a circle, as seen in the earths rotating frame. What is the
radius of the circle? What is the frequency of the motion? Assume that the radius
of the circle is small compared to the radius of the earth.
Week 54 (9/22/03)
Rolling the die
Two players alternately roll an N-sided die. The player who fails to improve upon
the previous roll loses. What is the probability that the rst player wins?
Week 55 (9/29/03)
Fixed highest point
Consider a top made of a uniform disk of radius R, connected to the origin by a
massless stick (which is perpendicular to the disk) of length , as shown below.
Paint a dot on the top at its highest point, and label this as point P. You wish to
set up uniform circular precession, with the stick making a constant angle with
the vertical, and with P always being the highest point on the top. What relation
between R and must be satised for this motion to be possible? What is the
frequency of precession, ?
l
P

R
Week 56 (10/6/03)
Stirlings formula
Using N! =

0
x
N
e
x
dx (which you can prove by induction), derive Stirlings
formula,
N! N
N
e
N

2N.
Also, nd the order-1/N correction to this (and the order-1/N
2
correction, if you
really want to).
Week 57 (10/13/03)
Throwing a beach ball
A beach ball is thrown upward with initial speed v
0
. Assume that the drag force
from the air is F = mv. What is the speed of the ball, v
f
, when it hits the
ground? (An implicit equation is sucient.) Does the ball spend more time or less
time in the air than it would if it were thrown in vacuum?
Week 58 (10/20/03)
Coins and Gaussians
Flip a coin 2N times, where N is large. Let P(x) be the probability of obtaining
exactly N +x heads. Show that
P(x)
e
x
2
/N

N
.
Week 59 (10/27/03)
Getting way ahead
A rocket with proper length L accelerates from rest, with proper acceleration g
(where gL c
2
). Clocks are located at the front and back of the rocket. If we look
at this setup in the frame of the rocket, then the general-relativistic time-dilation
eect tells us that the times on the two clocks are related by t
f
= (1 + gL/c
2
)t
b
.
Therefore, if we look at things in the ground frame, then the times on the two clocks
are related by
t
f
= t
b

1 +
gL
c
2

Lv
c
2
,
where the last term is the standard special-relativistic lack-of-simultaneity result.
Derive the above relation by working entirely in the ground frame.
Note: You may nd this relation surprising, because it implies that the front
clock will eventually be an arbitrarily large time ahead of the back clock, in the
ground frame. (The subtractive Lv/c
2
term is bounded by L/c and will therefore
eventually become negligible compared to the additive, and unbounded, (gL/c
2
)t
b
term.) But both clocks are doing basically the same thing relative to the ground
frame, so how can they eventually dier by so much? Your job is to nd out.
Week 60 (11/3/03)
Cereal box prizes
In each box of a certain cereal there is a prize, which is one of N colors. Assuming
that the prize in each box has equal odds of being any color (even after youve
bought many boxes), how many boxes do you need to buy, on average, before you
collect all the dierent colors?
Let P(n) be the probability that you collect the nal color in the nth box. For what
value of n, in terms of N, is P(n) maximum? Assume that N is large, and ignore
terms in your answer that are of subleading order in N.
Week 61 (11/10/03)
Falling rope
(a) A rope of length L lies in a straight line on a frictionless table, except for a
very small piece at one end which hangs down through a hole in the table.
This piece is released, and the rope slides down through the hole. What is the
speed of the rope at the instant it loses contact with the table?
(b) A rope of length L lies in a heap on a table, except for a very small piece at
one end which hangs down through a hole in the table. This piece is released,
and the rope unravels and slides down through the hole. What is the speed of
the rope at the instant it loses contact with the table? (Assume that the rope
is greased, so that it has no friction with itself.)
Week 62 (11/17/03)
Leftover dental oss
Two rolls of dental oss initially have equal lengths, L. Each day, a person chooses
one of the rolls at random and cuts o a xed small length, d. This continues until
one of the rolls runs out of oss. How much oss, on average, is left on the other roll
at this time? Assume that N L/d is very large, and ignore terms in your answer
that are of subleading order in N.
Week 63 (11/24/03)
Minimal surface
Consider a soap bubble that stretches between two identical circular rings of radius
r, as shown below. The planes of the rings are parallel, and the distance between
them is 2. Find the shape of the soap bubble.
What is the largest value of /r for which a stable soap bubble exists? You will have
to solve something numerically here.
r r
2l
Week 64 (12/1/03)
Attracting bugs
N bugs are initially located at the vertices of a regular N-gon, whose sides have
length . At a given moment, they all begin crawling with equal speeds in the
clockwise direction, directly toward the adjacent bug. They continue to walk directly
toward the adjacent bug, until they nally all meet at the center of the original N-
gon. What is the total distance each bug travels? How many times does each bug
spiral around the center?
Week 65 (12/8/03)
Relativistic cart
A long cart moves at relativistic speed v. Sand is dropped into the cart at a rate
dm/dt = in the ground frame. Assume that you stand on the ground next to
where the sand falls in, and you push on the cart to keep it moving at constant
speed v. What is the force between your feet and the ground? Calculate this force
in both the ground frame (your frame) and the cart frame, and show that the results
are equal (as should be the case for longitudinal forces).
Week 66 (12/15/03)
Bowl of spaghetti
A bowl contains N spaghetti noodles. You reach into the bowl and grab two free
ends at random and attach them. You do this N times until there are no free ends
left. On average, how many loops are formed by this process?
Week 67 (12/22/03)
Inverted pendulum
A pendulum consists of a mass m at the end of a massless stick of length . The
other end of the stick is made to oscillate vertically with a position given by y(t) =
Acos(t), where A . It turns out that if is large enough, and if the pendulum
is initially nearly upside-down, then it will, surprisingly, not fall over as time goes
by. Instead, it will (sort of) oscillate back and forth around the vertical position.
Explain why the pendulum doesnt fall over, and nd the frequency of the back and
forth motion.
Week 68 (12/29/03)
Tower of circles
Consider N circles stacked on top of each other inside an isosceles triangle, as shown
below for the case N = 4. Let A
C
be the sum of the areas of the N circles, and let
A
T
be the area of the triangle. In terms of N, what should the vertex angle, , be
so that the ratio A
C
/A
T
is maximized? Assume that N is large, and ignore terms
in your answer that are of subleading order in N.

Week 69 (1/5/04)
Compton scattering
A photon collides with a stationary electron. If the photon scatters at an angle ,
show that the resulting wavelength,

, is given in terms of the original wavelength,

, by

= +
h
mc
(1 cos ),
where m is the mass of the electron. Note: The energy of a photon is E = h = hc/.
Week 70 (1/12/04)
Painting a funnel
Consider the curve y = 1/x, from x = 1 to x = . Rotate this curve around the
x-axis to create a funnel-like surface of revolution. The volume of this funnel is
V =

x
2
dx = ,
which is nite. The surface area, however, is
A =

1
2

1 +y
2
x
dx >

1
2
x
dx,
which is innite. So it seems like you can ll up the funnel with paint, but you cant
paint it. What is the solution to this apparent paradox?
Week 71 (1/19/04)
Maximum trajectory length
A ball is thrown at speed v from zero height on level ground. At what angle should
it be thrown so that the distance traveled through the air is maximum. (You will
have to solve something numerically.)
Week 72 (1/26/04)
Find the foci
Using a straightedge and compass, construct (1) the foci of a given ellipse, (2) the
focus of a given parabola, and (3) the foci of a given hyperbola.
Week 73 (2/2/04)
Chain on a scale
A chain of length L and mass density is held such that it hangs vertically just
above a scale. It is then released. What is the reading on the scale, as a function of
the height of the top of the chain?
Week 74 (2/9/04)
Comparing the numbers
The numbers 1 through N are put in a hat. You and N 1 other people each pick
a number. You then compare your number with the others, one at a time, until you
nd one that is smaller than yours. This procedure is repeated many times. How
many numbers, on average, will you have to check in order to nd one that is smaller
than yours? (Ignore the situations where you have the number 1.) Consider two
cases:
(a) You ask the other people randomly. That is, at all times you have equal
probabilities of asking each person. This could be arranged, for example, by
demanding that you have a very bad memory, so that you may ask a given
person more than once.
(b) You have a good memory. In other words, you dont ask a given person more
than once.
Week 75 (2/16/04)
Hanging chain
A chain of uniform mass density per unit length hangs between two given points on
two walls. Find the shape of the chain. (Aside from an arbitrary additive constant,
the function describing the shape should contain one unknown constant.)
Week 76 (2/23/04)
Crawling ant
A rubber band with initial length L has one end tied to a wall. At t = 0, the other
end is pulled away from the wall at speed V (assume that the rubber band stretches
uniformly). At the same time, an ant located at the end not attached to the wall
begins to crawl toward the wall, with speed u relative to the band. Will the ant
reach the wall? If so, how much time will it take?
Week 77 (3/1/04)
Relativistic momentum paradox
Two equal masses are connected by a massless string with tension T. (By massless,
we mean that it has no mass in its unstretched, zero-length state.) The masses
are constrained to move with speed v along parallel lines, as shown below. The
constraints are then removed, and the masses are drawn together. They collide and
make one blob which continues to move to the right. Is the following reasoning
correct? If your answer is no, then state what is invalid about whichever of the
four sentences is/are invalid.
The forces on the masses point in the y-direction. Therefore, there is no change
in momentum in the x-direction. But the mass of the resulting blob is greater than
the sum of the initial masses (because they collided with some relative speed).
Therefore, the speed of the resulting blob must be less than v (to keep p
x
constant),
so the whole apparatus slows down in the x-direction.
v
v
T
Week 78 (3/8/04)
Innite square roots
Find the value of

17
16

17
16

1 .
Assume, when taking the limit, that deep within the nested square roots we choose
to begin with a 1 (starting with a 17/16 would yield a complex number).
Week 79 (3/15/04)
Propelling a car
For some odd reason, you decide to throw baseballs at a car of mass M, which is free
to move frictionlessly on the ground. You throw the balls at the back of the car at
speed u, and at a mass rate of kg/s (assume the rate is continuous, for simplicity).
If the car starts at rest, nd its speed and position as a function of time, assuming
that the back window is open, so that the balls collect inside the car.
Week 80 (3/22/04)
Nine divisible by 9
Given any seventeen integers, show that there is at least one subset of nine integers
whose sum is divisible by 9.
Week 81 (3/29/04)
Rainbows
Assuming that the index of refraction of water is 4/3 and that raindrops are spher-
ical, show that the location of a rainbow is approximately 42

above the line from

the sun to you. If you see a double rainbow, what is the angle of the second one?
Even triple rainbows are possible, although they are dicult to see; where is the
third one?
Week 82 (4/5/04)
Standing in a line
N people are standing in a line, facing forward down the line. How many of them,
on average, are able to say, I am taller than everyone in front of me.?
Week 83 (4/12/04)
The brachistochrone
A bead is released from rest at the origin and slides down a frictionless wire that
connects the origin to a given point, as shown. What shape should the wire take so
that the bead reaches the endpoint in the shortest possible time?
x
y
Week 84 (4/19/04)
Poisson and Gaussian
Throw N balls at random into B boxes. Let a be the average number of balls, N/B,
in a box. Let P(x) be the probability that a given box has exactly x balls in it.
(a) Show that
P(x)
a
x
e
a
x!
.
Certain assumptions are needed for this expression to be valid. What are
they?
(b) Show that if a is large, the above Poisson distribution essentially becomes a
Gaussian distribution,
P(x) =
a
x
e
a
x!

e
(xa)
2
/2a

2a
.
Week 85 (4/26/04)
Tower of cylinders
Consider the innitely tall system of identical massive cylinders and massless planks
shown below. The moment of inertia of the cylinders is I = MR
2
/2. There are two
cylinders at each level, and the number of levels is innite. The cylinders do not
slip with respect to the planks, but the bottom plank is free to slide on a table. If
you pull on the bottom plank so that it accelerates horizontally with acceleration a,
what is the horizontal acceleration of the bottom row of cylinders?
.

.

.

.
.

.

.

.
a
Week 86 (5/3/04)
Shifted intervals
Let 1/N. Choose a number at random between 0 and 1. Choose a second number
between and 1 + . Choose a third number between 2 and 1 + 2. Continue this
process, until you choose an Nth number between 1 and 2 . What is the
probability that the rst number you choose is the smallest of all the numbers?
Assume that N is very large, and make suitable approximations.
Week 87 (5/10/04)
Leaving the hemisphere
A point particle of mass m sits at rest on top of a frictionless hemisphere of mass
M, which rests on a frictionless table, as shown. The particle is given a tiny kick
and slides down the hemisphere. At what angle (measured from the top of the
hemisphere) does the particle lose contact with the hemisphere?
In answering this question for m = M, it is sucient for you to produce an
equation that must satisfy (it will be a cubic). However, for the special case of
m = M, this equation can be solved without too much diculty; nd the angle in
this case.
M
m
Week 88 (5/17/04)
Ladder envelope
A ladder initially stands vertically against a wall. Its bottom end is given a sideways
kick, causing the ladder to slide down. Assume that the bottom end is constrained
to keep contact with the ground, and the top end is constrained to keep contact
with the wall. Describe the envelope of the ladders positions.
Week 89 (5/24/04)
Rope between inclines
A rope rests on two platforms which are both inclined at an angle (which you are
free to pick), as shown. The rope has uniform mass density, and its coecient of
friction with the platforms is 1. The system has left-right symmetry. What is the
largest possible fraction of the rope that does not touch the platforms? What angle
allows this maximum value?

Week 90 (5/31/04)
The game of NIM
Determine the best strategy for each player in the following two-player game. There
are three piles, each of which contains some number of coins. Players alternate turns,
each turn consisting of removing any (non-zero) number of coins from a single pile.
The goal is to be the person to remove the last coin(s).
Solution
Week 1 (9/16/02)
Basketball and tennis ball
(a) For simplicity, assume that the balls are separated by a very small distance,
so that the relevant bounces happen a short time apart. This assumption isnt
necessary, but it makes for a slightly cleaner solution.
Just before the basketball hits the ground, both balls are moving downward
with speed (using mv
2
/2 = mgh)
v =

2gh. (1)
Just after the basketball bounces o the ground, it moves upward with speed
v, while the tennis ball still moves downward with speed v. The relative speed
is therefore 2v. After the balls bounce o each other, the relative speed is still
2v. (This is clear if you look at things in the frame of the basketball, which is
essentially a brick wall.
1
) Since the upward speed of the basketball essentially
stays equal to v, the upward speed of the tennis ball is 2v + v = 3v. By
conservation of energy, it will therefore rise to a height of H = d +(3v)
2
/(2g).
But v
2
= 2gh, so we have
H = d + 9h. (2)
(b) Just before B
1
hits the ground, all of the balls are moving downward with
speed v =

2gh.
We will inductively determine the speed of each ball after it bounces o the
one below it. If B
i
achieves a speed of v
i
after bouncing o B
i1
, then what
is the speed of B
i+1
after it bounces o B
i
? The relative speed of B
i+1
and
B
i
(right before they bounce) is v + v
i
. This is also the relative speed after
they bounce. Since B
i
is still moving upwards at essentially speed v
i
, the nal
upward speed of B
i+1
is therefore (v + v
i
) + v
i
. Thus,
v
i+1
= 2v
i
+ v. (3)
Since v
1
= v, we obtain v
2
= 3v (in agreement with part (a)), v
3
= 7v,
v
4
= 15v, etc. In general,
v
n
= (2
n
1)v, (4)
which is easily seen to satisfy eq. (3), with the initial value v
1
= v.
From conservation of energy, B
n
will bounce to a height of
H = +
((2
n
1)v)
2
2g
= + (2
n
1)
2
h. (5)
1
It turns out that the relative speed is the same before and after any elastic collision, independent
of what the masses are. This is easily seen by working in the center-of-mass frame, where the masses
simply reverse their velocities.
1
If h is 1 meter, and we want this height to equal 1000 meters, then (assuming
is not very large) we need 2
n
1 >

1000. Five balls wont quite do the

trick, but six will, and in this case the height is almost four kilometers.
Escape velocity from the earth (which is v
esc
=

2gR 11, 200 m/s) is reached

when
v
n
v
esc
= (2
n
1)

2gh

2gR = n ln
2

R
h
+ 1

. (6)
With R = 6.4 10
6
m and h = 1 m, we nd n 12. Of course, the elasticity
assumption is absurd in this case, as is the notion that one can nd 12 balls
with the property that m
1
m
2
m
12
.
2
Solution
Week 2 (9/23/02)
Green-eyed dragons
Lets start with a smaller number of dragons, N, instead of one hundred, to get
a feel for the problem.
If N = 1, and you tell this dragon that at least one of the dragons has green
eyes, then you are simply telling him that he has green eyes, so he must turn into a
sparrow at midnight.
If N = 2, let the dragons be called A and B. After your announcement that at
least one of them has green eyes, A will think to himself, If I do not have green
eyes, then B can see that I dont, so B will conclude that she must have green
eyes. She will therefore turn into a sparrow on the rst midnight. Therefore, if B
does not turn into a sparrow on the rst midnight, then on the following day A will
conclude that he himself must have green eyes, and so he will turn into a sparrow
on the second midnight. The same thought process will occur for B, so they will
both turn into sparrows on the second midnight.
If N = 3, let the dragons be called A, B, and C. After your announcement,
C will think to himself, If I do not have green eyes, then A and B can see that
I dont, so as far as they are concerned, they can use the reasoning for the N = 2
situation, in which case they will both turn into sparrows on the second midnight.
Therefore, if A and B do not turn into sparrows on the second midnight, then on
the third day C will conclude that he himself must have green eyes, and so he will
turn into a sparrow on the third midnight. The same thought process will occur for
A and B, so they will all turn into sparrows on the third midnight. The pattern
now seems clear.
Claim: Consider N dragons, all of whom have green eyes. If you announce to all
of them that at least one of them has green eyes, they will all turn into sparrows on
the Nth midnight.
Proof: We will prove this by induction. We will assume the result is true for N
dragons, and then we will show that it is true for N + 1 dragons. We saw above
that it holds for N = 1, 2, 3.
Consider N + 1 dragons, and pick one of them, called A. After your announce-
ment, she will think to herself, If I do not have green eyes, then the other N dragons
can see that I dont, so as far as they are concerned, they can use the reasoning
for the situation with N dragons, in which case they will all turn into sparrows on
the Nth midnight. Therefore, if they do not all turn into sparrows on the Nth
midnight, then on the (N + 1)st day A will conclude that she herself must have
green eyes, and so she will turn into a sparrow on the (N + 1)st midnight. The
same thought process will occur for the other N dragons, so they will all turn into
sparrows on the (N + 1)st midnight.
Hence, in our problem all one hundred dragons will turn into sparrows on the
100th midnight.
Although weve solved the problem, you may be troubled by the fact that your
seemingly useless information did indeed have major consequences. How could this
be, when surely all the dragons already knew what you told them? Did you re-
ally give them new information? The answer is yes. Lets see what this new
information is.
Consider the case N = 1. Here it is clear that you provided new information,
since you essentially told the one dragon that he has green eyes. But for the cases
N 2, the new information is slightly more subtle.
Consider the case N = 2. Prior to your announcement, A knows that B has
green eyes, and B knows that A has green eyes. That is the extent of the knowledge,
and they cant conclude anything else from it. But after you tell them that at least
one of them has green eyes, then A knows two things: He knows that B has green
eyes, and he knows that B knows that there is at least one dragon with green eyes
(because A knows that B heard your information). B gains a similar second piece
of information. This second piece of information is critical, as we saw above in the
reasoning for the N = 2 case.
Consider the case N = 3. A knows that B green eyes, and he also knows that
B knows that there is at least one dragon with greens eyes (because A can see that
B can see C). So the two bits of information in the N = 2 case above are already
known before you speak. What new information is gained after you speak? Only
after you speak is it true that A knows that B knows that C knows that there is at
least one dragon with green eyes.
The analogous result holds for a general number N. There is no paradox here.
Information is gained by your speaking. More information is added to the world
than the information you gave.
1
And it turns out, as seen in the proof of Claim 1,
that the new information is indeed enough to allow all the dragons to eventually
gure out their eye color.
To sum up: Before you make your announcement, the following statement is
true for N dragons: A
1
knows that A
2
knows that A
3
knows that . . . that A
N2
knows that A
N1
knows that there is at least one dragon with green eyes. This is
true because A
N1
can see A
N
; and A
N2
can see that A
N1
can see A
N
; and so
on, until lastly A
1
can see that A
2
can see that . . . that A
N1
can see A
N
. The
same result holds, of course, for any group of N 1 dragons. The point is that it
is only after you make your announcement that the chain is extended the nal step
to the Nth dragon. The fact that the Nth dragon heard your statement is critical
to the truth of this complete chain.
So, in the end, it turns out to be of great importance how far the chain, A
knows that B knows that C knows that . . . goes.
Note that if one of the dragons missed your farewell announcement (which was
At least one the 100 dragons on this island has green eyes), then they will all
happily remain dragons throughout the ages.
1
For example, A knows that you made your statement while stepping onto your boat and wearing
a blue shirt. Or, more relevantly, A knows that you made your statement in front of all the other
dragons. In short, its not just what you say; its how you say it.
Solution
Week 3 (9/30/02)
Balancing a pencil
(a) The component of gravity in the tangential direction is mg sin mg. There-
fore, the tangential F = ma equation is mg = m

, which may be written as

= (g/). The general solution to this equation is

(t) = Ae
t/
+ Be
t/
, where

/g. (1)
The constants A and B are found from the initial conditions,
(0) =
0
= A + B =
0
,

(0) =
0
= (AB)/ =
0
. (2)
Solving for A and B, and then plugging into eq. (1) gives
(t) =
1
2
(
0
+
0
) e
t/
+
1
2
(
0

0
) e
t/
. (3)
(b) The constants A and B will turn out to be small (they will each be of order

h). Therefore, by the time the positive exponential has increased enough to
make of order 1, the negative exponential will have become negligible. We
will therefore ignore the latter term from here on. In other words,
(t)
1
2
(
0
+
0
) e
t/
. (4)
The goal is to keep small for as long as possible. Hence, we want to min-
imize the coecient of the exponential, subject to the uncertainty-principle
constraint, (
0
)(m
0
) h. This constraint gives
0
h/(m
2

0
). Hence,
(t)
1
2

0
+
h
m
2

e
t/
. (5)
Taking the derivative with respect to
0
to minimize the coecient, we nd
that the minimum value occurs at

0
=

h
m
2
. (6)
Substituting this back into eq. (5) gives
(t)

h
m
2
e
t/
. (7)
Setting 1, and then solving for t gives (using

/g)
t
1
4

g
ln

m
2

3
g
h
2

. (8)
With the given values, m = 0.01 kg and = 0.1 m, along with g = 10 m/s
2
and h = 1.06 10
34
Js, we obtain
t
1
4
(0.1 s) ln(9 10
61
) 3.5 s. (9)
No matter how clever you are, and no matter how much money you spend
on the newest, cutting-edge pencil-balancing equipment, you can never get a
pencil to balance for more than about four seconds.
Remarks: This smallness of this answer is quite amazing. It is remarkable that a
quantum eect on a macroscopic object can produce an everyday value for a time
scale. Basically, the point here is that the fast exponential growth of (which gives
rise to the log in the nal result for t) wins out over the smallness of h, and produces
a result for t of order 1. When push comes to shove, exponential eects always win.
The above value for t depends strongly on and g, through the

/g term. But
the dependence on m, , and g in the log term is very weak. If m were increased
by a factor of 1000, for example, the result for t would increase by only about 10%.
Note that this implies that any factors of order 1 that we neglected throughout this
problem are completely irrelevant. They will appear in the argument of the log term,
and will thus have negligible eect.
Note that dimensional analysis, which is generally a very powerful tool, wont get you
too far in this problem. The quantity

/g has dimensions of time, and the quantity

m
2

3
g/h
2
is dimensionless (it is the only such quantity), so the balancing time
must take the form
t

g
f(), (10)
where f is some function. If the leading term in f were a power (even, for example,
a square root), then t would essentially be innite (t 10
30
s for the square root).
But f in fact turns out to be a log (which you cant determine without solving the
problem), which completely cancels out the smallness of h, reducing an essentially
innite time down to a few seconds.
Solution
Week 4 (10/7/02)
Passing the spaghetti
(a) For the case of n = 3, it is obvious that the two people not at the head of the
table have equal 1/2 chances of being the last served (BTLS).
For the case of n = 4, label the diners as A,B,C,D (with A being the head),
and consider Ds probability of BTLS. The various paths of spaghetti that
allow D to be the last served are:
ABC..., ABABC..., ABABABC..., etc. (1)
The sum of the probabilities of these is
1
2
2
+
1
2
4
+
1
2
6
+ =
1/4
1 1/4
=
1
3
. (2)
By symmetry, B also has a 1/3 chance of BLTS, and then that leaves a 1/3
chance for C. Hence, B, C, and D all have equal 1/3 chances of BLTS.
The probabilities for n = 5 are a bit tedious to calculate in this same manner,
so at this point we will (for lack of a better option) make the following guess:
Claim: For arbitrary n, all diners not at the head of the table have equal
1/(n 1) chances of being the last served (BTLS).
This seems a bit counterintuitive (because you might think that the diners
further from the head have a greater chance of BTLS), but it is in fact correct.
Proof: Two things must happen to a given diner for BTLS:
(1) The plate must approach the given diner from the right or left and reach
the person next to that diner.
(2) The plate must then reverse its direction and make its way (in whatever
manner) all the way around the table until it reaches the person on the
other side of the given diner.
For any of the (non-head) diners, the probability that the rst of these condi-
tions will be satised is 1. This condition will therefore not dierentiate the
probabilities of BTLS.
Given that (1) has happened, there is some denite probability of (2) hap-
pening, independent of where the diner is located. This is true because the
probability of traveling all the way around the table does not depend on where
this traveling starts. Hence, (2) also does not dierentiate between the n 1
(non-head) probabilities of BTLS.
Thus, all the n1 (non-head) probabilities of BTLS are equal, and are there-
fore equal to 1/(n 1).
(b) This problem is equivalent to asking how many steps it takes, on average, for
a random walk in one dimension to hit n sites.
Let f
n
be this expected number of steps. And let g
n
be dened as follows.
Assume that n sites have been visited, and that the present position is at one
of the ends of this string of n sites. Then g
n
is the expected number of steps
it takes to reach a new site.
We then have
f
n
= f
n1
+ g
n1
. (3)
This is true because in order to reach n sites, you must rst reach n 1 sites
(which takes f
n1
steps, on average). And then you must reach one more site,
starting at the end of the string of n 1 sites (which takes g
n1
steps, on
average).
Claim: g
n
= n.
Proof: Let the sites which have been visited be labeled 1, 2, . . . , n. Let the
present position be site 1.
There is a 1/2 chance that the next step will be to site number 0, in which
case it only takes one step to reach a new site.
There is a 1/2 chance that the next step will be to site number 2. By con-
sidering this site to be an end-site of the string 2, 3, . . . , n 1 (which has size
n 2), we see that it takes g
n2
steps (on average) to reach sites 1 or n. And
then from each of these, it takes of g
n
steps (on average) to reach a new site.
Putting this together gives g
n
=
1
2
(1) +
1
2
(1 + g
n2
+ g
n
), or
g
n
= g
n2
+ 2. (4)
Since we obviously have g
1
= 1, and since it is easy to see from the above
reasoning that g
2
= 2 (or equivalently, that g
0
= 0), we inductively obtain
g
n
= n.
Therefore, f
n
= f
n1
+ (n 1). Using f
1
= 0, we see by induction that
f
n
=
n(n 1)
2
. (5)
Solution
Week 5 (10/14/02)
The raindrop
Let be the mass density of the raindrop, and let be the average mass density
in space of the water droplets. Let r(t), M(t), and v(t) be the radius, mass, and
speed of the raindrop, respectively.
We need three equations to solve for the above three unknowns. The equations
we will use are two dierent expressions for dM/dt, and the F = dp/dt expression
for the raindrop.
The rst expression for

M is obtained by simply taking the derivative of M =
(4/3)r
3
, which gives

M = 4r
2
r (1)
= 3M
r
r
. (2)
The second expression for

M is obtained by noting that the change in M is due
to the acquisition of water droplets. The raindrop sweeps out volume at a rate given
by its cross-sectional area times its velocity. Therefore,

M = r
2
v. (3)
The force on the droplet is Mg, and the momentum is Mv. Therefore, F = dp/dt
gives
Mg =

Mv + M v. (4)
We now have three equations involving the three unknowns, r, M, and v.
1
Our goal is to nd v. We will do this by rst nding r. Eqs. (1) and (3) give
v =
4

r (5)
= v =
4

r. (6)
Plugging eqs. (2, 5, 6) into eq. (4) gives
Mg =

3M
r
r

+ M

. (7)
Therefore,
gr = 12 r
2
+ 4r r, (8)
where we have dened g g/, for convenience. The only parameter in eq. (8) is
g. Therefore, r(t) can depend only on g and t. Hence, by dimensional analysis, r
must take the form
r = A gt
2
, (9)
1
Note that we cannot write down the naive conservation-of-energy equation (which would say
that the decrease in the waters potential energy equals the increase in its kinetic energy), because
mechanical energy is not conserved. The collisions between the raindrop and the droplets are
completely inelastic. The raindrop will, in fact, heat up. See the remark at the end of the solution.
where A is a numerical constant, to be determined. Plugging this expression for r
into eq. (8) gives
g(A gt
2
) = 12(2A gt)
2
+ 4(A gt
2
)(2A g)
= A = 48A
2
+ 8A
2
. (10)
Therefore, A = 1/56, and so r = 2A g = g/28 = g/28. Eq. (6) then gives the
acceleration of the raindrop as
v =
g
7
, (11)
independent of and .
Remark: A common invalid solution to this problem is the following, which (incor-
rectly) uses conservation of energy.
The fact that v is proportional to r (shown in eq. (5)) means that the volume swept
out by the raindrop is a cone. The center-of-mass of a cone is 1/4 of the way from the base
to the apex (as you can show by integrating over horizontal circular slices). Therefore, if M
is the mass of the raindrop after it has fallen a height h, then an (incorrect) application of
conservation of energy gives
1
2
Mv
2
= Mg
h
4
= v
2
=
gh
2
. (12)
Taking the derivative of this (or equivalently, using v
2
= 2ad), we obtain
v =
g
4
. (Incorrect) (13)
The reason why this solution is invalid is that the collisions between the raindrop and
the droplets are completely inelastic. Heat is generated, and the overall kinetic energy of
the raindrop is smaller than you would otherwise expect.
Lets calculate how much mechanical energy is lost (and therefore how much the raindrop
heats up) as a function of the height fallen. The loss in mechanical energy is
E
lost
= Mg
h
4

1
2
Mv
2
. (14)
Using v
2
= 2(g/7)h, this becomes
E
int
= E
lost
=
3
28
Mgh, (15)
where E
int
is the gain in internal thermal energy.
The energy required to heat 1g of water by 1 C

is 1 calorie (= 4.18 Joules). Therefore,

the energy required to heat 1 kg of water by 1 C

is 4200 J. In other words,

E
int
= 4200 M T, (16)
where M is measured in kilograms, and T is measured in Celsius. Eqs. (15) and (16) give
the increase in temperature as a function of h,
4200 T =
3
28
gh. (17)
How far must the raindrop fall before it starts to boil? If we assume that the water
droplets temperature is near freezing, then the height through which the raindrop must fall
to have T = 100 C

is found to be
h 400 km. (18)
We have, of course, idealized the problem. But needless to say, there is no need to worry
about getting burned by the rain.
A typical value for h more like a few kilometers, which would raise the temperature only
by about one degree. This eect, of course, is washed out by many other factors.
Solution
Week 6 (10/21/02)
Flipping a coin
(a) There is a 1/2 chance that you win one dollar, a 1/4 chance that you win two
dollars, a 1/8 chance that you win three dollars, etc. Therefore, the average
value of your winnings is
1
2
+
2
4
+
3
8
+
4
16
+ . (1)
This may be written as

1
2
+
1
4
+
1
8
+
1
16
+

1
4
+
1
8
+
1
16
+

1
8
+
1
16
+

+ , (2)
which equals
(1) +

1
2

1
4

+ = 2. (3)
So you expect to win an average of two dollars each time you play this game.
(b) There is a 1/2 chance that you win one dollar, a 1/4 chance that you win two
dollars, a 1/8 chance that you win four dollars, etc. So the average value of
your winnings is now
1
2
+
2
4
+
4
8
+
8
16
+ =
1
2
+
1
2
+
1
2
+
1
2
+ = . (4)
You will quickly discover, however, that you will not win an innite amount
of money playing this game. We seem to have a paradox. The expectation
value is innite, but certainly no one is going to put up an innite amount of
money, or even a million dollars, for the opportunity to play this game once.
What is the solution to this paradox?
The answer is that an expectation value is dened to be an average over an
innite number of trials (or the limit towards an innite number), but you
are simply not going to play an innite number of games. In other words,
the calculated expectation value doesnt agree with your experiment, because
your experiment has nothing whatsoever to do with the precise denition of an
expectation value. To be sure, if you did somehow play an innite number of
games, then you would indeed have an innite average for your winnings. The
whole paradox arises from trying to make expectation value mean something
it doesnt.
This might not be a very satisfying explanation, so let us get a better feeling
for the problem by looking at a situation where someone plays N = 2
n
games.
How much money would a reasonable person be willing to put up front for
the opportunity to play these N games?
Well, in about 2
n1
games he will win one dollar; in about 2
n2
he will win two
dollars; in about 2
n3
games he will win four dollars; etc., until in about one
game he will win 2
n1
dollars. In addition, there are the fractional numbers
of games where he wins much larger quantities of money (for example, in
half a game he will win 2
n
dollars, etc.), and this is indeed where the innite
expectation value comes from, in the calculation above. But let us forget about
these for the moment, in order to just get a lower bound on what a reasonable
person should put on the table. Adding up the above cases gives the total
winnings as 2
n1
(1) +2
n2
(2) +2
n3
(4) + +1(2
n1
) = 2
n1
n. The average
value of these winnings in the N = 2
n
games is therefore 2
n1
n/2
n
= n/2 =
(log
2
N)/2.
A reasonable person should therefore expect to win at least (log
2
N)/2 dollars
per game. (By expect, we mean that if the player plays a very large number
of sets of N games, and then takes an average over these sets, he will win
at least 2
n1
n dollars per set.) This clearly increases with N, and goes to
innity as N goes to innity. It is nice to see that we can obtain this innite
limit without having to worry about what happens in the innite number
of fractional games. Remember, though, that this quantity, (log
2
N)/2, has
nothing to do with a true expectation value, which is only dened for N .
Someone may still not be satised and want to ask, But what if I play only
N games? I will never ever play another game. How much money do I expect
to win? The proper answer is that the question has no meaning. It is not
possible to dene how much one expects to win, if one is not willing to take
an average over a arbitrarily large number of trials.
Solution
Week 7 (10/28/02)
Mountain climber
(a) We will take advantage of the fact that a cone is at, in the sense that we
can make one out of a piece of paper, without crumpling the paper.
Cut the cone along a straight line emanating from the peak and passing
through the knot of the lasso, and roll the cone at onto a plane. Call the
resulting gure, which is a sector of a circle, S.
P P

If the cone is very sharp, then S will look like a thin pie piece. If the cone is
very wide, with a shallow slope, then S will look like a pie with a piece taken
out of it. Points on the straight-line boundaries of the sector S are identied
with each other. Let P be the location of the lassos knot. Then P appears
on each straight-line boundary, at equal distances from the tip of S. Let be
the angle of the sector S.
The key to this problem is to realize that the path of the lassos loop must
be a straight line on S, as shown in the above gure. (The rope will take the
shortest distance between two points since there is no friction. And rolling
the cone onto a plane does not change distances.) A straight line between the
two identied points P is possible if and only if the sector S is smaller than a
semicircle. The restriction on the mountain is therefore < 180

.
What is this restriction, in terms of the angle of the peak, ? Let C denote
a cross-sectional circle, a distance d (measured along the cone) from the top
of the mountain. A semicircular S implies that the circumference of C equals
d. This then implies that the radius of C equals d/2. Therefore,
sin(/2) <
d/2
d
=
1
2
= < 60

(1)
is the condition under which the mountain is climbable. In short, having
< 60

guarantees that there is a loop around the cone of shorter length than
the distance straight to the peak and back.
Remark: When viewed from the side, the rope will appear perpendicular to the
side of the mountain at the point opposite the lassos knot. A common mistake is to
assume that this implies < 90

. This is not the case, because the loop does not lie
in a plane. Lying in a plane, after all, would imply an elliptical loop; but the loop
must certainly have a kink in it where the knot is, since there must exist a vertical
component to the tension. (If we had posed the problem with a planar, triangular
mountain, then the answer would be < 90

.)
(b) Use the same strategy. Roll the cone onto a plane. If the mountain is very
steep, then the climbers position can fall by means of the loop growing larger.
If the mountain has a shallow slope, the climbers position can fall by means
of the loop growing smaller. The only situation in which the climber will not
fall is the one where the change in position of the knot along the mountain is
exactly compensated by the change in length of the loop.
In terms of the sector S in a plane, this condition requires that if we move
P a distance up (or down) along the mountain, the distance between the
identied points P decreases (or increases) by . In the above gure, we must
therefore have an equilateral triangle, and so = 60

.
What peak-angle does this correspond to? As above, let C be a cross-
sectional circle, a distance d (measured along the cone) from the top of the
mountain. Then = 60

implies that the circumference of C equals (/3)d.

This then implies that the radius of C equals d/6. Therefore,
sin(/2) =
d/6
d
=
1
6
= 19

(2)
is the condition under which the mountain is climbable. We see that there is
exactly one angle for which the climber can climb up along the mountain.
Remark: Another way to see the = 60

result is to note that the three directions

of rope emanating from the knot must all have the same tension, since the deluxe
lasso is one continuous piece of rope. They must therefore have 120

angles between
themselves (to provide zero net force on the massless knot). This implies that = 60

.
Further remarks: For each type of lasso, for what angles can the mountain be
climbed if the lasso is looped N times around the top of the mountain? The solution
here is similar to that above. For the cheap lasso of part (a), roll the cone N times
onto a plane, as shown for N = 4.
P P
N=4
The resulting gure S
N
is a sector of a circle divided into N equal sectors, each
representing a copy of the cone. As above, S
N
must be smaller than a semicircle.
The circumference of the circle C (dened above) must therefore be less than d/N.
Hence, the radius of C must be less than d/2N. Thus,
sin(/2) <
d/2N
d
=
1
2N
= < 2 sin
1

1
2N

. (3)
For the deluxe lasso of part (b), again roll the cone N times onto a plane. From
the reasoning in part (b), we must have N = 60

. The circumference of C must

therefore be d/3N, and so its radius must be d/6N. Therefore,
sin(/2) =
d/6N
d
=
1
6N
= = 2 sin
1

1
6N

. (4)
Solution
Week 8 (11/4/02)
Sub-rectangles
Put the rectangle in the x-y plane, with its sides parallel to the x and y axes,
and consider the function,
f(x, y) = e
2ix
e
2iy
. (1)
Claim: The integral of f(x, y) over a rectangle, whose sides are parallel to the
axes, is zero if and only if at least one pair of sides has integer length.
Proof:

b
a

d
c
e
2ix
e
2iy
dxdy =

b
a
e
2ix
dx

d
c
e
2iy
dy (2)
=
1
4
2

e
2ib
e
2ia

e
2id
e
2ic

=
e
2ia
e
2ic
4
2

e
2i(ba)
1

e
2i(dc)
1

.
This equals zero if and only if at least one of the factors in parentheses is zero. In
other words, it equals zero if and only if at least one of (b a) and (d c) is an
integer.
Since we are told that each of the smaller rectangles has at least one integer pair
of sides, the integral of f over each of these smaller rectangles is zero. Therefore,
the integral of f over the whole rectangle is also zero. Hence, from the Claim, we
see that the whole rectangle has at least one integer pair of sides.
Remarks:
1. This same procedure works for the analogous problem in higher dimensions. Given
an N-dimensional rectangular parallelepiped which is divided into smaller ones, each
of which has the property that at least one edge has integer length, then the original
parallelepiped must also have this property. In three dimensions, for example, this
follows from considering integrals of the function f(x, y, z) = e
2ix
e
2iy
e
2iz
, in the
same manner as above.
2. We can also be a bit more general and make the following statement. Given an
N-dimensional rectangular parallelepiped which is divided into smaller ones, each of
which has the property that at least n non-equivalent edges (that is, ones that
arent parallel) have integer lengths, then the original parallelepiped must also have
this property.
This follows from considering the

N
m

functions (where m N + 1 n) of the form,

f(x
j
1
, x
j
2
, . . . , x
j
m
) = e
2ix
j
1
e
2ix
j
2
e
2ix
j
m
, (3)
where the indices j
1
, j
2
, . . . j
m
are a subset of the indices 1, 2, . . . , N. (For example, the
previous remark deals with n = 1, m = N, and only

N
N

= 1 function.) Well leave it

to you to show that the integrals of these

N
m

functions, over an N-dimensional rect-

angular parallelepiped, are all equal to zero if and only if at least n non-equivalent
edges of the parallelepiped have integer length.
Solution
Week 9 (11/11/02)
Fractal moment
The strategy here will be to compute the moment of inertia by using scaling
arguments, along with the parallel-axis theorem. To do this, we will need to compare
the I for our triangle of side to that of a triangle of side 2. So let us scale up our
triangle by a factor of 2 and examine what happens to the integral I =

r
2
dm. We
get a simple factor of 2
2
from the r
2
, but happens to the dm?
A solid triangle would yield a factor of 2
2
= 4 in the dm (since area is propor-
tional to length squared), but our fractal object is a bit dierent. The mass scales
in a strange way. Doubling the size of our triangle increases its mass by a factor of
only 3. This is true because the doubled triangle is simply made up of three of the
smaller ones, plus an empty triangle in the middle. Thus, the dm picks up a factor
of 3, and so the I for a fractal triangle of side 2 is 4 3 = 12 times that of a fractal
triangle of side (where the axes pass through any two corresponding points).
In what follows, well use pictures to denote the Is of the fractal objects around
the dots shown. In terms of these pictures, we have,
l
m
=
=
12
2
3
2
(
(
)
)
+
___
l
l
= 3
The rst line comes from the scaling argument, the second is obvious (moments
of inertia simply add), and the third comes from the parallel-axis theorem (you can
show that the distance between the dots is /

3). Equating the right-hand sides of

the rst two equations, and then using the third to eliminate , gives
l
l
m
=
2
_ 1
9
Remarks: This result is larger than the I for a uniform triangle (which happens to be
m
2
/12), because the mass is generally further away from the center in the fractal case.
When we increase the side length of our fractal triangle by a factor of 2, the factor of
3 in the dm is between the factor of 2
1
= 2 relevant to a one-dimensional object, and the
factor of 2
2
= 4 relevant to a two-dimensional object. So in some sense our object has a
dimension between 1 and 2. It is reasonable to dene the dimension, d, of an object as the
number for which f
d
is the increase in volume when the dimensions are increased by a
factor f. For our fractal triangle, we have 2
d
= 3, and so d = log
2
3 1.58.
Solution
Week 10 (11/18/02)
Product of lengths
(This clever solution and generalization comes from Mike Robinson.)
Put the circle in the complex plane, with its center at the origin. Let the given
vertex of the N-gon be located at the point (1, 0). Let a e
2i/N
, so that a
N
= 1.
Then the other vertices are located at the points a
n
, where n = 1, . . . , N 1.
Let the distance between the vertex at (1, 0) and the vertex at a
n
be
n
. Then
the desired product (call it P
N
) of the N 1 segments from the given vertex to the
other vertices is
P
N
=
1

2
. . .
N1
= |1 a||1 a
2
| |1 a
N1
|
= (1 a)(1 a
2
) (1 a
N1
), (1)
where the third line comes from the fact that the product is real, because (1a
k
) is
the complex conjugate of (1 a
Nk
), so the phases in the product cancel in pairs.
Consider the function,
F(z) z
N
1. (2)
One factorization of F(z) is
F(z) = (z 1)(z
N1
+ z
N2
+ + 1). (3)
Another factorization is
F(z) = (z 1)(z a)(z a
2
) (z a
N1
), (4)
because the righthand side is simply the factorization that yields the zeros of z
N
1
(namely, all the numbers of the form a
n
). These two factorizations give
(z a)(z a
2
) (z a
N1
) = z
N1
+ z
N2
+ + 1. (5)
This equality holds for any value of z. In particular, if we set z = 1 we obtain
P
N
= N, as desired.
Remark: Consider the product of the N lengths from an arbitrary point z in the
complex plane, to all N vertices of the N-gon.
N=10
z
This product equals the absolute value of the righthand side of eq. (4). Hence, it equals
|F(z)| = |z
N
1|. Note what this gives in the N limit. If z equals any of the Nth
roots of 1, we obtain zero, of course. But if z is any point inside the unit circle, we obtain
|0 1| = 1, independent of both z and N.
The product of all the lengths except the length to the point (1,0), as in the original
statement of the problem, equals

z
N
1
z 1

= |z
N1
+ z
N2
+ + 1|. (6)
Note that as N , this goes to innity for z = 1 (its just our original result of N, by
looking at the righthand side). But it goes to the nite number 1/|z 1| for any z inside
the unit circle (by looking at the lefthand side).
Solution
Week 11 (11/25/02)
Break or not break?
There are two possible reasonings, which seem to create a paradox:
To an observer in the original rest frame, the spaceships stay the same distance,
d, apart. Therefore, in the frame of the spaceships, the distance between them,
d

, must equal d. This is true because d

is the distance that gets length-

contracted down to d. After a long enough time, will dier appreciably from
1, so the string will be stretched by a large factor. Therefore, it will break.
Let A be the rear spaceship, and let B be the front spaceship. From As point
of view, it looks like B is doing exactly what he is doing (and vice versa). A
says that B has the same acceleration that he has. So B should stay the same
distance ahead of him. Therefore, the string should not break.
The rst reasoning is correct. The string will break. So thats the answer to our
problem. But as with any good relativity paradox, we shouldnt feel at ease until
weve explained whats wrong with the wrong reasoning.
The problem with the second reasoning is that A does not see B doing exactly
what he is doing. Rather, A sees Bs clock running fast. Perhaps the easiest way to
show this is via the gravitational time-dilation eect. Since A and B are accelerating,
they may be considered (by the Equivalence Principle) to be in a gravitational eld,
with B higher in the eld. But high clocks run fast in a gravitational eld. Hence,
A sees Bs clock running fast (and B sees As clock running slow). A therefore sees
Bs engine running faster, and so B pulls away from A. Therefore, the string
eventually breaks.
Remarks:
1. There is one slight (inconsequential) aw in the rst reasoning above. There is not
one frame of the spaceships. Their frames dier, since they measure a relative speed
between themselves. It is therefore not clear exactly what is meant by the length
of the string, because it is not clear what frame the measurement should take place
in. This ambiguity, however, does not change the fact that A and B observe their
separation to be (essentially) d.
If we want there to eventually be a well-dened frame of the spaceships, we can sim-
ply modify the problem by stating that after a while, the spaceships stop accelerating
simultaneously, as measured by someone in the original inertial frame. (Equivalently,
A and B turn o their engines after equal proper times.) What A will see is the
following. B pulls away from A. B then turns o his engine. The gap continues to
widen. But A continues to re his engine until be reaches Bs speed. They then sail
onward, in a common frame, keeping a constant separation (which is greater than the
original separation, by a factor .)
2. The main issue in this problem is that it depends exactly on how we choose to accel-
erate an extended object. If we accelerate a stick by pushing on the back end (or by
pulling on the front end), its length will remain essentially the same in its own frame,
and it will become shorter in the original frame. But if we arrange for each end (or
perhaps a number of points on the stick) to speed up in such a way that they always
move at the same speed with respect to the original frame, then the stick will be torn
apart.
3. There is no need to invoke the Equivalence Principle to show that A sees Bs clock run
fast. We can demonstrate this eect completely within the realm of special relativity.
Consider the following Minkowski diagram (about which well invoke a few standard
facts).
x
x
d
t ad
ct
c
c
ct
A
A
____
t
A
B
After a small time t, the x
A
-axis (which consists of simultaneous events, as observed
by A) tilts upward with slope v/c = a t/c. It therefore intersects Bs worldline at a
ct value of
c t +
ad t
c
= c t

1 +
ad
c
2

. (1)
A therefore sees Bs clock run fast by a factor (1 + ad/c
2
), which is the standard
gravitational time-dilation result.
Another derivation of the ad/c
2
result is the following. Consider the situation a short
time after the start. An outside observer sees As and Bs clocks showing the same
time. Therefore, by the usual vd/c
2
loss-of-simultaneity result in special relativity,
Bs clock must read vd/c
2
more that As, in the moving frame. The increase per unit
time, as viewed by A, must therefore be (vd/c
2
)/t = ad/c
2
, as above.
At any later time, we can repeat (roughly) either of the above two derivations in
the instantaneous rest frame of A. Note that any special-relativistic time-dilation or
length-contraction eects will be second order in (v/c), and hence negligible for small
v.
Solution
Week 12 (12/2/02)
Decreasing numbers
First Solution: Let E(x) be the expected number of numbers you have yet to pick,
given that you have just picked the number x. Then, for example, E(0) = 1, because
the next number you pick is guaranteed to be greater than x = 0, whereupon the
game stops. Lets calculate E(x).
Imagine picking the next number, having just picked x. There is a (1x) chance
that this next number is greater than x, in which case the game stops. So in this
case it takes you just one pick after the number x. If, on the other hand, you pick
a number, y, which is less than x, then you can expect to pick E(y) numbers after
that. So in this case it takes you an average of E(y) +1 total picks after the number
x. These two scenarios my be combined to give the equation,
E(x) = 1 (1 x) +

x
0
(E(y) + 1) dy
= 1 +

x
0
E(y) dy (1)
Dierentiating this with respect to x gives E

(x) = E(x). Therefore, E(x) = Ae

x
,
where A is some constant. The condition E(0) = 1 gives A = 1. Hence
E(x) = e
x
. (2)
The total number of picks, T, is simply T = E(1), because the rst pick is auto-
matically less than 1, so the number of picks after starting a game with the number
1 is equal to the total number of picks in a game starting with a random number.
Since E(1) = e, we have
T = e. (3)
Second Solution: Let the rst number you pick be x
1
, the second x
2
, the third
x
3
, and so on. There is a p
2
= 1/2 chance that x
2
< x
1
. There is a p
3
= 1/3! chance
that x
3
< x
2
< x
1
. There is a p
4
= 1/4! chance that x
4
< x
3
< x
2
< x
1
, and so on.
You must make at least two picks in this game. The probability that you make
exactly two picks is equal to the probability that x
2
> x
1
, which is 1 p
2
= 1/2.
The probability that you make exactly three picks is equal to the probability
that x
2
< x
1
and x
3
> x
2
. This equals the probability that x
2
< x
1
minus the
probability that x
3
< x
2
< x
2
, that is, p
2
p
3
.
The probability that you make exactly four picks is equal to the probability that
x
3
< x
2
< x
1
and x
4
> x
3
. This equals the probability that x
3
< x
2
< x
1
minus
the probability that x
4
< x
3
< x
2
< x
2
, that is, p
3
p
4
.
Continuing in this manner, we nd that the expected total number of picks, T,
is
T = 2(1 p
2
) + 3(p
2
p
3
) + 4(p
3
p
4
) +
= 2

1
1
2!

+ 3

1
2!

1
3!

+ 4

1
3!

1
4!

+
= 2 +
(3 2)
2!
+
(4 3)
3!
+
(5 4)
4!
+
= 1 + 1 +
1
2!
+
1
3!
+
1
4!
+
= e. (4)
Third Solution: Let p(x) dx be the probability that a number between x and
x + dx is picked as part of the decreasing sequence. Then we may nd p(x) by
adding up the probabilities, p
j
(x) dx, that a number between x and x+dx is picked
on the jth pick.
The probability that such a number is picked rst is dx. The probability that it
is picked second is (1 x)dx, because 1 x is the probability that the rst number
is greater than x. The probability that it is picked third is (1/2)(1 x)
2
dx, because
(1 x)
2
is the probability that the rst two numbers are greater than x, and 1/2
is the probability that these numbers are picked in decreasing order. Likewise, the
probability that it is picked fourth is (1/3!)(1x)
3
dx. Continuing in this manner, we
see that the probability that it is picked sooner or later in the decreasing sequence
is
p(x) dx =

1 + (1 x) +
(1 x)
2
2!
+
(1 x)
3
3!
+

dx
= e
1x
dx. (5)
The expected number of numbers picked in the decreasing sequence is therefore

1
0
e
1x
dx = e1. Adding on the last number picked (which is not in the decreasing
sequence) gives a total of e numbers picked, as above.
Remarks:
1. What is the average value of the smallest number you pick? The probability that the
smallest number is between x and x +dx equals e
1x
(1 x) dx. This is true because
p(x) dx = e
1x
dx is the probability that you pick a number between x and x +dx as
part of the decreasing sequence (from the third solution above), and then (1 x) is
the probability that the next number you pick is larger. The average value, s, of the
smallest number you pick is therefore s =

1
0
e
1x
(1 x)xdx. Letting y 1 x for
convenience, and integrating (say, by parts), we have
s =

1
0
e
y
y(1 y) dy
= (y
2
e
y
+ 3ye
y
3e
y
)

1
0
= 3 e
0.282. (6)
Likewise, the average value of the nal number you pick is

1
0
e
1x
(1x)(1+x)/2 dx,
which you can show equals 2 e/2 0.64. The (1 + x)/2 in this integral arises
from the fact that if you do pick a number greater than x, its average value will be
(1 + x)/2.
2. We can also ask questions such as: Continue the game as long as x
2
< x
1
, and
x
3
> x
2
, and x
4
< x
3
, and x
5
> x
4
, and so on, with the numbers alternating in size.
What is the expected number of numbers you pick?
We can apply the method of the rst solution here. Let A(x) be the expected number
of numbers you have yet to pick, for x = x
1
, x
3
, x
5
, . . .. And let B(x) be the expected
number of numbers you have yet to pick, for x = x
2
, x
4
, x
6
, . . .. From the reasoning
in the rst solution, we have
A(x) = 1 (1 x) +

x
0
(B(y) + 1) dy = 1 +

x
0
B(y) dy,
B(x) = 1 x +

1
x
(A(y) + 1) dy = 1 +

1
x
A(y) dy. (7)
Dierentiating these two equations yields A

(x) = B(x) and B

(x) = A(x). If we
then dierentiate the rst of these and substitute the result into the second, we obtain
A

(x) = A(x). Likewise, B

(x) = B(x). The solutions to these equations may be

written as
A(x) = sinx + cos x and B(x) = cos x sinx. (8)
The condition A(0) = 1 yields = 1. The condition B(1) = 1 then gives =
(1 + sin1)/ cos 1. The desired answer to the problem equals B(0), since we could
imagine starting the game with someone picking a number greater than 0, which is
guaranteed. (Similarly, the desired answer also equals A(1).) So the expected total
number of picks is B(0) = (1 + sin1)/ cos 1. This has a value of about 3.41, which is
greater than the e 2.72 answer to our original problem.
Solution
Week 13 (12/9/02)
Unchanged velocity
Our strategy will be to produce (and equate) two dierent expressions for the
total change in the angular momentum of the ball (relative to its center). The rst
comes from the eects of the friction force on the ball. The second comes from
looking at the initial and nal motion.
To produce our rst expression for L, note that the normal force provides no
torque, so we may ignore it. The friction force, F, from the paper changes both p
and L, according to,
p =

Fdt,
L =

dt =

(R z) Fdt = (R z)

Fdt. (1)
Both of these integrals run over the entire slipping time, which may include time on
the table after the ball leaves the paper. In the second line above, we have used the
fact that the friction force always acts at the same location, namely (R z), relative
to the center of the ball. The two above equations yield
L = (R z) p. (2)
To produce our second equation for L, lets examine how L is related to p
when the ball is rolling without slipping, which is the case at both the start and the
nish. When the ball is not slipping, we have the following situation (assume the
ball is rolling to the right):
L
p
(top view)
The magnitudes of p and L are given by
p = mv,
L = I =
2
5
mR
2
=
2
5
Rm(R) =
2
5
Rmv =
2
5
Rp, (3)
where we have used the non-slipping condition, v = R. (The actual I = (2/5)mR
2
value for a solid sphere will not be important for the nal result.) It is easy to see
that the directions of L and p can be combined with the above L = 2Rp/5 scalar
relation to give
L =
2
5
R z p, (4)
where z points out of the page. Since this relation is true at both the start and the
nish, it must also be true for the dierences in L and p. That is,
L =
2
5
R z p. (5)
Eqs. (2) and (5) give
(R z) p =
2
5
R z p
= 0 = z p. (6)
There are three ways this cross product can be zero:
p is parallel to z. But it isnt, since p lies in the horizontal plane.
z = 0. Not true.
p = 0. So this must be true. Therefore, v = 0, as we wanted to show.
Solution
Week 14 (12/16/02)
Find the angles
Although this problem seems simple at rst glance, angle chasing wont provide the
answer. Something a bit more sneaky is required. At the risk of going overboard,
well give four solutions. You can check that all of the solutions rely on the equality
of the two given 50

angles, and on the fact that 2(80

) + 20

= 180

.
First Solution: In the gure below, note that

ACD = 60

and

ABD = 30

.
Let AC and BD intersect at E. Draw the angle bisectors of triangle ACD. They
meet at the incenter, I, located along segment ED. Since

ECI = 30

=

EBA,
triangles ECI and EBA are similar. Therefore, triangles EBC and EAI are also
similar. Thus,

EBC =

EAI = 10

. We then easily nd

ECB = 60

.
50
30
50 10
30
10
30
A
I
E
B
D
C
Second solution: In the gure below, note that

ABD = 30

. Let AC and BD
intersect at E. Draw segment AF, with F on BE, such that

EAF = 50

. We
then have

FAB = 30

. So triangle FAB is isosceles, with FA = FB.

Since

EDC =

EAF, triangles EDC and EAF are similar. Therefore, tri-
angles EAD and EFC are also similar. Hence,

ECF = 50

, and triangle FCA

is isosceles with FC = FA. Thus, FC = FA = FB, and so triangle FBC is also
isosceles, with

FBC =

FCB. Since it is easy to show that these two angles must
sum to 20

, they must each be 10

. Therefore,

FBC = 10

and

ECB = 60

.
1
50
50
30
50
30
A
E
F
B
D
C
Third Solution: In the gure below, note that

ACD = 60

. Reect triangle
ABC across AB to yield triangle ABG. Note that D, A, and G are collinear. From
the law of sines in triangle DBC, we have
sin50

BC
=
sin(60

+)
BD
. (1)
From the law of sines in triangle DBG, we have
sin50

BG
=
sin
BD
. (2)
But BC = BG, so we have sin(60

+) = sin. Therefore, 60

+ and must be
supplementary angles, which gives = 60

. We then easily obtain

DBC = 10

.
50 50
80
80
60
50
A G
B
D
C

2
Fourth Solution: We now present the brute-force method using the law of sines,
just to show that it can be done.
In the gure below, let AC and BD intersect at E. Let the length of AB be 1
unit. Then the law of sines in triangle AED gives
a =
sin50

sin110

, and d =
sin20

sin110

. (3)
The law of sines in triangles AEB and DEC then gives
b =

sin80

sin30

sin50

sin110

, and c =

sin50

sin60

sin20

sin110

. (4)
The law of sines in triangle BEC nally gives

sin80

sin50

sin30

sin110

sin =

sin50

sin20

sin60

sin110

sin. (5)
Substituting 70

for yields (after some algebra)

tan =
sin60

sin80

sin70

sin60

sin80

cos 70

+ sin30

sin20

. (6)
Using sin20

= 2 sin10

cos 10

= 2 sin10

sin80

, along with sin30

= 1/2, gives
tan =
sin60

sin70

sin60

cos 70

+ sin10

. (7)
Finally, expanding sin10

= sin(70

60

) gives the result

tan = tan60

. (8)
Hence = 60

, and so = 10

.
a
b
c
d
1

50
110
80
50
20
60
30
A
E
B
D
C
3
Solution
Week 15 (12/23/02)
Maximal gravity
Assume that the material has been shaped and positioned so that the eld at
P is maximum. Let this eld point in the x-direction. The key to this problem
is to realize that all the small elements of mass dm on the surface of the material
must give equal contributions to the x-component of the eld at P. If this were not
the case, then we could simply move a tiny piece of the material from one point
on the surface to another, thereby increasing the eld at P, in contradiction to our
assumption that the eld at P is maximum.
Label the points on the surface by their distance r from P, and by the angle
that the line of this distance subtends with the x-axis. Then a small mass dm on
the surface provides an x-component of the gravitational eld equal to
F
x
= cos
Gdm
r
2
. (1)
Since this contribution cannot depend on the location of the mass dm on the surface,
we must have r
2
cos . The surface may therefore be described by the equation,
r
2
= a
2
cos , (2)
where the constant a
2
depends on the volume of the material. The desired shape
clearly exhibits cylindrical symmetry around the x-axis, so let us consider a cross
section in the x-y plane. In terms of x and y (with x
2
+ y
2
= r
2
and cos = x/r),
eq. (2) becomes
r
3
= a
2
x = r
2
= a
4/3
x
2/3
= y
2
= a
4/3
x
2/3
x
2
. (3)
To get a sense of what this surface looks like, note that dy/dx = at both x = 0
and x = a (the point on the surface furthest from P). So the surface is smooth and
has no cusps. We can easily calculate the volume in terms of a, and we nd
V =

a
0
y
2
dx =

a
0
(a
4/3
x
2/3
x
2
) dx =
4
15
a
3
. (4)
Since the diameter of a sphere of volume V is (6V/)
1/3
, we see that a sphere with
the same volume would have a diameter of (8/5)
1/3
a 1.17a. Hence, our shape is
squashed by a factor of (5/8)
1/3
0.85 along the x-direction, compared to a sphere
of the same volume.
We may also calculate the maximum height in the y-direction. You can show
that is occurs at x = 3
3/4
a 0.44a and has a value of 2(4/27)
1/3
a 1.24a. Hence,
our shape is stretched by a factor of 2(4/27)
1/4
(5/8)
1/3
1.24/1.17 1.06 in the
y-direction, compared to a sphere of the same volume. Cross sections of our shape
and a sphere with the same volume are shown below.
1
d
d d
d
.85
1.06
2
Solution
Week 16 (12/30/02)
Letters in envelopes
First Solution: (This solution is due to Aravi Samuel.) We will use induction on
N. Let B
N
denote the number of bad congurations where none of the N letters
end up in the correct envelope. We claim that B
N+1
= N(B
N
+ B
N1
). This can
be seen as follows. In proceeding inductively from N to N +1 letters, there are two
possible ways we can generate bad congurations:
Given a bad conguration with N letters, we can create a bad conguration
with N +1 letters by simply placing down the (N +1)st letter in its envelope,
and then trading that letter with any of the other N letters. This provides us
with NB
N
bad congurations.
We can also create a bad conguration with N +1 letters by taking a congu-
ration of N letters where exactly one letter is in the correct envelope (there are
NB
N1
such congurations) and then trading that letter with the (N + 1)st
letter. This provides us with NB
N1
bad congurations.
We therefore see that B
N+1
= N(B
N
+ B
N1
). The probability of obtaining a bad
conguration with N letters is P
N
= B
N
/N!. Hence, B
N
= N! P
N
, and so
(N + 1)! P
N+1
= N
_
N! P
N
+ (N 1)! P
N1
_
= (N + 1)P
N+1
= NP
N
+ P
N1
. (1)
To solve this recursion relation, we can write it in the more suggestive form,
P
N+1
P
N
=
1
N + 1
_
P
N
P
N1
_
. (2)
Since P
1
= 0 and P
2
= 1/2, we have P
2
P
1
= 1/2. We then nd inductively that
P
k
P
k1
= (1)
k
/k! . Therefore,
P
N
= P
1
+
N

k=2
(P
k
P
k1
)
= 1 1 +
N

k=2
(1)
k
k!
=
N

k=0
(1)
k
k!
. (3)
This is simply the partial series expansion for e
1
. So for large N, it approaches
1/e 37%. This series expansion for 1/e converges very rapidly, so N does not
have to be very large for the approximation P
N
1/e to be valid. For example, if
N = 5 we have P
5
1/e 0.001.
Remark: This 1/e result in the large-N limit can also be seen in the following way. The
probability that a given letter does not end up in its corresponding envelope is 1 1/N.
1
Therefore, if we ignore the fact that the placements of the letters are related (they are
related because two letters cannot end up in the same envelope), then the probability that
no letter ends up in the correct envelope is
_
1
1
N
_
N

1
e
. (4)
It is not obvious that the correlations between the letters can be neglected here, but in view
of the above result, this is apparently the case.
Second Solution: Let P
N
be the probability that none of the N letters end up
in the correct envelope. Let L
i
and E
i
denote the ith letter and corresponding
envelope, respectively.
Consider a given letter, L
a
1
, and assume that no letter ends up in the correct
envelope. Then L
a
1
must end up in some E
a
2
, with a
2
= a
1
. L
a
2
will then end
up in some E
a
3
. L
a
3
will then end up in some E
a
4
, and so on. Eventually, one of
the envelopes in this chain must be E
a
1
. Let it be E
a
n+1
. We may describe this
situation by saying that L
a
1
belongs to a loop of length n. If no letter ends up in
the correct envelope, then n can be any number from 2 to N.
Claim: The probability that the loop containing L
a
1
has length n is equal to 1/N,
independent of n.
Proof: L
a
1
has an (N 1)/N probability of ending up in some E
a
2
, with a
2
=
a
1
. L
a
2
then has an (N 2)/(N 1) probability of ending up in some E
a
3
, with
a
3
= a
2
, a
1
. This continues until L
a
n1
then has an (N (n 1))/(N (n 2))
probability of ending up in some E
an
, with a
n
= a
n1
, , a
2
, a
1
. Finally, L
an
has a
1/(N (n 1)) probability of ending up in E
a
n+1
= E
a
1
. The probability that L
a
1
belongs to a loop of length n is therefore equal to
_
N 1
N
__
N 2
N 1
_

_
N (n 1)
N (n 2)
__
1
N (n 1)
_
=
1
N
. (5)
Given that a loop of length n is formed, which happens with probability 1/N,
the probability that all the N n other letters end up in the wrong envelopes is
simply P
Nn
. We therefore arrive at the relation,
P
N
=
1
N
_
P
N2
+ P
N3
+ + P
1
+ P
0
_
. (6)
There is no P
N1
term in this equation, because a loop of length 1 would mean that
L
a
1
went into E
a
1
. Note that P
1
= 0, and that P
0
= 1 here.
Multiplying eq. (6) through by N, and then subtracting the analogous equation
for P
N1
(after multiplying through by N 1), gives
NP
N
(N 1)P
N1
= P
N2
= P
N
P
N1
=
1
N
(P
N1
P
N2
) . (7)
This is the same as eq. (2), with N + 1 replaced by N. The solution proceeds as
above.
2
Third Solution: We will nd P
N
by counting the number of cases that have no
letter in the correct envelope, and then dividing this by the total number of possible
arrangements, N! .
We may count these cases in the following manner. There are N! total com-
binations. To count the number that have no letter in the correct envelope, we
must subtract from N! the number of combinations with, for example, (at least)
L
1
in the correct envelope; there are (N 1)! of these combinations. Likewise for
the situations where another letter is in the correct envelope. So there seem to
be N! N(N 1)! combinations with no letter in the correct envelope. However,
we have double-counted some of the cases. For example, a combination which has
(at least) L
1
and L
2
in the correct envelopes has been subtracted twice; there are
(N 2)! of these. Likewise for all the other pairs of letters. So we must add on
_
N
2
_
(N 2)! combinations. But now a combination which has (at least) L
1
, L
2
, and
L
3
in the correct envelopes has not been counted at all (because we have subtracted
it o three times, and then added it on three times); there are (N 3)! of these.
Likewise for the other triplets. So we must subtract o
_
N
3
_
(N 3)! combinations.
Now, however, the combinations with (at least) L
1
, L
2
, L
3
, and L
4
in the correct
envelopes have been counted
_
4
1
_
+
_
4
2
_

_
4
3
_
= 2 times. Likewise for the other
quadruplets. So we must add on
_
N
4
_
(N 4)! combinations.
In general, if we have done this procedure up to (k 1)-tuplets, then a combina-
tion having (at least) k letters in the correct envelopes has been counted T times,
where
T =
_
k
1
_
+
_
k
2
_
+ (1)
k1
_
k
k 1
_
. (8)
However, the binomial expansion gives
0 = (1 1)
k
= 1
_
k
1
_
+
_
k
2
_
+ + (1)
k1
_
k
k 1
_
+ (1)
k
= 1 + T + (1)
k
. (9)
Therefore, T = 2 for even k, and T = 0 for odd k. So we have either undercounted
by one, or overcounted by one. Hence, the total number of combinations having no
letter in the correct envelope is
N!
_
N
1
_
(N 1)! +
_
N
2
_
(N 2)! + =
N

k=0
(1)
k
N!
k!
. (10)
To obtain the probability, P
N
, that no letter is in the correct envelope, we must
divide this result by N! . Therefore,
P
N
=
N

k=0
(1)
k
k!
. (11)
3
Remarks:
1. What is the probability (call it P
l
N
) that exactly l out of the N letters end up in the
correct envelopes? (With this notation, P
0
N
equals the P
N
from above.) We can nd
P
l
N
as follows.
The probability that a given set of l letters goes into the correct envelopes is 1/
_
N(N
1)(N 2) (N l +1)
_
. The probability that the remaining N l letters all go into
the wrong envelopes is P
0
Nl
. This situation can happen in
_
N
l
_
ways. Therefore,
P
l
N
=
_
N
l
_
N(N 1) (N l + 1)
P
0
Nl
=
1
l!
P
Nl
. (12)
Hence, using eq. (3),
P
l
N
=
1
l!
Nl

k=0
(1)
k
k!
. (13)
For large N, we have P
l
N
1/(l!e). The fact that this falls o so rapidly with l means
that we are essentially guaranteed of having just a few letters in the correct envelopes.
For example, we nd (for large N) that the probability of having four or fewer letters
in the correct envelopes is about 99.7%.
2. It is interesting to note that the equality, P
l
N
=
1
l!
P
Nl
, may directly yield the large-N
result, P
N
1/e, without having to go through all the work of the original problem.
To see this, note that
1 =
N

l=0
P
l
N
=
N

l=0
1
l!
P
Nl
. (14)
Since the terms with small l values dominate this sum, we may (for large N) replace
the P
Nl
values with lim
M
P
M
. Hence,
1
N

l=0
1
l!
_
lim
M
P
M
_
. (15)
Therefore,
lim
M
P
M

_
N

l=0
1
l!
_
1

1
e
. (16)
3. Lets check that the P
l
N
given by eq. (13) do indeed satisfy

N
l=0
P
l
N
= 1. This may
be done as follows:
N

l=0
P
l
N
=
N

l=0
Nl

k=0
1
l!
(1)
k
k!
=
N

l=0
N

s=l
1
l!
(1)
sl
(s l)!
(with s = l + k)
=
N

s=0
s

l=0
1
l!
(1)
sl
(s l)!
(rewriting the limits in the s, l plane)
4
=
N

s=0
1
s!
s

l=0
s!
l!(s l)!
(1)
sl
=
N

s=0
1
s!
(1 1)
s
= 1 (only s = 0 contributes). (17)
4. What is the average number, A, of letters in the correct envelopes? If the setup of
the problem is repeated many times, then the average number of times a given letter
ends up in the correct envelope is 1/N. Since there are N letters, the average total
number of correct envelopes is N(1/N) = 1.
You can check that the expression for P
l
N
in eq. (13) leads to A = 1. For nite N,
the sum gets a little messy, but the result in eq. (17) will help simplify things a bit if
you want to work it out.
For large N, where we have P
l
N
1/(l!e), the sum is easy, and we obtain
A =
N

l=0
lP
l
N

1
e
N

l=1
1
(l 1)!
1. (18)
5
Solution
Week 17 (1/6/03)
Icosahedron of resistors
First Solution: We will calculate the eective resistance between vertices 1 and
2 in the gure below. When the icosahedron is viewed from the angle shown, four
vertices lie directly behind four other vertices; each of these pairs is represented by
a dot inside a circle. And thirteen edges lie directly behind thirteen other edges;
each of these pairs is represented by a bold line. The remaining two (of the 30 total)
edges not represented in the gure are the ones connecting vertices 5 and 6, and 7
and 8. For future reference, points X and Y are dened to be the midpoints of the
edges shown.
2 1
3 4
7,8 5,6
9,10
11,12
X
Y
If a potential dierence is created between vertices 1 and 2, then the potentials
at vertices 5 and 6 are equal, and likewise for the other three pairs of vertices. We
may therefore bring each of these four pairs of points together and identify each pair
as one point. The resulting circuit is simply the gure above, now planar, where
each bold line represents a (1/2) resistor (because it arises from two 1 resistors
in parallel).
We now note that all points on the vertical bisector of the circuit (that is, X, Y ,
the 9-10 pair, and the 11-12 pair) are at equal potentials, so we may bring them all
together and identify them as one point. Since X and Y split the top and bottom
resistors into two (1/2) resistors, we arrive at the following circuit, where every
line in the gure represents a (1/2) resistor.
1 2
1
This circuit may be reduced as follows:
1/2 1/2
3/16 3/16
1/4 1/4
1/2
1/2 1/2
1/2
1/4 1/4
1/4 1/4
1/4 1/4
1/2 1/2
1/4 1/4
1/4 1/4
3/4 3/4
1/4 1/4
11/16 11/16
11/60 11/60
11/30
The eective resistance between two adjacent vertices is therefore (11/30).
Second Solution: Let vertices 1 and 2 be adjacent. Consider the setup where a
current 1A enters through vertex 1, and a current (1/11)A leaves through the other
11 vertices. Note that, due to symmetry, a current (1/5)A ows through each of the
5 edges leaving vertex 1. Hence, the voltage dierence between vertices 1 and 2 is
V
1
V
2
=

1
5
A

(1) =
1
5
V. (1)
Consider a second setup, where a current 1A leaves through vertex 2, and a
current (1/11)A enters through the other 11 vertices. Again, note that a current
(1/5)A ows through each of the 5 edges entering vertex 2. Hence, the voltage
dierence between vertices 1 and 2 is
V
1
V
2
=

1
5
A

(1) =
1
5
V. (2)
If we superimpose these two setups on each other, then we arrive at the setup
where:
A current (12/11)A enters through vertex 1,
2
A current (12/11)A leaves through vertex 2,
No current enters or leaves through the other 10 vertices, and
The potential dierence between vertices 1 and 2 is
V
1
V
2
=
1
5
V +
1
5
V =
2
5
V. (3)
We have therefore constructed precisely the experimental setup that serves to de-
termine the eective resistance between vertices 1 and 2. That is, we have put a
current in at vertex 1, taken it out at vertex 2, and measured the voltage dierence
between the two points. The eective resistance between vertices 1 and 2 is therefore
given by
V = IR =
2
5
V =

12
11
A

R
e
= R
e
=
11
30
. (4)
Remark: Note that the sum of the eective resistances across all of the 30 edges in the
icosahedron is 11 = (N 1), where N is the number of vertices. You can quickly use the
method of the second solution above to calculate the sum of the eective resistances across
all of the edges in any of the other regular polyhedra. You will nd that the result is always
(N 1). Does this result hold for more general gures? Stay tuned for a future problem
of the week...
3
Solution
Week 18 (2/13/03)
Distribution of primes
A necessary and sucient condition for N to be prime is that N have no prime
factors less than or equal to

N. Therefore, under the assumption that a prime p
divides N with probability 1/p, the probability that N is prime is
P(N) =

1
1
2

1
1
3

1
1
5

1
1
7

1
1
p
(

N)

, (1)
where p
(

N)
denotes the largest prime less than or equal to

N. Our strategy for
solving for P(N) will be to produce a dierential equation for it.
Consider P(N +n), where n is an integer that satises

N n N. We have
P(N +n) =

1
1
2

1
1
3

1
1
5

1
1
7

1
1
p
(

N+n)

, (2)
where p
(

N+n)
denotes the largest prime less than or equal to

N +n. Eq. (2) may

be written as
P(N +n) = P(N)

1
1
p
1

1
1
p
2

1
1
p
(

N+n)

, (3)
where the p
i
are all the primes between

N and

N +n. Let there be k of these
primes. Since n N, we have

N +n/

N 1. Therefore, the p
i
are multiplica-
tively all roughly the same. To a good approximation, we may therefore set them
all equal to

N in eq. (3). This gives
P(N +n) P(N)

1
1

k
. (4)
We must now determine k. The number of numbers between

N and

N +n is

N +n

N =

N

1 +
n
N

1 +
n
2N

N
=
n
2

N
. (5)
Each of these numbers has roughly a P(

N) chance of being prime. Therefore,

there are approximately
k
P(

N)n
2

N
(6)
prime numbers between

N and

N +n.
1
Since n N, we see that k

N. Therefore, we may approximate the

(1 1/

N)
k
term in eq. (4) by 1 k/

N. Using the value of k from eq. (6), and

writing P(N +n) P(N) +P

(N)n, we can rewrite eq. (4) as

P(N) +P

(N)n P(N)

1
P(

N)n
2N

. (7)
We therefore arrive at the dierential equation,
P

(N)
P(N)P(

N)
2N
. (8)
It is easy to check that the solution for P is
P(N)
1
lnN
, (9)
as we wanted to show.
Remarks:
1. It turns out (under the assumption that a prime p divides N with probability 1/p)
that the probability that N has exactly n prime factors is
P
n
(N)
(lnlnN)
n1
(n 1)! ln N
. (10)
Our original problem dealt with the case n = 1, and eq. (10) does indeed reduce to
eq. (9) when n = 1. Eq. (10) can be proved by induction on n, but the proof I have
is rather messy. If anyone has a clean proof, let me know.
2. We should check that P
1
(N) + P
2
(N) + P
3
(N) + = 1. The sum must equal 1, of
course, because every number N has some number of divisors. Indeed (letting the
sum go to innity, with negligible error),

n=1
P
n
(N) =

n=1
(lnlnN)
n1
(n 1)! ln N
=
1
lnN

m=0
(lnlnN)
m
m!
=
e
ln ln N
lnN
= 1. (11)
3. We can also calculate the expected number, n, of divisors of N. To do this, lets
calculate n 1 (which is a little cleaner), and then add 1.
n 1 =

n=1
(n 1)P
n
(N)

n=2
(lnlnN)
n1
(n 2)! ln N
=
lnln N
lnN

k=0
(lnln N)
k
k!
= lnlnN. (12)
2
We can now add 1 to this to obtain n. However, all our previous results have been
calculated to leading order in N, so we have no right to now include an additive term
of 1. To leading order in N, we therefore have
n lnln N. (13)
4. There is another way to calculate n, without using eq. (10). Consider a group of M
numbers, all approximately equal to N. The number of prime factors among all of
these M numbers (which equals Mn by denition) is given by
1
Mn =
M
2
+
M
3
+
M
5
+
M
7
+ . (14)
Since the primes in the denominators occur with frequency 1/ lnx, this sum may be
approximated by the integral,
Mn M

N
1
dx
xlnx
= M lnlnN. (15)
Hence, n lnlnN, in agreement with eq. (13).
5. For which n is P
n
(N) maximum? Since P
n+1
(N) = (lnlnN/n)P
n
(N), we see that
increasing n increases P
n
(N) if n < lnln N. But increasing n decreases P
n
(N) if
n > lnlnN. So the maximum P
n
(N) is obtained when
n lnln N. (16)
6. The probability distribution in eq. (10) is a Poisson distribution, for which the results
in the previous remarks are well known. A Poisson distribution is what arises in a
random process such as throwing a large number of balls into a group of boxes. For
the problem at hand, if we take M(lnlnN) primes and throw them down onto M
numbers (all approximately equal to N), then the distribution of primes (actually,
the distribution of primes minus 1) will be (roughly) correct.
1
Weve counted multiple factors of the same prime only once. For example, weve counted 16 as
having only one prime factor. To leading order in N, this method of counting gives the same n as
assigning four prime factors to 16 gives (due to the fact that

(1/p
k
) converges for k 2).
3
Solution
Week 19 (1/20/03)
Block and bouncing ball
(a) Consider one of the collisions. Let it occur at a distance from the wall, and
let v and V be the speeds of the ball and block, respectively, after the collision.
We claim that the quantity (v V ) is invariant. That is, it is the same for
each collision. This can be seen as follows.
The time to the next collision is given by V t + vt = 2 (because the sum of
the distances traveled by the two objects is 2). Therefore, the next collision
occurs at a distance

from the wall, where

= V t =
2V
V + v
=
(v V )
v + V
. (1)
Therefore,

(v + V ) = (v V ). (2)
We now make use of the fact that in an elastic collision, the relative speed
before the collision equals the relative speed after the collision. (This is most
easily seen by working in the center of mass frame, where this scenario clearly
satises conservation of E and p.) Therefore, if v

and V

are the speeds after

the next collision, then
v + V = v

. (3)
Using this in eq. (2) gives

(v

) = (v V ), (4)
as we wanted to show.
What is the value of this invariant? After the rst collision, the block continues
to move at speed V
0
, up to corrections of order m/M. And the ball acquires a
speed of 2V
0
, up to corrections of order m/M. (This can be seen by working
in the frame of the heavy block, or equivalently by using eq. (3) with V

V = V
0
and v = 0.) Therefore, the invariant (v V ) is essentially equal to
L(2V
0
V
0
) = LV
0
.
Let L
min
be the closest distance to the wall. When the block reaches this
closest point, its speed is (essentially) zero. Hence, all of the initial kinetic
energy of the block now belongs to the ball. Therefore, v = V
0

M/m, and
our invariant tells us that LV
0
= L
min
(V
0

M/m0). Thus,
L
min
= L

m
M
. (5)
(b) (This solution is due to Slava Zhukov)
With the same notation as in part (a), conservation of momentum in a given
collision gives
MV mv = MV

+ mv

. (6)
1
This equation, along with eq. (3),
1
allows us to solve for V

and v

in terms
of V and v. The result, in matrix form, is

Mm
M+m
2m
M+m
2M
M+m
Mm
M+m

V
v

. (7)
The eigenvectors and eigenvalues of this transformation are
A
1
=

1
i

M
m

,
1
=
(M m) + 2i

Mm
M + m
e
i
,
A
2
=

1
i

M
m

,
2
=
(M m) 2i

Mm
M + m
e
i
, (8)
where
arctan

Mm
M m

m
M
. (9)
The initial conditions are

V
v

V
0
0

=
V
0
2
(A
1
+ A
2
). (10)
Therefore, the speeds after the nth bounce are given by

V
n
v
n

=
V
0
2
(
n
1
A
1
+
n
2
A
2
)
=
V
0
2

e
in

1
i

M
m

+ e
in

1
i

M
m

= V
0

cos n

M
m
sinn

. (11)
Let the block reach its closest approach to the wall at the Nth bounce. Then
V
N
= 0, and so eq. (11) tells us that N = /2. Using the denition of from
eq. (9), we have
N =
/2
arctan
2

Mm
Mm

M
m
. (12)
Remark: This solution is exact, up to the second line in eq. (12), where we nally
use M m. We can use the rst line of eq. (12) to determine the relation between
m and M for which the Nth bounce leaves the block exactly at rest at its closest
1
Alternatively, you could use conservation of energy, but this is a quadratic statement in the
velocities, which makes things messy. Conservation of energy is built into the linear eq. (3).
2
approach to the wall. For this to happen, we need the N in eq. (12) to be an integer.
Letting m/M r, we can rewrite eq. (12) as
2

r
1 r
= tan

2N

1 cos
1 + cos
, (13)
where we have used the tan half-angle formula, with /N. Squaring both sides
and solving the resulting quadratic equation for r gives
r =
3 + cos 2

2 + 2 cos
1 cos
. (14)
If we want the block to come to rest after N = 1 bounce, then = gives r = 1,
which is correct. If we want N = 2, then = /2 gives r = 3 2

2 0.172. If we
want N = 3, then = /3 gives r = 7 4

3 0.072. For general N, eq. (14) must

be computed numerically. For large N, the second line in eq. (12) shows that r goes
like 1/N
2
. More precisely, r
2
/(16N
2
).
3
Solution
Week 20 (1/27/03)
Collinear points
Draw all the lines determined by the points. From the assumption of the problem,
there are at least three points on each of these lines. Consider all of the distances
between any of the points and any of the lines. (Many of these distances are zero,
of course, for points lying on a given line.) Since there is a nite number of points
and lines, there is a nite number of these distances.
Assume that all the points do not lie on one line, so that some of these distances
are nonzero. Then there is a smallest nonzero distance, d
min
(it may occur more
than once). Consider a point P, and line L, associated with d
min
, as shown.
P
Q A
L
l
d
A A
1 2 3
min
Let Q be the projection of point P onto L. Since L contains at least three
points, at least two of them must lie on the same side of Q (or one may coincide
with Q). Call these points A
1
and A
2
. Let be the line through P and A
1
. Then
the distance from A
2
to is less than d
min
(because this distance is less than or equal
to the distance from Q to , which is strictly less than the distance from Q to P).
But this contradicts our assumption that d
min
was the smallest nonzero distance.
Hence, there can be no smallest nonzero distance. Therefore, all the distances are
zero, and all the points must lie on one line.
1
Solution
Week 21 (2/3/03)
Ball on turntable
Let the angular velocity of the turntable be z, and let the angular velocity of
the ball be . If the ball is at position r (with respect to the lab frame), then its
velocity (with respect to the lab frame) may be broken up into the velocity of the
turntable (at position r) plus the balls velocity with respect to the turntable. The
non-slipping condition says that this latter velocity is given by (a z), where a is
the radius of the ball. The balls velocity with respect to the lab frame is therefore
v = ( z) r + (a z). (1)
The important point to realize in this problem is that the friction force from
the turntable is responsible for changing both the balls linear momentum and its
angular momentum. In particular, F = dp/dt gives
F = m
dv
dt
. (2)
And the angular momentum of the ball is L = I, so = dL/dt (relative to the
center of the ball) gives
(a z) F = I
d
dt
, (3)
because the force is applied at position a z relative to the center.
We will now use the previous three equations to demonstrate that the ball un-
dergoes circular motion. Our goal will be to produce an equation of the form,
dv
dt
=

z v, (4)
since this describes circular motion, with frequency

(to be determined). Plugging

the expression for F from eq. (2) into eq. (3) gives
(a z)

m
dv
dt

= I
d
dt
=
d
dt
=

am
I

z
dv
dt
. (5)
Taking the derivative of eq. (1) gives
dv
dt
= z
dr
dt
+
d
dt
(a z)
= z v

am
I

z
dv
dt

(a z). (6)
Since the vector dv/dt lies in the horizontal plane, it is easy to work out the cross-
product in the right term (or just use the identity (AB)C = (AC)B(BC)A)
to obtain
dv
dt
= z v

ma
2
I

dv
dt
=
dv
dt
=

1 + (ma
2
/I)

z v. (7)
1
For a uniform sphere, I = (2/5)ma
2
, so we obtain
dv
dt
=

2
7

z v. (8)
Therefore, in view of eq. (4), we see that the ball undergoes circular motion, with
a frequency equal to 2/7 times the frequency of the turntable. This result for the
frequency does not depend on initial conditions.
Remarks:
1. Integrating eq. (8) from the initial time to some later time gives
v v
0
=

2
7

z (r r
0
). (9)
This may be written (as you can show) in the more suggestive form,
v =

2
7

r
0
+
7
2
( z v
0
)

. (10)
This equation describes circular motion, with the center located at the point,
r
c
= r
0
+
7
2
( z v
0
), (11)
and with radius,
R = |r
0
r
c
| =
7
2
| z v
0
| =
7v
0
2
. (12)
2. There are a few special cases to consider:
If v
0
= 0 (that is, if the spinning motion of the ball exactly cancels the rotational
motion of the turntable), then R = 0 and the ball remains in the same place (of
course).
If the ball is initially not spinning, and just moving along with the turntable,
then v
0
= r
0
. The radius of the circle is therefore R = (7/2)r
0
, and its center
is located at (from eq. (11))
r
c
= r
0
+
7
2
(r
0
) =
5r
0
2
. (13)
The point on the circle diametrically opposite to the initial point is therefore at
a radius r
c
+ R = (5/2)r
0
+ (7/2)r
0
= 6r
0
.
If we want the center of the circle be the center of the turntable, then eq. (11)
says that we need (7/2) z v
0
= r
0
. This implies that v
0
has magnitude
v
0
= (2/7)r
0
and points tangentially in the same direction as the turntable
moves. (That is, the ball moves at 2/7 times the velocity of the turntable
beneath it.)
3. The fact that the frequency (2/7) is a rational multiple of means that the ball will
eventually return to the same point on the turntable. In the lab frame, the ball will
trace out two circles in the time it takes the turntable to undergo seven revolutions.
From the point of view of someone on the turntable, the ball will spiral around ve
times before returning to the original position.
4. If we look at a ball with moment of inertia I = ma
2
(so a uniform sphere has
= 2/5), then you can show that the 2/7 in eq. (8) gets replaced by /(1 + ).
If a ball has most of its mass concentrated at its center (so that 0), then the
frequency of the circular motion goes to 0, and the radius goes to (as long as
v
0
= 0).
2
Solution
Week 22 (2/10/03)
Trading envelopes
(a) Let your envelope contain N dollars. Then the other envelope contains either
2N or N/2 dollars. If you switch, the expected value of your assets is
1
2
(2N) +
1
2
(N/2) = 5N/4. This is greater than N. Therefore, you should switch.
(b) There are (at least) two possible modes of reasoning, yielding dierent results:
It seems that we should be able to use the same reasoning as in part (a).
If you have N dollars in your envelope, then the other one has either
2N or N/2. Since you had a 50-50 chance of picking either envelope,
the other envelope should have a 50-50 chance of containing 2N or N/2
dollars. If you switch, there is a 1/2 chance you win N dollars, and a 1/2
chance you lose N/2 dollars. Therefore, the expectation value for your
gain is N/4 dollars. So you should switch.
If the correct strategy is to switch (that is, if there is an average gain
from trading), then if person A picks one envelope and person B picks
the other, then they are both better o if they switch. This cannot be
true. Likewise, it cannot be true that they are both better o if they do
not switch. Therefore, it doesnt matter whether or not they switch.
The second reasoning is correct. The aw in the rst reasoning is that the other
envelope does not have a 50-50 chance of containing 2N or N/2 dollars. Such
a 50-50 distribution would yield a zero probability of the envelopes containing
a nite and nonzero quantity (as well explain below). In a nutshell, it is
incorrect to assume that because you have a 50-50 chance of picking each
envelope, the envelope you dont pick has a 50-50 chance of having twice or
half the amount in your envelope.
(c) As we have stated, the fundamental dierence between the scenarios in parts
(a) and (b) is that the second envelope in scenario (b) does not have a 50-50
chance of containing half or twice the amount in your envelope. Lets look at
this further.
Consider the following slightly modied game, which has all the essentials of
the original one. Consider a game where powers of 2 (positive, negative, or
zero) are the only numbers of dollars allowed in the envelopes.
The fact that in scenario (b) there is not a 50-50 chance that the other en-
velope has 2N or N/2 dollars is most easily seen by looking at the simplest
distribution of money in the envelopes, the case where only two quantities are
used. Lets say that I always put $4 in one envelope and $8 in the other. (And
assume that you have a bad memory and cant remember anything from one
game to the next.) If your strategy is to switch, and if you initially have $4,
then you will denitely win $4 on the switch. And if you initially have $8,
then you will denitely lose $4 on the switch. Since you have a 50-50 chance
1
of starting with the $4 or $8 envelope, you will on average neither win nor lose
any money. In this example, it is clear that if you have, for example, $4, there
is not a 50-50 chance that the other envelope contains $2 or $8. Rather, there
is a 100% chance that it contains $8.
You can try make a situation in scenario (b) that comes close to always
having a 50-50 chance that the other envelope has twice or half the amount
in your envelope. For example, let there be a 1/n chance that the envelopes
contain 2
k
and 2
k+1
dollars, for all k from 1 to n. Then indeed if there are 2
m
dollars in your envelope, for m = 2, . . . , n1, then there is a 50-50 chance that
the other envelope has twice or half that amount. In all these n2 cases, you
will win money, on average, if you switch. And you will certainly win money if
you switch in the case where you have the minimum amount, 2
1
dollars. You
will, however, lose a great deal of money if you happen to start out with 2
n+1
dollars. This only happens 1/(2n) of the time, but it in fact precisely cancels,
on average, the winnings from all the other n 1 cases (as you can show).
Therefore, it doesnt matter if you switch.
If you want to produce a 50-50 chance that the other envelope has twice or
half the amount in your envelope for all m, then you have to assign equal prob-
abilities to all of the (2
k
, 2
k+1
) pairs, for < k < . But the assignment
of equal probabilities to an innite set requires that all of these probabilities
are zero, which means that there is a zero chance of putting a nite amount of
money in the envelopes. Since its stated that there is some amount of money
in the envelopes, we conclude that all the probabilities of the (2
k
, 2
k+1
) pairs
are not equal. The setup in part (b) is therefore not the same as in part (a),
and there is no paradox.
2
Solution
Week 23 (2/17/03)
V (x) versus a hill
Quick solution: Consider the normal force, N, acting on the bead at a given
point. Let be the angle that the tangent to V (x) makes with the horizontal, as
shown.
N
V(x)
mg

The horizontal F = ma equation is

N sin = m x. (1)
The vertical F = ma equation is
N cos mg = m y = N cos = mg + m y. (2)
Dividing eq. (1) by eq. (2) gives
tan =
x
g + y
. (3)
But tan = V

(x). Therefore,
x = (g + y)V

. (4)
We see that this is not equal to gV

. In fact, there is in general no way to
construct a curve with height y(x) which gives the same horizontal motion that a
1-D potential V (x) gives, for all initial conditions. We would need (g + y)y

= V

,
for all x. But at a given x, the quantities V

and y

are xed, whereas y depends

on the initial conditions. For example, if there is a bend in the wire, then y will be
large if y is large. And y depends (in general) on how far the bead has fallen.
Eq. (4) holds the key to constructing a situation that does give the x = gV

result for a 1-D potential V (x). All we have to do is get rid of the y term. So heres
what we do. We grab our y = V (x) wire and then move it up (and/or down) in
precisely the manner that makes the bead stay at the same height with respect to
the ground. (Actually, constant vertical speed would be good enough.) This will
make the y term vanish, as desired. (Note that the vertical movement of the curve
doesnt change the slope, V

, at a given value of x.)
Remark: There is one case where x is (approximately) equal to gV

, even when
the wire remains stationary. In the case of small oscillations of the bead near a minimum
of V (x), y is small compared to g. Hence, eq. (4) shows that x is approximately equal to
1
gV

. Therefore, for small oscillations, it is reasonable to model a particle in a 1-D potential
mgV (x) as a particle sliding in a valley whose height is given by y = V (x).
Long solution: The component of gravity along the wire is what causes the change
in speed of the bead. That is,
g sin =
dv
dt
, (5)
where is given by
tan = V

(x) = sin =
V

1 + V
2
, cos =
1

1 + V
2
. (6)
We are, however, not concerned with the rate of change of v, but rather with the rate
of change of x. In view of this, let us write v in terms of x. Since x = v cos , we have
v = x/ cos = x

1 + V
2
. (Dots denote d/dt. Primes denote d/dx.) Therefore, eq.
(5) becomes
gV

1 + V
2
=
d
dt

1 + V
2

= x

1 + V
2
+
xV

(dV

/dt)

1 + V
2
. (7)
Hence, x is given by
x =
gV

1 + V
2

xV

(dV

/dt)
1 + V
2
. (8)
Well simplify this in a moment, but rst a remark.
Remark: A common incorrect solution to this problem is the following. The accelera-
tion along the curve is g sin = g(V

/

1 + V
2
). Calculating the horizontal component of
this acceleration brings in a factor of cos = 1/

1 + V
2
. Therefore, we might think that
x =
gV

1 + V
2
. (9)
But we have missed the second term in eq. (8). Where is the mistake? The error is that
we forgot to take into account the possible change in the curves slope. (Eq. (9) is true
for straight lines.) We addressed only the acceleration due to a change in speed. We forgot
about the acceleration due to a change in the direction of motion. (The term we missed
is the one with dV

/dt.) Intuitively, if we have sharp enough bend in the wire, then x
can change at an arbitrarily large rate. In view of this fact, eq. (9) is denitely incorrect,
because it is bounded (by g/2, in fact).
To simplify eq. (8), note that V

dV/dx = (dV/dt)/(dx/dt)

V / x. Therefore,
xV

dV

dt
= xV

d
dt

V
x

= xV

V

V x
x
2

= V

V V

x

V
x

= V

V V
2
x. (10)
2
Substituting this into eq. (8), we obtain
x = (g +

V )V

, (11)
in agreement with eq. (4), since y(x) = V (x).
Eq. (11) is valid for a curve V (x) that remains xed. If we grab the wire and
start moving it up and down, then the above solution is invalid, because the starting
point, eq. (5), rests on the assumption that gravity is the only force that does work
on the bead. But if we move the wire, then the normal force also does work.
It turns out that for a moving wire, we simply need to replace the

V in eq.
(11) by y. This can be seen by looking at things in the (instantaneously inertial)
vertically-moving frame in which the wire is at rest. In this new frame, the normal
force does no work, so the above solution is valid. And in this new frame, y =

V .
Eq. (11) therefore becomes x = (g + y)V

. Shifting back to the lab frame (which
moves at constant speed with respect to the instantaneous inertial frame of the
wire) doesnt change y. We thus arrive at eq. (4), valid for a stationary or vertically
moving wire.
3
Solution
Week 24 (2/24/03)
Verifying weights
(a) Each weight may be used in three dierent ways. It may be put on the left
side, the right side, or not used at all. Therefore, if we have n weights, they
may be combined in 3
n
ways. This is true because in adding the weights, there
are three choices for the coecient of each weight in the sum: a plus sign if
its on the left, a minus sign if its on the right, and a zero if its not used at
all.
There are, however, duplicates among these 3
n
combinations. For every pos-
itive number, there is its negative (where the left and right scales are simply
reversed), which represents the same weight on the balance scale. Since it is,
in principle, possible for the number 0 to not be repeated, an upper bound
on the number of positive integer weights that may be weighed with n xed
weights is (3
n
1)/2. Therefore, to weigh all weights up to 121, we must have
n 5.
We claim that 5 weights are sucient. For 5 weights to do the job, we must
have no duplications. Therefore, the m smallest xed weights must yield all
values up to (3
m
1)/2. Hence, the rst weight must be 1 (to get values up to
1), the second weight must be 3 (to get values up to 4), the third weight must
be 9 (to get values up to 13), the fourth weight must be 27 (to get values up
to 40), and the fth weight must be 81 (to get values up to 121). The xed
weights are therefore powers of 3.
(b) In general, the n weights of 1, 3, 9, . . . , 3
n1
may be used to verify all integral
weights from 1 to W
n
(3
n
1)/2. This may be prove by induction on n, as
follows.
Assume that all weights up to W = (3
n
1)/2 can be veried with n xed
weights. If the next xed weight is chosen to be 3
n
, then we can additionally
verify all weights from 3
n
W
n
to 3
n
+W
n
. Using the above form of W
n
, this
range can be rewritten as (3
n
+ 1)/2 to (3
n+1
1)/2. But this is exactly the
range needed to be able to verify any weight up to W
n+1
= (3
n+1
1)/2.
Therefore, we have shown that if we can verify up to W
n
by using n xed
weights, then we can verify up to W
n+1
by using n+1 xed weights. Since we
can clearly verify up to W
1
= 1 by using one weight, the result holds for all n.
W
n
(3
n
1)/2 is the largest number for which this result holds, by the
upper-bound argument in part (a). Note that the inductive process shows
that all the xed weights must be powers of 3.
Remarks:
(a) To rephrase the inductive argument in part (b), powers of 3 have the relevant
property that (1 + 3 + 3
2
+ 3
n1
) + 1 = 3
n
(1 + 3 + 3
2
+ 3
n1
). The
importance of this can be seen by looking at, for example, n = 4. We see that 41,
which is not quite obtainable with four weights, can be obtained by taking the
1
fth weight, 81, and subtracting o the highest possible sum with four weights,
namely 40.
(b) Consider the following question: You wish to pick n pairs of equal weights (for
example, if n = 3, you might pick weights of 1, 1, 4, 4, 11, 11) such that you can
verify any positive integer weight up to W. How should you choose the weights
in order to maximize W?
Along the lines of the reasoning in part (a), we observe that each pair of weights
may be used in ve ways: We may put (1) both on the left side, (2) one on the
left and none on the right, (3) one on each side or use neither, (4) one on the
right and none on the left, or (5) both on the right.
Therefore, the weights may be used in 5
n
ways. As above, however, there are
duplicates among these 5
n
combinations. For every positive number, there is its
negative, which represents the same weight on the balance scale. Since it is, in
principle, possible for the number 0 to not be repeated, an upper bound on the
number of positive integer weights that may be weighed with n pairs of xed
weights is (5
n
1)/2.
It is indeed possible to achieve W = (5
n
1)/2, by choosing the weights to be
powers of 5. The weights should be 1, 1, 5, 5, 25, 25, . . .. Powers of 5 have the
relevant property that 2(1+5+5
2
+ 5
n1
) +1 = 5
n
2(1+5+5
2
+ 5
n1
).
For example, if n = 3. We see that 63, which is not quite obtainable with
three pairs of weights, can be obtained by taking a weight from the fourth pair,
namely 125, and subtracting o the highest possible sum obtainable with three
pairs, namely 62.
In general, if we use n k-tuplets of weights, it is possible to verify any positive
integer weight up to [(2k +1)
n
1]/2. The weights should be powers of (2k +1).
2
Solution
Week 25 (3/3/03)
Maximum deection angle
First Solution: Although it is possible to solve this problem by working in the lab
frame (see the second solution below), it is much easier to make use of the center-
of-mass frame. Let M have initial speed V in the lab frame. Then the CM moves
with speed
V
CM
=
MV
M +m
, (1)
as shown.
V
V
M m
MV
M+m
____
CM
CM
=
The speeds of the masses in the CM frame are therefore equal to
U = V V
CM
=
mV
M +m
, and u = | V
CM
| =
MV
M +m
, (2)
as shown.
M
U u
m
MV
M+m
____ mV
M+m
____
CM
= =
In the CM frame, the collision is simple. The particles keep the same speeds, but
simply change their directions (while still moving in opposite directions), as shown.
U
U
u
u

The angle is free to have any value. This scenario clearly satises conservation of
energy and momentum; therefore, it is what happens.
The important point to note is that since can have any value, the tip of the
U velocity vector can be located anywhere on a circle of radius U. If we then shift
back to the lab frame, we see that the nal velocity of M with respect to the lab
frame, V
lab
, is obtained by adding V
CM
to the vector U (which can point anywhere
on the dotted circle below). A few possibilities for V
lab
are shown.
1
U V
CM
V
lab
The largest angle of deection is obtained when V
lab
is tangent to the dotted circle,
in which case we have the following situation.
MV
M+m
M+m
mV
____
____

The maximum angle of deection, , is therefore given by

sin =
U
V
CM
=
mV
M+m
MV
M+m
=
m
M
. (3)
Second Solution: Let V

and v

be the nal speeds, and let and be the

scattering angles of M and m, respectively, in the lab frame. Then conservation of
p
x
, p
y
, and E give
MV = MV

cos +mv

cos , (4)
0 = MV

sin mv

sin, (5)
1
2
MV
2
=
1
2
MV
2
+
1
2
mv
2
. (6)
Putting the terms on the left-hand sides of eqs. (4) and (5), and then squaring
and adding these equations, gives
M
2
(V
2
+V
2
2V V

cos ) = m
2
v
2
. (7)
Equating this expression for m
2
v
2
with the one obtained by multiplying eq. (6)
through by m gives
M(V
2
+V
2
2V V

cos ) = m(V
2
V
2
)
= (M +m)V
2
(2MV cos )V

+ (M m)V
2
= 0. (8)
2
A solution to this quadratic equation in V

exists if and only if the discriminant is

non-negative. Therefore, we must have
(2MV cos )
2
4(M +m)(M m)V
2
0
= m
2
M
2
(1 cos
2
)
= m
2
M
2
sin
2

=
m
M
sin. (9)
Remarks: If M < m, then eq. (9) says that any value of is possible. In particular, it
it possible for M to bounce directly backwards. In the language of the rst solution above,
if M < m then V
CM
< U, so the dotted circle passes to the left of the left vertex of the
triangle. This means that can take on any value.
The method of the rst solution provides an easy way to demonstrate the result that
if the two masses are equal, then they always scatter at a relative angle of 90

(a familiar
result in billiards). If M = m, then
u = U = V
CM
=
MV
M +M
=
V
2
. (10)
Therefore, the u and U vectors in the gure below form a diameter of the dotted circle,
which means that the nal velocities of M and m in the lab frame are perpendicular.
U
u
m
V
CM
V
lab
V
lab
M
3
Solution
Week 26 (3/10/03)
Drunken walk
(a) First Solution: Imagine a large number of copies of the given setup proceed-
ing simultaneously. After each drunk takes his rst step in all of the copies,
the average position of all of them remains the same (namely, n steps from the
river), because each one had a 50-50 chance of moving either way. Likewise,
the average position remains unchanged after each successive step. This is
true because the drunks who are still moving wont change the average po-
sition (because of their random motion), and the drunks who have stopped
at an end of the street certainly wont change the average position (because
they arent moving). Therefore, the average position is always n steps from
the river.
Let the drunks keep moving until all of them have stopped at either end.
Let P
r
(n) and P
p
(n) be the probabilities of ending up at the river and police
station, respectively, having started n steps from the river. Then after all the
drunks have stopped, their average distance from the river is 0P
r
(n)+NP
p
(n).
But this must equal n. Hence, P
p
(n) = n/N, and so P
r
(N) = 1 (n/N).
Second Solution: Let the river and police station be located at positions 0
and N, respectively. Let P
r
(n) be the probability of ending up at the river,
given a present position of n. Since after one step the drunk is equally likely
to be at n 1 and n + 1, we must have
P
r
(n) =
1
2
P
r
(n 1) +
1
2
P
r
(n + 1). (1)
Therefore, P
r
(n) is a linear function of n. Invoking the requirements that
P
r
(0) = 1 and P
r
(N) = 0, we nd P
r
(n) = 1 n/N. The probability of
ending up at the police station, P
p
(n), is then P
p
(n) = 1 P
r
(n) = n/N.
(b) Let g(k) be the expected number of steps it takes to reach an end of the street,
having started k steps from the river. After one step from position k, there is
a 1/2 chance of being at position k 1, and a 1/2 chance of being at position
k + 1. Therefore, the g(k) are related by
g(k) =
1
2
g(k 1) +
1
2
g(k + 1) + 1, (2)
where g(0) = g(N) = 0. We must now solve this recursion relation.
Multiplying through by 2, and then summing all the eqs. (2) for values of k
from 1 to m gives
g(1) + g(m) = g(m + 1) + 2m
= g(m + 1) = g(1) + g(m) 2m. (3)
1
Note that if we set m = N 1, we obtain 0 = g(1) + g(N 1) 2(N 1).
Since g(1) = g(N 1) by symmetry, we nd g(1) = g(N 1) = N 1.
Summing all the eqs. (3) for values of m from 1 to n 1 gives
g(n) = n g(1) 2
n1

1
m
= n g(1) n(n 1). (4)
Using g(1) = N 1, we nd
g(n) = n(N 1) n(n 1)
= n(N n). (5)
This can be written as
g(n) =

N
2

N
2
n

2
, (6)
which is just an inverted parabola.
2
Solution
Week 27 (3/17/03)
Relativistic cookies
Let the diameter of the cookie cutter be L, and consider the two following reasonings.
In the lab frame, the dough is length-contracted, so the diameter L corresponds
to a distance larger than L (namely L) in the doughs frame. Therefore, when
you buy a cookie, it is stretched out by a factor in the direction of the belt.
1
In the frame of the dough, the cookie cutter is length-contracted in the direc-
tion of motion. It has length L/. So in the frame of the dough, the cookies
have a length of only L/. Therefore, when you buy a cookie, it is squashed
by a factor in the direction of the belt.
Which reasoning is correct? The rst one is. The cookies are stretched out. The
fallacy in the second reasoning is that the various parts of the cookie cutter do not
strike the dough simultaneously in the dough frame. What the dough sees is this:
The cutter moves to, say, the left. The right side of the cutter stamps the dough,
then nearby parts of the cutter stamp it, and so on, until nally the left side of the
cutter stamps the dough. But by this time the front (that is, the left) of the cutter
has moved farther to the left. So the cookie turns out to be longer than L. It takes
a little work to demonstrate that the length is actually L, but lets do that now
(by working in the dough frame).
Consider the moment when the the rightmost point of the cutter strikes the
dough. In the dough frame, a clock at the rear (the right side) of the cutter reads
Lv/c
2
more than a clock at the front (the left side). The front clock must therefore
advance by Lv/c
2
by the time it strikes the dough. (This is true because all points
on the cutter strike the dough simultaneously in the cutter frame. Hence, all cutter
clocks read the same when they strike.) But due to time dilation, this takes a
time (Lv/c
2
) in the dough frame. During this time, the cutter travels a distance
v(Lv/c
2
). Since the front of the cutter was initially a distance L/ (due to length
contraction) ahead of the back, the total length of the cookie in the dough frame
equals
=
L

+ v

Lv
c
2

= L

2
+
v
2
c
2

= L

1
v
2
c
2

+
v
2
c
2

= L,
as we wanted to show.
1
The shape is an ellipse, since thats what a stretched-out circle is. The eccentricity of an ellipse
is the focal distance divided by the semi-major axis length. As an exercise, you can show that this
equals v/c here.
1
Solution
Week 28 (3/24/03)
Rectangle in a circle
In the gure below, let the incenters of triangles ADB and ADC be X and Y ,
respectively.
A
B
C
P
X
Y
D
Angle

XAY can be written as

XAY =

XAD

Y AD
=
1
2

BAD
1
2

CAD
=
1
2

BAC
=
1
4
(BC

).
A similar argument (with A, B, X interchanged with D, C, Y ) shows that angle

Y DX also equals (1/4)(BC

). This equality of angles

XAY and

Y DX implies
that triangles XAP and Y DP are similar. This in turn implies that triangles
PXY and PAD are similar. Therefore,

PXY =

PAD. These results may be
summarized in the following gure.
A
P
X
Y
D

1
We may now repeat the above procedure with the incenters (Y and Z) of triangles
DCA and DCB. The result is two more pairs of equal angles, as shown.
A
B
C
D
X
Y
Z

The four angles shown have the values,

= (1/4)(BC

),
= (1/2)

CAD = (1/4)(CD

),
= (1/4)(AB

),
= (1/2)

ACD = (1/4)(AD

).
Therefore,
+ + + =
1
4
(BC

+CD

+AB

+AD

) =
1
4
(360

) = 90

. (1)
We now note that angle

XY Z is given by

XY Z = 360

XY D

ZY D
= 360

(180

) (180

)
= + + +
= 90

.
The same reasoning holds for the other three vertices of the incenter quadrilateral.
Therefore, this quadrilateral is a rectangle, as we wanted to show.
2
Solution
Week 29 (3/31/03)
Balls in a semicircle
(a) Let M/N be the mass of each ball in the semicircle. We need the deection
angle in each collision to be = /N. However, if the ratio /m is too small,
then this angle of deection is not possible.
From Problem of the Week #25, the maximal angle of deection in each
collision is given by sin = /m. (Well just invoke this result here.) Since we
want = /N here, this sin /m condition becomes (using sin , for
the small angle )


m
=

N

M/N
m
=
M
m
. (1)
(b) Referring back to the solution to Problem #25, we see that ms speed after
the rst bounce is obtained from the following gure.
mV
m+
V
m+
____
____
V

max
f
Looking at the right triangle, we see that the speed after the bounce is
V
f
= V

m
2

2
m+
. (2)
To rst order in the small quantity /m, this equals
V
f

mV
m+
V

1

m

. (3)
The same reasoning holds for each successive bounce, so the speed decreases
by a factor of (1 /m) after each bounce. In the minimum M/m case found
in part (a), we have

m
=
M/N
m
=
M/m
N
=

N
. (4)
Therefore, the ratio of ms nal speed to initial speed equals
V
nal
V
initial

1

N

N
e

. (5)
Thats a nice result, if there ever was one! Since e

is roughly equal to 1/23,

only about 4% of the initial speed remains.
1
Solution
Week 30 (4/7/03)
Dierence of Powers
A value of 26 is obtainable with m = n = 1. By considering the remainder when
33
m
7
n
is divided by certain numbers, we will show that no value smaller than 26
is possible. We will use the mod notation for convenience, where a b (mod c)
means that a leaves a remainder of b when divided by c.
Consider divisions by 16. We have 33 1 (mod 16), and 7
n
7 or 1 (mod
16) because 7
2
1 (mod 16). Therefore, 33
m
7
n
0 or 10 (mod 16). So
the only possible answers to the problem are 0, 10, 16, and 26.
Now consider divisions by 3. We have 33 0 (mod 3), and 7 1 (mod 3).
Therefore, 33
m
7
n
2 (mod 3). This leaves 26 as the only possibility.
1
Solution
Week 31 (4/14/03)
Simultaneous claps
First Solution: The relative speed of A and B is obtained from the velocity-
addition formula, which yields

rel
=
1

3
+
1

3
1 +
1

3

1

3
=

3
2
. (1)
In Bs frame, As clock runs slow by a factor = 1/
_
1
2
rel
= 2. Hence, if A claps
her hands when her clock reads T (which happens to be L/

2c, but we wont need

the actual value), then B claps her hands when her clock reads 2T. But likewise,
in As frame, Bs clock runs slow by a factor = 2. Hence, A will make her second
clap at 4T, and so on. A and B therefore increase their separation by a factor of 4
between successive claps of A. Their separation after As nth clap therefore equals
d
n
= 4
n1
L. (2)
In the case where A and B move at a general speed v, their relative speed equals

rel
=
2
1 +
2
. (3)
The associated factor is
=
1
_
1
_
2
1+
2
_
2
=
1 +
2
1
2
. (4)
From the reasoning above, the separation after As nth clap equals
d
n
= L
_
1 +
2
1
2
_
2(n1)
. (5)
This agrees with eq. (2) when = 1/

3.
Second Solution: We can also solve this problem by using Minkowski diagrams.
Such diagrams show what the x and ct axes of one frame look like with respect to
the x and ct axes of another frame. As we will see below, these diagrams make
things very easy to visualize, and provide an easy geometrical way of determining
various quantities.
A basic Minkowski diagram is shown below. This diagram is what arises when
A travels to the right at speed v. Her x
A
and ct
A
axes are tilted inward at an angle
with respect to the x
G
and ct
G
axes of the ground frame. The angle satises
tan = v/c. These facts (and many others) of a Minkowski diagram can be derived
from the Lorentz transformations,
x
A
= (x
G
ct
G
)
ct
A
= (ct
G
x
G
), (6)
where v/c.
1
x
ct
A
ct
G
A
x
G

An important point to note in this diagram is that all events on the x

A
-axis
have t
A
= 0. Therefore, all events on the x
A
-axis are simultaneous in As frame.
Likewise, all events on any line parallel to the x
A
-axis are simultaneous with each
other in As frame. Armed with this fact, we can easily solve the given problem.
The following diagram shows what the axes of As and Bs frames look like with
respect to the ground frame.
x
ct
A
ct
G
ct
B
A
x
G
x
B
30 30
4L
L
A
1
A
P
2
B
1
B
2
A makes her rst clap at event A
1
. At this event, she is a distance L from B, as
measured in the ground frame. Bs rst clap (which occurs simultaneously with A
1
,
as measured by B) is obtained by drawing a line through A
1
parallel to the x
B
-axis.
Since the x
B
-axis is perpendicular to the ct
A
-axis (due to the 30

= tan
1
(1/

3)
angles in the diagram), we obtain the right angle shown.
Similarly, As second clap (which occurs simultaneously with B
1
, as measured
by A) is obtained by drawing a line through B
1
parallel to the x
A
-axis. Since the
x
A
-axis is perpendicular to the ct
B
-axis, we obtain the right angle shown.
2
Continuing in this manner, we can locate all subsequent claps. Making use of
the plethora of 30

-60

-90

triangles in the gure, we see that PB

1
= 2PA
1
, and
PA
2
= 2PB
1
, and so on. Therefore, PA
2
= 4PA
1
, and PA
3
= 4PA
2
, and so on.
Looking now at the equilateral triangles whose top sides (the bold lines in the gure)
give the distances between A and B (as measured in the ground frame), we see that
these distances increase by a factor of 4 during each interval between As claps. The
separation after As nth clap therefore equals
d
n
= 4
n1
L, (7)
in agreement with eq. (2).
In the case where A and B move at a general speed v, we obtain the following
gure.
x
ct
A
ct
G
ct
B
A
x
G
x
B
A
1
A
P
2
A
3
A
4
B
1
B
2
B
3


Using the same reasoning above, but now with a plethora of right triangles with
an angle of 2, we see that PA
2
= PA
1
/ cos
2
2, and so on. Looking now at the
isosceles triangles with vertex angle 2, whose top sides (the bold lines in the gure)
give the distances between A and B (as measured in the ground frame), we see that
these distances increase by a factor of 1/ cos
2
2 during each interval between As
claps. The separation after As nth clap therefore equals
d
n
=
L
(cos 2)
2(n1)
. (8)
Using tan = , you can show that cos 2 = (1
2
)/(1 +
2
). Therefore,
d
n
= L
_
1 +
2
1
2
_
2(n1)
, (9)
3
in agreement with eq. (5).
Remark: In the above problem, each person clapped her hands simultaneously (as
measured by her) with the other persons clap. We may also consider the (entirely dierent)
setup where each person claps her hands when she sees the other persons clap. In this new
setup, we are concerned with the actual travel of photons from one person to the other,
whereas in the original problem, the travel of any photons was irrelevant.
The ratio of the Bs clock reading when she makes her rst clap, to As clock reading
when she makes her rst clap, is simply the (inverse of the) longitudinal doppler factor
_
(1
rel
)/(1 +
rel
). (This is clear if you imagine A continually sending out a series of
light ashes to B.) Using the
rel
from eq. (3), we nd this ratio to be (1 + )/(1 ).
We can use this same reasoning on successive claps, and so the separation between A and
B after As nth clap equals
d
n
= L
_
1 +
1
_
2(n1)
. (10)
We can also solve this new setup by using a Minkowski diagram. The solution proceeds
just as it did with the Minkowski diagrams above, except that now we must draw the
45

lines which describe a photons travel, instead of the lines of simultaneity (which were
parallel to the x
A
and x
B
axes) that we drew above. These 45

lines appear as follows.

x
ct
A
ct
G
ct
B
A
x
G
x
B
A
1
A
P
2
B
1


We must nd the ratio of PB
1
to PA
1
. You can show that the law of signs in triangle
PB
1
A
1
gives
PB
1
PA
1
=
sin(135

)
sin(45

)
. (11)
Using the trig sum formula for sine, we nd
PB
1
PA
1
=
cos + sin
cos sin
=
1 + tan
1 tan
=
1 +
1
. (12)
4
Likewise, this is the ratio of PA
2
to PB
1
. Therefore, PA
2
/PA
1
= (1 + )
2
/(1 )
2
, and
so the separation between A and B after As nth clap equals
d
n
= L
_
1 +
1
_
2(n1)
, (13)
in agreement with eq. (10). As you can check, this is larger than the answer to the original
problem, given in eq. (9). This must be the case, because the photons take some nonzero
time to travel between the A and B.
5
Solution
Week 32 (4/21/03)
The game show
If you do not switch, your probability of winning equals 1/3. No actions taken by
the host can change the fact that if you play a large number, N, of these games,
then (roughly) N/3 of them will have the prize behind the door you pick.
If you switch, your probability of winning turns out to be greater. It increases
to 2/3. This can be seen as follows. Without loss of generality, assume that you
pick the rst door. There are three equally likely possibilities for what is behind the
three doors: PGG, GPG, and GGP (where P denotes prize, and G denotes goat).
If you do not switch, then in only the rst of these three cases do you win, so your
odds of winning are 1/3. If you do switch, then in the rst case you lose, but in
the other two you win (because the door not opened by the host has the prize).
Therefore, your odds of winning are 2/3, so you do in fact want to switch.
Remarks:
1. After the host reveals a goat, there is one prize and one goat behind the
unopened doors. You might think that this implies that the probability of
winning the prize is 1/2, independent of whether or not a switch is made.
This is incorrect, because the above reasoning shows that there is only a 1/3
chance that the door you initially picked has the prize, and a 2/3 chance
that the other unopened door has the prize. (The fact that there are two
possibilities doesnt mean that their probabilities have to be equal, of course.)
2. It should be no surprise that the odds are dierent for the two strategies after
the host has opened a door (the odds are obviously the same, equal to 1/3,
whether or not a switch is made before the host opens a door), because the
host gave you some of the information he had about the locations of things.
3. To make the above reasoning more believable, imagine a situation with 1000
doors (containing one prize and 999 goats). After you pick a door, the host
opens 998 other doors to reveal 998 goats. In this setup, if you do not switch,
your chances of winning are 1/1000. If you do switch, your chances of winning
are 999/1000 (which can be seen by listing out the 1000 cases, as we did with
the three cases above). In this case it is clear that the switch should be made,
because the only case where you lose after you switch is the case where you
had initially picked the prize (and that happens only 1/1000 of the time).
4. The clause in the statement of the problem, The host announces that after
you select a door (without opening it), he will open one of the other two doors
and reveal a goat, is crucial. If it is omitted, and it is simply stated that,
The host then opens one of the other doors and reveals a goat, then it
is impossible to state a preferred strategy. If the host doesnt announce his
actions beforehand, then for all you know, he always reveals a goat (in which
case you should switch). Or he randomly opens a door, and just happened to
pick a goat (in which case it doesnt matter whether or not you switch, as you
1
can show). Or he opens a door and reveals a goat if and only if your initial
door has the prize (in which case you denitely should not switch).
5. This problem is infamous for the intense arguments it so easily lends itself to.
The common incorrect answer is that there are equal 1/2 chances of winning
whether or not you switch. Now, theres nothing bad about getting the wrong
answer, nor is there anything bad about not believing the correct answer for a
while. But concerning the arguments that drag on and on, I think it should be
illegal to argue about this problem for more than ten minutes, because at that
point everyone should simply stop and play the game. Three coins with a dot
on the bottom of one of them is all you need. Not only will the actual game
give the correct answer (if you play enough times so that things average out),
but the patterns that form when playing will undoubtedly convince the skeptic
of the correct reasoning. Arguing endlessly about an experiment, when you
can actually do the experiment, is as silly as arguing endlessly about whats
behind a door, when you can simply open the door.
2
Solution
Week 33 (4/28/03)
Ball rolling in a cone
It turns out that the ball can move arbitrarily fast around the cone. As we will see,
the plane of the contact circle (represented by the chord in the gure below) will
need to be tilted downward from the contact point, so that the angular momentum
has a rightward horizontal component, as shown.

L
r
F
R
Lets rst look at F = ma along the plane. Let be the angular frequency of
the balls motion around the cone. Then the balls horizontal acceleration is m
2
to the left. So F = ma along the plane gives (where F
f
is the friction force)
mg sin + F
f
= m
2
cos . (1)
Now lets look at = dL/dt. To get a handle on how fast the ball is spinning,
consider what the setup looks like in the rotating frame in which the center of the
ball is stationary (so the ball just spins in place as the cone spins around). Since
there is no slipping, the contact points on the ball and the cone must have the same
speed. That is,
r = = =

r
, (2)
where is the angular speed of the ball in the rotating frame, and r is the radius of
the contact circle on the ball.
1
The angular momentum of the ball in the lab frame
equals L = I (at least for the purposes here
2
), and it points in the direction shown
above.
1
If the center of the ball travels in a circle of radius , then the here should actually be replaced
with + Rsin , which is the radius of the contact circle on the cone. But since were assuming
that R , we can ignore the Rsin part.
2
This L = I result isnt quite correct, because the angular velocity of the ball in the lab frame
equals the angular velocity in the rotating frame (which tilts downwards with the magnitude we
just found) plus the angular velocity of the rotating frame with respect to the lab frame (which
points straight up with magnitude ). This second part of the angular velocity simply yields an
additional vertical component of the angular momentum. But the vertical component of L doesnt
change with time as the ball moves around the cone. It is therefore irrelevant, since we will be
concerned only with dL/dt in what follows.
1
The L vector precesses around a cone in L-space with the same frequency, , as
the ball moves around the cone. Only the horizontal component of L changes, and
it traces out a circle of radius L
hor
= Lsin, at frequency . Therefore,

dL
dt

= L
hor
= (I sin) =
I
2
sin
r
, (3)
and the direction of dL/dt is into the page.
The torque on the ball (relative to its center) is due to the friction force, F
f
.
Hence, || = F
f
R, and its direction is into the page. Therefore, = dL/dt gives
(with I = mR
2
, where = 2/5 in this problem)
F
f
R =
I
2
sin
r
= F
f
=
mR
2
sin
r
. (4)
Using this F
f
in eq. (1) gives
mg sin +
mR
2
sin
r
= m
2
cos . (5)
Solving for gives

2
=
g sin

cos
Rsin
r
. (6)
We see that it is possible for the ball to move around the cone innitely fast if
cos =
sin
x
, (7)
where x r/R. But from the above gure, we see that is given by
= sin
1
(r/R). (8)
Therefore, eq. (7) gives
cos =

x
sin( sin
1
x)
= xcos = sin cos(sin
1
x) cos sin(sin
1
x)
= xcos = sin

1 x
2
cos x
= x(1 + ) cos = sin

1 x
2
. (9)
Squaring and solving for x
2
gives
x
2
=

2
sin
2

(1 + )
2
cos
2
+
2
sin
2

. (10)
In the problem at hand, we have = 2/5, so
r
R
x =

4 sin
2

49 cos
2
+ 4 sin
2

. (11)
2
Remarks:
1. What value of allows largest the tilt angle of the contact circle (that is, the largest
)? From eq. (7), we see that maximizing is equivalent to maximizing xcos , or
equivalently x
2
cos
2
. Using the value of x
2
in eq. (10), we see that we want to
maximize
x
2
cos
2
=

2
sin
2
cos
2

(1 + )
2
cos
2
+
2
sin
2

. (12)
Taking the derivative with respect to and going through a bit of algebra, we nd
that the maximum is achieved when
sin =

1 +
1 + 2
=

7
9
= = 61.9

. (13)
You can then show that
sin
max
=
1
1 + 2
=
5
9
=
max
= 33.7

. (14)
2. Lets consider three special cases for the contact circle, namely, when it is a horizontal
circle, a great circle, or a vertical circle.
(a) Horizontal circle: In this case, we have = 0, so eq. (6) gives

2
=
g tan

. (15)
In this case, L points vertically, which means that dL/dt is zero, which means
that the torque is zero, which means that the friction force is zero. Therefore,
the ball moves around the cone with the same speed as a particle sliding without
friction. (You can show that such a particle does indeed have
2
= g tan/.)
The horizontal contact-point circle ( = 0) is the cuto case between the sphere
moving faster or slower than a frictionless particle.
(b) Great circle: In this case, we have r = R and = (90

). Hence, sin =
cos , and eq. (6) gives

2
=
g tan
(1 + )
. (16)
This reduces to the frictionless-particle case when = 0, as it should.
(c) Vertical circle: In this case, we have r = Rcos and = 90

, so eq. (6) gives

2
=
g tan

1 +

cos
2

. (17)
Again, this reduces to the frictionless-particle case when = 0, as it should.
But for 90

, goes to zero, whereas in the other two cases above, goes

to .
3
Solution
Week 34 (5/5/03)
Counterfeit coin
(a) Note that there are three possible outcomes to each weighing: left side heavier,
right side heavier, or both sides equal. In order to do the given task in as few
weighings as possible, we will need as much information from each weighing as
possible. Hence, all three possibilities should be realizable for each weighing
(except for the nal weighing in some scenarios, as we will see below). So,
for example, an initial weighing of six coins against six coins is probably not
a good idea, because it is not possible for the scale to balance. We should
expect to have to switch coins from one side of the scale to the other, from
one weighing to the next, in order to make the three possibilities realizable for
a given weighing. Having said that, here is one scheme that does the task in
three weighings (there are other variations that also work):
Weigh four coins (labelled A
1
, A
2
, A
3
, A
4
) against four others (B
1
, B
2
, B
3
,
B
4
). The remaining four will be labelled C
1
, C
2
, C
3
, C
4
. There are three
possible outcomes to this weighing:
(1) The A group is heavier than the B group. We know in this case that the
C coins are good, and the bad coin is either an A or a B. If the bad
coin is an A, it is heavy. If the bad coin is a B, it is light.
Now weigh (A
1
, A
2
, B
1
) against (A
3
, A
4
, B
2
). There are three possible
outcomes:
If the (A
1
, A
2
, B
1
) side is heavier, the bad coin must be A
1
, A
2
, or
B
2
. Weigh A
1
against A
2
. If A
1
is heavier, it is the bad (heavy) coin;
if A
2
is heavier, it is the bad (heavy) coin; if they are equal, B
1
is
the bad (light) coin.
If the (A
3
, A
4
, B
2
) side is heavier, the bad coin must be A
3
, A
4
, or
B
1
. Use the same strategy as in the previous case.
If they are equal, the bad coin must be B
3
or B
4
. Simply weigh B
3
against a good coin.
(2) The B group is heavier than the A group. This case is the same as the
previous one, but with heavy switched with light.
(3) The A and B groups balance. So the bad coin is a C.
Weigh (C
1
, C
2
) against (C
3
, good-coin). There are three possible out-
comes:
If the (C
1
, C
2
) side is heavier, weigh C
1
against C
2
. If C
1
is heavier,
it is the bad (heavy) coin; if C
2
is heavier, it is the bad (heavy) coin;
if they are equal, C
3
is the bad (light) coin.
If the (C
3
, good-coin) side is heavier, this is equivalent to the previous
case, with heavy switched with light.
If they are equal, the bad coin is C
4
. Weigh C
4
against a good coin
to determine if it is heavy or light.
1
(b) Lemma: Let there be N coins, about which our information is the following:
The N coins may be divided into two sets, {H} and {L}, such that i) if a
coin is in {H} and it turns out to be the bad coin, it is heavy; and ii) if a
coin is in {L} and it turns out to be the bad coin, it is light. Then, given n
weighings, the maximum value of N for which we can identify the bad coin,
and also determine whether it is heavy or light, is N = 3
n
.
Proof: For the case n = 0, the lemma is obviously true, because by assump-
tion we know which of the two sets, {H} and {L}, the one coin is in. We will
show by induction that the lemma is true for all n.
Assume the lemma true for n weighings. Let us show that it is then true for
n + 1 weighings. We will do this by rst showing that N = 3
n+1
is solvable,
and then showing that N = 3
n+1
+ 1 is not always solvable.
By assumption, the N = 3
n+1
coins are divided into {H} and {L} sets. On
both sides of the scale, put h coins from {H} and l coins from {L}, with
h +l = 3
n
. (Either h or l may be zero, if necessary.) There are then 3
n
coins
left over.
There are three possible outcomes to this weighing:
If the left side is heavier, the bad coin must be one of the h H-coins from
the left or one of the l L-coins from the right.
If the right side is heavier, the bad coin must be one of the h H-coins
from the right or one of the l L-coins from the left.
If the scale balances, the bad coin must be one of the 3
n
leftover coins.
In each of these cases, the problem is reduced to a setup with 3
n
coins which
are divided into {H} and {L} sets. But this is assumed to be solvable with n
weighings, by induction. Therefore, since N = 3
0
= 1 is solvable for n = 0, we
conclude that N = 3
n
is solvable for all n.
Let us now show that N = 3
n+1
+1 is not always solvable with n+1 weighings.
Assume inductively that N = 3
n
+ 1 is not always solvable with n weighings.
(N = 3
0
+ 1 = 2 is certainly not solvable for n = 0.) For the rst weighing,
the leftover pile can have at most 3
n
coins in it, since the bad coin may end up
being there. There must therefore be at least 2 3
n
+1 total coins on the scale
(which then implies that there must be at least 2 3
n
+ 2 total coins on the
scale, since the number must be even). Depending on how the {H} and {L}
coins are distributed on the scale, the rst weighing will (assuming the scale
doesnt balance) tell us that the bad coin is either in a subset containing s
coins, or in the complementary subset containing (2 3
n
+2) s coins. One of
these sets will necessarily have at least 3
n
+1 coins in it, which by assumption
is not necessarily solvable.
Returning to the original problem, let us rst consider a modied setup where
we have an additional known good coin at our disposal.
Claim: Given N coins and W weighings, and given an additional known
good coin, the maximum value for N for which we can identify the bad coin,
2
and also determine whether it is heavy or light, is N
g
W
= (3
W
1)/2, where
the superscript g signies that we have a known good coin available.
Proof: The claim is true for W = 1. Assume inductively that it is true for
W weighings. We will show that it is then true for W + 1 weighings.
In the rst of our W+1 weighings, we may have (by the inductive assumption)
at most (3
W
1)/2 leftover coins not involved in the weighing, since the bad
coin may end up being there (in which case we have many good coins from
the scale at out disposal).
From the above lemma, we may have at most 3
W
suspect coins on the scale.
We can indeed have this many, if we bring in a known good coin to make the
number of weighed coins, 3
W
+1, even (so that we can have an equal number
on each side). If the scale doesnt balance, the 3
W
suspect coins satisfy the
hypotheses of the lemma (they can be divided into {H} and {L} sets), so if
the bad coin is among these 3
W
coins, it can be determined in W weighings.
Therefore,
N
g
W+1
= N
g
W
+ 3
W
=
3
W
1
2
+ 3
W
=
3
W+1
1
2
, (1)
as we wanted to show.
Corollary: Given N coins and W weighings (and not having an additional
known good coin available), the maximum value of N for which we can identify
the bad coin, and also determine whether it is heavy or light, is
N
ng
w
=
3
W
1
2
1, (2)
where the superscript ng signies that we do not have a known good coin
available.
Proof: If we are not given a known good coin, the only modication to the
reasoning in the above claim is that we cant put a total of 3
W
suspect coins
on the scale, because 3
W
is odd. We are limited to a total of 3
W
1 coins on
the scale, so we now obtain
N
ng
W+1
= N
g
W
+ (3
W
1) =
3
W
1
2
+ (3
W
1) =
3
W+1
1
2
1. (3)
Note that if the scale balances, so that we know the bad coin is a leftover coin,
then from that point on, we do indeed have a known good coin at our disposal
(any coin on the scale), so N
g
W
= (3
W
1)/2 is indeed what appears in the
above equation.
Therefore, N
W
is decreased by one if we dont have a known good coin at the
start.
Remark: It is possible to write down an upper bound for N
g
W
and N
ng
W
, without
going through all of the above work. We may do this by considering the number of
possible outcomes of the W weighings. There are 3 possibilities for each weighing
3
(left side heavier, right side heavier, or both sides equal), so there are at most 3
W
possible outcomes. Each of these outcomes may be labelled by a string of W letters,
for example, LLRER if W = 5 (with L for left, etc.).
However, the EEEEE string (where the scale always balances) does not give enough
information to determine whether the bad coin is heavy or light. Also, the mir-
ror image outcome (namely RRLEL, for the LLRER case above) corresponds to
equivalent information, with left and right simply reversed. Therefore, there
are (3
W
1)/2 eectively dierent strings. Hence, we can start with no more than
(3
W
1)/2 coins (because we can imagine initially labelling each coin with a string,
so that if that particular string is the one obtained, then the corresponding coin is
the bad coin).
As we saw above, this upper bound of (3
W
1)/2 is obtainable if we have an additional
known good coin at our disposal. But we fall short of the bound by one, if we initially
do not have an additional good coin.
Note that although there may be dierent possible strategies for coin placement at
various points in the weighing process (so that there are actually far more than 3
W
possible outcomes, taking into consideration dierent placements of coins), only 3
W
of the possible outcomes are realizable in a given scheme. Whatever weighing strategy
you pick, you can write down an if, then tree before you start the process. Once
you pick a scheme, there are 3
W
possible outcomes (and thus (3
W
1)/2 eectively
dierent outcomes).
4
Solution
Week 35 (5/12/03)
Rising hoop
Let be the angle through which the bead has fallen, and let N be the normal
force from the hoop on the bead, with inward taken to be positive. Then the radial
F = ma equation for the bead is
N + mg cos =
mv
2
R
. (1)
The height the bead has fallen is R Rcos , so conservation of energy gives
mv
2
2
= mgR(1 cos ) = v
2
= 2gR(1 cos ). (2)
Therefore, the radial F = ma equation becomes
N =
mv
2
R
mg cos
= 2mg(1 cos ) mg cos
= mg(2 3 cos ). (3)
By Newtons third law, this is the force from the bead on the hoop, with outward
taken to be positive. Note that this force is positive (that is, the bead pulls outward
on the hoop) if > cos
1
(2/3) 48.2

.
Since there are two beads, the total upward force on the hoop from the beads is
2N cos = 2mg(2 3 cos ) cos . (4)
The that yields the maximum value of this upward force is obtained by taking the
derivative, which gives
0 =
d
d
(2 cos 3 cos
2
)
= 2 sin + 6 sin cos . (5)
Therefore, the maximum value is achieved when cos = 1/3, in which case the
upward force equals
2mg

2 3

1
3

1
3

=
2mg
3
. (6)
The hoop will rise up o the ground if this maximum upward force is larger than
the weight of the hoop. That is, if
2mg
3
> Mg =
m
M
>
3
2
. (7)
Remark: Alternatively, we can solve for the minimum value of m/M by setting the upward
force, 2mg(2 3 cos ) cos , equal to the weight of the hoop, Mg, and then using the
quadratic formula to solve for cos . A solution for cos exists only if the discriminant is
positive, which is the case only if m/M > 3/2.
1
Solution
Week 36 (5/19/03)
Monochromatic Triangle
(a) Let us try to avoid forming a monochromatic triangle, a task that we will show
is impossible. Consider one point and the sixteen lines drawn from it to the
other sixteen points. From the pigeonhole principle (if you have n pigeons and
n 1 pigeonholes, then at least two pigeons must go in one pigeonhole), we
see that at least six of these lines must be of the same color. Let this color be
red.
Now consider the six points at the ends of these red lines. Look at the lines
going from one of these points to the other ve. In order to not form a red
triangle, each of these ve lines must be either green or blue. Hence (by the
pigeonhole principle) at least three of them must be of the same color. Let
this color be green.
Finally, consider the three points at the ends of the three green lines. If any
one of the three lines connecting them is red, a red triangle is formed. And if
any one of the three lines connecting them is green, a green triangle is formed.
Therefore, they must all be blue, and a blue triangle is formed.
(b) Consider the problem for the case of n = 4, in order to get an idea of how the
solution generalizes. We claim that 66 points will necessitate a monochromatic
triangle. As in the case of n = 3, isolate one point and paint all the lines from it
to the other points. Since we have 65 other points and 4 colors, the pigeonhole
principle requires that at least 17 of these lines be of the same color. In order
to not have a monochromatic triangle, the lines joining the endpoints of these
17 lines must use only the remaining three colors, and the problem is reduced
to the case of n = 3.
Generalizing this reasoning yields the following result:
Claim: If n colors and P
n
points necessitate a monochromatic triangle, then
n + 1 colors and
P
n+1
= (n + 1)(P
n
1) + 2 (1)
points also necessitate a monochromatic triangle.
Proof: Isolate one point, and paint each of the (n + 1)(P
n
1) + 1 lines to
the other points one of n+1 colors. From the pigeonhole principle, at least P
n
of these lines must be the same color. In order to not have a monochromatic
triangle, the points at the ends of these P
n
lines must be joined by the n other
colors. But by hypothesis, there must then be a monochromatic triangle.
If we use the above recursion relation in itself (that is, if we write the P
n
in
eq. (1) in terms of P
n1
, and then write P
n1
in terms of P
n2
, and so on),
the pattern becomes clear. Using the initial condition P
1
= 3, we arrive at the
1
following expression for P
n
:
P
n
= n!

1 +
1
1!
+
1
2!
+ +
1
n!

+ 1, (2)
as you can easily verify. The sum in the parentheses is smaller than e by a
margin that is less than 1/n!. Therefore, P
n
does indeed equal the smallest
integer greater than n!e.
Remark: For n = 1, 2, 3, the numbers n!e (which equal 3,6,17, respectively) are
the smallest numbers which necessitate a monochromatic triangle. (For n = 1, two
points dont even form a triangle. And for n = 2, you can easily construct a diagram
that doesnt contain a monochromatic triangle. For n = 3, things are much more
dicult, but in 1968 Kalbeisch and Stanton showed that 16 points do not necessitate
a monochromatic triangle.) For n 4, the problem of nding the smallest number of
points that necessitate a monochromatic triangle is unsolved, I believe.
2
Solution
Week 37 (5/26/03)
Bouncing down a plane
Let us ignore the tilt of the plane for a moment and determine how the
f
and v
f
after a bounce are related to the
i
and v
i
before the bounce (where v denotes the
velocity component parallel to the plane). Let the positive directions of velocity
and force be to the right along the plane, and let the positive direction of angular
velocity be counterclockwise. If we integrate the force and torque over the small
time of a bounce, we obtain
F =
dP
dt
=

F dt = P,
=
dL
dt
=

dt = L. (1)
But = RF. And since R is constant, we have
L =

RF dt = R

F dt = RP. (2)
Therefore,
I(
f

i
) = Rm(v
f
v
i
). (3)
But conservation of energy gives
1
2
mv
2
f
+
1
2
I
2
f
=
1
2
mv
2
i
+
1
2
I
2
i
= I(
2
f

2
i
) = m(v
2
i
v
2
f
). (4)
Dividing this equation by eq. (3) gives
1
R(
f
+
i
) = (v
f
+ v
i
). (5)
We can now combine this equation with eq. (3), which can be rewritten (using
I = (2/5)mR
2
) as
2
5
R(
f

i
) = v
f
v
i
. (6)
Given v
i
and
i
, the previous two equations are two linear equations in the two
unknowns, v
f
and
f
. Solving for v
f
and
f
, and then writing the result in matrix
notation, gives

v
f
R
f

=
1
7

3 4
10 3

v
i
R
i

v
i
R
i

. (7)
1
We have divided out the trivial
f
=
i
and v
f
= v
i
solution, which corresponds to slipping
motion on a frictionless plane. The nontrivial solution we will nd shortly is the non-slipping one.
Basically, to conserve energy, there must be no work done by friction. But since work is force
times distance, this means that either the plane is frictionless, or that there is no relative motion
between balls contact point and the plane. Since we are given that the plane has friction, the latter
(non-slipping) case must be the one we are concerned with.
1
Note that
A
2
=
1
49

49 0
0 49

= I. (8)
Now let us consider the eects of the tilted plane. Since the balls speed perpen-
dicular to the plane is unchanged by each bounce, the ball spends the same amount of
time in the air between any two successive bounces. This time equals T = 2V/g cos ,
because the component of gravity perpendicular to the plane is g cos . During this
time, the speed along the plane increases by (g sin)T = 2V tan V
0
.
Let Q denote the (v, R) vector at a given time (where v denotes the veloc-
ity component parallel to the plane). The ball is initially projected with Q = 0.
Therefore, right before the rst bounce, we have Q
before
1
= (V
0
, 0) V
0
. (We have
used the fact that doesnt change while the ball is in the air.) Right after the
rst bounce, we have Q
after
1
= AV
0
. We then have Q
before
2
= AV
0
+ V
0
, and so
Q
after
2
= A(AV
0
+V
0
). Continuing in this manner, we see that
Q
before
n
= (A
n1
+ +A+I)V
0
, and
Q
after
n
= (A
n
+ +A
2
+A)V
0
. (9)
However, A
2
= I, so all the even powers of A equal I. The value of Q after the nth
bounce is therefore given by
n even = Q
after
n
=
n
2
(A+I)V
0
.
n odd = Q
after
n
=
1
2

(n + 1)A+ (n 1)I

V
0
. (10)
Using the value of A dened in eq. (7), we nd
n even =

v
n
R
n

=
n
7

5 2
5 2

V
0
0

.
n odd =

v
n
R
n

=
1
7

5n 2 2n 2
5n 5 2n 5

V
0
0

. (11)
Therefore, the speed along the plane after the nth bounce equals (using V
0

2V tan)
v
n
=
10nV tan
7
(n even), and v
n
=
(10n 4)V tan
7
(n odd). (12)
Remark: Note that after an even number of bounces, eq. (11) gives v = R. This is
the rolling condition. That is, the angular speed exactly matches up with the translation
speed, so v and are unaected by the bounce. (The vector (1, 1) is an eigenvector of
A.) At the instant that an even-n bounce occurs, the v and are the same as they would
be for a ball that simply rolls down the plane. At the instant after an odd-n bounce, the v
is smaller than it would be for the rolling ball, but the is larger. (Right before an odd-n
bounce, the v is larger but the is smaller.)
2
Solution
Week 38 (6/2/03)
Sum over 1
(a) First Solution: We will use the following fact: Given n random numbers
between 0 and 1, the probability P
n
(1) that their sum does not exceed 1 equals
1/n!.
1
To prove this, let us prove a slightly stronger result:
Theorem: Given n random numbers between 0 and 1, the probability P
n
(s)
that their sum does not exceed s equals s
n
/n! (for all s 1).
Proof: Assume inductively that the result holds for a given n. (It clearly
holds for all s 1 when n = 1.) What is the probability that n + 1 numbers
sum to no more than t (with t 1)? Let the (n +1)st number have the value
x. Then the probability P
n+1
(t) that all n + 1 numbers sum to no more than
t equals the probability P
n
(t x) that the rst n numbers sum to no more
than t x. But P
n
(t x) = (t x)
n
/n!. Integrating this probability over all
x from 0 to t gives
P
n+1
(t) =
_
t
0
(t x)
n
n!
dx =
(t x)
n+1
(n + 1)!

t
0
=
t
n+1
(n + 1)!
. (1)
We see that if the theorem holds for n, then it also holds for n+1. Therefore,
since the theorem holds for all s 1 when n = 1, it holds for all s 1 for any
n.
We are concerned with the special case s = 1, for which P
n
(1) = 1/n!.
The probability that it takes exactly n numbers for the sum to exceed 1 equals
1/(n1)! 1/n!. This is true because the rst n1 numbers must sum to less
than 1, and the nth number must push the sum over 1, so we must subtract
o the probability that is does not.
The expected number of numbers, N, to achieve a sum greater than 1, is
therefore
N =

2
n
_
1
(n 1)!

1
n!
_
=

2
1
(n 2)!
= e. (2)
Could it really have been anything else?
Second Solution: We will use the result, P
n
(s) = s
n
/n!, from the rst
solution. Let F
n
(s) ds be the probability that the sum of n numbers is between
1
The number 1/n! is the volume of the region in n dimensions bounded by the coordinate planes
and the hyperplane x1 +x2 + +xn = 1. For example, in two dimensions we have a triangle with
area 1/2; in three dimensions we have a pyramid with volume 1/6; etc.
1
s and s +ds. Then F
n
(s) is simply the derivative of P
n
(s), with respect to s.
Therefore, F
n
(s) = s
n1
/(n 1)!.
In order for it to take exactly m numbers for the sum to exceed 1, two things
must happen: (1) the sum of the rst m 1 numbers must equal a number,
s, less than 1; this occurs with probability density s
m2
/(m 2)!. And (2)
the mth number must push the sum over 1, that is, the mth number must be
between 1 s and 1; this occurs with probability s.
The probability that it takes exactly m numbers for the sum to exceed 1 is
therefore
_
1
0
s(s
m2
/(m2)!) ds. The expected number of numbers, N, to
achieve a sum greater than 1, therefore equals
N =

2
m
_
1
0
s
m1
(m2)!
ds
=

2
m
1
m(m2)!
= e. (3)
(b) First Solution: After n numbers have been added, the probability that their
sum is between s and s +ds is, from above, F
n
(s) = s
n1
/(n1)! (for s 1).
There is a probability of s that the (n+1)st number pushes the sum over 1. If
this happens, then (since this last number must be between 1s and 1, and is
evenly distributed) the average result will be equal to s +(1 s/2) = 1 +s/2.
The expected sum therefore equals
S =

1
_
_
1
0
_
1 +
s
2
_
s
s
n1
(n 1)!
ds
_
=

1
3n
2
+ 5n
2(n + 2)!
=

1
3(n + 2)(n + 1) 4(n + 2) + 2
2(n + 2)!
=

1
_
3
2n!

2
(n + 1)!
+
1
(n + 2)!
_
=
3
2
(e 1) 2(e 2) + (e 5/2)
=
e
2
. (4)
Second Solution: Each of the random numbers has an average value of 1/2.
Therefore, since it takes (on average) e numbers for the sum to exceed 1, the
average value of the sum will be e/2.
This reasoning probably strikes you as being either completely obvious or com-
pletely mysterious. In the case of the latter, imagine playing a large number
of games in succession, writing down each of the random numbers in one long
sequence. (You can note the end of each game by, say, putting a mark after
2
the nal number, but this is not necessary.) If you play N games (with N very
large), then the result from part (a) shows that there will be approximately
Ne numbers listed in the sequence. Each number is a random number between
0 and 1, so the average value is 1/2. The sum of all of the numbers in the
sequence is therefore approximately Ne/2. Hence, the average total per game
is e/2.
3
Solution
Week 39 (6/9/03)
Viewing the spokes
The contact point on the ground does not look blurred, because it is instantaneously
at rest. But although this is the only point on the wheel that is at rest, there will
be other locations in the picture where the spokes do not appear blurred.
The relevant property of a point in the picture where a spoke does not appear
blurred is that the point lies on the spoke throughout the duration of the cameras
exposure. (The point in the picture need not, however, actually correspond to the
same atom on the spoke.)
At a given time, consider a spoke in the lower half of the wheel. A short time
later, the spoke will have moved (via both translation and rotation), but it will
intersect its original position. The spoke will not appear blurred at this intersection
point. We must therefore nd the locus of these intersections. We can do this in
two ways.
First method: Let R be the radius of the wheel. Consider a spoke that makes an
angle of with the vertical at a given time. If the wheel then rolls through a small
angle d, the center moves a distance Rd. The spokes motion is a combination
of a translation through this distance Rd, plus a rotation through the angle d
(around the top end).
Let r be the radial position of the intersection of the initial and nal positions of
the spoke. In the following gure, we can write down two expressions for the short
segment drawn perpendicular to the initial position of the spoke.
Rd
d

r
The two expression are (Rd) cos , and r d (to rst order in d). Equating these
gives r = Rcos . This describes a circle whose diameter is the lower vertical radius
of the wheel, as shown below.
1
R

r=Rcos
wheel
Second method: Since the wheels contact point with the ground is instanta-
neously at rest, the wheel may be considered to be instantaneously rotating around
this point. This means that every atom in the wheel (both in the spokes and the
rim) instantaneously traces out the arc of a circle centered at the contact point. For
the point P in the gure below, this circle is the dotted one drawn (with Q being
the contact point). A spoke will not appear blurred at the point where this circular
motion is along the direction of the spoke. That is, a spoke will not appear blurred
at the point where the dotted circle is tangent to the spoke, as shown.
R

wheel
P
O
Q
We are therefore concerned with the locus of all points P such that the segments
PQ and PO are perpendicular. As seen in the previous diagram above, this locus
is the circle whose diameter is the lower vertical radius of the wheel.
2
Solution
Week 40 (6/16/03)
Staying ahead
Consider a two-dimensional lattice on which a vote for A is signied by a unit step in
the positive x-direction, and a vote for B by a unit step in the positive y-direction.
Then the counting of the votes until A has a votes and B has b votes corresponds
to a path from the origin to the point (a, b), with a > b. There are
_
a+b
a
_
such paths
(because any a steps of the total a + b steps can be chosen to be the ones in the
x-direction). All of these paths from the origin to (a, b) are equally likely, as you
can show.
1
The probability that a particular path corresponds to the way the votes
are counted is thus 1/
_
a+b
a
_
.
The problem can therefore be solved by nding the number, N
g
, of paths that
reach the point (a, b) without passing through the region y > x. (Well call these
the good paths; hence the subscript g.)
It will actually be easier to nd the number, N
b
, of paths that reach the point
(a, b) and that do pass through the region y > x. (Well call these the bad paths.)
The desired probability that As sub-total is always greater than or equal to Bs
sub-total equals 1 N
b
/
_
a+b
a
_
Claim: The number of bad paths from the origin to (a, b) (that is, the number
of paths that pass through the region y > x) equals N
b
=
_
a+b
b1
_
.
Proof: The number of bad paths from (0, 0) to (a, b) equals the number of bad
paths from (0, 1) to (a, b) plus the number of bad paths from (1, 0) to (a, b). Lets
look at these two classes of paths.
The number of bad paths from (0, 1) to (a, b) is simply all of the paths from
(0, 1) to (a, b), which equals
_
a+b1
b1
_
.
The number of bad paths from (1, 0) to (a, b) equals the number of paths from
(1, 2) to (a, b). This follows from the fact that any bad path from (1, 0) to
(a, b) must proceed via a point on the line y = x + 1, as shown below. Hence,
there is a one-to-one correspondence between the bad paths starting at (1, 0)
and all of the paths starting at (1, 2). This correspondence is obtained by
reecting, in the line y = x + 1, the portion of the bad path between (1, 0)
and the rst point of contact with y = x + 1. An example of such a reection
is shown below. The number of paths from (1, 2) to (a, b) is
_
a+b1
b2
_
.
1
The probabilities are all equal to a!b!/(a + b)!, as you can see by picking an arbitrary path and
writing down the product of the probabilities of each step. This result can be written as 1/
_
a+b
a
_
,
which makes sense, because
_
a+b
a
_
is the total number of paths.
1
y=x
y=x+1
(a,b)
(-1,2)
(1,0)
The total number of bad paths from (0, 0) to (a, b) is therefore
N =
_
a +b 1
b 1
_
+
_
a +b 1
b 2
_
=
_
a +b
b 1
_
. (1)
The probability that As sub-total is always greater than or equal to Bs sub-total
is therefore
P
AB
= 1
_
a +b
b 1
_
_
_
a +b
a
_
= 1
b
a + 1
. (2)
Remarks:
1. We may also ask what the probability is that As sub-total is always strictly greater
than Bs. The same reection reasoning holds, except that now we must reect
across the line y = x, because any path that touches the line y = x is now bad.
The two classes of paths in the above proof are now identical (theyre both the ones
that go from (0, 1) to (a, b)). Therefore, the number of bad paths is twice the number
of paths from (0, 1) to (a, b), that is, 2
_
a+b1
b1
_
.
The probability that As sub-total is always strictly greater than Bs sub-total is
therefore
P
A>B
= 1 2
_
a +b 1
b 1
_
_
_
a +b
b
_
=
a b
a +b
. (3)
2. Lets consider the special case (of the original problem) where a = b, that is, where
the election ends in a tie. From eq. (2), we see that that if a = b n, then
P
AB
= 1/(n + 1). There is another very nice method of solving this special case. It
is the method of generating functions, and it proceeds as follows.
Let A
n
be the total number of good paths that go from (0, 0) to (n, n), and let B
n
be the number of good paths that touch the line y = x for the rst time on the
2nth step. (The A
n
and B
n
here have nothing to do with the people A and B given in
the problem.) We will nd a relation between A
n
and B
n
, and then nd a recursion
relation for A
n
.
The relation between the A
n
and B
n
may be found as follows. The number, B
n
,
of good paths touching y = x for the rst time after 2n steps equals the number of
paths which go from (1, 0) to (n, n1) without touching the line y = x. If we imagine
shifting out coordinate system, we see that this equals the number of paths that go
from (0, 0) to (n1, n1) without going above the line y = x. But this is just A
n1
.
So we have
B
n
= A
n1
. (4)
2
Now lets nd a recursion relation for A
n
. We may categorize the A
n
good paths
from (0, 0) to (n, n) according to when they rst touch the line y = x. There will be,
for example, some paths that touch this line after 2 steps. The number of these is
B
1
A
n1
. This is true because there are B
1
(= 1) paths that touch the line for the
rst time at the point (1,1), and then there are A
n1
good paths that go from (1, 1)
to (n, n). Likewise, there are B
2
A
n2
paths that touch y = x for the rst time at
(2, 2), on the way to (n, n). Continuing in this fashion, we nd
A
n
= B
1
A
n1
+B
2
A
n2
+ +B
n1
A
1
+B
n
A
0
, (5)
where A
0
is dened to be 1.
Using eq. (4), we obtain the recursion relation
A
n
= A
0
A
n1
+A
1
A
n2
+ +A
n2
A
1
+A
n1
A
0
, (6)
We can calculate the rst few of these. They are A
0
= 1, A
1
= 1, A
2
= 2, A
3
= 5,
A
4
= 14, and A
5
= 42. A lucky guess can produce the result A
n
=
_
2n
n
_
/(n + 1), but
lets obtain this in a more deductive way.
A nice way to solve recursion relations such as the one in eq. (6) is to use generating
functions. By this, we mean consider the polynomial
F(x) = A
0
+A
1
x +A
2
x
2
+A
3
x
3
+ . (7)
The key in this problem is to note that if we square F(x) and use eq. (6), we obtain
(F(x))
2
= A
0
A
0
+ (A
0
A
1
+A
1
A
0
)x + (A
0
A
2
+A
1
A
1
+A
2
A
0
)x
2
+(A
0
A
3
+A
1
A
2
+A
2
A
1
+A
3
A
0
)x
3
+
= A
1
+A
2
x +A
3
x
2
+A
4
x
3
+
= (F(x) 1)/x. (8)
Therefore, F(x) satises the equation,
xF
2
F + 1 = 0. (9)
The solution to this is
F =
1

1 4x
2x
. (10)
We must now expand the square root in a Taylor series in order to obtain the coe-
cients of the various powers of x. It is clear that we must choose the minus sign if we
want the coecient of x
0
(which is A
0
) to be 1. By taking a plethora of derivatives,
you can show that the Taylor expansion of

1 +y can be written as
_
1 +y = 1 +

n=1
1
n!
__
1
2
__
1
2
__
3
2
__
5
2
_

_
(2n 3)
2
__
y
n
= 1 +

n=1
_
(1)
n1
(2(n 1))!
2
2n1
n!(n 1)!
_
y
n
. (11)
Therefore, with y (4x), eq. (10) gives
F =

m=0
_
(2m)!
m!(m+ 1)!
_
x
m
. (12)
Thus,
A
n
=
(2n)!
n!(n + 1)!
=
1
n + 1
_
2n
n
_
(13)
is the number of good paths that go from (0, 0) to (n, n). Since the total number
of paths from the origin to (n, n) is
_
2n
n
_
, the probability of a good path is 1/(n + 1),
as we wanted to show.
3
Solution
Week 41 (6/23/03)
Speedy travel
Let the earths mass be M, and let its radius be R. Consider the object when it is
a distance x from the center of the tube, at radius r, as shown.
R
m
x
r

The gravitational force on the object is due to the mass of the earth that is inside
the radius r. Since mass is proportional to volume, the mass inside the radius r is
M(r/R)
3
. The force on the object is therefore
F =
G

M(r/R)
3

m
r
2
=
GMmr
R
3
. (1)
We are interested in the component of the force along the tube, which is
F
x
= F sin =
GMmr
R
3

x
r

GMm
R
3

x. (2)
Therefore, F = ma along the tube gives

GMm
R
3

x = m x = x =

GM
R
3

x. (3)
This equation describes simple harmonic motion, with frequency
=

GM
R
3
. (4)
Therefore, the roundtrip time is
T =
2

= 2

R
3
GM
= 2

(6.38 10
6
m)
3
(6.67 10
11
m
3
/kg s
2
)(5.98 10
24
kg)
5060 s
84 minutes. (5)
1
The one-way time to the other end of the tube is therefore 42 minutes. Thats quick!
Note that this result is independent of where chord is. The chord can be a diameter
of the earth, or it can be a straight tube spanning the 20-foot width of a room.
Neglecting friction, an object will take 42 minutes to go from one side of the room
to the other. The very slight component of gravity along the tube is enough to do
the trick.
2
Solution
Week 42 (6/30/03)
How much change?
If the item costs between N/2 and N dollars, then you can buy only one item. These
extremes will produce remainders of N/2 and 0, respectively. The average amount
of money left over in this region, which has length N(1 1/2) = N/2, is therefore
N/4.
Likewise, if the item costs between N/3 and N/2 dollars, then you can buy only
two items. These extremes will produce remainders of N/3 and 0, respectively. The
average amount of money left over in this region, which has length N(1/2 1/3) =
N/6, is therefore N/6.
Continuing in this manner, we see that if the item costs between N/(n +1) and
N/n, then you can buy only n items. These extremes will produce remainders of
N/(n+1) and 0, respectively. The average amount of money left over in this region,
which has length N(1/n 1/(n + 1)) = N/n(n + 1), is therefore N/2(n + 1).
The expected amount, M, of money left over is therefore
M =

n=1

1
n

1
n + 1

N
2(n + 1)
=
N
2

n=1

1
n(n + 1)

1
(n + 1)
2

=
N
2

n=1

1
n

1
n + 1

1
(n + 1)
2

=
N
2

2
6
1

= N

1

2
12

.
In the third line, we have used the fact that the sum in brackets telescopes to 1, and
also that

n=1
1/n
2
=
2
/6. Since
2
/12 0.82, the amount of money left over is
roughly (0.18)N dollars. Note that what we have essentially done in this problem
is nd the area under the graph in the following gure.
price/N
change
N/2
1/2 1 1/3
N/3
1/4
N/4
.
.
.
.
1
Solution
Week 43 (7/7/03)
Innite Atwoods machine
First Solution: If the strength of gravity on the earth were multiplied by a factor
, then the tension in all of the strings in the Atwoods machine would likewise be
multiplied by . This is true because the only way to produce a quantity with the
units of tension (that is, force) is to multiply a mass by g. Conversely, if we put
the Atwoods machine on another planet and discover that all of the tensions are
multiplied by , then we know that the gravity there must be g.
Let the tension in the string above the rst pulley be T. Then the tension in
the string above the second pulley is T/2 (because the pulley is massless). Let the
downward acceleration of the second pulley be a
2
. Then the second pulley eectively
lives in a world where gravity has strength g a
2
.
Consider the subsystem of all the pulleys except the top one. This innite
subsystem is identical to the original innite system of all the pulleys. Therefore,
by the arguments in the rst paragraph above, we must have
T
g
=
T/2
g a
2
, (1)
which gives a
2
= g/2. But a
2
is also the acceleration of the top mass, so our answer
is g/2.
Remarks: You can show that the relative acceleration of the second and third pulleys is
g/4, and that of the third and fourth is g/8, etc. The acceleration of a mass far down in
the system therefore equals g(1/2 + 1/4 + 1/8 + ) = g, which makes intuitive sense.
Note that T = 0 also makes eq. (1) true. But this corresponds to putting a mass of zero
at the end of a nite pulley system (see the following solution).
Second Solution: Consider the following auxiliary problem.
Problem: Two setups are shown below. The rst contains a hanging mass m.
The second contains a pulley, over which two masses, m
1
and m
2
, hang. Let both
supports have acceleration a
s
downward. What should m be, in terms of m
1
and
m
2
, so that the tension in the top string is the same in both cases?
1 2
m
m m
a
s
a
s
Answer: In the rst case, we have
mg T = ma
s
. (2)
1
In the second case, let a be the acceleration of m
2
relative to the support (with
downward taken to be positive). Then we have
m
1
g
T
2
= m
1
(a
s
a),
m
2
g
T
2
= m
2
(a
s
+ a). (3)
Note that if we dene g

g a
s
, then we may write the above three equations as
mg

= T,
m
1
g

=
T
2
m
1
a,
m
2
g

=
T
2
+ m
2
a. (4)
Eliminating a from the last two of these equations gives 4m
1
m
2
g

= (m
1
+ m
2
)T.
Using this value of T in the rst equation then gives
m =
4m
1
m
2
m
1
+ m
2
. (5)
Note that the value of a
s
is irrelevant. (We eectively have a xed support in a
world where the acceleration from gravity is g

.) This auxiliary problem shows that

the two-mass system in the second case may be equivalently treated as a mass m,
given by eq. (5), as far as the upper string is concerned.
Now lets look at our innite Atwoods machine. Start at the bottom. (Assume that
the system has N pulleys, where N .) Let the bottom mass be x. Then the
auxiliary problem shows that the bottom two masses, m and x, may be treated as
an eective mass f(x), where
f(x) =
4mx
m + x
=
4x
1 + (x/m)
. (6)
We may then treat the combination of the mass f(x) and the next m as an eective
mass f(f(x)). These iterations may be repeated, until we nally have a mass m and
a mass f
(N1)
(x) hanging over the top pulley. So we must determine the behavior
of f
N
(x), as N . This behavior is clear if we look at the following plot of f(x).
2
m
m
2m
3m
4m
2m 3m 4m 5m
x
f (x)
f (x)
=
x
Note that x = 3m is a xed point of f. That is, f(3m) = 3m. This plot shows
that no matter what x we start with, the iterations approach 3m (unless we start
at x = 0, in which case we remain there). These iterations are shown graphically by
the directed lines in the plot. After reaching the value f(x) on the curve, the line
moves horizontally to the x value of f(x), and then vertically to the value f(f(x))
on the curve, and so on.
Therefore, since f
N
(x) 3m as N , our innite Atwoods machine is
equivalent to (as far as the top mass is concerned) just two masses, m and 3m. You
can then quickly show that that the acceleration of the top mass is g/2.
Note that as far as the support is concerned, the whole apparatus is equivalent
to a mass 3m. So 3mg is the upward force exerted by the support.
3
Solution
Week 44 (7/14/03)
Relatively prime numbers
The probability that two random numbers both have a given prime p as a factor is
1/p
2
. The probability that they do not have p as a common factor is thus 1 1/p
2
.
Therefore, the probability that two numbers have no common prime factors is
P = (1 1/2
2
)(1 1/3
2
)(1 1/5
2
)(1 1/7
2
)(1 1/11
2
) . (1)
Using
1
1 x
= 1 +x +x
2
+x
3
+ , (2)
this can be rewritten as
P =

(1 + 1/2
2
+ 1/2
4
+ )(1 + 1/3
2
+ 1/3
4
+ )

1
. (3)
By the Unique Factorization Theorem (every positive integer is expressible as the
product of primes in exactly one way), we see that the previous expression is equiv-
alent to
P = (1 + 1/2
2
+ 1/3
2
+ 1/4
2
+ 1/5
2
+ 1/6
2
+ )
1
. (4)
And since the sum of the squares of the reciprocals of all of the positive integers is
known to be
2
/6, the desired probability is P = 6/
2
61%.
Remarks:
1. The probability that n random numbers all have a given prime p as a factor is 1/p
n
.
So the probability that they do not all have p as a common factor is 1 1/p
n
. In
exactly the same manner as above, we nd that the probability, P
n
, that n numbers
have no common factor among all of them is
P
n
= (1 + 1/2
n
+ 1/3
n
+ 1/4
n
+ 1/5
n
+ 1/6
n
+ )
1
. (5)
This is, by denition, the Riemann zeta function, (n). It can be calculated exactly
for even values of n, but only numerically for odd values. For the case of n = 4, we
can use the known value (4) =
4
/90 to see that the probability that four random
numbers do not all have a common factor is P
4
= 90/
4
92%.
2. We can also perform the somewhat silly exercise of applying this result to the case
of n = 1. The question then becomes: What is the probability, P
1
, that a randomly
chosen positive integer does not have a factor? Well, 1 is the only positive integer
without any factors, so the probability is 1/= 0. And indeed,
P
1
= (1 1/2)(1 1/3)(1 1/5)(1 1/7) . . .
= (1 + 1/2 + 1/3 + 1/4 + 1/5 + 1/6 )
1
= 1/, (6)
because the sum of the reciprocals of all of the positive integers diverges.
1
3. Let (n) equal the number of integers less than n that are relatively prime to n. Then
(n)/n equals the probability that a randomly chosen integer is relatively prime to n.
(This is true because any integer is relatively prime to n if and only if its remainder,
when divided by n, is relatively prime to n.) The result of our original problem
therefore tells us that the average value of (n)/n is 6/
2
.
4. To be precise about what we mean by probabilities in this problem, we really should
word the question as: Let N be a very large integer. Pick two random integers less
than or equal to N. What is the probability that these numbers are relatively prime,
in the limit where N goes to innity?
The solution would then be slightly modied, in that the relevant primes p would be
cut o at N, and edge eects due to the nite size of N would have to be considered.
It is fairly easy to see that the answer obtained in this limit is the same as the answer
obtained above.
2
Solution
Week 45 (7/21/03)
Sliding along a plane
The normal force from the plane is N = mg cos . Therefore, the friction force on
the block is N = (tan)N = mg sin. This force acts in the direction opposite to
the motion. The block also feels the gravitational force of mg sin pointing down
the plane.
Because the magnitudes of the friction force and the gravitational force along the
plane are equal, the acceleration along the direction of motion equals the negative
of the acceleration in the direction down the plane. Therefore, in a small increment
of time, the speed that the block loses along its direction of motion exactly equals
the speed that it gains in the direction down the plane. Letting v be the speed of
the block, and letting v
y
be the component of the velocity in the direction down the
plane, we therefore have
v + v
y
= C, (1)
where C is a constant. C is given by its initial value, which is V +0 = V . The nal
value of C is V
f
+ V
f
= 2V
f
(where V
f
is the nal speed of the block), because the
block is essentially moving straight down the plane after a very long time. Therefore,
2V
f
= V = V
f
=
V
2
. (2)
1
Solution
Week 46 (7/28/03)
The birthday problem
(a) Given n people, the probability, P
n
, that there is not a common birthday
among them is
P
n
=
_
1
1
365
__
1
2
365
_

_
1
n 1
365
_
. (1)
The rst factor is the probability that two given people do not have the same
birthday. The second factor is the probability that a third person does not
have a birthday in common with either of the rst two. This continues until
the last factor is the probability that the nth person does not have a birthday
in common with any of the other n 1 people.
We want P
n
< 1/2. If we simply multiply out the above product with suc-
cessive values of n, we nd that P
22
= .524, and P
23
= .493. Therefore, there
must be at least 23 people in a room in order for the odds to favor at least
two of them having the same birthday.
Remark: This answer of n = 23 is much smaller than most people expect, so it
provides a nice betting opportunity. For n = 30, the odds of a common birthday
increase to 70.6%, and most people still nd it hard to believe that among 30 people
there are probably two who have the same birthday. The table below lists various
values of n and the probabilities, 1 P
n
, that at least two people have a common
birthday.
n 10 20 23 30 50 60 70 100
1 P
n
11.7% 41.1% 50.7% 70.6% 97.0% 99.4% 99.92% 99.9994%
Even for n = 50, most people would be happy to bet, at even odds, that no two
people have the same birthday. If they seem a bit hesitant, however, you can simply
oer them the irrefusable odds of 10 to 1. (That is, you put down, for example, $10
on the table, and they put down $1. If there is a common birthday, you take all the
money. If there is no common birthday, they take all the money.) Since there is a
97.0% chance of a common birthday among 50 people, a quick calculation shows that
you will gain, on average, 67 cents for every $10 you put down.
One reason why many people do not believe the n = 23 answer is correct is that they
are asking themselves a dierent question, namely, How many people need to be
present for there to be a 1/2 chance that someone else has my birthday? The answer
to this question is indeed much larger than 23. The probability that no one out of n
people has a birthday on a given day is (1 1/365)
n
. For n = 252, this is just over
1/2. And for n = 253, it is just under 1/2. Therefore, you need to come across 253
other people in order to expect that at least one of them has your birthday.
(b) First Solution: Given n people, and given N days in a year, the reasoning in
part (a) shows that the probability that no two people have the same birthday
is
P
n
=
_
1
1
N
__
1
2
N
_

_
1
n 1
N
_
. (2)
1
If we take the natural log of this equation and use the expansion,
ln(1 x) = (x +x
2
/2 + ), (3)
then the requirement P
n
1/2 becomes

_
1
N
+
2
N
+
n 1
N
_

1
2
_
1
N
2
+
4
N
2
+
(n 1)
2
N
2
_
ln2. (4)
Using the sums,
n

1
k =
n(n + 1)
2
, and
n

1
k
2
=
n(n + 1)(2n + 1)
6
, (5)
we can rewrite eq. (4) as
n(n 1)
2N
+
n(n 1)(2n 1)
12N
2
+ ln2. (6)
For large N, the rst term will be of order 1 when n

N, in which case
the second- and higher-order terms are negligible. Therefore, keeping only the
rst term (which behaves like n
2
/2N), we nd that P
n
is equal to 1/2 when
n

2 ln2

N. (7)
Lets look at a few cases:
For N = 365, eq. (7) gives n = 22.5. Since we must have an integral
number of people, this agrees with the exact result, n = 23.
For N = 24 365 = 8760 (that is, for births in the same hour), we nd
n = 110.2. This agrees with the exact result, n = 111, obtained by
multiplying out eq. (2).
For N = 60 24 365 = 525, 600 (that is, for births in the same minute),
we nd n = 853.6. This agrees with the exact result, n = 854, obtained
by multiplying out eq. (2) (not by hand!). This is a very small number
compared to the more than half a million minutes in a year.
Remarks:
(1) If we wish to ask how many people need to be in a room in order for the
probability to be at least p that some two have the same birthday, then the
above derivation is easily modied to yield
n
_
2 ln(
1
1p
)

N. (8)
(2) The alternative question introduced in part (a), How many people need to be
present for there to be a 1/2 chance that someone else has my birthday?, may
also be answered in the large-N limit. The probability that no one out of n
people has a birthday on a given day is
_
1
1
N
_
n
=
_
1
1
N
_
N(n/N)
e
n/N
. (9)
2
Therefore, if n > N ln2, you can expect that at least one of the n people has
your birthday. For N = 365, we nd that N ln2 is slightly less than 253, so
this agrees with the result obtained in part (a). Note that this result is linear in
N, whereas the result of the original problem in eq. (7) behaves like

N. The
reason for this square-root behavior can be seen in the next solution.
Second Solution: Given a large number, n, of people, there are
_
n
2
_
=
n(n 1)/2 n
2
/2 pairs of people. The probability that a given pair has
the same birthday is 1/N, so the probability that they do not have the same
birthday is 1 1/N.
1
Therefore, the probability that no pair has a common
birthday is
P
n
=
_
1
1
N
_
n
2
/2

_
1
1
N
_
N(n
2
/2N)
e
n
2
/(2N)
. (10)
This equals 1/2 when
n

2 ln2

N. (11)
in agreement with eq. (7).
Remark: We assumed above that all the pairs are independent, as far as writing
down the 1 1/N probability goes. Let us now show that this approximately true.
We will show that for large N and n, the assumptions on the coincidence of birthdays
in some pairs do not signicantly aect the probability of coincidence in other pairs.
More precisely, the relation P
n
e
n
2
/(2N)
is true if n N
2/3
. The reasoning is as
follows.
Consider n people and N days in a year. Assumptions on the coincidence of birthdays
in some pairs will slightly aect the probability of coincidence in other pairs, because
the given assumptions restrict the possible birthdays of these other pairs. For example,
if it is given that A and B do not have the same birthday, and also that B and C do
not, then the probability that A and C do not have the same birthday is 1 1/364
(instead of 1 1/365), because they are both restricted from having a birthday on
Bs birthday, whatever it may be.
The maximal restriction that can be placed on the possible birthdays of a given pair
occurs when it is known that neither of them has a birthday in common with any of
the other n2 people. In this case, the possible birthdays for the last two people are
restricted to N (n 2) days of the year. Therefore, the probability that the last
pair does not have a common birthday is 1 1/(N n +2) > 1 1/(N n). This is
the most that any such probability can dier from the naive 1 1/N. Therefore, we
may say that
_
1
1
N n
_
n
2
/2
P
n

_
1
1
N
_
n
2
/2
= e
n
2
/2(Nn)
P
n
e
n
2
/(2N)
. (12)
The ratio of these upper and lower bounds on P
n
is
exp
_

n
2
2N
+
n
2
2(N n)
_
= exp
_
n
3
2N(N n)
_
. (13)
1
This is not quite correct for all the pairs, because two pairs are not independent if, for example,
they share a common person. But it is accurate enough for our purposes in the large-N limit. See
the remark below.
3
For large N, this ratio is essentially equal to 1, provided that n N
2/3
. Therefore,
P
n
e
n
2
/(2N)
if n N
2/3
. And since the result in eq. (11) is in this realm, it is
therefore valid.
Extension: We can also ask the following question: How many people must
be in a room in order for the probability to be greater than 1/2 that at least b
of them have the same birthday? (Assume that there is a very large number,
N, of days in a year, and ignore eects that are of subleading order in N.)
We can solve this problem in the manner of the second solution above. Given
a large number, n, of people, there are
_
n
b
_
groups of b people. This is approx-
imately equal to n
b
/b! (assuming that b n).
The probability that a given group of b people all have the same birthday
is 1/N
b1
, so the probability that they do not all have the same birthday is
1 (1/N
b1
).
2
Therefore, the probability, P
(b)
n
, that no group of b people all
have the same birthday is
P
(b)
n

_
1
1
N
b1
_
n
b
/b!
e
n
b
/b!N
b1
. (14)
This equals 1/2 when
n (b! ln2)
1/b
N
11/b
. (15)
Remarks:
(1) Eq. (15) holds in the large-N limit. If we wish to make another approximation,
that of large b, we see that the quantity (b! ln2)
1/b
goes like b/e, for large b. (This
follows from Sterlings formula, m! m
m
e
m

2m.) Therefore, for large N,

n, and b (with b n N), we have P
(b)
n
= 1/2 when
n (b/e)N
11/b
. (16)
(2) The right-hand side of equation eq. (15) scales with N according to N
11/b
.
This means that if we look at the numbers of people needed to have a greater
than 1/2 chance that pairs, triplets, etc., have common birthdays, we see that
these numbers scale like
N
1/2
, N
2/3
, N
3/4
, . (17)
For large N, these numbers are multiplicatively far apart. Therefore, there are
values of n for which we can say, for example, that we are virtually certain that
there are pairs and triplets with common birthdays, but also that we are virtually
certain that there are no quadruplets with a common birthday. For example, if
n = N
17/24
(which satises N
2/3
< n < N
3/4
), then eq. (14) shows that the
probability that there is a common birthday triplet is 1 e

1
6
N
1/8
1, whereas
the probability that there is a common birthday quadruplet is 1 e

1
24
N
1/6

1
24
N
1/6
0.
2
Again, this is not quite correct, because the groups are not all independent. But I believe it is
accurate enough for our purposes in the large-N limit. I have a proof which I think is correct, but
which is very ugly. If anyone has a nice clean proof, let me know.
4
Solution
Week 47 (8/4/03)
Sliding ladder
The important point to realize in this problem is that the ladder loses contact with
the wall before it hits the ground. Lets nd where this loss of contact occurs.
Let r = /2, for convenience. While the ladder is in contact with the wall, its
CM moves in a circle of radius r. (This follows from the fact that the median to the
hypotenuse of a right triangle has half the length of the hypotenuse). Let be the
angle between the wall and the radius from the corner to the CM. (This is also the
angle between the ladder and the wall.)
r
r r

Well solve this problem by assuming that the CM always moves in a circle, and
then determining the position at which the horizontal CM speed starts to decrease,
that is, the point at which the normal force from the wall would have to become
negative. Since the normal force of course cant be negative, this is the point where
the ladder loses contact with the wall.
By conservation of energy, the kinetic energy of the ladder equals the loss in
potential energy, which is mgr(1 cos ). This kinetic energy may be broken up
into the CM translational energy plus the rotation energy. The CM translational
energy is simply mr
2

2
/2 (because the CM travels in a circle of radius r). The
rotational energy is I

2
/2. (The same

applies here as in the CM translational
motion, because is the angle between the ladder and the vertical, and thus is the
angle of rotation of the ladder.) Letting I mr
2
to be general ( = 1/3 for our
ladder), the conservation of energy statement is (1 + )mr
2

2
/2 = mgr(1 cos ).
Therefore, the speed of the CM, which is v = r

, equals
v =

2gr
1 +

(1 cos ) . (1)
The horizontal component of this is
v
x
=

2gr
1 +

(1 cos ) cos . (2)

Taking the derivative of

(1 cos ) cos , we see that the horizontal speed is max-

imum when cos = 2/3. Therefore the ladder loses contact with the wall when
cos =
2
3
= 48.2

. (3)
1
Note that this is independent of . This means that, for example, a dumbbell (two
masses at the ends of a massless rod, with = 1) will lose contact with the wall at
the same angle.
Plugging this value of into eq. (2), and using = 1/3, we obtain a nal
horizontal speed of
v
x
=

2gr
3

g
3
. (4)
Note that this is 1/3 of the

2gr horizontal speed that the ladder would have if
it were arranged (perhaps by having the top end slide down a curve) to eventually
slide horizontally along the ground.
Remark: The normal force from the wall is zero at the start and nish, so it must reach a
maximum at some intermediate value of . Lets nd this . Taking the derivative of v
x
in
eq. (2) to nd a
x
, and then using

1 cos from eq. (1), we see that the force from

the wall is proportional to
a
x

sin(3 cos 2)

1 cos

sin(3 cos 2). (5)

Taking the derivative of this, we nd that the force from the wall is maximum when
cos =
1 +

19
6
= 26.7

. (6)
2
Solution
Week 48 (8/11/03)
The hotel problem
In guring out the probability for success (choosing the cheapest hotel), it is conve-
nient to organize the dierent cases according to what the highest ranking hotel (in
order of cheapness) in the rst fraction x is. Let H
1
denote the cheapest hotel, H
2
the second cheapest, etc.
Assume that H
1
is among the rst x, which happens with probability x. In this
case, there is failure.
Assume that H
2
is the cheapest among the rst x, which happens with proba-
bility x(1x). This is the probability that H
2
is in the rst x, times the probability
that H
1
is not.
1
In this case, we have success.
Assume that H
3
is the cheapest among the rst x, which happens with proba-
bility x(1 x)
2
(again, see the remark below). This is the probability that H
3
is in
the rst x, times the probability that H
2
is not, times the probability that H
1
also
is not. In this case, we have success 1/2 of the time, because only 1/2 of the time
H
1
will come before H
2
.
Continuing in this fashion, we see that the probability for success, P, is
P(x) = x(1 x) +
1
2
x(1 x)
2
+
1
3
x(1 x)
3
+
=

k=1
1
k
x(1 x)
k
. (1)
The 1/k factor comes from the probability that H
1
is rst among the top k hotels
which lie in the nal (1 x) fraction. Using the expansion ln(1 y) = (y +y
2
/2 +
y
3
/3 + ), with y = 1 x, we obtain
P(x) = xlnx. (2)
Taking the derivative, we see that P(x) is maximized when x = 1/e, in which case
the value is 1/e. Therefore, you want to pass up 1/e 37% of the hotels, and then
pick the next one that is better than all the ones youve seen. Your chance of getting
the best one is 1/e 37%.
Remark: For suciently large N, the probabilities in eq. (1) are arbitrarily close to
x(1 x)
k
/k, for small values of k. But each successive term in eq. (1) is suppressed by a
factor of at least (1 1/e). The terms therefore become negligibly small at a k value that is
independent of N. The only signicant contribution to the sum therefore comes from small
values of k, for which the given probabilities are essentially correct.
1
The (1 x) factor is technically not correct, because there are only N 1 spots available for
H1, given that H2 has been placed. But the error is negligible for large N. See the remark below.
1
Solution
Week 49 (8/18/03)
Falling chimney
Before we start dealing with the forces in the rods, lets rst determine

as a
function of (the angle through which the chimney has fallen). Let be the height
of the chimney. Then the moment of inertia around the pivot point on the ground
is m
2
/3 (if we ignore the width), and the torque (around the pivot point) due to
gravity is = mg(/2) sin. Therefore, = dL/dt gives mg(/2) sin = (1/3)m
2

,
or

=
3g sin
2
. (1)
Lets now determine the forces in the rods. Our strategy will be to imagine
that the chimney consists of a chimney of height h, with another chimney of height
h placed on top of it. Well nd the forces in the rods connecting these two
sub-chimneys, and then well maximize one of these forces (T
2
, dened below) as
a function of h.
The forces on the top piece are gravity and the forces from the two rods at
each end of the bottom board. Lets break these latter forces up into transverse
and longitudinal forces along the chimney. Let T
1
and T
2
be the two longitudinal
components, and let F be the sum of the transverse components, as shown.
T
2
T
l-h
1
r
r
F
h

We have picked the positive directions for T

1
and T
2
such that positive T
1
corre-
sponds to a compression in the left rod, and positive T
2
corresponds to a tension in
the right rod (which is what the forces will turn out to be, as well see). It turns
out that if the width (which well call 2r) is much less than the height, then T
2
F
(as we will see below), so the tension in the right rod is essentially equal to T
2
. We
will therefore be concerned with maximizing T
2
.
In writing down the force and torque equations for the top piece, we have three
equations (the radial and tangential F = ma equations, and = dL/dt around the
CM), and three unknowns (F, T
1
, and T
2
). If we dene the fraction f h/, then
the top piece has length (1f) and mass (1f)m, and its CM travels in a circle of
radius (1 +f)/2, Therefore, our three force and torque equations are, respectively,
T
2
T
1
+ (1 f)mg cos = (1 f)m

(1 + f)
2

2
,
F + (1 f)mg sin = (1 f)m

(1 + f)
2

,
1
(T
1
+ T
2
)r F
(1 f)
2
= (1 f)m

(1 f)
2

2
12

. (2)
At this point, we could plow forward and solve this system of three equations in
three unknowns. But things simplify greatly in the limit where r . The third
equation says that T
1
+ T
2
is of order 1/r, and the rst equation says that T
2
T
1
is of order 1. These imply that T
1
T
2
, to leading order in 1/r. Therefore, we may
set T
1
+ T
2
2T
2
in the third equation. Using this approximation, along with the
value of

from eq. (1), the second and third equations become
F + (1 f)mg sin =
3
4
(1 f
2
)mg sin,
2rT
2
F
(1 f)
2
=
1
8
(1 f)
3
mg sin. (3)
This rst of these equations gives
F =
mg sin
4
(1 + 4f 3f
2
), (4)
and then the second gives
T
2

mg sin
8r
f(1 f)
2
. (5)
As stated above, this is much greater than F (because /r 1), so the tension in
the right rod is essentially equal to T
2
. Taking the derivative of T
2
with respect to
f, we see that it is maximum at
f
h

=
1
3
. (6)
Therefore, the chimney is most likely to break at a point one-third of the way up
(assuming that the width is much less than the height). Interestingly, f = 1/3
makes the force F in eq. (4) exactly equal to zero.
2
Solution
Week 50 (8/25/03)
Equal segments
The construction will proceed inductively. Given a segment divided into N equal
segments, we will show how to divide it into N + 1 equal segments. For purposes
of concreteness and having manageable gures, we will just do the case N = 3.
Generalization to arbitrary N will be clear.
In the gure below, let the segment AB be divided into three equal segments by
D
1
and D
2
. From an arbitrary point P (assume that P is on the side of AB opposite
to the innite line L, although it need not be), draw lines through A and B, which
meet L at points M and N, respectively. Draw segments MD
1
, MD
2
, MB, NA,
ND
1
, and ND
2
. Let the resulting intersections (the ones closest to segment AB)
be Q
1
, Q
2
, and Q
3
, as shown.
A B
L
D D
M N
P
Q
1 2
3
Q
2
Q
1
Q
0
Q
4
Claim: The lines PQ
1
, PQ
2
, and PQ
3
divide AB into four equal segments.
Proof: First, note that Q
1
, Q
2
, and Q
3
are collinear on a line parallel to AB and
L. This is true because the ratio of similar triangles AQ
1
D
1
and NQ
1
M is the
same as the ratio of similar triangles D
1
Q
2
D
2
and NQ
2
M (because AD
1
= D
1
D
2
).
Therefore, the altitude from Q
1
to AD
1
equals the altitude from Q
2
to D
1
D
2
. The
same reasoning applies to Q
3
, so all the Q
i
are equal distances from AB. Let the
line determined by the Q
i
intersect PM and PN at Q
0
and Q
4
, respectively.
We now claim that the distances Q
i
Q
i+1
, i = 0, . . . , 3 are equal. They are equal
because the ratio of similar triangles Q
0
AQ
1
and MAN is the same as the ratio of
similar triangles Q
1
D
1
Q
2
and MD
1
N (because the ratio of the altitudes from A in
the rst pair is the same as the ratio of the altitudes from D
1
in the second pair).
Hence, Q
0
Q
1
= Q
1
Q
2
. Likewise for the other Q
i
Q
i+1
.
Therefore, since Q
0
Q
4
is parallel to AB, the intersections of the lines PQ
i
with
AB divide AB into four equal segments.
1
Remark: To divide AB into ve equal segments, we can use the same gure, with most
of the work having already been done. The only new lines we need to draw are NQ
0
and
MQ
4
, to give a total of four intersections on a horizontal line one level below the Q
i
. If
we continue with this process, we obtain gures looking like the one below. The horizontal
lines in this gure are divided into equal parts by the intersections of the diagonal lines.
The initial undivided segment is the top one.
1
We leave for you the following exercise: Given a segment of length , and a line parallel to
it, construct a segment with a length equal to an arbitrary multiple of , using only a ruler.
1
This segment must be used again for the N = 2 segment, because the above procedure yields
only one point Q
1
, and this single point doesnt determine a line parallel to L.
2
Solution
Week 51 (9/1/03)
Accelerating spaceship
We will solve this problem by considering two nearby times and using the velocity-
addition formula,
v =
v
1
+ v
2
1 + v
1
v
2
/c
2
. (1)
Using the denition of the proper acceleration, a, we have (with v
1
v(t) and
v
2
a dt)
v(t + dt) =
v(t) + a dt
1 + v(t)a dt/c
2
. (2)
Expanding both sides to rst order in dt yields
1
dv
dt
= a

1
v
2
c
2

. (3)
Separating variables and integrating gives, using 1/(1z
2
) = 1/2(1z)+1/2(1+z),

v
0

1
1 v/c
+
1
1 + v/c

dv = 2a

t
0
dt. (4)
This yields ln((1+v/c)/(1v/c)) = 2at/c. Exponentiating, and solving for v, gives
v(t) = c

e
2at/c
1
e
2at/c
+ 1

= c tanh(at/c). (5)
Note that for small a or small t (more precisely, for at/c 1), we obtain v(t) at,
as we should. And for at/c 1, we obtain v(t) c, as we should.
Remarks: If a happens to be a function of time, a(t), then we cant move the a outside
the integral in eq. (4), so we instead end up with the general formula,
v(t) = c tanh

1
c

t
0
a(t) dt

. (6)
If we dene the rapidity, , by
(t)
1
c

t
0
a(t) dt, (7)
then we have
v = c tanh tanh =
v
c
. (8)
Note that whereas v has c as a limiting value, can become arbitrarily large. The
associated with a given v is simply 1/mc times the time integral of the force (felt by
the astronaut) needed to bring the astronaut up to speed v. By applying a force for an
arbitrarily long time, we can make arbitrarily large.
1
Equivalently, just take the derivative of (v + w)/(1 + vw/c
2
) with respect to w, and then set
w = 0.
1
The quantity is very useful because many expressions in relativity (which well just
invoke here) take on a particularly nice form when written in terms of . Consider, for
example, the velocity-addition formula. Let
1
= tanh
1
and
2
= tanh
2
. Then if we add

1
and
2
using the velocity-addition formula, eq. (1), we obtain

1
+
2
1 +
1

2
=
tanh
1
+ tanh
2
1 + tanh
1
tanh
2
= tanh(
1
+
2
), (9)
where we have used the addition formula for tanh (which can be proved by writing things
in terms of the exponentials e

). Therefore, while the velocities add in the strange manner

of eq. (1), the rapidities add by standard addition.
The Lorentz transformation,

x
ct

ct

, (10)
also takes a nice form when written in terms of the rapidity. Note that can be written as

1

1
2
=
1

1 tanh
2

= cosh , (11)
and so

1
2
=
tanh

1 tanh
2

= sinh. (12)
Therefore, the Lorentz transformation becomes

x
ct

cosh sinh
sinh cosh

ct

. (13)
This looks similar to a rotation in a plane, which is given by

x
y

cos sin
sin cos

, (14)
except that we now have hyperbolic trig functions instead of the usual trig functions. The
fact that the invariant interval, s
2
c
2
t
2
x
2
, does not depend on the frame is clear from
eq. (13), because the cross terms in the squares cancel, and cosh
2
sinh
2
= 1. (Compare
with the invariance of r
2
x
2
+ y
2
for rotations in a plane.)
Quantities associated with a Minkowski diagram also take a nice form when written in
terms of the rapidity. In particular, the angle between the axes of the two relevant frames
happens to be tan = , where c is the relative speed between the frames. But = tanh,
so the angle between the axes is given by
tan = tanh. (15)
The integral

a(t) dt (which is c times the rapidity) may be described as the naive,
incorrect speed. That is, it is the speed the astronaut might think he has, if he has his eyes
closed and knows nothing about the theory of relativity. (And indeed, his thinking would be
essentially correct for small speeds.) The quantity

a(t) dt seems like a reasonably physical
thing, so if there is any justice in the world,

a(t) dt =

F(t) dt/m should have some
meaning. And indeed, although it doesnt equal v, all you have to do to get v is take a tanh
and throw in some factors of c.
The fact that rapidities add via simple addition when using the velocity-addition for-
mula, as we saw in eq. (9), is evident from eq. (6). There is really nothing more going on
here than the fact that

t
2
t
0
a(t) dt =

t
1
t
0
a(t) dt +

t
2
t
1
a(t) dt. (16)
2
To be explicit, let a force be applied from t
0
to t
1
that brings a mass up to speed

1
= tanh
1
= tanh(

t
1
t
0
a dt), and then let an additional force be applied from t
1
to t
2
that
adds on an additional speed of
2
= tanh
2
= tanh(

t
2
t
1
a dt) (relative to the speed at t
1
).
Then the resulting speed may be looked at in two ways: (1) it is the result of relativistically
adding the speeds
1
= tanh
1
and
2
= tanh
2
, and (2) it is the result of applying the
force from t
0
to t
2
(you get the same nal speed, of course, whether or not you bother to
record the speed along the way at t
1
), which is = tanh(

t
2
t0
adt) = tanh(
1
+
2
), where
the last equality comes from the obvious statement, eq. (16). Therefore, the relativistic
addition of tanh
1
and tanh
2
gives tanh(
1
+
2
), as we wanted to show.
3
Solution
Week 52 (9/8/03)
Construct the center
Pick an arbitrary point A on the circle, as shown below. Construct points B and
C on the circle, with AB = AC = , where is arbitrary.
1
Construct point D with
DB = DC = . Let the distance DA be r. If O is the location (which we dont know
yet) of the center of the circle, then triangles AOB and ABD are similar isosceles
triangles (because they have

BAD in common). Therefore, OA/BA = BA/DA,
which gives r =
2
/R.
A
C
B
D
O
l
r
R
The above construction shows that if we are given a length and a circle of
radius R, then we can construct the length
2
/R. Therefore, we can produce the
length R by simply repeating the above construction with the same length , but
now with a circle of radius
2
/R (which we just produced). In the following gure,
we obtain GA =
2
/(
2
/R) = R. Hence, G is the center of the circle.
A
C
B
E
F
D
G
l
l /R
2
1
However, this construction will not work if is too large or too small. We will determine these
bounds below.
1
If you want to go through the similar-triangles argument, note that triangles ADE
and AEG are similar isosceles triangles (because they have

EAD in common).
Therefore, DA/EA = EA/GA, which gives GA =
2
/(
2
/R) = R.
Restrictions: In order for this construction to work, it is necessary (and sucient)
for R/2 < < 2R. The upper limit on comes from the requirement that a circle of
radius (centered at A) intersects the given circle of radius R.
2
This gives < 2R.
The lower limit on comes from the requirement that a circle of radius (centered
at A) intersects the circle of radius
2
/R (centered at D). This gives < 2
2
/R,
which implies R/2 < .
remark: The above solution can be extended to solve the following problem: Given three
points, construct the circle passing through them.
The solution proceeds along the lines of the above solution. In the gure below, construct
point D with DB = AB =
1
and DC = AC =
2
. Let O be the location (which we
dont know yet) of the center of the desired circle. Then

BOA = BA

. Also,

DCA =
2(

BCA) = 2(BA

/2) = BA

. Therefore,

BOA =

DCA, and so triangles BOA and
DCA are similar isosceles triangles. Hence,
BO
BA
=
DC
DA
=
R

1
=

2
DA
= DA =

1

2
R
. (1)
A
C
B
D
O
R
l
2
l
1
As in the above solution, we can apply this construction again, with the same lengths

1
and
2
, but now with a circle of radius
1

2
/R (which we just produced). In the following
gure, we obtain GA =
1

2
/(
1

2
/R) = R. And having found R, we can easily construct
the center of the circle.
2
The construction still works even if the intersection points are nearly diametrically opposite to
A.
2
A
C
B
D
G
E
F
l
2
l
1
l
2
/ R
l
1
Restrictions: In order for this construction to work, we must be able to construct points
E and F on the circle of radius
1

2
/R (centered at D). In order for these points to exist, the
diameter of this circle must be larger than both
1
and
2
. That is, 2
1

2
/R > max(
1
,
2
).
This can be rewritten as min(
1
,
2
) > R/2. This is the condition that must be satised.
(There are no upper bounds on
1
and
2
.)
What do we do if this lower bound is not satised? Simply construct more points on
the circle until some three of them satisfy the constraint. For example, as shown below,
construct point B
1
with B
1
B = CA and B
1
A = CB. Then triangle B
1
BA is congruent to
triangle CAB, so point B
1
also lies on the circle. In a similar manner we can construct B
2
as shown, and then B
3
, etc., to obtain an arbitrary number of points on the circle. After
constructing many points, we will eventually be able to pick three of them that satisfy the
constraint, min(
1
,
2
) > R/2.
A
C
B
B
B
B
1
2
3
Of course, after constructing these new points on the circle, it is easy to pick three of them
that have
1
=
2
(for example B, B
2n
, and B
4n
, in the notation in the gure). We can
then use the easier, symmetrical solution in part (a) to nd the center of the circle.
3
Solution
Week 53 (9/15/03)
Circles on the ice
By construction, the normal force from the ice exactly cancels all eects of the
gravitational and centrifugal forces in the rotating frame of the earth (because the
plumb bob hangs in the direction of the eective gravity force, which is the sum of
the gravitational and centrifugal forces). We therefore need only concern ourselves
with the Coriolis force. This force equals F
cor
= 2m v.
Let the angle down from the north pole be (we assume the circle is small enough
so that is essentially constant throughout the motion). Then the component of
the Coriolis force that points horizontally along the surface has magnitude f =
2mv cos and is perpendicular to the direction of motion. (The vertical component
of the Coriolis force will simply modify the required normal force.) Because this
force is perpendicular to the direction of motion, v does not change. Therefore, f is
constant. But a constant force perpendicular to the motion of a particle produces
a circular path. The radius of the circle is given by
2mv cos =
mv
2
r
= r =
v
2 cos
. (1)
The frequency of the circular motion is

=
v
r
= 2 cos . (2)
Remarks: To get a rough idea of the size of the circle, you can show (using 7.3
10
5
s
1
) that r 10 km when v = 1 m/s and = 45

. Even the tiniest bit of friction will

clearly make this eect essentially impossible to see.
For the /2 (that is, near the equator), the component of the Coriolis force along
the surface is negligible, so r becomes large, and

goes to 0.
For the 0 (that is, near the north pole), the Coriolis force essentially points along the
surface. The above equations give r v/(2), and

2. For the special case where the

center of the circle is the north pole, this

2 result might seem incorrect, because you

might want to say that the circular motion should be achieved by having the puck remain
motionless in the inertial fame, while the earth rotates beneath it (thus making

= ).
The error in this reasoning is that the level earth is not spherical, due to the non-radial
direction of the eective gravity (the combination of the gravitational and centrifugal forces).
If the puck starts out motionless in the inertial frame, it will be drawn toward the north
pole, due to the component of gravity in that direction. In order to not fall toward the pole,
the puck needs to travel with frequency (relative to the inertial frame) in the direction
opposite to the earths rotation.
1
The puck therefore moves at frequency 2 relative to the
frame of the earth.
1
In the rotating frame of the puck, the puck then feels the same centrifugal force that it would
feel if it were at rest on the earth, spinning with it. It therefore happily stays at the same value
on the level surface, just as a puck at rest on the earth does.
1
Solution
Week 54 (9/22/03)
Rolling the die
To get a feel for the problem, you can work out the answer for small values of N.
For N = 1, the probability that the rst player wins is 1. For N = 2, it is 3/4. And
for N = 3, it is 19/27. A pattern in these numbers is more evident if we instead
list the probabilities that the rst player loses. These are 0, 1/4, and 8/27. (And if
you work things out for N = 4, youll obtain 81/256.) We therefore guess that the
probability, P
L
, that the rst player loses is
P
L
=

1
1
N

N
. (1)
Well prove this by proving the following more general claim. Eq. (1) is the special
case of the claim with r = 0.
Claim: Let L
r
be the probability that a player loses, given that a roll of r has just
occurred. Then
L
r
=

1
1
N

Nr
. (2)
Proof: Assume that a roll of r has just occurred. To determine the probability,
L
r
, that the player who goes next loses, lets consider the probability, 1 L
r
, that
she wins. In order to win, she must roll a number, a, greater than r (each of which
occurs with probability 1/N); and her opponent must then lose, given that he has
to beat a roll of a (this occurs with probability L
a
). So the probability of winning,
given that one must beat a roll of r, is
1 L
r
=
1
N
(L
r+1
+ L
r+2
+ + L
N
). (3)
If we write down the analogous equation using r 1 instead of r,
1 L
r1
=
1
N
(L
r
+ L
r+1
+ + L
N
), (4)
and then subtract eq. (4) from eq. (3), we obtain
L
r1
=

1
1
N

L
r
, (5)
for all r from 1 to N. Using L
N
= 1, we nd that
L
r
=

1
1
N

Nr
(0 r N). (6)
We may consider the rst player to start out with a roll of r = 0 having just occurred.
Therefore, the probability that the rst player wins is 1 (1 1/N)
N
. For large
N, this probability approaches 1 1/e 63.2%. For a standard die with N = 6, it
takes the value 1 (5/6)
6
66.5%.
1
Note that the probability that the rst player wins can be written as
1

1
1
N

N
=
1
N

1
1
N

N1
+

1
1
N

N2
+ +

1
1
N

1
+ 1

. (7)
The right-hand side shows explicitly the probabilities of winning, depending on what
the rst roll is. For example, the rst term on the right-hand side is the probability,
1/N, that the rst player rolls a 1, times the probability, (1 1/N)
N1
, that the
second player loses given that he must beat a 1.
2
Solution
Week 55 (9/29/03)
Fixed highest point
For the desired motion, the important thing to note is that every point in the top
moves in a xed circle around the z-axis. Therefore, points vertically. Hence, if
is the frequency of precession, we have = z, as shown.

x
3
x
2
l
P
R

Well need to use = dL/dt to solve this problem, so lets rst calculate L. With
the pivot as the origin, the principle axes are shown above (and x
1
points into the
page, but it wont come into play here). The principal moments are
I
3
=
MR
2
2
, and I I
1
= I
2
= M
2
+
MR
2
4
, (1)
where we have used the parallel-axis theorem to obtain the latter. The components
of along the principal axes are
3
= cos , and
2
= sin. Therefore, we have
L = I
3

3
x
3
+ I
2

2
x
2
= I
3
cos x
3
+ Isin x
2
, (2)
where we have kept things in terms of the moments, I
3
and I, to be general for now.
The horizontal component of L is
L

= L
3
sin L
2
cos
= (I
3
cos ) sin (Isin) cos
= (I
3
I)cos sin, (3)
with rightward taken to be positive. This horizontal component spins around in a
circle with frequency . Therefore, dL/dt has magnitude

dL
dt

= L

=
2
sin cos (I
3
I), (4)
and it is directed into the page (or out of the page, if this quantity is negative).
dL/dt must equal the torque, which has magnitude || = Mg sin and is directed
into the page. Therefore,
=

Mg
(I
3
I) cos
. (5)
1
We see that for a general symmetric top, the desired precessional motion (where the
same side always points up) is possible only if
I
3
> I. (6)
Note that this condition is independent of . For the problem at hand, I
3
and I are
given in eq. (1), so we nd
=

4g
(R
2
4
2
) cos
. (7)
The necessary condition for the desired motion to exist is therefore
R > 2. (8)
Remarks:
1. Given that the desired motion does indeed exist, it is intuitively clear that should
become very large as /2. But it is by no means intuitively clear (at least to me)
that such motion should exist at all for angles near /2.
2. approaches a non-zero constant as 0, which isnt entirely obvious.
3. If both R and are scaled up by the same factor, we see that decreases. This also
follows from dimensional analysis.
4. The condition I
3
> I can be understood in the following way. If I
3
= I, then L ,
and so L points vertically along . If I
3
> I, then L points somewhere to the right
of the z-axis (at the instant shown in the gure above). This means that the tip of L
is moving into the page, along with the top. This is what we need, because points
into the page. If, however, I
3
< I, then L points somewhere to the left of the z-axis,
so dL/dt points out of the page, and hence cannot be equal to .
2
Solution
Week 56 (10/6/03)
Stirlings formula
Lets rst prove the result,
N! =
_

0
x
N
e
x
dx I
N
. (1)
The proof by induction proceeds as follows. Integrating by parts gives
_

0
x
N
e
x
dx = x
N
e
x
dx

0
+ N
_

0
x
N1
e
x
dx. (2)
The rst term on the right-hand side is zero, so we have I
N
= NI
N1
. Therefore,
if I
N1
= (N 1)!, then I
N
= N!. Since it is indeed true that I
0
= 0! = 1, we see
that I
N
= N! for all N.
Let us now write x
N
e
x
as exp(N lnx x) exp(f(x)), and then expand f(x)
in a Taylor series about its maximum, which occurs at x = N. Computing the
rst two derivatives of f(x), evaluated at N, gives the following Taylor series in the
exponent.
N! =
_

0
exp
_
N + N lnN
(x N)
2
2N
+
_
dx
N
N
e
N
_

0
exp
_

(x N)
2
2N
_
dx. (3)
If N is very large, we can let the integral run from to , with negligible error.
Letting y x N, we have
N! N
N
e
N
_

e
y
2
/2N
dy
= N
N
e
N

2N, (4)
as desired.
Remark: We have used the fact that
_

e
y
2
dy =

. This can be proved in the
following way, where we make use of a change of variables from cartesian to polar coordinates.
Let I be the desired integral. Then
I
2
=
_

e
x
2
dx
_

e
y
2
dy
=
_

e
(x
2
+y
2
)
dxdy
=
_
2
0
_

0
e
r
2
r dr d
= 2
_

e
r
2
2
_

0
= . (5)
1
A change of variables then gives
_

e
y
2
/n
dy =

n.
The calculation of the higher-order corrections is a bit messier, because we have to
keep track of more terms in the Taylor expansion of f(x). Lets nd the order-1/N
correction rst. Our strategy will be to write the integrand in eq. (1) as a gaussian
plus small corrections. Computing the rst four derivatives of f(x), evaluated at N,
gives us the following Taylor series in the exponent (letting y x N, and letting
the limits of integration run from to ).
N! =
_

exp
_
N + N lnN
y
2
2N
+
y
3
3N
2

y
4
4N
3
+
_
dy
N
N
e
N
_

exp
_

y
2
2N
_
exp
_
y
3
3N
2

y
4
4N
3
_
dy
N
N
e
N
_

exp
_

y
2
2N
__
1 +
_
y
3
3N
2

y
4
4N
3
_
(6)
+
1
2
_
y
3
3N
2

y
4
4N
3
_
2
+
_
dy.
Since terms with odd powers of y integrate to zero, we obtain (to leading orders in
1/N),
N! N
N
e
N
_

exp
_

y
2
2N
_
_
_
1
y
4
4N
3
+
1
2
_
y
3
3N
2
_
2
+
_
_
dy. (7)
At this point, we need to know how to calculate integrals of the form
_

x
2n
e
ax
2
dx.
Using
_

e
ax
2
dx =

a
1/2
, and successively dierentiating with respect to a,
we obtain
_

e
ax
2
dx =

a
1/2
,
_

x
2
e
ax
2
dx =
1
2

a
3/2
,
_

x
4
e
ax
2
dx =
3
4

a
5/2
,
_

x
6
e
ax
2
dx =
15
8

a
7/2
. (8)
Letting a 1/2N here, eq. (7) gives
N! N
N
e
N

_
(2N)
1/2

1
4N
3
3
4
(2N)
5/2
+
1
18N
4
15
8
(2N)
7/2
_
= N
N
e
N

2N
_
1 +
1
12N
_
. (9)
Note that to obtain all the terms of order 1/N, it is necessary to include the
(y
3
/3N
2
)
2
term in eq. (7). This is an easy term to forget.
If you like these sorts of calculations, you can go a step further and nd the order-
1/N
2
correction. It turns out that you have to keep terms out to the y
6
/6N
5
term
2
in the expansion in the rst line of eq. (6). Furthermore, you must keep terms out
to the [ ]
4
/4! term in the expansion in the last line of eq. (6). The relevant extra
terms that take the place of the in eq. (7) then turn out to be
_

y
6
6N
5
_
+
1
2
_
_
_

y
4
4N
3
_
2
+ 2
_
y
3
3N
2
__
y
5
5N
4
_
_
_
+
1
3!
_
_
3
_
y
3
3N
2
_
2
_

y
4
4N
3
_
_
_
+
1
4!
_
_
_
y
3
3N
2
_
4
_
_
, (10)
where we have grouped these terms via square brackets according to which term in
the e
z
series expansion in the last line of eq. (6) they come from. To do all of the
necessary integrals in the modied eq. (7), well need the next three integrals in the
list in eq. (8). They are
_

x
8
e
ax
2
dx =
3 5 7
2
4

a
9/2
,
_

x
10
e
ax
2
dx =
3 5 7 9
2
5

a
11/2
,
_

x
12
e
ax
2
dx =
3 5 7 9 11
2
6

a
13/2
. (11)
Putting the terms of eq. (10) in place of the in eq. (7), we nd that the
coecient of N
N
e
N

2N equals 1/N
2
times

1
6
3 5
2
3
2
3
+
1
2
_
1
16
+
2
15
_
3 5 7
2
4
2
4

1
3!
3
36
3 5 7 9
2
5
2
5
+
1
4!
1
81
3 5 7 9 11
2
6
2
6
=
1
288
. (12)
Therefore, we may write Stirlings formula as
N! N
N
e
N

2N
_
1 +
1
12N
+
1
288N
2
_
. (13)
This result of 1/288 is rather fortuitous, because it is the third term in the Taylor
series for e
1/12
. This means that we can write N! as
N! = N
N
e
N

2N
_
e
1/12N
+O(1/N
3
)
_
N
N
e
N+1/12N

2N. (14)
It turns out that the order-1/N
3
correction is not equal to 1/(3! 12
3
), which is the
next term in the expansion for e
1/12
.
3
Solution
Week 57 (10/13/03)
Throwing a beach ball
On both the way up and the way down, the total force on the ball is
F = mg mv. (1)
On the way up, v is positive, so the drag force points downward, as it should. And
on the way down, v is negative, so the drag force points upward.
Our strategy for nding v
f
will be to produce two dierent expressions for the
maximum height, h, and then equate them. Well nd these two expressions by
considering the upward and then the downward motion of the ball. In doing so, we
will need to write the acceleration of the ball as
a =
dv
dt
=
dy
dt
dv
dy
= v
dv
dy
. (2)
For the upward motion, F = ma gives
mg mv = mv
dv
dy
=

h
0
dy =

0
v
0
v dv
g + v
. (3)
where we have taken advantage of the fact that we know that the speed of the ball
at the top is zero. Writing v/(g + v) as [1 g/(g + v)]/, we may evaluate the
integral to obtain
h =
v
0

2
ln

1 +
v
0
g

. (4)
Now let us consider the downward motion. Let v
f
be the nal speed, which is
a positive quantity. The nal velocity is then the negative quantity, v
f
. Using
F = ma, we similarly obtain

0
h
dy =

v
f
0
v dv
g + v
. (5)
Performing the integration (or just replacing the v
0
in eq. (4) with v
f
) gives
h =
v
f

2
ln

1
v
f
g

. (6)
Equating the expressions for h in eqs. (4) and (6) gives an implicit equation for v
f
in terms of v
0
,
v
0
+ v
f
=
g

ln

g + v
0
g v
f

. (7)
Remarks: In the limit of small (more precisely, in the limit v
0
/g 1), we can use
ln(1+x) = xx
2
/2+ to obtain approximate values for h in eqs. (4) and (6). The results
are, as expected,
h
v
2
0
2g
, and h
v
2
f
2g
. (8)
1
We can also make approximations for large (or large v
0
/g). In this limit, the log
term in eq. (4) is negligible, so we obtain h v
0
/. And eq. (6) gives v
f
g/, because
the argument of the log must be very small in order to give a very large negative number,
which is needed to produce a positive h on the left-hand side. There is no way to relate v
f
and h is this limit, because the ball quickly reaches the terminal velocity of g/ (which is
the velocity that makes the net force equal to zero), independent of h.
Lets now nd the times it takes for the ball to go up and to go down. Well present
two methods for doing this.
First method: Let T
1
be the time for the upward path. If we write the acceleration
of the ball as a = dv/dt, then F = ma gives
mg mv = m
dv
dt
=

T
1
0
dt =

0
v
0
dv
g + v
. (9)
T
1
=
1

ln

1 +
v
0
g

. (10)
In a similar manner, we nd that the time T
2
for the downward path is
T
2
=
1

ln

1
v
f
g

. (11)
Therefore,
T
1
+ T
2
=
1

ln

g + v
0
g v
f

. (12)
Using eq. (7), we have
T
1
+ T
2
=
v
0
+ v
f
g
. (13)
This is shorter than the time in vacuum (namely 2v
0
/g) because v
f
< v
0
.
Second method: The very simple form of eq. (13) suggests that there is a cleaner
way to calculate the total time of ight. And indeed, if we integrate mdv/dt =
mg mv with respect to time on the way up, we obtain v
0
= gT
1
h
(because

v dt = h). Likewise, if we integrate mdv/dt = mg mv with respect
to time on the way down, we obtain v
f
= gT
2
+ h (because

v dt = h).
Adding these two results gives eq. (13). This procedure only works, of course,
because the drag force is proportional to v.
Remarks: The fact that the time here is shorter than the time in vacuum isnt obvious. On
one hand, the ball doesnt travel as high in air as it would in vacuum (so you might think
that T
1
+ T
2
< 2v
0
/g). But on the other hand, the ball moves slower in air (so you might
think that T
1
+T
2
> 2v
0
/g). It isnt obvious which eect wins, without doing a calculation.
For any , you can use eq. (10) to show that T
1
< v
0
/g. But T
2
is harder to get a
handle on, because it is given in terms of v
f
. But in the limit of large , the ball quickly
reaches terminal velocity, so we have T
2
h/v
f
(v
0
/)/(g/) = v
0
/g. Interestingly, this
is the same as the downward (and upward) time for a ball thrown in vacuum.
2
Solution
Week 58 (10/20/03)
Coins and Gaussians
There are
_
2N
N+x
_
ways to obtain N +x heads in 2N ips. Therefore, the probability
of obtaining N + x heads is
P(x) =
1
2
2N
_
2N
N + x
_
. (1)
Our goal is to nd an approximate expression for
_
2N
N+x
_
when N is large. Using
Stirlings formula,
1
N! N
N
e
N

2N, we have
_
2N
N + x
_
=
(2N)!
(N + x)!(N x)!

(2N)
2N

2N
(N + x)
N+x
(N x)
Nx

N
2
x
2
=
2
2N

N
_
1 +
x
N
_
N+x
_
1
x
N
_
N+x

N
2
x
2
. (2)
Now,
_
1 +
x
N
_
N+x
= exp
_
(N + x) ln
_
1 +
x
N
_
_
= exp
_
(N + x)
_
x
N

x
2
2N
2
+
_
_
exp
_
x +
x
2
2N
_
. (3)
Likewise,
_
1
x
N
_
N+x
exp
_
x +
x
2
2N
_
. (4)
Therefore, eq. (2) becomes
_
2N
N + x
_

2
2N
e
x
2
/N

N
, (5)
where we have set

N
2
x
2
N, because the exponential factor shows that only
x up to order

N contribute signicantly. Note that it is necessary to expand the
log in eq. (3) to second order to obtain the correct result. Using eq. (5) in eq. (1)
gives the desired result,
P(x)
e
x
2
/N

N
. (6)
Note that if we integrate this probability over x, we do indeed obtain 1, because
_

e
x
2
/N
dx =

N (see the solution to Problem of the Week 56).

1
From Problem of the Week 56.
1
Remark: To nd where eq. (6) is valid, we can expand the log factor in eq. (3) to fourth
order in x. We then obtain
P(x)
e
x
2
/N

N
e
x
4
/6N
3
. (7)
Therefore, when x N
3/4
, eq. (6) is not valid. However, when x N
3/4
, the e
x
2
/N
factor
in P(x) makes it negligibly small, so P(x) is essentially zero in any case.
2
Solution
Week 59 (10/27/03)
Getting way ahead
The explanation of why the two clocks show dierent times in the ground frame is the
following. The rocket becomes increasingly length contracted in the ground frame,
which means that the front end isnt traveling as fast as the back end. Therefore,
the time-dilation factor for the front clock isnt as large as that for the back clock.
So the front clock loses less time relative to the ground, and hence ends up ahead
of the back clock. Of course, its not at all obvious that everything works out
quantitatively, and that the front clock eventually ends up an arbitrarily large time
ahead of the back clock. In fact, its quite surprising that this is the case, since
the above dierence in speeds is rather small. But lets now show that the above
explanation does indeed account for the dierence in the clock readings.
Let the back of the rocket be located at position x. Then the front is located at
position x+L

1 v
2
, due to the length contraction. (We will set c = 1 throughout
this solution.) Taking the time derivatives of the two positions, we see that the
speeds of the back and front are (with v dx/dt)
1
v
b
= v, v
f
= v(1 L v), where 1/

1 v
2
. (1)
We must now nd v in terms of the time, t, in the ground frame. The quickest way
to do this is to use the fact that longitudinal forces are independent of the frame.
2
The force on, say, the astronaut, is f = mg in the spaceship frame, so it must also
be mg in the ground frame. Therefore, F = dp/dt in the ground frame gives (using
the fact that g is constant)
mg =
d(mv)
dt
= gt = v = v =
gt

1 + (gt)
2
. (2)
Remark: We can also nd v(t) by using the result of Problem of the Week 51, which says
that in terms of the proper time, , of a uniformly-accelerated particle, its speed is given by
v() = tanh(g). This yields = 1/

1 v
2
= cosh(g). If we then integrate dt = d, we
obtain gt = sinh(g). This gives
v(t) = tanh(g) =
sinh(g)
cosh(g)
=
sinh(g)

1 + sinh
2
(g)
=
gt

1 + (gt)
2
. (3)
Having found v, we must now nd the -factors associated with the speeds of
the front and back of the rocket. The -factor associated with the speed of the back
(namely v) is

b
=
1

1 v
2
=

1 + (gt)
2
. (4)
1
Since these speeds are not equal, there is of course an ambiguity concerning which speed we
should use in the length-contraction factor,

1 v
2
. Equivalently, the rocket actually doesnt have
one inertial frame that describes all of it. But you can show that any dierences arising from this
ambiguity are of higher order in gL/c
2
than we need to be concerned with.
2
This takes a little eort to prove, but well just accept it here. If you want a self-contained
method for nding v, well give another one in the remark below.
1
The -factor associated with the speed of the front, v
f
= v(1 L v), is a little
harder to obtain. We must rst calculate v. From eq. (2), we nd v = g/(1+g
2
t
2
)
3/2
,
which gives
v
f
= v(1 L v) =
gt

1 + (gt)
2

1
gL
1 + g
2
t
2

. (5)
The -factor (or rather 1/, which is what well be concerned with) associated with
this speed can now be found as follows. In the rst line below, we ignore the higher-
order (gL)
2
term, because it is really (gL/c
2
)
2
, and we are assuming that gL/c
2
is small. And in obtaining the third line, we use the Taylor-series approximation

1 1 /2.
1

f
=

1 v
2
f

1
g
2
t
2
1 + g
2
t
2

1
2gL
1 + g
2
t
2

=
1

1 + g
2
t
2

1 +
2g
3
t
2
L
1 + g
2
t
2

1 + g
2
t
2

1 +
g
3
t
2
L
1 + g
2
t
2

. (6)
We can now calculate the time that each clock shows, at time t in the ground
frame. The time on the back clock changes according to dt
b
= dt/
b
, so eq. (4)
gives
t
b
=

t
0
dt

1 + g
2
t
2
. (7)
The integral
3
of 1/

1 + x
2
is sinh
1
x. Letting x gt, this gives
gt
b
= sinh
1
(gt), (8)
in agreement with the results in the above remark. The time on the front clock
changes according to dt
f
= dt/
f
, so eq. (6) gives
t
f
=

t
0
dt

1 + g
2
t
2
+

t
0
g
3
t
2
Ldt
(1 + g
2
t
2
)
3/2
. (9)
The integral
4
of x
2
/(1 + x
2
)
3/2
is sinh
1
x x/

1 + x
2
. Letting x gt, this gives
gt
f
= sinh
1
(gt) + (gL)

sinh
1
(gt)
gt

1 + g
2
t
2

. (10)
Using eqs. (8) and (2), we may rewrite this as
gt
f
= gt
b
(1 + gL) gLv. (11)
Dividing by g, and putting the cs back in to make the units correct, gives
t
f
= t
b

1 +
gL
c
2

Lv
c
2
, (12)
as we wanted to show.
3
To derive this, make the substitution x sinh .
4
Again, to derive this, make the substitution x sinh .
2
Solution
Week 60 (11/3/03)
Cereal box prizes
First Solution: Assume that you have collected c of the colors, and let B
c
be the
number of boxes it takes to get the next color. The average value of B
c
, which we
will denote by B
c
, may be found as follows. The probability that a box yields a new
color is 1 c/N, and the probability that it does not is c/N. The expected number
of boxes to get the next prize is therefore
B
c
= 1
_
1
c
N
_
+ 2
_
c
N
__
1
c
N
_
+ 3
_
c
N
_
2
_
1
c
N
_
+ . (1)
Letting x c/N, we have
B
c
= (1 x)(1 + 2x + 3x
2
+ 4x
3
+ )
= (1 x)
_
(1 + x + x
2
+ ) + (x + x
2
+ ) + (x
2
+ ) +
_
= (1 x)
_
1
1 x
+
x
1 x
+
x
2
1 x
_
+
=
1
1 x
. (2)
Therefore,
B
c
=
N
N c
. (3)
Note that the above 1 + 2x + 3x
2
+ = 1/(1 x)
2
result, for 0 x < 1, may also
be derived by taking the derivative of the equality 1 + x + x
2
+ = 1/(1 x).
We see that (with c = 0) it takes one box, of course, to get the rst color. Then
(with c = 1) it takes an average of N/(N 1) boxes to get the second color. Then
(with c = 2) it takes an average of N/(N 2) boxes to get the third color. This
continues until (with c = N 1) it takes an average of N boxes to get the last color.
We wish to nd the average, B, of the total number of boxes needed to get all the
colors. B can be viewed as the average of the sum of the waiting times for each new
color. But this equals the sum of the average waiting times, which were calculated
above. In other words, letting B be the total number of boxes in a particular trial,
and letting B
0
, B
1
, . . . , B
N1
be the waiting times during that particular trial, we
have (with a bar denoting average)
B = (B
0
+ B
1
+ + B
N1
)
= B
0
+ B
1
+ + B
N1
= N
_
1
N
+
1
N 1
+ +
1
2
+ 1
_
. (4)
For large N, this goes like
B N(lnN + ), (5)
where 0.577 is Eulers constant.
1
Second Solution: We will calculate the probability, P(n), that the nal color is
obtained in the nth box. The desired expectation value is then

0
nP(n).
Claim: P(n) =
N1

k=1
(1)
k1
_
N 1
k 1
_
_
1
k
N
_
n1
(n = 2, 3, . . .).
Proof: Let us rst calculate the probability, p(n), that you have obtained all
the colors by the time you have looked in the nth box. P(n) is then given by
P(n) = p(n) p(n 1).
Assume that n boxes have been bought. Then there is a total of N
n
equally
likely possibilities for the way the colors can turn up in these n boxes. In nding
p(n), we will need to subtract from N
n
the number of possibilities that do not have
at least one prize of each color. We can count this number is the following way.
If (at least) color 1 is missing, then there are (N 1)
n
dierent combinations.
Likewise for the situations where another color is missing. So there seem to be
N(N 1)
n
combinations missing a color. However, we have double-counted some
of the cases. For example, a combination which has (at least) 1 and 2 missing has
been counted twice; there are (N 2)
n
of these. Likewise for all the other pairs of
colors. So we must subtract o
_
N
2
_
(N 2)
n
combinations. But now a combination
which has (at least) 1, 2, and 3 missing has not been counted at all (because we have
included it three times, and then subtracted it o three times); there are (N3)
n
of
these. Likewise for the other triplets. So we must add on
_
N
3
_
(N3)
n
combinations.
Now, however, the combinations with (at least) 1, 2, 3, and 4 missing have been
counted
_
4
1
_

_
4
2
_
+
_
4
3
_
= 2 times. Likewise for the other quadruplets. So we must
subtract o
_
N
4
_
(N 4)
n
combinations.
In general, if we have done this procedure up to (k 1)-tuplets, then the com-
binations missing (at least) k of the colors have been counted T times, where
T =
_
k
1
_

_
k
2
_
+ + (1)
k
_
k
k 1
_
. (6)
However, the binomial expansion gives
0 = (1 1)
k
= 1
_
k
1
_
+
_
k
2
_
+ + (1)
k1
_
k
k 1
_
+ (1)
k
= 1 T + (1)
k
. (7)
Therefore, T = 2 for even k, and T = 0 for odd k. So we have either overcounted
by one, or undercounted by one. Hence, the total number of combinations missing
at least one color is
_
N
1
_
(N 1)
n

_
N
2
_
(N 2)
n
+ + (1)
N
_
N
N 1
_
(1)
n
. (8)
To obtain the probability, p(n), that you have collected all the colors by the nth
box, we must subtract this number from N
n
, and then divide the result by N
n
. We
2
obtain
p(n) =
N1

k=0
(1)
k
_
N
k
_
_
1
k
N
_
n
, (n = 1, 2, 3, . . .). (9)
Remark: p(n) must of course equal zero, for n = 1, 2, . . . , N 1, because you need to buy
at least N boxes to get all the colors. To show this explicitly, rst note that p(n) may be
written as
p(n) =
N

k=0
(1)
k
_
N
k
__
1
k
N
_
n
, (10)
where we have let the sum run up to k = N, because the k = N term is zero anyway. It
is therefore sucient to demonstrate that

N
k=0
(1)
k
_
N
k
_
k
m
= 0, for m = 0, 1, . . . N 1,
because this identity will make all the separate terms arising from the binomial expansion
of (1 k/N)
n
in eq. (10) equal to zero on their own. You can prove this identity by taking
successive derivatives of the relation (1 x)
N
=

(1)
k
_
N
k
_
x
k
, and setting x = 1.
The probability, P(n), that the nal color is obtained in the nth box is
P(n) = p(n) p(n 1)
=
N1

k=0
(1)
k
_
N
k
_
_
1
k
N
_
n1
__
1
k
N
_
1
_
=
N1

k=0
(1)
k
N!
k!(N k)!
_
1
k
N
_
n1
_

k
N
_
=
N1

k=1
(1)
k1
_
N 1
k 1
_
_
1
k
N
_
n1
, (n = 2, 3, . . .). (11)
P(n) is zero for n = 2, 3, . . . , N 1. And P(1) is of course also zero, but it cannot
be expressed as in eq. (11), because eq. (9) is not valid for n = 0.
Having found P(n), we can write the average number, B, of required boxes as
B =

n=2
n
N1

k=1
(1)
k1
_
N 1
k 1
_
_
1
k
N
_
n1
. (12)
We have let the sum over n range from 2 to , instead of N to , because it will
simplify our calculations a bit (and all the terms from n = 2 to n = N 1 are
zero). Switching the order of the sums, and performing the sum over n by using a
technique similar to the one used in eq. (2), we obtain
B =
N1

k=1
(1)
k1
_
N 1
k 1
__
N
2
k
2
1
_
=
N1

k=1
(1)
k1
(N 1)!
(k 1)!(N k)!
_
(N + k)(N k)
k
2
_
=
N1

k=1
(1)
k
_
N 1
k
_
_
1 +
N
k
_
3
=
N1

k=1
(1)
k
_
N 1
k
_

N1

k=1
(1)
k
_
N 1
k
_
N
k
=
_
(1 1)
N1
1
_
N
N1

k=1
(1)
k
_
N 1
k
_
1
k
.
= 1 N
N1

k=1
(1)
k
_
N 1
k
_
1
k
. (13)
At this point, we need the following claim.
Claim:
M

k=1
(1)
k
_
M
k
_
1
k
=
_
1 +
1
2
+
1
3
+ +
1
M
_
.
Proof: We will prove this by using a technique similar to the one mentioned in the
above Remark, except that now we will nd it necessary to take an integral instead
of a derivative, because we want to produce a k in the denominator. Starting with
the binomial expansion, we have
M

k=0
(1)
k
_
M
k
_
x
k
= (1 x)
M
(14)
=
_
M

k=1
(1)
k
_
M
k
_
x
k1
dx =
_
1
x
_
(1 x)
M
1
_
dx
=
M

k=1
(1)
k
_
M
k
_
x
k
k
=
_
1 (1 x)
M
1 (1 x)
dx
=
M

k=1
(1)
k
_
M
k
_
x
k
k
=
_
_
1 + (1 x) + + (1 x)
M1
_
dx
=
M

k=1
(1)
k
_
M
k
_
x
k
k
= (1 x) +
1
2
(1 x)
2
+ +
1
M
(1 x)
M
+ C,
where C is a constant of integration. Setting x = 0 in the last line yields
C =
_
1 +
1
2
+
1
3
+ +
1
M
_
. (15)
Setting x = 1 in the last line then proves the claim.
Using the claim in eq. (13) gives
B = 1 + N
_
1
N 1
+ +
1
2
+ 1
_
= N
_
1
N
+
1
N 1
+ +
1
2
+ 1
_
, (16)
in agreement with eq. (4).
4
Now let us maximize P(n), as given in eq. (11). For convenience, we will instead
maximize P(n +1), so that there will be a simpler n in the exponent in eq. (11),
instead of n1. Of course, such a distinction is irrelevant in the limit of large N.
Let us assume for the moment (we will justify this below) that the values of k that
contribute signicantly to the sum in eq. (11) are small enough so that we may use
the approximation,
_
1
k
N
_
n

_
e
n/N
_
k
. (17)
The expression for P(n + 1) in eq. (11) then becomes (letting the sum run up to
k = N, since the k = N term is zero)
P(n + 1) e
n/N
N

k=1
(1)
k1
_
N 1
k 1
_
_
e
n/N
_
k1
= e
n/N
N1

j=0
(1)
j
_
N 1
j
_
_
e
n/N
_
j
= e
n/N
_
1 e
n/N
_
N1
e
n/N
e
(N1)e
n/N
. (18)
To maximize P(n), we therefore want to minimize n/N +(N 1)e
n/N
. Taking the
derivative of this with respect to n, we nd that the minimum is achieved when
n
max
= N ln(N 1) N lnN, (19)
where we have dropped terms of order 1. Using this result in eq. (18), we see that
the maximum value of P(n), which is obtained at n N lnN, equals 1/(Ne).
Remarks:
1. Let us now justify eq. (17). Using
ln
_
1
k
N
_
n
= nln
_
1
k
N
_
n
_

k
N

k
2
2N
2

_
, (20)
we have
_
1
k
N
_
n
e
nk/N
e
nk
2
/2N
2
. (21)
Therefore, eq. (17) is valid if nk
2
/N
2
1. For n near N lnN, which is the
general size of n we are concerned with, this requires that k
_
N/ lnN.
But for large N and for n N lnN, the rst few terms in the second (and
hence rst) line of eq. (18) dominate the sum. This is true because the
binomial coecient goes like N
j
/j!, and the (e
n/N
)
j
term goes like 1/N
j
(using n N lnN). Hence, the the jth term goes like 1/j!. Therefore, only
the rst few terms contribute, so the relevant k values easily satisfy the bound
k
_
N/ lnN.
Note also that the step in going from the third to fourth line in eq. (18) is
valid because e
n/N
1/N is suciently small.
5
2. For large N, the average number of boxes, and the point where P(n) is maxi-
mum (given in eqs. (5) and (19), respectively) are
B = N(lnN + ) and n
max
= N lnN. (22)
The dierence between these is N (0.577)N. For example, when N = 100,
the maximum occurs at N lnN 460, while the average number is N lnN +
N 518.
3. If we let n N lnN + xN, then we can rewrite eq. (18), to leading order in
N, as
NP(x) e
xe
x
. (23)
A plot of NP(x) vs. x is shown below. This is simply a plot of NP(n) vs. n,
centered at N lnN, and with the horizontal axis on a scale of N.

x
N P(x)
N P() = 0.32
1 -1 2 0 -2 3
0.1
0.2
0.3
N P(0) = 1/e = 0.37
-x-e
-x
e

For all large N, we obtain the same shape. Virtually all of the graph is
contained within a few units of N from the maximum. Compared to a P(n)
vs. n graph, this graph has its vertical axis multiplied by N and its horizontal
axis multiplied by 1/N, so the total integral should still be 1. And indeed,
_

e
xe
x
dx = e
e
x

= 1. (24)
How much area under the curve lies to the left of, say, x = 2? Letting
x = a to be general, the integral in eq. (24) gives an area (in other words,
a probability) of e
e
x
|
a

= e
e
a
. This decreases very rapidly as a grows.
Even for a = 2, we nd that there is a probability of only 0.06% of obtaining
all the colors before you hit the n = N lnN 2N box.
How much area under the curve lies to the right of, say, x = 3? Letting x = b
to be general, we nd an area of e
e
x
|

b
= 1 e
e
b
1 (1 e
b
) = e
b
.
This decreases as b grows, but not as rapidly as in the above case. For b = 3,
we nd that there is a probability of 5% that you havent obtained all the
colors by the time you hit the n = N lnN + 3N box.
6
4. You might be tempted to ddle around with a saddle-point approximation in
this problem. That is, you might want to approximate P(n) as a Gaussian
around its maximum at n
max
N lnN. However, this will not work in this
problem, because for any (large) N, P(n) will always keep its same lopsided
shape. The average will always be a signicant distance (namely N, which
is comparable to the spread of the graph) from the maximum, and the ratio
of the height at the average to the height at the maximum will always be
P()
P(0)
=
e
e

e
0e
0
0.87. (25)
7
Solution
Week 61 (11/10/03)
Falling rope
(a) First Solution: Let be the mass density of the rope. From conservation of
energy, we know that the ropes nal kinetic energy, which is (L)v
2
/2, equals
the loss in potential energy. This loss equals (L)(L/2)g, because the center
of mass falls a distance L/2. Therefore,
v =

gL. (1)
This is the same as the speed obtained by an object that falls a distance L/2.
Note that if the initial piece hanging down through the hole is arbitrarily short,
then the rope will take an arbitrarily long time to fall down. But the nal
speed will be always be (arbitrarily close to)

gL.
Second Solution: Let x be the length that hangs down through the hole.
The gravitational force on this length, which is (x)g, is responsible for chang-
ing the momentum of the entire rope, which is (L) x. Therefore, F = dp/dt
gives (x)g = (L) x, which is simply the F = ma equation. Hence, x =
(g/L)x, and the general solution to this equation is
x(t) = Ae
t

g/L
+ Be
t

g/L
. (2)
Note that if is the initial value for x, then A = B = /2 satises the ini-
tial conditions x(0) = and x(0) = 0, in which case we may write x(t) =
cosh(t

g/L). But we wont need this information in what follows.

Let T be the time for which x(T) = L. If is very small, then T will be very
large. But for large t,
1
we may neglect the negative-exponent term in eq. (2).
We then have
x Ae
t

g/L
= x Ae
t

g/L

g/L x

g/L (for large t).

(3)
When x = L, we obtain
x(T) = L

g/L =

gL, (4)
in agreement with the rst solution.
(b) Let be the mass density of the rope, and let x be the length that hangs
down through the hole. The gravitational force on this length, which is (x)g,
is responsible for changing the momentum of the rope. This momentum is
(x) x, because only the hanging part is moving. Therefore, F = dp/dt gives
xg =
d(x x)
dt
= xg = x x + x
2
. (5)
1
More precisely, for t

L/g.
1
Note that F = ma gives the wrong equation, because it neglects the fact that
the moving mass, x, is changing. It therefore misses the second term on
the right-hand side of eq. (5). In short, the momentum of the rope increases
because it is speeding up (which gives the x x term) and because additional
mass is continually being added to the moving part (which gives the x
2
term,
as you can show).
To solve eq. (5) for x(t), note that g is the only parameter in the equation.
Therefore, the solution for x(t) can involve only gs and ts.
2
By dimensional
analysis, x(t) must then be of the form x(t) = bgt
2
, where b is a numerical
constant to be determined. Plugging this expression for x(t) into eq. (5) and
dividing by g
2
t
2
gives b = 2b
2
+4b
2
. Therefore, b = 1/6, and our solution may
be written as
x(t) =
1
2

g
3

t
2
. (6)
This is the equation for something that accelerates downward with acceleration
g

= g/3. The time the rope takes to fall a distance L is then given by
L = g

t
2
/2, which yields t =

2L/g

. The nal speed in thus

v = g

t =

2Lg

2gL
3
. (7)
This is smaller than the

gL result from part (a). We therefore see that

although the total time for the scenario in part (a) is very large, the nal
speed in that case is in fact larger than that in the present scenario.
Remarks: Using eq. (7), you can show that 1/3 of the available potential energy
is lost to heat. This inevitable loss occurs during the abrupt motions that suddenly
bring the atoms from zero to non-zero speed when they join the moving part of the
rope. The use of conservation of energy is therefore not a valid way to solve this
problem.
You can show that the speed in part (a)s scenario is smaller than the speed in part
(b)s scenario for x less than 2L/3, but larger for x greater than 2L/3.
2
The other dimensionful quantities in the problem, L and , do not appear in eq. (5), so they
cannot appear in the solution. Also, the initial position and speed (which will in general appear in
the solution for x(t), because eq. (5) is a second-order dierential equation) do not appear in this
case, because they are equal to zero.
2
Solution
Week 62 (11/17/03)
Leftover dental oss
Let (x, y) denote the occurrence where x segments have been cut o the left roll,
and y segments have been cut o the right roll. In solving this problem, well need
to rst calculate the probability that the process ends at (N, n), which leaves a
length (N n)d on the right roll. For this to happen, the process must rst get to
(N1, n), and then the left roll must be chosen for the last segment. The probability
of reaching the point (N 1, n) is
P
N1,n
=
1
2
N1+n

N 1 + n
n

, (1)
because the binomial coecient gives the number of dierent ways the N 1 + n
choices of roll (each of which occurs with probability 1/2) can to end up at the point
(N 1, n). The probability of then choosing the left roll for the next piece is 1/2.
Therefore, the probability of ending the process at (N, n) is
P
end
N,n
=
1
2
N+n

N 1 + n
n

. (2)
The leftover length in this case is (Nn)d, so we see that the average leftover length
at the end of the process is
= 2d
N1

n=0
(N n)P
end
N,n
= 2d
N1

n=0
N n
2
N+n

N 1 + n
n

, (3)
where the factor of 2 out front comes from the fact that either roll may be the one
that runs out.
In order to simplify this result, we will use the standard result that for large N,
a binomial coecient can be approximated by a Gaussian function. From Problem
of the Week 58, we have, for large N and x N,

2M
M x

2
2M

M
e
x
2
/M
. (4)
To make use of this, well rst need to rewrite eq. (3) as
= 2d
N1

n=0
N n
2
N+n
N
N + n

N + n
n

= 2d
N

z=1
z
2
2Nz
N
2N z

2N z
N z

(with z N n)
= 2d
N

z=1
z
2
2Nz
N
2N z

2(N z/2)
(N z/2) z/2

. (5)
1
Using eq. (4) to rewrite the binomial coecient gives (with M N z/2, and
x z/2)
2d
N

z=1
Nz
2N z
e
z
2
/4(Nz/2)

(N z/2)

N
N

z=0
ze
z
2
/(4N)

0
ze
z
2
/(4N)
dz
= 2d

. (6)
In obtaining the second line above, we have kept only the terms of leading order
in N. The exponential factor guarantees that only z values up to order

N will
contribute. Hence, z is negligible when added to N.
Without doing all the calculations, its a good bet that the answer should go like

N d, but it takes some eort to show that the exact coecient is 2/

.
In terms of the initial length of oss, L Nd, the average leftover amount can
be written as (2/

Ld. So we see that is proportional to the geometric

mean of L and d.
2
Solution
Week 63 (11/24/03)
Minimal surface
First Solution: By symmetry, the surface is obtained by rotating a certain function
y(x) around the x-axis. Our goal is to nd y(x). Consider a thin vertical cross-
sectional ring of the surface, as shown below. The ratio of the circumferences of the
circular boundaries of the ring is y
2
/y
1
.
1

1
2
y 2
y
The condition that the bubble be in equilibrium is that the tension (force per unit
length) throughout the surface is constant, because otherwise there would be a net
force on some little patch. Therefore, the requirement that the horizontal forces on
the ring cancel is y
1
cos
1
= y
2
cos
2
, where the s are the angles of the surface,
as shown. In other words, y cos is constant throughout the surface. But cos =
1/

1 + y
2
, so we have
y

1 + y
2
= Constant = 1 + y
2
= By
2
, (1)
where B is some constant. At this point, motivated by the facts that 1 + sinh
2
z =
cosh
2
z and d(coshz)/dz = sinhz, we can guess that the solution to this dierential
equation is
y(x) =
1
b
coshb(x + d), (2)
where b =

B, and d is a constant of integration. Or, we can do things from
scratch by solving for y

dy/dx and then separating variables to obtain (again

with b =

B)
dx =
dy

(by)
2
1
. (3)
We can then use the fact that the integral of 1/

z
2
1 is cosh
1
z, to obtain the
same result as in eq. (2).
Eq. (2) gives the general solution in the case where the rings may have unequal
radii. The constants b and d are determined by the boundary conditions (the facts
that the y values equal the radii of the rings at the x-values of the rings). In the
special case at hand where the radii are equal, the two boundary conditions give
r = (1/b) coshb( + d), where x = 0 has been chosen to be midway between the
rings. Therefore, d = 0, and so the constant b is determined from
r =
1
b
coshb. (4)
1
Our solution for y(x) then
y(x) =
1
b
coshbx. (5)
There is, however, an ambiguity in this solution, in that there may be two solutions
for b in eq. (4). Well comment on this in the rst Remark at the end of the
problem.
Second Solution: We can also solve this problem by using a principle of least
action type of argument, which takes advantage of the fact that the surface is the
one with the minimum area. There are two ways of going about this. One is quick,
and the other is lengthy. A sketch of the lengthy way is the following. The area of
the surface in the following gure is

2y

1 + y
2
dx, (6)
where y

dy/dx.
r
x
y
r
l l
In analogy with the principle of least action, our Lagrangian (from a physicists
point of view) is L = 2y

1 + y
2
, and in order for the area to be minimized, L
must satisfy the Euler-Lagrange equation,
d
dx

L
y

=
L
y
. (7)
It is, alas, rather tedious to work through all of the necessary dierentiations here.
If you so desire, you can show that this equation does in fact lead to eq. (1). But
lets instead just do things the quick way. If we consider x to be a function of y
(theres no need to worry about any double-value issues, because the Euler-Lagrange
formalism deals with local variations), we may write the area as

2y

1 + x
2
dy, (8)
where x

dx/dy. Our Lagrangian is now L = 2y

1 + x
2
, and the Euler-
Lagrange equation gives
d
dy

L
x

=
L
x
=
d
dy

yx

1 + x
2

= 0. (9)
The zero on the right-hand side makes everything nice and easy, because it tells us
that yx

1 + x
2
is constant. Dening this constant to be 1/b, and then solving
2
for x

and separating variables, gives

dx =
dy

(by)
2
1
, (10)
which is identical to eq. (3). The solution then concludes as above.
Let us now determine the maximum value of /r for which the minimal surface
exists. If /r is too large, then we will see that there is no solution for b in eq. (4).
In short, the minimal surface turns out to be the two given circles, attached by a
line, which isnt a nice two-dimensional surface. If you perform the experiment with
soap bubbles (which want to minimize their area), and if you pull the rings too far
apart, then the surface will break and disappear, as it tries to form the two circles.
Dene the dimensionless quantities,


r
, and z br. (11)
Then eq. (4) becomes
z = coshz. (12)
If we make a rough plot of the graphs of w = z and w = coshz for a few values
of , as shown below, we see that there is no solution for z if is too big. The
limiting value of for which there exists a solution occurs when the curves w = z
and w = coshz are tangent; that is, when the slopes are equal in addition to the
functions being equal.
w
w
z
z
z cosh( )
=
w=
Let
0
be the limiting value of , and let z
0
be the place where the tangency occurs.
Then equality of the values and the slopes gives
z
0
= cosh(
0
z
0
), and 1 =
0
sinh(
0
z
0
). (13)
Dividing the second of these equations by the rst gives
1 = (
0
z
0
) tanh(
0
z
0
). (14)
This must be solved numerically. The solution is

0
z
0
1.200. (15)
Plugging this into the second of eqs. (13) gives

max

0
0.663. (16)
3
Note also that z
0
= 1.200/
0
= 1.810. We see that if /r is larger than 0.663, then
there is no solution for y(x) that is consistent with the boundary conditions. Above
this value of /r, the soap bubble minimizes its area by heading toward the shape
of just two disks, but it will pop well before it reaches that conguration.
Remarks:
1. As mentioned at the end of the rst solution above, there may be more than one
solution for the constant b in eq. (5). In fact, the preceding graph shows that for
any < 0.663, there are two solutions for z in eq. (12), and hence two solutions for
b in eq. (4). This means that there are two possible surfaces that might solve our
problem. Which one do we want? It turns out that the surface corresponding to the
smaller value of b is the one that minimizes the area, while the surface corresponding
to the larger value of b is the one that (in some sense) maximizes the area.
We say in some sense because the surface is actually a saddle point for the area.
It cant be a maximum, after all, because we can always make the area larger by
adding little wiggles to it. Its a saddle point because there does exist a class of
variations for which is has the maximum area, namely ones where the dip in the
curve is continuously made larger (just imagine lowering the midpoint in a smooth
manner). The reason why this curve arises in the rst solution above is that we
simply demanded that the surface be in equilibrium; it just happens to be an unstable
equilibrium in this case. The reason why it arises in the second solution above is that
the Euler-Lagrange technique simple sets the derivative equal to zero and thus does
not dierentiate between maxima, minima, and saddle points.
2. How does the area of the limiting surface (with
0
= 0.663) compare with the area of
the two circles? The area of the two circles is
A
c
= 2r
2
. (17)
The area of the limiting surface is
A
s
=

2y

1 + y
2
dx. (18)
Using eq. (5), this becomes
A
s
=

2
b
cosh
2
bxdx
=

b
(1 + cosh2bx) dx
=
2
b
+
sinh2b
b
2
. (19)
But from the denitions of and z, we have =
0
r and b = z
0
/r for the limiting
surface. Therefore, A
s
can be written as
A
s
= r
2

2
0
z
0
+
sinh2
0
z
0
z
2
0

. (20)
Plugging in the numerical values (
0
0.663 and z
0
1.810) gives
A
c
(6.28)r
2
, and A
s
(7.54)r
2
. (21)
The ratio of A
s
to A
c
is approximately 1.2 (its actually
0
z
0
, as you can show). The
limiting surface therefore has a larger area. This is expected, of course, because for
/r >
0
the surface tries to run o to one with a smaller area, and there are no other
stable congurations besides the cosh solution we found.
4
3. How does the area of the surface change if we gradually transform it from a cylinder
to the two disks? There are many ways to go about doing this transformation, but
lets just be vague and say that we pick a nice smooth method that passes through
the two cosh solutions (if < 0.663) that we found above. The transformation might
look something like:
2l
r
The area of the starting cylinder is A
i
= (2)(2r) = 4r, and the area of the ending
two disks is A
f
= 2r
2
. Note that the ratio of these is A
i
/A
f
= 2/r 2. For
a given /r, what does the plot of the changing area look like? Below are four
qualitative plots, for four values of . Weve imagined changing by keeping r (and
hence A
f
) xed and changing . Ignore the actual measure along the axes; just look
at the general shape.
= 0.1
A
i
A
i
A
i
A
i
A
f
A
f
A
f
A
f
= 0.5
= 0.6 = 0.663
,
cylinder 2 disks
cylinder 2 disks cylinder 2 disks
cylinder 2 disks
We see that = 0.663 is the value for which the maximum and minimum merge into
one point with zero slope. For higher values of , there are no points on the curve
that have zero slope.
5
The actual values of the area along these curves is nebulous, because we havent
been quantitative about exactly how were varying the surface. But the area at the
maximum and minimum (at which points we have one of our cosh surfaces) can be
found from eq. (20), which says that
A
s
= r
2

2
z
+
sinh2z
z
2

, (22)
for general /r and z br. For a given , the two solutions for z are found from
eq. (12). Plugging each of these into eq. (22) gives the areas at the maximum and
minimum.
6
Solution
Week 64 (12/1/03)
Attracting bugs
First Solution: In all of these solutions, the key point to realize is that at any
time, the bugs form the vertices of a regular N-gon, as shown below for N = 6. This
is true because this is the only conguration that respects the symmetry of the N
bugs. The N-gon will rotate and shrink until it becomes a point at the center.
N = 6
l
v
The important quantity in this rst solution is the relative speed of two adjacent
bugs. This relative speed is constant, because the relative angle of the bugs motions
is always the same. If the bugs speed is v, then we see from the gure below that
the relative speed is v
r
= v(1 cos ), where = 2/N. This is the rate at which
the separation between two adjacent bugs decreases.
1
N = 6
v v cos

v
For example, if N = 3 we have v
r
= 3v/2; if N = 4 we have v
r
= v; and if N = 6
we have v
r
= v/2. Note also that for N = 2 (which does not give not much of a
polygon, being just a straight line) we have v
r
= 2v, which is correct for two bugs
walking directly toward each other. And if N we have v
r
0, which is correct
for bugs walking around in a circle.
If two bugs start a distance apart, and if they always walk at a relative speed
of v(1 cos ), then the time it takes for them to meet is t = /(v(1 cos )).
Therefore, since the bugs walk at speed v, they will each travel a total distance of
vt =

1 cos(2/N)
. (1)
1
The transverse v sin component of the front bugs velocity is irrelevant here, because it provides
no rst-order change in the distance between the bugs, for small increments in time.
1
Note that for a square, this distance equals the length of a side, . For large N, the
approximation cos 1
2
/2 gives vt N
2
/(2
2
).
The bugs will spiral around an innite number of times. This is true because
the future path of the bugs at any time must simply be a scaled-down version of
the future path at the start (because any point in time may be considered to be
the start time, with a scaled-down version of the initial separation). This would
not be possible if the bugs hit the center after spiraling around only a nite number
of times. We will see in the third solution below that the bugs distance from the
center deceases by a factor e
2 tan(/N)
after each revolution.
2
Second Solution: In this solution, we will determine how quickly the bugs ap-
proach the center of the N-gon. A bugs velocity may be separated into radial and
tangential components, v
R
and v
T
, as shown below. Because at any instant the bugs
all lie on the vertices of a regular N-gon, they always walk at the same angle relative
to a circular motion. Therefore, the magnitudes of v
R
and v
T
remain constant.
N = 6
v
v
v
/N
R
T
What is the radial component, v
R
, in terms of v? The angle between a bugs motion
and a circular motion is /N, so we have
v
R
= v sin(/N). (2)
What is the radius, R
0
, of the initial N-gon? A little geometry shows that
R
0
=

2 sin(/N)
. (3)
The time it takes a bug to reach the center is then t = R
0
/v
R
= (/v)/(2 sin
2
(/N)) .
Therefore, each bug travels a total distance of
vt =

2 sin
2
(/N)
. (4)
2
Of course, bugs of non-zero size would hit each other before they reach the center. If the bugs
happen to be very, very small, then they would eventually require arbitrarily large friction with the
oor, in order to provide the centripetal acceleration needed to keep them in a spiral with a very
small radius.
2
This agrees with the result of the rst solution, due to the half-angle formula,
sin
2
(/2) = (1 cos )/2. The same reasoning used in the rst solution shows that
the bugs spiral around an innite number of times.
Third Solution: In this solution, we will parametrize a bugs path, and then
integrate the dierential arclength. Let us nd a bugs distance, R(), from the
center, as a function of the angle through which it has travelled. The angle
between a bugs motion and a circular motion is /N. Therefore, the change in
radius, dR, divided by the change in arclength along the circle, Rd, is dR/(Rd) =
tan(/N). Separating variables and integrating gives

R
R
0
dR
R
=

0
tan(/N) d
= ln(R/R
0
) = tan(/N)
= R() = R
0
e
tan(/N)
, (5)
where R
0
is the initial distance from the center, equal to /(2 sin(/N)). We now
see, as stated in the rst solution, that one revolution decreases R by the factor
e
2 tan(/N)
, and that an innite number of revolutions is required for R to become
zero. Having found R(), we may nd the total distance travelled by integrating
the arclength:

(Rd)
2
+ (dR)
2
=

R
2
+ (dR/d)
2
d
=

0
R
0
e
tan(/N)
d
cos(/N)
=

2 sin
2
(/N)
. (6)
Remark: In the rst solution, we found that for large N the total distance travelled
goes like N
2
/(2
2
). This result can also be found in the following manner. For large
N, a bugs motion can be approximated by a sequence of circles, C
n
, with radii R
n
=
R
0
e
n(2) tan(/N)
R
0
e
n(2
2
/N)
. To leading order in N, the total distance travelled is
therefore the sum of the geometric series,

n=0
2R
n

n=0
2R
0
e
n(2
2
/N)
=
2R
0
1 e
2
2
/N

2R
0
2
2
/N

N
2

2
2
, (7)
where we have used R
0
= /(2 sin(/N)) N/(2).
3
Solution
Week 65 (12/8/03)
Relativistic cart
Ground frame (your frame): In your frame, the force at your feet is responsible
for changing the momentum of the cart-plus-sand-inside system. Lets label this
system as C.
To nd the dp/dt of C, lets determine how fast the mass of C increases. We
claim that the rate of mass increase is . This can be seen as follows. Assume
that C has a mass M at a given time. A mass dt falls into the cart during a time
dt. The energy of the resulting C

is M + dt (well drop the cs here), while the

momentum is still Mv. Using E
2
= p
2
+m
2
, we see that the resulting mass equals
M

(M + dt)
2
(Mv)
2

M
2
+ 2M dt, (1)
where we have dropped the second-order dt
2
terms. Using the Taylor series

1 +
1 + /2, we may approximate M

as
M

1 +
2 dt
M
M

1 +
dt
M

= M + dt. (2)
Therefore, Cs mass increases at a rate .
1
Intuitively, this rate of increase must
certainly be greater than the nonrelativistic result of , because heat is generated
during the collision, and this heat shows up as mass in the nal object.
Having found the rate at which the mass increases, we see that the rate at which
the momentum increases is (using the fact that v is constant)
dp
dt
=

dM
dt

v = ()v =
2
v. (3)
Since F = dp/dt, this is the force that you exert on the cart. Therefore, it is also
the force that the ground exerts on your feet (because the net force on you is zero).
Cart frame: The sand-entering-cart events happen at the same location in the
ground frame, so time dilation says that the sand enters the cart at a slower rate
in the cart frame; that is, at a rate /. The sand ies in at speed v, and then
eventually comes at rest on the cart, so its momentum decreases at a rate (/)v =
v. This is the force that your hand applies to the cart.
If this were the only change in momentum in the problem, then we would have
a problem, because the force on your feet would be v in the cart frame, whereas
we found above that it is
2
v in the ground frame. This would contradict the fact
that longitudinal forces are the same in dierent frames. What is the resolution of
this apparent paradox?
1
This result is easier to see if we work in the frame where C is at rest. In this frame, a mass dt
comes ying in with energy dt, and essentially all of this energy shows up as mass (heat) in the
nal object. That is, essentially none of it shows up as overall kinetic energy of the object, which
is a general result for when a small object hits a stationary large object.
1
The resolution is that while you are pushing on the cart, your mass is decreasing.
You are moving with speed v in the cart frame, and mass is continually being
transferred from you (who are moving) to the cart (which is at rest). This is the
missing change in momentum we need. Lets be quantitative about this.
Go back to the ground frame for a moment. We found above that the mass of
C increases at rate in the ground frame. Therefore, the energy of C increases at
a rate () in the ground frame. The sand provides of this energy, so you must
provide the remaining (
2
1) part. Therefore, since you are losing energy at this
rate, you must also be losing mass at this rate in the ground frame (because you are
at rest there).
Now go back to the cart frame. Due to time dilation, you lose mass at a rate
of only (
2
1)/. This mass goes from moving at speed v (that is, along with
you), to speed zero (that is, at rest on the cart). Therefore, the rate of decrease in
momentum of this mass is ((
2
1)/)v = (
2
1)v.
Adding this result to the v result due to the sand, we see that the total rate of
decrease in momentum is
2
v. This is therefore the force that the ground applies
to your feet, in agreement with the calculation in the ground frame.
2
Solution
Week 66 (12/15/03)
Bowl of spaghetti
Assume that we have reached into the bowl and pulled out one end. Then there are
2N 1 free ends left in the bowl. Therefore, there is a 1/(2N 1) chance that a
loop is formed by choosing the other end of the noodle that we are holding. And
there is a (2N 2)/(2N 1) chance that a loop is not formed. In the former case,
we end up with one loop and N 1 strands. In the latter case, we just end up with
N 1 strands, because we have simply created a strand of twice the original length,
and the length of a strand is irrelevant in this problem.
Therefore, after the rst step, we see that no matter what happens, we end
up with N 1 strands and, on average, 1/(2N 1) loops. We can now repeat
this reasoning with N 1 strands. After the second step, we are guaranteed to be
left with N 2 strands and, on average, another 1/(2N 3) loops. This process
continues until we are left with one strand, whereupon the nal Nth step leaves us
with zero strands, and we (denitely) gain one more loop.
Adding up the average number of loops gained at each stage, we obtain an
average total number of loops equal to
n =
1
2N 1
+
1
2N 3
+ +
1
3
+ 1. (1)
This grows very slowly with N. It turns out that we need N = 8 noodles in order
to expect at least two loops. If we use the ordered pair (n, N) to signify that N
noodles are needed in order to expect n loops, we can numerically show that the
rst few integer-n ordered pairs are: (1, 1), (2, 8), (3, 57), (4, 419), and (5, 3092).
For large N, we can say that the average number of loops given in eq. (1) is
roughly equal to 1/2 times the sum of the reciprocals up to 1/N. So it approximately
equals (lnN)/2. To get a better approximation, let S
N
denote the sum of the integer
reciprocals up to 1/N. Then we have (using S
N
lnN + , where 0.577 is
Eulers constant)
n +

1
2
+
1
4
+ +
1
2N 2
+
1
2N

= S
2N
= n +
1
2
S
N
= S
2N
= n +
1
2
(lnN + ) ln(2N) +
= n
1
2
(lnN + ln4 + )
= N
e
2n
4
. (2)
You can show that this relation between n and N agrees with the above numerical
results, except for the n = 1 case. You will need to use the more precise value of
0.5772 for the n = 5 case.
1
Solution
Week 67 (12/22/03)
Inverted pendulum
Let be dened as shown below. Well use the Lagrangian method to determine
the equation of motion for .

l
m
y(t)
With y(t) = Acos(t), the position of the mass m is given by
(X, Y ) = ( sin, y + cos ). (1)
Taking the derivatives of these coordinates, we see that the square of the speed is
V
2
=

X
2
+

Y
2
=
2

2
+ y
2
2 y

sin. (2)
The Lagrangian is therefore
L = K U =
1
2
m(
2

2
+ y
2
2 y

sin) mg(y + cos ). (3)
The equation of motion for is
d
dt

=
L

y sin = g sin. (4)

Plugging in the explicit form of y(t), we have

+ sin

A
2
cos(t) g

= 0. (5)
In retrospect, this makes sense. Someone in the reference frame of the support,
which has acceleration y = A
2
cos(t), may as well be living in a world where
the acceleration from gravity is g A
2
cos(t) downward. Eq. (5) is simply the
F = ma equation in the tangential direction in this accelerated frame.
Assuming is small, we may set sin , which gives

a
2
cos(t)
2
0

= 0, (6)
where
0

g/, and a A/. Eq. (6) cannot be solved exactly, but we can still
get a good idea of how depends on time. We can do this both numerically and
(approximately) analytically.
The gures below show how depends on time for parameters with values =
1 m, A = 0.1 m, and g = 10 m/s
2
(so a = 0.1, and
2
0
= 10 s
2
). In the rst plot,
= 10 s
1
. And in the second plot, = 100 s
1
. The stick falls over in rst case,
but undergoes oscillatory motion in the second case. Apparently, if is large enough
the stick will not fall over.
1
0.2 0.4 0.6 0.8 1 1.2 1.4
t
0.1
0.05
0.05
0.1
theta
0.2 0.4 0.6 0.8 1 1.2 1.4
t
0.25
0.5
0.75
1
1.25
1.5
1.75
theta
Lets now explain this phenomenon analytically. At rst glance, its rather sur-
prising that the stick stays up. It seems like the average (over a few periods of the
oscillations) of the tangential acceleration in eq. (6), namely (a
2
cos(t)
2
0
),
equals the positive quantity
2
0
, because the cos(t) term averages to zero (or so
it appears). So you might think that there is a net force making increase, causing
the stick fall over.
The fallacy in this reasoning is that the average of the a
2
cos(t) term is not
zero, because undergoes tiny oscillations with frequency , as seen below. Both
of these plots have a = 0.005,
2
0
= 10 s
2
, and = 1000 s
1
(well work with small
a and large from now on; more on this below). The second plot is a zoomed-in
version of the rst one near t = 0.
2 4 6 8
t
0.1
0.05
0.05
0.1
theta
0.02 0.04 0.06 0.08 0.1
t
0.098
0.0985
0.099
0.0995
theta
The important point here is that the tiny oscillations in shown in the second
plot are correlated with cos(t). It turns out that the value at the t where
cos(t) = 1 is larger than the value at the t where cos(t) = 1. So there is a
net negative contribution to the a
2
cos(t) part of the acceleration. And it may
indeed be large enough to keep the pendulum up, as we will now show.
To get a handle on the a
2
cos(t) term, lets work in the approximation
where is large and a A/ is small. More precisely, we will assume a 1 and
a
2

2
0
, for reasons we will explain below. Look at one of the little oscillations
in the second of the above plots. These oscillations have frequency , because they
are due simply to the support moving up and down. When the support moves
up, increases; and when the support moves down, decreases. Since the average
position of the pendulum doesnt change much over one of these small periods, we
can look for an approximate solution to eq. (6) of the form
(t) C +b cos(t), (7)
2
where b C. C will change over time, but on the scale of 1/ it is essentially
constant, if a A/ is small enough.
Plugging this guess for into eq. (6), and using a 1 and a
2

2
0
, we nd
that b
2
cos(t) +Ca
2
cos(t) = 0, to leading order.
1
So we must have b = aC.
Our approximate solution for is therefore
C

1 +a cos(t)

. (8)
Lets now determine how C gradually changes with time. From eq. (6), the
average acceleration of , over a period T = 2/, is

a
2
cos(t)
2
0

1 +a cos(t)

a
2
cos(t)
2
0

= C

a
2

2
cos
2
(t)
2
0

= C

a
2

2
2

2
0

C
2
, (9)
where
=

a
2

2
2

g

. (10)
But if we take two derivatives of eq. (7), we see that

simply equals

C. Equating
this value of

with the one in eq. (9) gives

C(t) +
2
C(t) 0. (11)
This equation describes nice simple-harmonic motion. Therefore, C oscillates sinu-
soidally with the frequency given in eq. (10). This is the overall back and forth
motion seen in the rst of the above plots. Note that we must have a >

2
0
if
this frequency is to be real so that the pendulum stays up. Since we have assumed
a 1, we see that a
2

2
> 2
2
0
implies a
2

2
0
, which is consistent with our
initial assumption above.
If a
0
, then eq. (10) gives a/

2. Such is the case if we change

the setup and simply have the pendulum lie at on a horizontal table where the
acceleration from gravity is zero. In this limit where g is irrelevant, dimensional
analysis implies that the frequency of the C oscillations must be a multiple of ,
because is the only quantity in the problem with units of frequency. It just so
happens that the multiple is a/

2.
1
The reasons for the a 1 and a
2

2
0
qualications are the following. If a
2

2
0
, then the
a
2
cos(t) term dominates the
2
0
term in eq. (6). The one exception to this is when cos(t) 0,
but this occurs for a negligibly small amount of time if a
2

2
0
. If a 1, then we can legally
ignore the

C term when eq. (7) is substituted into eq. (6). We will nd below, in eq. (9), that our
assumptions lead to

C being roughly proportional to a
2

2
. Since the other terms in eq. (6) are
proportional to a
2
, we need a 1 in order for the

C term to be negligible. In short, a 1 is the
condition under which C varies slowly on the time scale of 1/.
3
Solution
Week 68 (12/29/03)
Tower of circles
Let the bottom circle have radius 1, and let the second circle have radius r. From
the following gure, we have
sin =
1 r
1 + r
, where /2. (1)

= /2
r
r
1-r
1
In solving this problem, it is easier to work with r, instead of the angle . So
we will nd the value of r for which A
C
/A
T
is maximum, and then use eq. (1) to
obtain . Note that r is the ratio of the radii of any two adjacent circles. This
follows from the fact that we could have drawn the above thin little right triangle
by using any two adjacent circles. Alternatively, it follows from the fact that if we
scale up the top N 1 circles by the appropriate factor, then we obtain the bottom
N 1 circles.
The area, A
T
, of the triangle may be calculated in terms of r and N as follows.
Since we could imagine stacking an innite number of circles up to the vertex of the
triangle, we see that the the height of the triangle is
h = 2 + 2r + 2r
2
+ 2r
3
+ =
2
1 r
. (2)
The length of half the base, b, of the triangle is give by b/2 = htan. But from eq.
(1) we have tan = (1 r)/(2

r). Therefore,
b =
2h
tan
=
2

r
. (3)
The area of the triangle is then
A
T
=
bh
2
=
2
(1 r)

r
. (4)
1
The total area of the circles is
A
C
=
_
1 + r
2
+ r
4
+ r
2(N1)
_
(5)
=
1 r
2N
1 r
2
. (6)
Therefore, the ratio of the areas is
A
C
A
T
=

2

r(1 r
2N
)
1 + r
. (7)
Setting the derivative of this equal to zero to obtain the maximum, we nd
(1 r) (4N + 1)r
2N
(4N 1)r
2N+1
= 0. (8)
In general, this can only be solved numerically for r. But if N is very large, we can
obtain an approximate solution. To leading order in N, we may set 4N 1 4N.
We may also set r
2N+1
r
2N
, because r must be very close to 1 (otherwise there
would be nothing to cancel the 1 term in eq. (8)). For convenience, let us write
r 1 , where is very small. Eq. (8) then yields
8N(1 )
2N
. (9)
But (1 )
2N
e
2N
.
1
Hence,
e
2N

8N
. (10)
Taking the log of both sides gives

1
2N
ln
_
8N

1
2N
ln
_
_
8N
1
2N
ln
_
8N

_
_
_
, etc. (11)
Therefore, to leading order in N, we have

1
2N
ln
_
16N
2
O(lnN)
_
=
lnN O(lnlnN) +
N

lnN
N
. (12)
Note that for large N, this is much less than 1/

N, so eq. (10) is indeed valid.

Hence, r = 1 1 (lnN)/N. Eq. (1) then gives
= 2 2 sin = 2
1 r
1 + r

2
2
= , (13)
and so

lnN
N
. (14)
This is the desired answer to leading order in N, in the sense that as N becomes
very large, this answer becomes multiplicatively arbitrarily close to the true answer.
1
This follows from taking the log of (1 )
2N
, to obtain ln((1)
2N
) = 2N ln(1) 2N( +

2
/2 + ). This is approximately equal to 2N, provided that the second term in the expansion
is small, which is the case when 1/

N, which we will nd to be true.

2
Remarks:
1. The radius of the top circle in the stack is
R
N
= r
N1
r
N
= (1 )
N
e
lnN
. (15)
Using eq. (10) and then eq. (12), we have
R
N

_

8N

lnN
2

2N
. (16)
2. The distance from the center of the top circle to the vertex equals
R
N
sin

R
N

ln N
2

2N
1
2N
lnN
=
1

2 lnN
, (17)
which goes to zero (but very slowly) for large N.
3. For r 1 (lnN)/N, eq. (7) yields A
C
/A
T
/4. This is the expected
answer, because if we look at a small number of adjacent circles, they appear
to be circles inside a rectangle (because the long sides of the isosceles triangle
are nearly parallel for small ), and it is easy to see that /4 is the answer for
the rectangular case.
4. Using eq. (7), along with r = (1 sin)/(1 + sin) from eq. (1), we can
make a plot of (4/)(A
C
/A
T
) as a function of sin. The gure below shows
the plot for N = 20. In the limit of very large N, the left part of the graph
approaches a vertical segment. The rest of the curve approaches a quarter
circle, as N goes to innity. That is, (4/)(A
C
/A
T
) cos , for N .
This is true because if N is large, and if is larger than order (1/N) lnN,
then we eectively have an innite number of circles in the triangle. In this
innite case, the ratio A
C
/A
T
is given by the ratio of the area of a circle to
the area of a circumscribing trapezoid whose sides are tilted at an angle .
You can show that this ratio is (/4) cos .
0.2 0.4 0.6 0.8 1
sin beta
0.2
0.4
0.6
0.8
1
3
5. We can also consider the more general case of higher dimensions. For example,
instead of stacking N circles inside a triangle, we can stack N spheres inside
a cone. Let be the angle at the peak of the cone. Then the for which the
ratio of the total volume of the spheres to the volume of the cone is maximum
is (2 lnN)/(3N). And the answer in the general case of d dimensions
(with d 2) is (2 lnN)/(dN). This agrees with eq. (14) for the d = 2
case. We can show this general result as follows.
As in the original problem, the height and base radius of the generalized cone
are still
h =
2
1 r
, and b =
2

r
. (18)
Therefore, the volume of the cone is proportional to
V
cone
b
d1
h
1
r
(d1)/2
(1 r)
. (19)
The total volume of the spheres is proportional to
V
spheres
1 + r
d
+ r
2d
+ r
(N1)d
(20)
=
1 r
Nd
1 r
d
. (21)
Therefore,
V
spheres
V
cone

(1 r
Nd
)r
(d1)/2
(1 r)
1 r
d
. (22)
To maximize this, it is easier to work with the small quantity 1 r. In
terms of , we have (using the binomial expansion)
V
spheres
V
cone

(1 (1 )
Nd
)(1 )
(d1)/2

1 (1 )
d
.

(1 e
Nd
)
_
1
_
d1
2
_
+
_
(d1)(d3)
8
_

2

_
d
_
1
_
d1
2
_
+
_
(d1)(d2)
6
_

2

_ . (23)
The terms in the square brackets in the numerator and the denominator dier
at order
2
, so we have
V
spheres
V
cone
(1 e
Nd
)(1 A
2
+ ), (24)
where A happens to equal (d
2
1)/24, but we wont need this. Taking the
derivative of eq. (24) with respect to to obtain the maximum, we nd
(1 e
Nd
)(2A) + Nde
Nd
(1 A
2
) = 0. (25)
The N in the second term tells us that e
Nd
must be at most order 1/N.
Therefore, we can set 1 e
Nd
1 in the rst term. Also, we can set
1 A
2
1 in the second term. This gives
e
Nd

2A
Nd
=
1
Nd
ln
_
Nd
2A
_
. (26)
4
In the same manner as in eq. (12), we nd, to leading order in N,

2 lnN
dN
. (27)
And since = from eq. (13), we obtain (2 lnN)/(dN), as desired.
6. Eq. (8) can be solved numerically for r, for any value of N. A few results are:
N r (deg) (rad) (lnN)/N (4/)(A
C
/A
T
)
1 .333 60 1.05 0 .770
2 .459 43.6 .760 .347 .887
3 .539 34.9 .609 .366 .931
10 .754 16.1 .282 .230 .987
100 .953 2.78 .0485 .0461 .999645
1000 .9930 .400 6.98 10
3
6.91 10
3
1 6.96 10
6
10
6
.9999864 7.76 10
4
1.36 10
5
1.38 10
5
1 2.47 10
11
For large N, we see that
(rad)
lnN
N
, and r 1
lnN
N
. (28)
Also, using eqs. (7) and (12), you can show that to leading order in N,
4

A
C
A
T
1
(lnN)
2
+ lnN
8N
2
, (29)
which agrees well with the numerical results, for large N.
5
Solution
Week 69 (1/5/04)
Compton scattering
We will solve this problem by making use of 4-momenta. The 4-momentum of a
particle is given by
P (P
0
, P
1
, P
2
, P
3
) (E, p
x
c, p
y
c, p
z
c) (E, pc). (1)
In general, the inner-product of two 4-vectors is given by
A B A
0
B
0
A
1
B
1
A
2
B
2
A
3
B
3
. (2)
The square of a 4-momentum (that is, the inner product of a 4-momentum with
itself) is therefore
P
2
P P = E
2
|p|
2
c
2
= m
2
c
4
. (3)
Lets now apply these idea to the problem at hand. We will actually be doing
nothing here other than applying conservation of energy and momentum. Its just
that the language of 4-vectors makes the whole procedure surprisingly simple. Note
that conservation of E and p during the collision can be succinctly written as
P
before
= P
after
. (4)
Referring to the gure below, the 4-momenta before the collision are
P

hc

,
hc

, 0, 0

, P
m
= (mc
2
, 0, 0, 0). (5)
And the 4-momenta after the collision are
P

hc

,
hc

cos ,
hc

sin, 0

, P

m
= (we won

t need this). (6)

m
m

'
x
y
If we wanted to, we could write P

m
in terms of its momentum and scattering angle.
But the nice thing about this 4-momentum method is that we dont need to introduce
any quantities that were not interested in.
1
Conservation of energy and momentum give P

+P
m
= P

+P

m
. Therefore,
(P

+P
m
P

)
2
= P
2
m
=P
2

+P
2
m
+P
2

+ 2P
m
(P

) 2P

= P
2
m
=0 +m
2
c
4
+ 0 + 2mc
2

hc

hc

2
hc

hc

(1 cos ) = m
2
c
4
. (7)
Multiplying through by

/(2hmc
3
) gives the desired result,

= +
h
mc
(1 cos ). (8)
The ease of this solution arose from the fact that all the unknown garbage in P

m
disappeared when we squared it.
Remarks:
1. If 0 (that is, not much scattering), then

, as expected.
2. If = (that is, backward scattering) and additionally h/mc (that is, mc
2

hc/ = E

), then

2h/mc, so
E

=
hc

hc
2h
mc
=
1
2
mc
2
. (9)
Therefore, the photon bounces back with an essentially xed E

, independent of the
initial E

(as long as E

is large enough). This isnt all that obvious.

2
Solution
Week 70 (1/12/04)
Painting a funnel
It is true that the volume is nite, and that you can ll it up with paint. It is also
true that the surface area is innite, but you actually can paint it.
The apparent paradox arises from essentially comparing apples and oranges. In
our case we are comparing areas (things of dimension two) with volumes (things
of dimension three). When someone says that the funnel cant be painted, she is
saying that it would take an innite volume of paint to cover it. But the fact that
the surface area is innite does not imply that it takes an innite volume of paint
to cover it. To be sure, if we try to paint the funnel with a given xed thickness of
paint, then we would indeed need an innite volume of paint. But is this case we
would essentially have a tube of paint of xed radius, for large values of x, with the
funnel taking up a negligible volume at the center of the tube. This tube certainly
has an innite volume.
But what if we paint the funnel with a decreasing thickness of paint, as x gets
larger? For example, if we make the thickness go like 1/x, then the volume of paint
goes like

1
(1/x
2
) dx, which is nite. In this manner, we can indeed paint the
funnel. To sum up, we buy paint by the gallon, not by the square meter. And a
gallon of paint can cover an innite area, as long as we make the thickness go to
zero fast enough.
1
Solution
Week 71 (1/19/04)
Maximum trajectory length
Let be the angle at which the ball is thrown. Then the coordinates are given by
x = (v cos )t and y = (v sin)t gt
2
/2. The ball reaches its maximum height at
t = v sin/g, so the length of the trajectory is
L = 2

v sin /g
0

dx
dt

2
+

dx
dt

2
dt
= 2

v sin /g
0

(v cos )
2
+ (v sin gt)
2
dt
= 2v cos

v sin /g
0

1 +

tan
gt
v cos

2
dt. (1)
Letting z tan gt/v cos , we obtain
L =
2v
2
cos
2

0
tan

1 + z
2
dz. (2)
Letting z tan, and switching the order of integration, gives
L =
2v
2
cos
2

0
d
cos
3

. (3)
You can either look up this integral, or you can derive it (see the remark at the end
of the solution). The result is
L =
2v
2
cos
2

g

1
2

sin
cos
2

+ ln

1 + sin
cos

=
v
2
g

sin + cos
2
ln

sin + 1
cos

. (4)
As a double-check, you can verify that L = 0 when = 0, and L = v
2
/g when
= 90

. Taking the derivative of eq. (4) to nd the maximum, we obtain

0 = cos 2 cos sin ln

1 + sin
cos

+ cos
2

cos
1 + sin

cos
2
+ (1 + sin) sin
cos
2

.
(5)
This reduces to
1 = sin ln

1 + sin
cos

. (6)
Finally, you can show numerically that the solution for is
0
56.5

.
A few possible trajectories are shown below. Since it is well known that = 45

provides the maximum horizontal distance, it follows from the gure that the
0
yielding the arc of maximum length must satisfy
0
45

. The exact angle, however,

requires the above detailed calculation.
1
y
x
45
path
=
Remark: Lets now show that the integral in eq. (3) is given by

d
cos
3

=
1
2

sin
cos
2

+ ln

1 + sin
cos

. (7)
Letting c cos and s sin for convenience, and dropping the d in the integrals, we
have

1
c
3
=

c
c
4
=

c
(1 s
2
)
2
=
1
4

1
1 + s
+
1
1 s

2
=
1
4

c
(1 + s)
2
+
c
(1 s)
2

+
1
2

c
(1 s
2
)
=
1
4

1
1 + s
+
1
1 s

+
1
4

c
1 + s
+
c
1 s

=
s
2(1 s
2
)
+
1
4

ln(1 + s) ln(1 s)

=
s
2c
2
+
1
4
ln

1 + s
1 s

=
1
2

s
c
2
+ ln

1 + s
c

, (8)
as we wanted to show.
2
Solution
Week 72 (1/26/04)
Find the foci
Ellipse: Let us rst nd the center of the ellipse. In the gure below, draw two
arbitrary parallel lines that each meet the ellipse at two points. Call these points
A
1
, A
2
on one line, and B
1
, B
2
on the other. Bisect segments A
1
A
2
and B
1
B
2
to
yield points A and B. Now repeat this construction with two other parallel lines to
give two new bisection points C and D.
A
A
C
B
D 1
C
1
C
2
D
1
D
2
A
2
B
1
B
2
Claim 1 The intersection of segments AB and CD is the center of the ellipse.
Proof: Given a circle, the segment joining the midpoints of two parallel chords
passes through the center of the circle. An ellipse is simply a stretched circle, and in
this stretching process, all midpoints of segments remain midpoints. (If this doesnt
satisfy you, well give an analytical proof when we get to the hyperbola case.)
Having found the center, we can now nd the major and minor axes by drawing
a circle, with its center at the center of the ellipse, which meets the ellipse at four
points (the vertices of a rectangle). The axes of the ellipse are the lines parallel to
the sides of this rectangle, through the center of the ellipse.
Having found the axes, the foci are the points on the major axis that are one-half
of a major-axis distance from the endpoints of the minor axis.
Parabola: Let us rst nd the axis of the parabola. Draw two arbitrary parallel
lines that each meet the parabola at two points. Call these points A
1
, A
2
on one
line, and B
1
, B
2
on the other. Bisect segments A
1
A
2
and B
1
B
2
to yield points A
and B.
1
A
A
B
1
A
2
B
1
B
2
Claim 2 Segment AB is parallel to the axis of the parabola.
Proof: This follows from the reasoning in the ellipse case, along with the fact that
a parabola is simply an ellipse with its center at innity. (Again, if this doesnt
satisfy you, well give an analytical proof when we get to the hyperbola case.)
To obtain the axis, draw a line perpendicular to AB, which meets the parabola
at points C and D. The perpendicular bisector of CD is the axis of the parabola.
Having found the axis, the focus may be found as follows. Call the axis the y-axis
of a coordinate system, with the parabola opening up in the positive y-direction.
Let the vertex of the parabola be at (0, 0), and let the focus be at (0, a). Then a
horizontal line through the focus must meet the parabola at points (2a, a), because
the absolute value of the x-coordinate (that is, the distance from these points to
the focus) must equal the distance from these points to the directrix (which is the
horizontal line located a distance a below the vertex), which equals 2a. This also
follows from writing the parabola in the form x
2
= 4ay, where a is the focal distance.
The focus of the parabola may therefore be found by drawing lines through the
vertex, with slopes 1/2 and 1/2. These two lines meet the parabola at points E
and F. The intersection of segment EF with the axis is the focus.
Hyperbola: Let is rst nd the center of the hyperbola. The same construction
works here as did for the ellipse, but we will now present an analytical proof.
Claim 3 The center of a conic section is the intersection of two lines, each con-
taining the midpoints of two parallel chords of the conic section.
Proof: Let the conic section be written as
rx
2
+ sy
2
= 1. (1)
This describes an ellipse if r and s are positive, and a hyperbola if r and s have
opposite sign. A parabola is obtained in the limit r/s 0, . Consider a line of
the form
y = ax + b. (2)
If this line meets the conic section in two points, you can show that the midpoint of
the resulting chord has coordinates

sab
r + sa
2
,
rb
r + sa
2

. (3)
2
Note that when solving the quadratic equation for the intersection of the line and
the conic section, we can ignore the discriminant in the quadratic formula, because
we are concerned only with the midpoint between the intersections. This simplies
things greatly.
The slope of the line joining this point to the center of the conic section (which
is the origin) equals r/(sa). This is independent of the constant b, so another
parallel chord (that is, another chord with the same a but a dierent b) will also
have its midpoint lying on this same line through the origin.
Having found the center, we can now nd the axes by drawing a circle, with its
center at the center of the hyperbola, which meets the hyperbola at four points (the
vertices of a rectangle). The axes of the hyperbola are the lines parallel to the sides
of this rectangle, through the center of the hyperbola.
Let us now, for convenience, assume that the hyperbola is written in the form
x
2
m
2

y
2
n
2
= 1. (4)
Then the focal length is well known to be c =

m
2
+ n
2
. We have already found
m, which is the distance from the center to an intersection of the major axis with
the hyperbola. So we simply need to nd n, which may be found by noting that
the point (

2m, n) is on the hyperbola. We may therefore construct the foci as

follows. Knowing the length m, we can construct the length

2m, and then the
point (

2m, 0). We can then draw a vertical line to obtain the point (

2m, n).
Then we can draw a horizontal line to obtain the point (m, n). This yields the
diagonal distance

m
2
+ n
2
. The foci are the points (

m
2
+ n
2
, 0).
3
Solution
Week 73 (2/2/04)
Chain on a scale
First solution: Let y be the height of the top of the chain, and let F be the
desired force applied by the scale. The net force on the whole chain is F (L)g
(with upward taken to be positive). The momentum of the chain is (y) y. Note
that this is negative, since y is negative. Equating the net force with the change in
momentum gives
F Lg =
d(y y)
dt
= y y + y
2
. (1)
The part of the chain that is still above the scale is in free-fall. Therefore, y = g.
And y =

2g(L y), which is the usual result for a falling object. Putting these
into eq. (1) gives
F = Lg yg + 2(L y)g
= 3(L y)g. (2)
This answer has the expected property of equaling zero when y = L, and also the
interesting property of equaling 3(L)g right before the last bit touches the scale.
Once the chain is completely on the scale, the reading will suddenly drop down to
the weight of the chain, namely (L)g.
Second solution: The normal force from the scale is responsible for doing two
things. It holds up the part of the chain that already lies on the scale, and it also
changes the momentum of the atoms that are suddenly brought to rest when they
hit the scale. The rst of these two parts of the force is simply the weight of the
chain already on the scale, which is F
weight
= (L y)g.
To nd the second part of the force, we need to nd the change in momentum,
dp, of the part of the chain that hits the scale during a given time, dt. The amount
of mass that hits the scale in a time dt is dm = |dy| = | y| dt = y dt. This
mass initially has velocity y, and then it is abruptly brought to rest. Therefore, the
change in momentum is dp = 0 (dm) y = y
2
dt. The force required to cause this
change in momentum is thus
F
dp/dt
=
dp
dt
= y
2
. (3)
But as in the rst solution, we have y =

2g(L y). Therefore, the total force
from the scale is
F = F
weight
+ F
dp/dt
= (L y)g + 2(L y)g
= 3(L y)g. (4)
1
Solution
Week 74 (2/9/04)
Comparing the numbers
(a) Let your number be n. We will average over the equally likely values of n
(excluding n = 1) at the end of the calculation.
For convenience, let p (n 1)/(N 1) be the probability that a person you
ask has a number smaller than yours. And let 1p (N n)/(N 1) be the
probability that a person you ask has a number larger than yours.
Let A
n
be the average number of people you have to ask in order to nd a
number smaller than yours, given that you have the number n. A
n
may be
calculated as follows.
There is a probability p that it takes only one check to nd a smaller number.
There is a probability 1 p that the rst person you ask has a larger number.
From this point on, you have to ask (by denition) an average of A
n
people
in order to nd a smaller number. In this scenario, you end up asking a total
of A
n
+ 1 people.
Therefore, it must be true that
A
n
= p 1 + (1 p)(A
n
+ 1). (1)
This gives
A
n
=
1
p
=
N 1
n 1
. (2)
All values of n, from 2 to N, are equally likely, so we simply need to nd the
average of the numbers A
n
, for n ranging from 2 to N. This average is
A =
1
N 1
N

n=2
N 1
n 1
= 1 +
1
2
+
1
3
+ +
1
N 1
. (3)
This expression for A is the exact answer to the problem. To obtain an ap-
proximate answer, note that the sum of the reciprocals of the numbers from 1
to M equals lnM +, where 0.577... is Eulers constant. So if N is large,
you have to check about lnN + other numbers before you nd one that is
smaller than yours.
(b) Let your number be n. As in part (a), we will average over the equally likely
values of n (excluding n = 1) at the end of the calculation.
Let B
N
n
be the average number of people you have to ask in order to nd a
smaller number, given that you have the number n among the N numbers.
B
N
n
may be calculated as follows.
There is a probability (n 1)/(N 1) that it takes only one check to nd a
smaller number.
1
There is a probability (N n)/(N 1) that the rst person you ask has a
larger number. From this point on, you have to ask (by denition) an average
of B
N1
n
people in order to nd a smaller number. In this scenario, you end
up asking a total of B
N1
n
+ 1 people.
Therefore, it must be true that
B
N
n
=
n 1
N 1
1 +
N n
N 1

B
N1
n
+ 1

= 1 +

N n
N 1

B
N1
n
. (4)
Using the fact that B
N
n
= 1 when N = n, we can use eq. (4) to inductively
increase N (while holding n constant) to obtain B
N
n
for N > n. If you work
out a few cases, you will quickly see that B
N
n
= N/n. We can then easily
check this by induction on N; it is true for N = n, so we simply need to verify
in eq. (4)) that
N
n
= 1 +

N n
N 1

N 1
n
, (5)
which is indeed true. Therefore,
B
N
n
=
N
n
. (6)
As in part (a), all values of n, from 2 to N, are equally likely, so we simply
need to nd the average of the numbers B
N
n
= N/n, for n ranging from 2 to
N. This average is
B =
1
N 1
N

n=2
N
n
=
N
N 1

1
2
+
1
3
+ +
1
N

. (7)
This expression for B is the exact answer to the problem. If N is large, then
the result is approximately equal to lnN + 1, due to the rst term of 1
missing in the parentheses. This result is one person fewer than the result in
part (a). So a good memory saves you, on average, one query.
Remark: The continuum version of this problem (in which case the quality of your
memory in irrelevant, to leading order) is the following.
Someone gives you a random number between 0 and 1, with a at distribution. Pick
successive random numbers until you nally obtain one that is smaller. How many
numbers, on average, will you have to pick?
This is simply the original problem, in the limit N . So the answer should be
ln(), which is innite. And indeed, from the reasoning in part (a), if you start with
the number x, then the average number of picks you need to make to nd a smaller
number is 1/x, from eq. (2). Averaging these waiting times of 1/x, over the equally
2
likely values of x, gives an average waiting time of
1

1
0
dx
x
= . (8)
We can get around this innite answer by changing the probability distribution on the
unit interval. For example, let the probability distribution be proportional to 1/

x.
Then the probability of someone having a number smaller than x is proportional to

x
0
dx/

x. Therefore, if you start with the number x, then the average number
of picks you need to make to nd a smaller number is proportional to 1/

x, from eq.
(2). Averaging these waiting times of 1/

x, over the equally likely values of x, gives

an average waiting time proportional to

1
0
1

x
dx = . (9)
In general, if the probability distribution is proportional to x
r
, then r = 0 (that is,
a at distribution) is the cuto case between having a nite or innite expectation
value for the number of necessary picks.
1
However, if you play this game a few times, you will quickly discover that the average number
of necessary picks is not innite. If you nd this unsettling, you are encouraged to look at Problem
of the Week 6.
3
Solution
Week 75 (2/16/04)
Hanging chain
Well present four solutions. The rst one involves balancing forces. The other three
involve various variations on a variational argument.
First solution: Let the chain be described by the function y(x), and let the
tension be described by the function T(x). Consider a small piece of the chain, with
endpoints at x and x + dx, as shown.

1
2
x+dx x
T(x)
T(x+dx)
Let the tension at x pull downward at an angle
1
with respect to the horizontal, and
let the tension at x + dx pull upward at an angle
2
with respect to the horizontal.
Balancing the horizontal and vertical forces on the small piece of chain gives
T(x + dx) cos
2
= T(x) cos
1
,
T(x + dx) sin
2
= T(x) sin
1
+
g dx
cos
1
, (1)
where is the mass per unit length. The second term on the right is the weight of
the small piece, because dx/ cos
1
(or dx/ cos
2
, which is essentially the same) is
its length. We must now somehow solve these two dierential equations for the two
unknown functions, y(x) and T(x). There are various ways to do this. Here is one
method, broken down into three steps.
First step: Squaring and adding eqs. (1) gives
(T(x + dx))
2
= (T(x))
2
+ 2T(x)g tan
1
dx +O(dx
2
). (2)
Writing T(x+dx) T(x)+T

(x) dx, and using tan

1
= dy/dx y

, we can simplify
eq. (2) to (neglecting second-order terms in dx)
T

= gy

. (3)
Therefore,
T = gy + c
1
, (4)
where c
1
is a constant of integration.
1
Second step: Lets see what we can extract from the rst equation in eqs. (1).
Using
cos
1
=
1
_
1 + (y

(x))
2
, and cos
2
=
1
_
1 + (y

(x + dx))
2
, (5)
and expanding things to rst order in dx, the rst of eqs. (1) becomes
T + T

dx
_
1 + (y

+ y

dx)
2
=
T
_
1 + y
2
. (6)
All of the functions here are evaluated at x, which we wont bother writing. Ex-
panding the rst square root gives (to rst order in dx)
T + T

dx
_
1 + y
2
_
1
y

dx
1 + y
2
_
=
T
_
1 + y
2
. (7)
To rst order in dx this yields
T

T
=
y

1 + y
2
. (8)
Integrating both sides gives
lnT + c
2
=
1
2
ln(1 + y
2
), (9)
where c
2
is a constant of integration. Exponentiating then gives
c
2
3
T
2
= 1 + y
2
, (10)
where c
3
e
c
2
.
Third step: We will now combine eq. (10) with eq. (4) to solve for y(x). Elim-
inating T gives c
2
3
(gy + c
1
)
2
= 1 + y
2
. We can rewrite this is the somewhat nicer
form,
1 + y
2
=
2
(y + h)
2
, (11)
where c
3
g, and h = c
1
/g. At this point we can cleverly guess (motivated by
the fact that 1 + sinh
2
z = cosh
2
z) that the solution for y is given by
y(x) + h =
1

cosh(x + a). (12)

Or, we can separate variables to obtain
dx =
dy
_

2
(y + h)
2
1
, (13)
and then use the fact that the integral of 1/

z
2
1 is cosh
1
z, to obtain the same
result.
The shape of the chain is therefore a hyperbolic cosine function. The constant h
isnt too important, because it simply depends on where we pick the y = 0 height.
Furthermore, we can eliminate the need for the constant a if we pick x = 0 to be
2
where the lowest point of the chain is (or where it would be, in the case where the
slope is always nonzero). In this case, using eq. (12), we see that y

(0) = 0 implies
a = 0, as desired. We then have (ignoring the constant h) the nice simple result,
y(x) =
1

cosh(x). (14)
Well show how to determine at the end of the solutions.
Second solution: We can also solve this problem by using a variational argument.
The chain will want to minimize its potential energy, so we want to nd the function
y(x) that minimizes the integral,
U =
_
(dm)gy =
_
_

_
1 + y
2
dx
_
gy = g
_
y
_
1 + y
2
dx, (15)
subject to the constraint that the length of the chain is some given length . That
is,
=
_
_
1 + y
2
dx. (16)
Without this constraint, we could nd y(x) by simply using the Euler-Lagrange
equation on the Lagrangian y
_
1 + y
2
given in eq. (15). But with the constraint,
we must use the method of Lagrange multipliers. This works for functionals in the
same way it works for functions. Basically, for any small variation in y(x) near the
minimum, we want the change in U to be proportional to the change in .
1
This
means that there exists a linear combination of U and that doesnt change, to rst
order in any small variation in y(x). In other words, the Lagrangian
2
L = y
_
1 + y
2
+ h
_
1 + y
2
= (y + h)
_
1 + y
2
(17)
satiss the Euler-Lagrange equation, for some value of h. Therefore,
d
dx
_
L
y

_
=
L
y
=
d
dx
_
(y + h)y

_
1 + y
2
_
=
_
1 + y
2
. (18)
We must now perform some straightforward (although tedious) dierentiations. Us-
ing the product rule on the left-hand side, and making copious use of the chain rule,
we obtain
y
2
_
1 + y
2
+
(y + h)y

_
1 + y
2

(y + h)y
2
y

(1 + y
2
)
3/2
=
_
1 + y
2
. (19)
Multiplying through by (1 + y
2
)
3/2
and simplifying gives
(y + h)y

= (1 + y
2
). (20)
1
The reason for this is the following. Assume that we have found the desired function y(x)
that minimizes U, and consider two dierent variations in y(x) that give the same change in ,
but dierent changes in U. Then the dierence in these variations will produce no change in ,
while yielding a nonzero rst-order change in U. This contradicts the fact that our y(x) yielded an
extremum of U.
2
Well use h for the Lagrange multiplier, to make the notation consistent with that in the rst
solution.
3
Having produced the Euler-Lagrange dierential equation, we must now integrate
it. If we multiply through by y

and rearrange, we obtain

y

1 + y
2
=
y

y + h
. (21)
Taking the dx integral of both sides gives (1/2) ln(1 + y
2
) = ln(y + h) + c
4
, where
c
4
is a constant of integration. Exponentiation then gives (with e
c
4
)
1 + y
2
=
2
(y + h)
2
. (22)
in agreement with eq. (11).
Third solution: Lets use a variational argument again, but now with y as the
independent variable. That is, let the chain be described by the function x(y). Then
the potential energy is
U =
_
(dm)gy =
_
_

_
1 + x
2
dy
_
gy = g
_
y
_
1 + x
2
dy. (23)
The constraint is
=
_
_
1 + x
2
dy. (24)
Using the method of Lagrange multipliers as in the second solution above, the
Lagrangian we want to consider is
L = y
_
1 + x
2
+ h
_
1 + x
2
= (y + h)
_
1 + x
2
. (25)
Our Euler-Lagrange equation is then
d
dy
_
L
x

_
=
L
x
=
d
dy
_
(y + h)x

1 + x
2
_
= 0. (26)
The zero on the right-hand side makes things nice and easy, because it means that
the quantity in parentheses is a constant. Calling this constant 1/ (to end up with
the notation in the second solution), we have (y + h)x

1 + x
2
. Therefore,
x

=
1
_

2
(y + h)
2
1
, (27)
which is equivalent to eq. (13).
Fourth solution: Note that our Lagrangian in the second solution above, which
is given in eq. (17) as
L = (y + h)
_
1 + y
2
, (28)
is independent of x. Therefore, in analogy with conservation of energy (which arises
from a Lagrangian that is independent of t), the quantity
E y

L
y

L =
y + h
_
1 + y
2
(29)
4
is independent of x. Call it 1/. Then we have reproduced eq. (11).
Remark: The constant can be determined from the locations of the endpoints and
the length of the chain. The position of the chain may be described by giving (1) the
horizontal distance, d, between the two endpoints, (2) the vertical distance, , between the
two endpoints, and (3) the length, , of the chain, as shown.
d

l
d-x -x x = 0
0 0
Note that it is not obvious what the horizontal distances between the ends and the minimum
point (which we have chosen as the x = 0 point) are. If = 0, then these distances are
simply d/2. But otherwise, they are not so clear.
If we let the left endpoint be located at x = x
0
, then the right endpoint is located at
x = d x
0
. We now have two unknowns, x
0
and . Our two conditions are
3
y(d x
0
) y(x
0
) = , (30)
along with the condition that the length equals , which takes the form (using eq. (14))
=
_
dx
0
x0
_
1 + y
2
dx
=
1

sinh(x)

dx
0
x
0
. (31)
Writing out eqs. (30) and (31) explicitly, using eq. (14), we have
cosh
_
(d x
0
)
_
cosh(x
0
) = , and
sinh
_
(d x
0
)
_
sinh(x
0
) = . (32)
If we take the dierence of the squares of these two equations, and use the hyperbolic
identities cosh
2
x sinh
2
x = 1 and coshxcoshy sinhxsinhy = cosh(x y), we obtain
2 2 cosh(d) =
2
(
2

2
). (33)
We can now numerically solve this equation for . Using a half-angle formula, you can
show that eq. (33) may also be written as
2 sinh(d/2) =
_

2
. (34)
We can check some limits here. If = 0 and = d (that is, the chain forms a horizontal
straight line), then eq. (34) becomes 2 sinh(d/2) = d. The solution to this is = 0,
which does indeed correspond to a horizontal straight line, because for small , eq. (14)
behaves like x
2
/2 (up to an additive constant), which varies slowly with x for small .
Another limit is where is much larger than both d and . In this case, eq. (34) becomes
2 sinh(d/2) . The solution to this is a very large , which corresponds to a droopy
chain, because eq. (14) varies rapidly with x for large .
3
Well take the right end to be higher than the left end, without loss of generality.
5
Solution
Week 76 (2/23/04)
Crawling ant
At time t, the movable end of the band is a distance (t) = L + V t from the wall.
Let the ants distance from the wall be x(t), and consider the fraction of the length
along the band, F(t) = x(t)/(t). The given question is equivalent to: For what
value of t does the fraction F(t) become zero (if at all)? To answer this, let us see
how F(t) changes with time.
After an innitesimal time, dt, the ants position, x, increases by (x/)V dt due
to the stretching, and decreases by udt due to the crawling. Therefore,
F(t + dt) =
x + (x/)V dt udt
+ V dt
=
x

udt
+ V dt
. (1)
To rst order in dt, this yields
F(t + dt) = F(t)
u

dt. (2)
In other words, F(t) decreases due to the fact that in a time dt the ant crawls a
distance udt relative to the band, which has a length (t). Eq. (2) gives
dF(t)
dt
=
u

. (3)
Using (t) = L + V t and integrating eq. (3), we obtain
F(t) = 1
u
V
ln

1 +
V
L
t

, (4)
where the constant of integration has been chosen to satisfy F(0) = 1. We now note
that for any positive values of u and V , we can make F(t) = 0 by choosing
t =
L
V

e
V/u
1

. (5)
For large V/u, the time it takes the ant to reach the wall becomes exponentially
large, but it does indeed reach it in a nite time. For small V/u, you can use
e

1 + to show that eq. (5) reduces to t L/u, as it should.

Remark: If u < V , then the ant will initially get carried away from the wall before it
eventually comes back and reaches it. What is the maximum distance it gets from the wall?
The ants distance from the wall is
x(t) = F(t)(t) =

1
u
V
ln

1 +
V
L
t

L + V t

. (6)
Setting the derivative of this equal to zero gives

1
u
V
ln

1 +
V
L
t

V u = 0. (7)
1
Note that we could have arrived at this by simply demanding that the speed of the ant be
zero, which means that the F(t)V speed due to the stretching cancels the u speed in the
other direction due to the crawling. The fraction F(t) is therefore simply u/V .
Solving eq. (7) for t gives
t
max
=
L
V

e
V/u1
1

. (8)
This holds only if V u. If V < u, then t
max
= 0 and x
max
= L. Plugging the t
max
from
eq. (8) into eq. (6) gives
x
max
=
u
V
L
e
e
V/u
(V u). (9)
If, for example, V = 2u, then x
max
(1.36)L. And if V = 10u, then x
max
(800)L.
Note that for large V/u, the t
max
in eq. (8) is approximately 1/e times the time it takes
the ant to reach the wall, given in eq. (5).
2
Solution
Week 77 (3/1/04)
Relativistic momentum paradox
The reasoning is not correct. The horizontal speed of the masses remains the same.
The system does not slow down in the x-direction. We can see that this must be the
case, by looking at the setup in the inertial frame that moves in the x-direction with
the v
x
that the masses have when they are still on the constraints (which happens
to be the same v
x
that they always have). In this moving frame, the masses feel
a force only in the y-direction when they come o the constraints, so they wont
move horizontally with respect to this frame. Therefore, they will always move with
constant v
x
in the lab frame.
Since v
x
is constant, the nal p
x
of the resulting blob is larger than the sum of
the initial p
x
s of the two masses, because the mass of the resulting blob is indeed
greater than the sum of the initial masses. Where does the extra momentum come
from? It comes from the string, which initially had some p
x
because it had stored
energy (because work was done to extend the string) and hence mass.
But where is the error in the reasoning stated in the problem? The rst sentence
is the one that is incorrect. In the lab frame, the force on each mass actually does
not point in the y-direction. The reason for this is the following.
It turns out that in relativity, a transverse force is decreased by a factor
when going from the particles frame to another frame. (Well just accept this fact
here.) More precisely, consider an object ying past you in some direction. In the
particles frame (S

), let the force components be F

in that direction, and F

in the
orthogonal direction. Then the force components that you measure on the particle
in the lab frame (S) are
F

= F

and F

=
F

. (1)
Lets see what this implies. After the masses have been drawn together by the
string a bit, they will be heading diagonally upward or downward, instead of directly
in the x-direction. Let the upper mass be traveling in the direction v shown below.
In the frame of the mass, the force F

points directly downward. But if we break this

force into the components along the motion and perpendicular to the motion, we see
that when transforming to the lab frame, the transverse component is decreased by
a factor , as shown. The force F in the lab frame therefore points slightly forward,
as shown. This forward component of the force is what increases the x-momentum
of the masses. The x-momentum of each mass does indeed increase, because p
x
takes the form p
x
= mv
x
, and increases due to the fact that the speed increases
because of the increasing v
y
.
1
F'
F'
=
F'
F'
v v
F
F F'
=
F
=
=
=
/

particle frame lab frame

2
Solution
Week 78 (3/8/04)
Innite square roots
Let
x =

17
16

17
16

1 . (1)
Then x =

17/16 x. Squaring a few times yields x

4
2x
2
+ x 1/16 = 0,
or equivalently,
(2x 1)(8x
3
+ 4x
2
14x + 1) = 0. (2)
Therefore, either x = 1/2, or x is a root of 8x
3
+ 4x
2
14x + 1 = 0. Solving this
cubic equation numerically, or ddling around with the values at a few points, shows
that it has three real roots. One is negative, between 1 and 2 ( 1.62). One
is slightly greater than 1 ( 1.05). And one is slightly greater than 1/14 ( .073).
The rst two of these cannot be the answer to the problem, because x must be
positive and less than one. So the only possibilities are x = 1/2 and x .073. A
double-check shows that neither was introduced in the squaring steps above; both
are correct solutions to x =

17/16 x. But x clearly has one denite value,

so which is it? A few iterations on a calculator, starting with a 1 under the
innermost radical, shows that x = 1/2 is denitely the answer. Why?
The answer lies in the behavior of f(x) =

17/16 x near the points

x = 1/2 and x = .07318 . . .. The point x = 1/2 is stable, in the sense that if
we start with an x value slightly dierent from 1/2, then f(x) will be closer to 1/2
than x was. The point x = , on the other hand, is unstable, in the sense that if
we start with an x value slightly dierent from , then f(x) will be farther from
than x was.
Said in another way, the slope of f(x) at x = 1/2 is less than 1 in absolute value
(it is in fact 2/3), while the slope at x = is greater than 1 in absolute value (it is
approximately 3.4). What this means is that points tend to head toward 1/2, but
away from , under iteration by f(x). In particular, if we start with the value 1, as
we are supposed to do in this problem, then we will eventually get arbitrarily close
to 1/2 after many applications of f. Therefore, 1/2 is the correct answer.
The above reasoning is perhaps most easily understood via the gure below. This
gure shows graphically what happens to two initial values of x, under iteration by
f. To nd what happens to a given point x
0
, draw a vertical line from x
0
to the curve
y = f(x); this gives f(x
0
). Then draw a horizontal line to the point (f(x
0
), f(x
0
))
on the line y = x. Then draw a vertical line to the curve y = f(x); this gives
f(f(x
0
)). Continue drawing these horizontal and vertical lines to obtain successive
iterations by f.
1
x
f (x)
f (x)
=
x
f (x)
=
.5 .25
.25
.5
.75
1
.75 1
1
-
17/16 - x
1/16
17/16
= .073
Using this graphical method, it is easy to see that if:
x
0
> 17/16, then we immediately get imaginary values.
< x
0
17/16, then iteration by f will lead to 1/2.
x
0
= exactly, then we stay at under iteration by f.
x
0
< , then we will eventually get imaginary values. If x
0
< 1/16, then
imaginary values will occur immediately, after one iteration; otherwise it will
take more than one iteration.
2
Solution
Week 79 (3/15/04)
Propelling a car
Let the speed of the car be v(t). Consider the time interval when a mass dm enters
the car. Conservation of momentum gives
(dm)u + mv = (m + dm)(v + dv)
= dm(u v) = mdv, (1)
where we have dropped the second-order dmdv term. Separating variables and
integrating gives

m
M
dm
m
=

v
0
dv
u v
= ln
(
m
M
)
= ln
(
u v
u
)
= m =
Mu
u v
. (2)
Note that m as v u, as it should.
How does m depend on time? Mass enters the car at a rate (uv)/u, because
although you throw the balls at speed u, the relative speed of the balls and the car
is only (u v). Therefore,
dm
dt
=
(u v)
u
. (3)
Substituting the m from eq. (2) into this equation gives

v
0
dv
(u v)
3
=

t
0
dt
Mu
2
=
1
2(u v)
2

1
2u
2
=
t
Mu
2
= v(t) = u

1
1

1 +
2t
M

. (4)
Note that v u as t , as it should. Integrating this speed to obtain the
position gives
x(t) = ut
Mu

1 +
2t
M
+
Mu

. (5)
We see that even though the speed approaches u, the car will eventually be an
arbitrarily large distance behind a ball with constant speed u. For example, pretend
that the rst ball missed the car and continued to travel forward at speed u.
1
Solution
Week 80 (3/22/04)
Nine divisible by 9
First, consider the following simpler problem:
Problem: Given any ve integers, show that there is at least one subset of three
integers whose sum is divisible by 3.
Solution: Let us try to nd a set of ve integers that contains no subset of three
integers whose sum is divisible by 3, a task that we will show is impossible. Each of
the ve integers is, for our purposes, equal to 0, 1, or 2, because we are concerned
only with divisions by 3. We cannot have one of each of these, because 0 + 1 + 2 is
divisible by 3. We must therefore have at most two of the types. But the pigeonhole
principle then implies that we have at least three of one of the types. The sum of
these three integers is divisible by 3.
Returning to the original problem, pick ve integers to obtain a triplet whose sum
is divisible by 3. Then pick another ve integers to obtain another such triplet. We
can continue to do this for a total of ve times, given the seventeen integers.
We now have ve triplets, each of whose sum is divisible by 3. As far as divisions
by 9 are concerned, these sums are equal to 0, 3, or 6. We can now use the same
reasoning as in our auxiliary problem above (but with everything scaled up by a
factor of 3) to show that we can nd a set of three triplets that has a sum divisible
by 9. In other words, we have found a set of nine integers whose sum is divisible by
9.
This result is a special case of the following theorem.
Theorem: Given any 2n 1 integers, there is at least one subset of n integers
whose sum is divisible by n.
We will prove this theorem by demonstrating two lemmas.
Lemma 1: If the theorem is true for integers n
1
and n
2
, then it is also true for
the product n
1
n
2
.
Proof: Consider a set of 2n
1
n
2
1 integers. Under the assumption that the theorem
is true for n
1
, we can certainly pick a subset of n
1
integers whose sum is divisible by
n
1
. From the remaining 2n
1
n
2
1 n
1
integers we can pick another such subset of
n
1
integers, and so on. We can continue to do this until we have obtained 2n
2
1
such subsets. This is true because after forming 2n
2
2 such subsets, there are
[2n
1
n
2
1] [(2n
2
2)n
1
] = 2n
1
1 integers left over, from which we can pick one
last such subset of n
1
integers.
Now consider these 2n
2
1 sums divided by n
1
. Assuming that the theorem
holds for n
2
, we can nd n
2
of these sums (divided by n
1
) that have a sum divisible
by n
2
. Bringing back in the factor of n
1
, we see that we have found a set of n
1
n
2
integers whose sum is divisible by n
1
n
2
.
1
In proving the general theorem, this rst lemma shows that it is sucient to prove
the theorem for primes, p:
Lemma 2: If p is prime, then given 2p 1 integers, there is at least one subset of
p integers whose sum is divisible by p.
Proof: Consider all the possible N

2p1
p

subsets of p integers. Label the sums

of these subsets as S
j
, where 1 j N (these may be indexed in an arbitrary
manner), and consider the sum
S =
N

j=1
S
(p1)
j
. (1)
Remark: The following proof isnt mine (Im not sure where it came from originally). At
rst glance, it might seem that adding up these (p 1)st powers is a little out of the blue,
but its actually a fairly reasonable thing to do. There are two types of sums: good ones
that are divisible by p, and bad ones that arent. It would be nice to label them all in a
sort of binary way, say, with a 0 for good, and a 1 for bad. Fermats Little Theorem
(which states that if a 0 (mod p) then a
p1
1 (mod p)) provides the perfect way for
doing this.
We will prove this second lemma by demonstrating two claims:
Claim 1: S is be divisible by p.
Proof: Let the 2p 1 integers be a
i
, where 1 i 2p 1. Expand all of the
S
(p1)
j
powers and collect all the like terms in S. The terms will have the form of
some coecient times a
b
1
i
1
a
b
2
i
2
a
b
k
i
k
. The number, k, of dierent a
i
s involved may
be any number from 1 to p 1, and the b
j
s must of course add up to (p 1).
We will now show that the coecient of an arbitrary a
b
1
i
1
a
b
2
i
2
a
b
k
i
k
term is di-
visible by p. The coecient will depend on the b
i
, but it will happen to always be
divisible by p. The coecient may be viewed as the product of two factors.
Firstly, there is a multinomial coecient from each S
(p1)
j
in which the given
a
b
1
i
1
a
b
2
i
2
a
b
k
i
k
occurs. This multinomial coecient is

p1
b
1
,b
2
,...,b
k

, but it will turn

out not to be important.
Secondly, we must count the number of dierent S
(p1)
j
in which the given
a
b
1
i
1
a
b
2
i
2
a
b
k
i
k
occurs. This number may be found as follows. We know that k
of the p integers in S
j
must be a
1
, a
2
, . . . , a
k
. The remaining p k integers
can be any subset of the other 2p 1 k integers. There are

2p1k
pk

such
subsets.
The coecient of the a
b
1
i
1
a
b
2
i
2
a
b
k
i
k
term in S is therefore

p1
b
1
,b
2
,...,b
k

2p1k
pk

. Writing
the second factor in this as

2p 1 k
p k

=
(2p 1 k)(2p 2 k) p
(p k)!
(2)
demonstrates that every coecient is divisible by p, independent of the values of
the b
i
. Therefore, S is divisible by p. Q.E.D.
2
Claim 2: If none of the S
j
are divisible by p, then S is not divisible by p.
Proof: Assume that none of the S
j
are divisible by p. Then by Fermats Little
Theorem (which states that if a 0 (mod p) then a
p1
1 (mod p)), we have
S

i=1
1

(mod p) N (mod p). (3)

We now note that
N

2p 1
p

=
(2p 1)(2p 2) (p + 1)
(p 1)!
, (4)
which is not divisible by p. Therefore, S is not divisible by p. Q.E.D.
These two claims show that at least one of the S
j
must be divisible by p.
Remark: For the case where n is a prime number, p, it is possible to say a bit more about
exactly how many of the S
j
are divisible by p. We claim that either 1, p + 1, 2p + 1, . . ., of
the S
j
are divisible by p. The reasoning is as follows.
Fermats Little Theorem implies that each S
j
that is not divisible by p contributes 1 to
S, while each S
j
that is divisible by p contributes 0 to S. Under the (incorrect) assumption
that none of the S
j
are divisible by p, eq. (3) states that S N (mod p). Using eq. (4),
and noting that
(2p 1)(2p 2) (p + 1) (p 1)! (mod p), (5)
we see that this assumption leads to the incorrect conclusion that S 1 (mod p). But since
we know from eq. (2) that S must actually be divisible by p, then either 1, p + 1, 2p + 1,
. . ., of the S
j
must be divisible by p, because each S
j
that is divisible by p will contribute 0,
instead of 1, to S. Thus, for example, given ve integers, there are either one, four, seven,
or ten subsets of three integers whose sum is divisible by 3.
It is easy to construct a case where only one of the S
j
is divisible by p. We may pick p1
of the 2p 1 integers to be congruent to each other modulo p, and then pick the remaining
p integers to also be congruent to each other (but not to the other p 1 integers) modulo p.
Then the subset consisting of these latter p integers is the only subset of p integers whose
sum is divisible by p.
3
Solution
Week 81 (3/29/04)
Rainbows
Rainbows exist due to the fact that raindrops scatter light preferentially in certain
directions. The eect of this focusing is to make the sky appear brighter in a
certain region, as we will see in detail below. This brightness is the rainbow that
you see. The colors of the rainbow are caused by the dierent indexes of refraction
of the dierent colors; more on this below.
Primary rainbow: The preferred direction of the scattered light depends on how
many internal reections the light ray undergoes in the raindrop. Lets rst consider
the case of one internal reection, as shown below. The light ray refracts into the
raindrop, then reects inside, and then refracts back out into the air. Let and
be dened as in the gure. Then the angles 2 and 2 are shown.
2-

What is as a function of ? Snells law at the refraction points gives

sin(2 ) =
4
3
sin. (1)
Solving for yields
= 2 arcsin

4
3
sin

. (2)
The plot of vs. looks like the following.
1
1
Note that cannot be greater than arcsin(3/4) 48.6

, because this would yield the acrsin of

a number larger than 1. 48.6

is the critical angle for the air/water interface.

1
30 15 45
15
30

We see that has a maximum of roughly

max
20

at
max
40

. To be more
precise, we can set d/d = 0. The result is
0 = 2
4
3
cos

1
16
9
sin
2

. (3)
Squaring and using cos
2
= 1 sin
2
yields
sin
max
=

5
12
=
max
40.2

. (4)
Substituting this back into eq. (2) gives

max
21.0

. (5)
The signicance of this maximum is not that it is the largest value of , but rather
that the slope of the () curve is zero, which means that there are many dierent
values of that yield essentially the same value ( 21

) of . The light therefore

2
gets focused into the total angle of
2
max
42

, (6)
so the sky appears brighter at this angle (relative to the line from the sun to you).
This brightness is the rainbow that you see. This reasoning is perhaps more clear if
we draw a diagram with the incoming angle of the rays constant (which is in fact the
case, because all of the suns rays are parallel). A rough diagram of this is shown
below. The bold line is the path with the maximum 2 angle of 42

. You can see

that the outgoing rays pile up at this angle.
to the sun
2
See the fourth remark below for elaboration on this reasoning.
2
Remarks:
1. The above reasoning explains why the rainbow is where it is. But why do we see the
dierent colors? The colors arise from the fact that dierent wavelengths of light have
dierent indexes of refraction at the air/water boundary. At one end of the visible
spectrum, violet light has an index of 1.344. And at the other end, red light has an
index of 1.332.
3
If you go through the above calculation, but now with these indexes
in place of the 4/3 we used above, you will nd that violet light appears at an angle
of 2
violet
max
40.5

, and red light appears at an angle of 2

red
max
42.2

. Since red
occurs at the larger angle, it is therefore the color at the top of the rainbow. Violet
is at the bottom, and intermediate wavelengths are in between. The fact that red is
at the top can be traced to the fact that 21

is the maximum value of .

2. A rainbow is actually a little wider than the approximately 2

spread we just found,

because the sun isnt a point source. It subtends an angle of about half a degree,
which adds half a degree to the rainbows spread. Also, the rainbows colors are
somewhat washed out on the scale of half a degree, so a rainbow isnt as crisp as it
would be if the sun were a point source.
3. In addition to the ordering of the colors, there is another consequence of the fact that
21

is the maximum value of . It is possible for a raindrop to scatter light at 2

values smaller than 42

(for one internal reection), but impossible for larger angles.

Therefore, the region in the sky below the rainbow appears brighter than the region
above it. Even though the focusing eect occurs only right at the rainbow, the simple
scattering of light through 2 values smaller than 42

makes the sky appear brighter

below the rainbow.
4. There is one slight subtlety in the above reasoning we should address, even though
turns out not to be important. We said above that if many dierent values corre-
spond to a certain value of , then there will be more light scattered into that value
of . However, the important thing is not how many values correspond to a certain
, but rather the cross section of light that corresponds to that . Since the light
hits the raindrop at an angle 2 with respect to the normal, the amount of light
that corresponds to a given d interval is decreased by a factor of cos(2 ). What
we did in the above solution was basically note that the interval d that corresponds
to a given interval d is
d =
d
d/d
, (7)
which diverges at
max
, because d/d = 0 there. But since we are in fact concerned
with the number of light rays that correspond to the interval d, we now note that
this number is proportional to
d cos(2 ) =
d
d/d
cos(2 ). (8)
But 2 = 90

at
max
, so this still diverges at
max
. Our answer therefore remains
the same.
Secondary rainbow: Now consider the secondary rainbow. This rainbow arises
from the fact that the light may undergo two reections inside the raindrop before
it refracts back out. This scenario is shown below. With dened as in the gure,
the 180

2 and 90

angles follow, and then the 3 90

angle follows.
3
The ends of the visible spectrum are somewhat nebulous, but I think these values are roughly
correct.
3

90- 390
390
180-2
Snells law at the refraction points gives
sin(3 90

+ ) =
4
3
sin. (9)
Solving for yields
= 90

3 + arcsin

4
3
sin

. (10)
Taking the derivative to nd the extremum (which is a minimum this time) gives
sin
min
=

65
128
=
min
45.4

. (11)
Substituting this back into eq. (10) gives

min
25.4

. (12)
Using the same reasoning as above, we see that the light gets focused into the total
angle of 2
max
51

, so the the sky appears brighter at this angle.

The fact that () has a minimum instead of a maximum means that the rainbow
is inverted, with violet now on top. More precisely, if you go through the above
calculation, but now with the red and violet indexes of refraction in place of the
4/3 used above, you will nd that violet light appears at an angle of 53.8

, and
red light appears at an angle of 50.6

. The spread here is a little larger than that

for the primary rainbow above.
The secondary rainbow is fainter than the primary one for three reasons. First,
the larger spread means that the light is distributed over a larger area. Second,
additional light is lost at the second internal reection. And third, the angle with
respect to the normal at which the light ray hits the raindrop is larger for the
secondary rainbow. For the primary rainbow it was 2 59

, while for the

secondary rainbow it is 3 90

+ 72

.
Tertiary rainbow: Now consider the tertiary rainbow. This rainbow arises from
three reections inside the raindrop, as shown below. With dened as in the
gure, the two 180

2 angles follow, and then the 4 180

angle follows.
4

4180
4180
180-2
180-2
Snells law at the refraction points gives
sin(4 180

+ ) =
4
3
sin. (13)
Solving for yields
= 180

4 + arcsin

4
3
sin

. (14)
Taking the derivative to nd the extremum (which is a minimum) gives
sin
min
=

8
15
=
min
46.9

. (15)
Substituting this back into eq. (10) gives

min
69.2

. (16)
Using the same reasoning as above, we see that the light gets focused into the total
angle of 2
max
138

, so the the sky appears brighter at this angle. The fact that
this angle is larger than 90

means that the rainbow is actually behind you (if you

are facing the primary and secondary rainbows), back towards the sun. It is a circle
around the sun, located at an angle of 180

138

= 42

relative to the line between

you and the sun.
This rainbow is much more dicult to see, in part because of the increased
eects of the three factors stated above in the secondary case, but also because you
are looking back toward the sun, which essentially drowns out the light from the
rainbow. Although I have never been able to seen this tertiary rainbow, it seems
quite possible under (highly improbable) ideal conditions, namely, having rain fall
generally all around you, except in a path between you and the sun, while at the
same time having the sun eclipsed by a properly sized cloud.
Nth-order rainbow: We can now see what happens in general, when the light
ray undergoes N reections inside the raindrop.
5

180-2
arcsin
(
sin
)
4
3
_
With dened as in the gure, the light gets deected clockwise by an angle of
1802 at each reection point. In addition, it gets deected clockwise by an angle
of arcsin(
4
3
sin) at the two refraction points. The total angle of deection is
therefore
= 2

arcsin

4
3
sin

+ N(180

2). (17)
Taking the derivative to nd the extremum gives
4
sin
0
=

9(N + 1)
2
16
16[(N + 1)
2
1]
. (18)
You can check that this agrees with the results in eqs. (4), (11), and (15). Substi-
tuting this back into eq. (17) gives the total angle of deection, , relative to the
direction from the sun to you. See the discussion below for an explanation on how
relates to the angle we used above. A few values of are given in the following
table.
N (mod 360

)
1 138

138

2 231

231

3 318

318

4 404

44

5 488

128

6 572

212

If you take the second derivative of , you will nd that it is always positive.
Therefore, the extremum is always a minimum. It takes a little thought, though, to
deduce the ordering of the colors from this fact. Since the red end of the spectrum
has the smallest index of refraction, it is bent the least at the refraction points.
Therefore, it has a smaller total angle of deection than the other colors. The red
light will therefore be located (approximately) at the angles given in the above table,
while the other colors will occur at larger angles. The locations of the focusing angles
4
Note that as N , we have sin 0 3/4. Therefore, approaches the critical angle, 48.6

,
which means that the light ray hits the raindrop at nearly 90

with respect to the normal. The

cross section of the relevant light rays is therefore very small, which contributes to the faintness of
the higher-order rainbows.
6
are shown below, where the arrows indicate the direction of the other colors relative
to the red. All the arrows point clockwise, because all the extrema are minima.
1
2
3
4
6
5
to the sun light ray
raindrop
138
Since the direction you see the light coming from is the opposite of the direction
in which the light travels, you see the light at the following angles. This diagram is
obtained by simply rotating the previous one by 180

.
1
2
3
4
6
5
to the sun
you
42
We now note that in the gure preceding eq. (17), we could have had the light
ray hitting the bottom part of the raindrop instead of the top, in which case the
word clockwise would have been replaced with counterclockwise. This would
have the eect of turning into . The light is, of course, reected in a whole
circle (as long as the ground doesnt get in the way), and this circle is the rainbow
that you see; the angles and simply represent the intersection of the circle
with a vertical plane. We can therefore label each rainbow with an angle between 0
and 180

. This simply means taking the dots and arrows below the horizontal line
and reecting them in the horizontal line. The result is the following gure, which
gives the locations and orientations (red on one end, violet on the other) of the rst
six rainbows.
7
1
2
3
4
6
5
to the sun
you
42
51
52
138
136
32
Note that in the N limit, the deection angle in eq. (17) takes a simple form.
In this limit, eq. (18) gives 48.6

. Therefore, arcsin(
4
3
sin) 90

, and eq. (17)

gives
2 (90

48.6

) + N(180

2 48.6

)
= (N + 1)(180

2 48.6

)
= (N + 1)82.8

. (19)
8
Solution
Week 82 (4/5/04)
Standing in a line
First solution: Let T
N
be the expected number of people who are able to make
the given statement. If we consider everyone except the last person in line, then
this group of N 1 people has by denition T
N1
people who are able to make the
statement. Let us now add on the last person. There is a 1/N chance that she is
the tallest, in which case she is able to make the statement. Otherwise she cannot.
Therefore, we have
T
N
= T
N1
+
1
N
. (1)
Starting with T
1
= 1, we therefore inductively nd
T
N
= 1 +
1
2
+
1
3
+ +
1
N
. (2)
For large N, this goes like lnN, which grows very slowly with N.
Second solution: Let T
N
be the desired average. Consider the location of the
tallest person. If he is the last person in the line (which occurs with probability
1/N), then the problem reduces to that for the N 1 people in front of him. So in
this case, we can expect 1+T
N1
people who are able to make the given statement.
If the tallest person is the second to last person in the line (which occurs with
probability 1/N), then the problem reduces to that for the N 2 people in front of
him (because the person behind him is not able to make the statement). So in this
case, we can expect 1 + T
N2
people who are able to make the given statement.
Continuing along these lines, and adding up all N possibilities for the location
of the tallest person, we nd
T
N
=
1
N

(1 + T
N1
) + (1 + T
N2
) + + (1 + T
1
) + (1 + T
0
)

= NT
N
= N + T
N1
+ T
N2
+ + T
1
. (3)
Writing down the analogous equation for N 1,
(N 1)T
N1
= (N 1) + T
N2
+ T
N2
+ + T
1
, (4)
and then subtracting this from eq. (3), yields
T
N
= T
N1
+
1
N
, (5)
which agrees with the rst solution.
1
Solution
Week 83 (4/12/04)
The brachistochrone
First solution: In the gure below, the boundary conditions are y(0) = 0 and
y(x
0
) = y
0
, with downward taken to be the positive y direction.
x
x
y
y
0 0
, ( )
From conservation of energy, the speed as a function of y is v =

2gy. The total
time is therefore
T =

x
0
0
ds
v
=

x
0
0

1 + y
2

2gy
dx. (1)
Our goal is to nd the function y(x) that minimizes this integral, subject to the
boundary conditions above. We can therefore apply the results of the variational
technique, with a Lagrangian equal to
L

1 + y
2

y
. (2)
The Euler-Lagrange equation is
d
dx

L
y

=
L
y
=
d
dx

y
y

1 + y
2

1 + y
2
2y

y
. (3)
Using the product rule on the three factors on the left-hand side, and making copious
use of the chain rule, we obtain

y
2
2y

1 + y
2
+
y

1 + y
2

y
2
y

y(1 + y
2
)
3/2
=

1 + y
2
2y

y
. (4)
Multiplying through by 2y

y(1 + y
2
)
3/2
and simplifying gives
2yy

= 1 + y
2
. (5)
We can integrate this equation if we multiply through by y

and rearrange to obtain

2y

1 + y
2
=

y
. = ln(1 + y
2
) = lny + A
= 1 + y
2
=
B
y
, (6)
1
where B e
A
. We must now integrate one more time. Solving for y

and separating
variables gives

y dy

B y
= dx. (7)
A helpful change of variables to get rid of the square root in the denominator is
y B sin
2
. Then dy = 2B sincos d, and eq. (7) simplies to
2B sin
2
d = dx. (8)
We can now make use of the relation sin
2
= (1 cos 2)/2 to integrate this. The
result is B(2 sin2) = 2x C, where C is an integration constant.
Now note that we may rewrite our denition of (which was y B sin
2
) as
2y = B(1 cos 2). If we then dene 2, we have
x = a( sin) d, y = a(1 cos ). (9)
where a B/2, and d C/2.
The particle starts at (x, y) = (0, 0). Therefore, starts at = 0, since this
corresponds to y = 0. The starting condition x = 0 then implies that d = 0. Also,
we are assuming that the wire heads down to the right, so we choose the positive
sing in the expression for x. Therefore, we nally have
x = a( sin), y = a(1 cos ). (10)
This is the parametrization of a cycloid, which is the path taken by a point on the
rim of a rolling wheel. The initial slope of the y(x) curve is innite, as you can
check.
Second solution: Lets use a variational argument again, but now with y as the
independent variable. That is, let the chain be described by the function x(y). The
arclength is now given by ds =

1 + x
2
dy. Therefore, instead of the Lagrangian
in eq. (2), we now have
L

1 + x
2

y
. (11)
The Euler-Lagrange equation is
d
dy

L
x

=
L
x
=
d
dy

y
x

1 + x
2

= 0. (12)
The zero on the right-hand side makes things nice and easy, because it means that
the quantity in parentheses is a constant. Call it D. We then have
1

y
x

1 + x
2
= D =
1

y
dx/dy

1 + (dx/dy)
2
= D
=
1

y
1

(dy/dx)
2
+ 1
= D. (13)
This is equivalent to eq. (6), and the solution proceeds as above.
2
Third solution: Note that the Lagrangian in the rst solution above, which is
given in eq. (2) as
L =

1 + y
2

y
, (14)
is independent of x. Therefore, in analogy with conservation of energy (which arises
from a Lagrangian that is independent of t), the quantity
E y

L
y

L =
y
2

1 + y
2

1 + y
2

y
=
1

1 + y
2
(15)
is a constant. We have therefore again reproduced eq. (6).
3
Solution
Week 84 (4/19/04)
Poisson and Gaussian
(a) Consider a given box. The probability that exactly x balls end up in it is
P(x) =
_
N
x
_
_
1
B
_
x
_
1
1
B
_
Nx
. (1)
This is true because the probability that a certain set of x balls ends up in the
given box is (1/B)
x
, and the probability that the other N x balls do not end
up in the box is (1 1/B)
Nx
, and there are
_
N
x
_
ways to pick this certain set
of x balls.
Let us now make approximations to P(x). If N and B are much larger than
x, then N!/(N x)! N
x
, and (1 1/B)
x
1 (well be more precise about
these approximations below). Therefore,
P(x) =
N!
(N x)!x!
_
1
B
_
x
_
1
1
B
_
Nx

N
x
x!
_
1
B
_
x
_
1
1
B
_
N

1
x!
_
N
B
_
x
e
N/B

a
x
e
a
x!
. (2)
This result is called the Poisson distribution. For what x is P(x) maximum?
If we set P(x) = P(x +1), we nd x = a 1. Therefore, we are most likely to
obtain a 1 or a balls in a box.
We can also consider eq. (2) to be a function of non-integer values of x, by
using Stirlings formula, x! x
x
e
x

2x. This is valid for large x, which is

generally the case we will be concerned with. (But note that eq. (2) is valid
for small x, too.) Allowing non-integer values of x, the maximum P(x) occurs
halfway between a 1 and a, that is, at x = a 1/2. You can also show this
by taking the derivative of eq. (2), with Stirlings expression in place of the
x!. Furthermore, you can show that x = a 1/2 leads to a maximum P(x)
value of P
max
1/

2a.
In the real world, x can take on only integer values, of course. So it should be
the case that the sum of the P(x) probabilities, from x = 0 to x = , equals
1. And indeed,

x=0
P(x) =

x=0
a
x
e
a
x!
= e
a

x=0
a
x
x!
= e
a
e
a
= 1. (3)
1
Lets now be precise about the approximations we made above.
N!/(N x)! N
x
, because
N!
(N x)!
= (1)
_
1
1
N
__
1
2
N
_

_
1
x 1
N
_
N
x
_
1
x
2
2N
_
. (4)
So we need x

N for this approximation to be valid.

(1 1/B)
x
1, because (1 1/B)
x
1 x/B. So we need x B for
this approximation to be valid.
In going from the second to the third line in eq. (2), we used the
approximation (1 1/B)
N
e
N/B
. A more accurate statement is
(1 1/B)
N
e
N/B
e
N/2B
2
, which you can verify by taking the log
of both sides. So we need N B
2
for this approximation to be valid.
In practice, the basic requirement is N B
2
. Given this, the x
max
a N/B
value that produces the maximum value of P(x) does satisfy both x
max
B
and x
max

N. So the approximations are valid in the region near x

max
.
These are generally the x values we are concerned with, because if x diers
much from x
max
, then P(x) is essentially zero anyway. This will be clear after
looking at the Gaussian expression that well derive in part (b).
(b) In showing that a Poisson distribution can be approximated by a Gaussian
distribution, it will be easier to work with the log of P(x). Let y xa. The
relevant y in this problem will turn out to be small compared to a, so we will
eventually expand things in terms of the small quantity y/a. Using Stirlings
formula, x! x
x
e
x

2x (for large x, as we are assuming here), and also

using the expansion ln(1 + )
2
/2 +
3
/3 , we have
lnP(x) = ln
_
a
x
e
a
x!
_
ln
_
a
x
e
a
x
x
e
x

2x
_
= xlna a xlnx + x ln

2x
= xln(a/x) + (x a) ln

2x
= (a + y) ln
_
1 +
y
a
_
+ y ln
_
2(a + y)
= (a + y)
_
y
a

y
2
2a
2
+
y
3
3a
3
+
_
+ y ln
_
2(a + y)

y
2
2a
+
y
3
6a
2
ln
_
2(a + y). (5)
The y
3
/6a
2
term is much smaller than the y
2
/2a term (assuming y a), so we
may ignore it. Also, we may set 2(a+y) 2a, with negligible multiplicative
2
error. Therefore, exponentiating bother sides of eq. (5) gives
P(x)
e
(xa)
2
/2a

2a
, (6)
which is the desired Gaussian distribution. Note that the maximum occurs at
x = a. If you want to be a little more accuarate, you can include the correction
from the ln
_
2(a + y) = ln

2a + ln
_
1 + (y/a) term in eq. (5), to show
that the Gaussian is actually centered at x = a 1/2, that is,
P(x)
e
(x(a1/2))
2
/2a

2a
. (7)
The location of the maximum now agrees with the x
max
= a1/2 result that
we obtained in part (a). However, for large a, the distinction between a and
a 1/2 is fairly irrelevant.
Note that the spread of the Gaussian is of the order

a. In our particular
case, if we let x (a 1/2) =

2a, then P(x) is decreased by a factor of

e

2
relative to the maximum. = 1 gives about 37% of the maximum, = 2
gives about 2%, and = 3 gives about .01%, which is quite negligible. If,
for example, a = 1000, then virtually all of the non-negligible part of the
graph is contained in the region 900 < x < 1100, as shown below. The solid
curve in these plots is the Poisson distribution from eq. (2), and the dotted
curve is the Gaussian distribution from eq. (7). The two curves are essentially
indistinguishable in the a = 1000 case.
5 10 15 20
x
0.02
0.04
0.06
0.08
0.1
0.12
P
80 100 120 140
x
0.01
0.02
0.03
0.04
P
900 1000 1100 1200
x
0.002
0.004
0.006
0.008
0.01
0.012
P
a = 10 a = 100
a = 1000
3
We should check that the Gaussian probability distribution in eq. (7) has an
integral equal to 1. In doing this, we can let the integral run from to
with negligible error. Using the general result,
_

e
y
2
/b
dy =

b, and
letting b 2a, we see that the integral is indeed equal to 1.
4
Solution
Week 85 (4/26/04)
Tower of cylinders
Both cylinders in a given row move in the same manner, so we may simply treat
them as one cylinder with mass m = 2M. Let the forces that the boards exert on
the cylinders be labelled as shown. F is the force from the plank below a given
cylinder, and G is the force from the plank above it.
F
G
F
G
n
a
n

n
n
n +1
n +1
Note that by Newtons third law, we have F
n+1
= G
n
, because the planks are
massless.
Our strategy will be to solve for the linear and angular accelerations of each
cylinder in terms of the accelerations of the cylinder below it. Since we want to solve
for two quantities, we will need to produce two equations relating the accelerations
of two successive cylinders. One equation will come from a combination of F = ma,
= I, and Newtons third law. The other will come from the nonslipping condition.
With the positive directions for a and dened as in the gure, F = ma on the
nth cylinder gives
F
n
G
n
= ma
n
, (1)
and = I on the nth cylinder gives
(F
n
+G
n
)R =
1
2
mR
2

n
= F
n
+G
n
=
1
2
mR
n
(2)
Solving the previous two equations for F
n
and G
n
gives
F
n
=
1
2

ma
n
+
1
2
mR
n

,
G
n
=
1
2

ma
n
+
1
2
mR
n

. (3)
But we know that F
n+1
= G
n
. Therefore,
a
n+1
+
1
2
R
n+1
= a
n
+
1
2
R
n
. (4)
We will now use the fact that the cylinders dont slip with respect to the boards.
The acceleration of the board above the nth cylinder is a
n
R
n
. But the accel-
eration of this same board, viewed as the board below the (n + 1)st cylinder, is
a
n+1
+R
n+1
. Therefore,
a
n+1
+R
n+1
= a
n
R
n
. (5)
1
Eqs. (4) and (5) are a system of two equations in the two unknowns, a
n+1
and
n+1
,
in terms of a
n
and
n
. Solving for a
n+1
and
n+1
gives
a
n+1
= 3a
n
+ 2R
n
,
R
n+1
= 4a
n
3R
n
. (6)
We can write this in matrix form as

a
n+1
R
n+1

3 2
4 3

a
n
R
n

. (7)
We therefore have

a
n
R
n

3 2
4 3

n1

a
1
R
1

. (8)
Consider now the eigenvectors and eigenvalues of the above matrix. The eigenvectors
are found via

3 2
4 3

= 0 =

= 3 2

2. (9)
The eigenvectors are then
V
+
=

, for
+
= 3 + 2

2,
V

, for

= 3 2

2. (10)
Note that |

| > 1, so
n

as n . This means that if the initial (a

1
, R
1
)
vector has any component in the V

direction, then the (a

n
, R
n
) vectors will head
to innity. This violates conservation of energy. Therefore, the (a
1
, R
1
) vector
must be proportional to V
+
.
1
That is, R
1
=

2a
1
. Combining this with the fact
that the given acceleration, a, of the bottom board equals a
1
+R
1
, we obtain
a = a
1
+

2a
1
= a
1
=
a

2 + 1
= (

2 1)a. (11)
Remark: Let us consider the general case where the cylinders have a moment of inertia
of the form I = MR
2
. Using the above arguments, you can show that eq. (7) becomes

a
n+1
R
n+1

=
1
1

(1 +) 2
2 (1 +)

a
n
R
n

. (12)
And you can show that the eigenvectors and eigenvalues are
V
+
=

, for
+
=

+ 1
,
V

, for

+ 1

1
. (13)
1
This then means that the (an, Rn) vectors head to zero as n , because |+| < 1. Also,
note that the accelerations change sign from one level to the next, because
+
is negative.
2
As above, we cannot have the exponentially growing solution, so we must have only the V
+
solution. We therefore have R
1
= a
1
/

. Combining this with the fact that the given

acceleration, a, of the bottom board equals a
1
+R
1
, we obtain
a = a
1
+
a
1

= a
1
=

1 +

a. (14)
You can verify that all of these results agree with the = 1/2 results obtained above.
Lets now consider a few special cases of the

+
=

+ 1
(15)
eigenvalue, which gives the ratio of the accelerations in any level to the ones in the next
level down.
If = 0 (all the mass of a cylinder is located at the center), then we have
+
= 1.
In other words, the accelerations have the same magnitudes but dierent signs from
one level to the next. The cylinders simply spin in place while their centers remain
xed. The centers are indeed xed, because a
1
= 0, from eq. (14).
If = 1 (all the mass of a cylinder is located on the rim), then we have
+
= 0. In
other words, there is no motion above the rst level. The lowest cylinder basically
rolls on the bottom side of the (stationary) plank right above it. Its acceleration is
a
1
= a/2, from eq. (14).
If (the cylinders have long massive extensions that extend far out beyond the
rim), then we have
+
= 1. In other words, all the levels have equal accelerations. This
fact, combined with the R
1
= a
1
/

0 result, shows that there is no rotational

motion at any level, and the whole system simply moves to the right as a rigid object
with acceleration a
1
= a, from eq. (14).
3
Solution
Week 86 (5/3/04)
Shifted intervals
Because is very small, let us discretize each of the intervals into tiny units of length
. If the rst number is in the smallest of its possible -units (that is, between 0 and
), then it is guaranteed to be the smallest of all the numbers. If it is in the second
smallest -unit (between and 2), there is a 1 chance that it is the smallest
of all the numbers, because this is the probability that the second number is larger
than it.
1
If it is in the third -unit, there is a (1 )(1 2) chance that it is the
smallest, because this is the probability that both the second and third numbers are
larger than it. In general, if the rst number is in the kth -unit, there is a
P
k
= (1 )(1 2)(1 3) (1 (k 1)) (1)
chance that it is the smallest. Since the rst number has an equal probability of
of being in any of the -units, the total probability that it is the smallest therefore
equals
P = + P
1
+ P
2
+ P
3
+ + P
1/
. (2)
For small , we can make an approximation to the P
k
s, as follows. Take the log of
P
k
in eq. (1) to obtain
lnP
k
= ln(1 ) + ln(1 2) + ln(1 3) + + ln(1 (k 1))

_


2
2
_
+
_
2
2
2

2
2
_
+
_
(k 1)
(k 1)
2

2
2
_
=
_
1 + 2 + + (k 1)
_

2
2
_
1 + 2
2
+ + (k 1)
2
_
=
_
k(k 1)
2
_

2
2
_
k(k 1)(2k 1)
6
_

k
2
2


2
k
3
6
. (3)
In going from the rst to the second line, we have used the rst two terms in the
Taylor series, ln(1 x) = xx
2
/2 . And in going from the fourth to the fth
line, we have used the fact that the k values we will be concerned with will generally
be large, so we have kept only the leading power of k. Exponentiating eq. (3) gives
P
k
e
k
2
/2
e

2
k
3
/6
. (4)
The second factor here is essentially equal to 1 if
2
k
3
1, that is, if k 1/
2/3
.
But we are only concerned with k values up to the order of 1/
1/2
, because if k is
much larger than this, the rst exponential factor in eq. (4) makes P
k
essentially
1
Technically, the probability is on average equal to 1/2, because the average value of the rst
number in this case is 3/2. But the /2 correction in this probability (and other analogous ones)
is inconsequential.
1
equal to zero. Since 1/
1/2
1/
2/3
, we see that whenever P
k
is not essentially
zero, we can set the second exponential factor equal to 1. So we have
P
k
e
k
2
/2
. (5)
Eq. (2) then becomes
P
_
1 + e
/2
+ e
2
2
/2
+ e
3
2
/2
+ + e
(1/)
2
/2
_
. (6)
Since is small, we may approximate this sum by an integral. And since the terms
eventually become negligibly small, we can let the integral run to innity. We then
have
P
_

0
e
(/2)x
2
dx. (7)
Using the general result,
_

e
y
2
/b
dy =

b, we have
P

2
_
2

=
_

2

_

2N
. (8)
Note that the P
k
in eq. (5) is negligibly small if k 1

N. Therefore, most
of the terms in the sum in eq. (6) are negligible. The fraction of the terms that
contribute goes like

N/N = 1/

N.
Remark: Eq. (8) shows that P scales like 1/

N. If we consider the dierent setup

where all the N intervals range from 0 to 1, instead of being successively shifted by , then
the probability that the rst number is the smallest is simply 1/N, because the smallest
number is equally likely to be in any of the N identical intervals. It makes sense that the
above 1/

N result for the shifted intervals is larger than the 1/N result for the non-shifted
intervals.
If you want to derive the non-shifted 1/N result by doing an integral, observe that
if the rst number equals x, then there is a (1 x)
N1
chance that all the other N 1
numbers are larger than x. Therefore,
P =
_
1
0
(1 x)
N1
dx =
1
N
. (9)
2
Solution
Week 87 (5/10/04)
Leaving the hemisphere
Assume that the particle slides o to the right. Let v
x
and v
y
be its horizontal
and vertical velocities, with rightward and downward taken to be positive, respec-
tively. Let V
x
be the velocity of the hemisphere, with leftward taken to be positive.
Conservation of momentum gives
mv
x
= MV
x
= V
x
=

m
M

v
x
. (1)
Consider the moment when the particle is located at an angle down from the top
of the hemisphere. Locally, it is essentially on a plane inclined at angle , so the
three velocity components are related by
v
y
v
x
+ V
x
= tan = v
y
= tan

1 +
m
M

v
x
. (2)
To see why this is true, look at things in the frame of the hemisphere. In this frame,
the particle moves to the right with speed v
x
+V
x
, and downward with speed v
y
. Eq.
(2) represents the constraint that the particle remains on the hemisphere, which is
inclined at an angle at the given location.
Let us now apply conservation of energy. In terms of , the particle has fallen a
distance R(1 cos ), so conservation of energy gives
1
2
m(v
2
x
+ v
2
y
) +
1
2
MV
2
x
= mgR(1 cos ). (3)
Using eqs. (1) and (2), we can solve for v
2
x
to obtain
v
2
x
=
2gR(1 cos )
(1 + r)

1 + (1 + r) tan
2

, where r
m
M
. (4)
This function of starts at zero for = 0 and increases as increases. It then
achieves a maximum value before heading back down to zero at = /2. However,
v
x
cannot actually decrease, because there is no force available to pull the particle
to the left. So what happens is that v
x
initially increases due to the non-zero normal
force that exists while contact remains. But then v
x
reaches its maximum, which
corresponds to the normal force going to zero and the particle losing contact with
the hemisphere. The particle then sails through the air with constant v
x
. Our goal,
then, is to nd the angle for which the v
2
x
in eq. (4) is maximum. Setting the
derivative equal to zero gives
0 =

1 + (1 + r) tan
2

sin (1 cos )(1 + r)

2 tan
cos
2

= 0 =

1 + (1 + r) tan
2

cos
3
2(1 + r)(1 cos )
= 0 = cos
3
+ (1 + r)(cos cos
3
) 2(1 + r)(1 cos )
= 0 = r cos
3
3(1 + r) cos + 2(1 + r). (5)
1
This is the desired equation that determines . It is a cubic equation, so in general
it cant be solved so easily for . But in the special case of r = 1, we have
0 = cos
3
6 cos + 4. (6)
By inspection, cos = 2 is an (unphysical) solution, so we nd
(cos 2)(cos
2
+ 2 cos 2) = 0. (7)
The physical root of the quadratic equation is
cos =

3 1 0.732 = 42.9

. (8)
Alternate solution: In the reference frame of the hemisphere, the horizontal
speed of the particle v
x
+ V
y
= (1 + r)v
x
. The total speed in this frame equals this
horizontal speed divided by cos , so
v =
(1 + r)v
x
cos
. (9)
The particle leaves the hemisphere when the normal force goes to zero. The radial
F = ma equation therefore gives
mg cos =
mv
2
R
. (10)
You might be concerned that we have neglected the sideways ctitious force in the
accelerating frame of the hemisphere. However, the hemisphere is not accelerating
beginning at the moment when the particle loses contact, because the normal force
has gone to zero. Therefore, eq. (10) looks exactly like it does for the familiar
problem involving a xed hemisphere; the dierence in the two problems is in the
calculation of v.
Using eqs. (4) and (9) in eq. (10) gives
mg cos =
m(1 + r)
2
Rcos
2

2gR(1 cos )
(1 + r)

1 + (1 + r) tan
2

. (11)
Simplifying this yields

1 + (1 + r) tan
2

cos
3
= 2(1 + r)(1 cos ), (12)
which is the same as the second line in eq. (5). The solution proceeds as above.
Remark: Lets look at a few special cases of the r m/M value. In the limit r 0 (in
other words, the hemisphere is essentially bolted down), eq. (5) gives
cos = 2/3 = 48.2

, (13)
a result which may look familiar to you. In the limit r , eq. (5) reduces to
0 = cos
3
3 cos + 2 = 0 = (cos 1)
2
(cos + 2). (14)
Therefore, = 0. In other words, the hemisphere immediately gets squeezed out very fast
to the left.
For other values of r, we can solve eq. (5) either by using the formula for the roots of a
cubic equation (very messy), or by simply doing things numerically. A few numerical results
are:
2
r cos
0 .667 48.2

1/2 .706 45.1

1 .732 42.9

2 .767 39.9

10 .858 30.9

100 .947 18.8

1000 .982 10.8

1 0

3
Solution
Week 88 (5/17/04)
Ladder envelope
Let the ladder have length 1, for simplicity. In the gure below, let the ladder slide
from segment AB to segment CD. Let CD make an angle with the oor, and let
AB make an angle +d, with d very small. The given problem is then equivalent
to nding the locus of intersections, P, of adjacent ladder positions AB and CD.
+d
B D
C
A
P
O
E
Put the ladder on a coordinate system with the oor as the x-axis and the wall as
the y-axis. Let a vertical line through B intersect CD at point E. We will nd the
x and y coordinates of point P by determining the ratio of similar triangles ACP
and BEP. We will nd this ratio by determining the ratio of AC to BE. AC is
given by
AC = sin( +d) sin cos d, (1)
which is simply the derivative of sin times d. Similarly,
BD = cos cos( +d) sin d, (2)
which is simply the negative of the derivative of cos times d. BE is then given
by
BE = BDtan tan sin d. (3)
The ratio of triangle ACP to triangle BEP is therefore
ACP
BEP
=
AC
BE

cos
tan sin
=
cos
2

sin
2

r. (4)
The x coordinate of P is then, using OB = cos( +d) cos ,
P
x
=
r
1 +r
(OB)
cos
2

sin
2
+ cos
2

(OB) cos
3
. (5)
Likewise, the y coordinate of P is P
y
= sin
3
. The envelope of the ladder may
therefore be described parametrically by
(x, y) = (cos
3
, sin
3
), /2 0. (6)
1
Equivalently, using cos
2
+ sin
2
= 1, the envelope may be described by the equa-
tion,
x
2/3
+y
2/3
= 1. (7)
2
Solution
Week 89 (5/24/04)
Rope between inclines
Let the total mass of the rope be m, and let a fraction f of it hang in the air.
Consider the right half of this section. Its weight, (f/2)mg, must be balanced by
the vertical component, T sin, of the tension at the point where it joins the part
of the rope touching the right platform. The tension at this point therefore equals
T = (f/2)mg/ sin.
Now consider the part of the rope touching the right platform. This part has
mass (1 f)m/2. The normal force from the platform is N = (1 f)(mg/2) cos ,
so the maximal friction force equals (1 f)(mg/2) cos , because = 1. This
ction force must balance the sum of the gravitational force component along the
plane, which is (1 f)(mg/2) sin, plus the tension at the lower end, which is the
(f/2)mg/ sin we found above. Therefore,
1
2
(1 f)mg cos =
1
2
(1 f)mg sin +
fmg
2 sin
, (1)
which gives
f =
F()
1 + F()
, where F() cos sin sin
2
. (2)
This expression for f is a monotonically increasing function of F(), as you can
check. The maximal f is therefore obtained when F() is as large as possible.
Using the double-angle formulas, we can rewrite F() as
F() =
1
2
(sin2 + cos 2 1). (3)
The derivative of this is cos 2sin2, which equals zero when tan2 = 1. Therefore,

max
= 22.5

. (4)
Eq. (3) then yields F(
max
) = (

2 1)/2, and so eq. (2) gives

f
max
=

2 1

2 + 1
= (

2 1)
2
= 3 2

2 0.172. (5)
1
Solution
Week 90 (5/31/04)
The game of NIM
As with many problems, this one can be solved in two possible ways. We can (1)
write down the correct answer, through some stroke of genius, and then verify that it
works, or (2) work out some simple cases, get a feel for the problem, and eventually
wind our way around to the correct answer. For the problem at hand, lets proceed
via the second method and try to arrive at the answer with some motivation.
Well start by working out what happens in particular cases of small numbers of
coins in the piles, and then well look for a pattern. A reasonable way to organize
the results is to determine which combinations of numbers are guaranteed losing
positions (assuming that both players are aware of the optimal strategy).
The most obvious losing position (LP) is piles with coins of numbers (1,1,0). If
you encounter this setup, you must pick one coin, and then your opponent will pick
the last one and thereby win. More generally, the combination (N, N, 0) is an LP,
because if you take o n coins from one pile, your opponent will take o n coins
from the other. He will keep matching you on each turn, until nally the situation is
(0,0,0), with his having removed the last coin(s). Note that situations of (N, M, 0)
and (N, N, M) are therefore winning positions (WP), because it is possible to turn
them into an LP with one move.
This reasoning utilizes the following two obvious properties that an LP must
have: (1) removal of any number of coins from one pile of an LP creates a non-LP,
and (2) on the next turn it is always possible to bring the situation back to an LP
again.
Consider now the cases where no two piles have the same numbers of coins. The
smallest case is (1,2,x). x = 3 is the rst possibility for an LP, and we quickly see
that it is indeed an LP, because the removal of any number of coins from any of the
piles yields a setup of the from (N, M, 0) or (N, N, M), which are WPs, as we saw
above.
Note that once we have found an LP, we know that any triplet that has two
numbers in common with the LP, with its remaining number larger than that in
the LP, must be a WP. This is true because it is possible to turn it into an LP by
removing coins from this last pile.
If we look at other triplets in which the rst number is 1, we nd that (1,4,5),
(1,6,7), (1,8,9), etc., are LPs, as you can check by showing that any move turns
them into a WP.
Now consider cases where 2 is the smallest number of coins in a pile. We nd,
after a little ddling, that (2,4,6), (2,5,7), (2,8,10), and (2,9,11), etc., are LPs, as
you can check.
Similar ddling, starting with a 3, gives (3,4,7), (3,5,6), (3,8,11), and (3,9,10),
etc., as LPs.
Lets now make a table of this hodgepodge of results, for up to seven coins in a
pile. The two axes will be the rst two numbers in an LP triplet, and the entry in
the table will be the third. The table is of course symmetric.
1
0 1 2 3 4 5 6 7
0 0 1 2 3 4 5 6 7
1 1 0 3 2 5 4 7 6
2 2 3 0 1 6 7 4 5
3 3 2 1 0 7 6 5 4
4 4 5 6 7 0 1 2 3
5 5 4 7 6 1 0 3 2
6 6 7 4 5 2 3 0 1
7 7 6 5 4 3 2 1 0
As a rst guess at the key to this table, we might say that two numbers in an LP
triplet must add up to the third. This, however, does not work for the (3,5,6) triplet.
It also does not work for the (3,9,10) triplet that we found above. Continuing on
to higher numbers, the guess seems to work for triplets starting with a 4. But then
if we start with a 5, we eventually nd the LP triplets (5,9,12) and (5,11,14), for
which the sum of two numbers doesnt equal the third.
In an eort to nd the key, let us exploit the patterns in the table, perhaps
brought out best by the following grouping:
0 1 2 3 4 5 6 7
1 0 3 2 5 4 7 6
2 3 0 1 6 7 4 5
3 2 1 0 7 6 5 4
4 5 6 7 0 1 2 3
5 4 7 6 1 0 3 2
6 7 4 5 2 3 0 1
7 6 5 4 3 2 1 0
The entries in the upper right 44 box are 4 more than the corresponding entries
in the upper left box. Likewise, within each 44 box, the entries in the upper right
2 2 box are 2 more than the entries in the upper left box. Similar results would
be evident if we doubled the size of the box (out to 15), where we would see 8 8
boxes having entries diering by 8.
All this suggests that powers of 2 are important in this problem. We therefore
should consider writing the numbers in a way where factors of 2 are evident, that
is, in base 2. There is no guarantee that this will help, but lets try it and see what
happens. Lets write down the troublesome triplets weve found (the ones for which
two of the numbers dont add up to the third) in base 2:
3: 11 3: 11 5: 101 5: 101
5: 101 9: 1001 9: 1001 11: 1011
6: 110 10: 1010 12: 1100 14: 1110
What property do these triplets have? When written in the above form, we see
that each column in base 2 contains an even number of 1s. After checking some
2
other triplets, this appears to be true in general for an LP. We will prove this with
the following theorem.
Theorem: Call a triplet an E-triplet (the E stands for even) if it has the
following property: When the three numbers are written in base 2, there is an even
number (that is, either zero or two) of 1s in each digits place. Then a triplet is a
losing position (LP) if and only if it is an E-triplet.
Proof: Let us establish the following three facts concerning E-triplets:
1. Removal of any number of coins from one pile of an E-triplet will turn the
triplet into a non-E-triplet.
2. Given a non-E-triplet, it is always possible to remove coins from one pile to
turn the triplet into an E-triplet.
3. (0,0,0) is an E-triplet.
These facts may be demonstrated as follows:
1. This is true because any two numbers in an E-triplet uniquely determine the
third.
2. We will describe how to turn any non-E-triplet into an E-triplet. Write the
three numbers of coins in base 2, and put them in a column, with the units
digits aligned, as we did above. Starting from the left, look at each digits
column until you nd a column with an odd number (that is, either one or
three) of 1s. Let this be the nth column (counting from the right).
If there is one 1 in the nth column, label the number containing this 1 as
A. If there are three 1s, then arbitrarily pick any of the numbers to be A.
Remove coins from A by switching the 1 in the nth column to a 0, and also by
switching any 1s to 0s, or 0s to 1s, in other columns to the right of the nth
column, in order to produce an even number or 1s is all columns. We have
now created an E-triplet.
Note that this switching of 1s and 0s does indeed correspond to removing
(as opposed to adding) coins from A, because even if all the columns to the
right of the nth column involve switching 0s to 1s, this addition of 2
n1
1
coins is still less than the subtraction of the 2
n1
coins arising from the 1-to-0
switch in the nth column.
3. This is true, by denition of an E-triplet.
The rst two of these facts show that if player X receives an E-triplet on a given
turn, then player Y can ensure that X receives an E-triplet on every subsequent
turn. Therefore, X must always create a non-E-triplet, by the rst of the three
facts. X therefore cannot take the last coin, because he cannot create the E-triplet
(0, 0, 0). Therefore, an E-triplet is a losing position.
The best strategy in this game is therefore to give your opponent an E-triplet
whenever you can. If both players are aware of this strategy, then the outcome of
3
the game is determined by the initial piles of coins. If they form an E-triplet, then
the player who goes rst loses. If they do not form an E-triplet, then the player who
goes rst wins, because he can always create an E-triplet to give to his opponent.
Remarks: If the starting numbers of coins are random, then the player who goes rst will
most likely win, because most triplets are not E-triplets. We can demonstrate this fact by
using the somewhat crude scenario where the three numbers are random numbers from 0 to
2
n
1, that is, they each have n digits in base 2 (many of which may be zero). Then there
are (2
n
)
3
possible triplets. But there are only 4
n
possible E-triplets, because each of the n
columns of three digits (when we write the three numbers on top of each other) has four
E-triplet possibilities: 0,0,0; 1,1,0; 1,0,1; and 0,1,1. The fraction of E-triplets is therefore
4
n
/2
3n
= 2
n
, which goes to zero for large n.
Note that there is nothing special about having three piles. We can have any number of
piles (but still two players), and all the above reasoning still holds. Losing positions are the
ones that have an even number of 1s in each column, when written in base 2. The three
facts in the above theorem still hold.
4