Estimation

Order-of-Magnitude Physics: Understanding the World with Dimensional
Analysis, Educated Guesswork, and White Lies

Peter Goldreich, California Institute of Technology
Sanjoy Mahajan, University of Cambridge
Sterl Phinney, California Institute of Technology
Draft of 1 August 1999
c 1999
Send comments to sanjoy@mrao.cam.ac.uk
ii
Contents
1 Wetting Your Feet 1
1.1 Warmup problems 1
1.2 Scaling analyses 13
1.3 What you have learned 21
2 Dimensional Analysis 23
2.1 Newtons law 23
2.2 Pendula 27
2.3 Drag in uids 31
3 Materials I 43
3.1 Sizes 43
3.2 Energies 51
3.3 Elastic properties 53
3.4 Application to white dwarfs 58
4 Materials II 63
4.1 Thermal expansion 63
4.2 Phase changes 65
4.3 Specic heat 73
4.4 Thermal diusivity of liquids and solids 77
4.5 Diusivity and viscosity of gases 79
4.6 Thermal conductivity 80
5 Waves 85
5.1 Dispersion relations 85
5.2 Deep water 88
5.3 Shallow water 106
5.4 Combining deep- and shallow-water gravity waves 108
5.5 Combining deep- and shallow-water ripples 108
5.6 Combining all the analyses 109
5.7 What we did 109
Bibliography 110
1
1 Wetting Your Feet
Most technical education emphasizes exact answers. If you are a
physicist, you solve for the energy levels of the hydrogen atom to
six decimal places. If you are a chemist, you measure reaction rates
and concentrations to two or three decimal places. In this book,
you learn complementary skills. You learn that an approximate
answer is not merely good enough; its often more useful than
an exact answer. When you approach an unfamiliar problem, you
want to learn rst the main ideas and the important principles,
because these ideas and principles structure your understanding
of the problem. It is easier to rene this understanding than to
create the rened analysis in one step.
The adjective in the title, order of magnitude, reects our
emphasis on approximation. An order of magnitude is a factor of
10. To be within an order of magnitude, or to estimate a quan-
tity to order of magnitude, means that your estimate is roughly
within a factor of 10 on either side. This chapter introduces the
art of determining such approximations.
Writers block is broken by writing; estimators block is broken
by estimating. So we begin our study of approximation using ev-
eryday examples, such as estimating budgets or annual production
of diapers. These warmups ex your estimation muscles, which
may have lain dormant through many years of traditional edu-
cation. After the warmup, we introduce a more subtle method:
scaling relations.
1.1 Warmup problems
Everyday estimations provide practice for our later problems, and
also provide a method to sanity check information that you see.
Suppose that a newspaper article says that the annual cost of
health care in the United States will soon surpass $1 trillion.
Whenever you read any such claim, you should automatically
think: Does this number seem reasonable? Is it far too small, or far
too large? You need methods for such estimations, methods that
we develop in several examples. We dedicate the rst example to
physicists who need employment outside of physics.
1.1.1 Armored cars
How much money is there in a fully loaded Brinks armored car?
The amount of money depends on the size of the car, the de-
nomination of the bills, the volume of each bill, the amount of air
between the bills, and many other factors. The question, at rst
glance, seems vague. One important skill that you will learn from
1. Wetting Your Feet 2
1. Once at a Fourth-of-July celebra-
tion, a reporter wondered and later
asked why Mr. Murphy (he was always
Mr. Murphy even to his closest asso-
ciates) did not join in the singing of the
National Anthem. Perhaps he didnt
want to commit himself, the bosss aide
explained. From the Introduction by
Arthur Mann, to William L. Riordan,
Plunkitt of Tammany Hall (New York:
E. P. Dutton, 1963), page ix.
this text, by practice and example, is what assumptions to make.
Because we do not need an exact answer, any reasonable set of
assumptions will do. Getting started is more important than dot-
ting every i; make an assumptionany assumptionand begin.
You can correct the gross lies after you have got a feeling for the
problem, and have learned which assumptions are most critical.
If you keep silent, rather than tell a gross lie, you never discover
anything.
Lets begin with our equality conventions, in ascending order
of precision. We use for proportionalities, where the units on the
left and right sides of the do not match; for example, Newtons
second law could read F m. We use for dimensionally correct
relations (the units do match), which are often accurate to, say, a
factor of 5 in either direction. An example is
kinetic energy Mv
2
. (1.1)
Like the sign, the sign indicates that weve left out a constant;
with , the constant is dimensionless. We use to emphasize that
the relation is accurate to, say, 20 or 30 percent. Sometimes,
relations are also that accurate; the context will make the distinc-
tion.
Now we return to the armored car. How much money does it
contain? Before you try a systematic method, take a guess. Make
it an educated guess if you have some knowledge (perhaps you
work for an insurance company, and you happened to write the
insurance policy that the armored-car company bought); make
it an uneducated guess if you have no knowledge. Then, after
you get a more reliable estimate, compare it to your guess: The
wonderful learning machine that is your brain magically improves
your guesses for the next problem. You train your intuition, and,
as we see at the end of this example, you aid your memory. As a
pure guess, lets say that the armored car contains $1 million.
Now we introduce a systematic method. A general method in
many estimations is to break the problem into pieces that we can
handle: We divide and conquer. The amount of money is large
by everyday standards; the largeness suggests that we break the
problem into smaller chunks, which we can estimate more reliably.
If we know the volume V of the car, and the volume v of a us bill,
then we can count the bills inside the car by dividing the two
volumes, N V/v. After we count the bills, we can worry about
the denominations (divide and conquer again). [We do not want
to say that N V/v. Our volume estimates may be in error easily
by 30 or 40 percent, or only a fraction of the storage space may
be occupied by bills. We do not want to commit ourselves.
1
]
We have divided the problem into two simpler subproblems:
determining the volume of the car, and determining the volume of
a bill. What is the volume of an armored car? The storage space
in an armored car has a funny shape, with ledges, corners, nooks,
and crannies; no simple formula would tell us the volume, even if
2 m
2 m
2 m
Figure 1.1. Interior of a Brinks ar-
mored car. The actual shape is irreg-
ular, but to order of magnitude, the
interior is a cube. A person can prob-
ably lie down or stand up with room
to spare, so we estimate the volume as
V 2 m2 m2 m 10 m
3
.
2. I seen my opportunities and I took
em.George Washington Plunkitt, of
Tammany Hall, quoted by Riordan [15,
page 3].
we knew the 50-odd measurements. This situation is just the sort
for which order-of-magnitude physics is designed; the problem is
messy and underspecied. So we lie skillfully: We pretend that
the storage space is a simple shape with a volume that we can nd.
In this case, we pretend that it is a rectangular prism (Figure 1.1).
To estimate the volume of the prism, we divide and conquer.
We divide estimating the volume into estimating the three dimen-
sions of the prism. The compound structure of the formula
V length width height (1.2)
suggests that we divide and conquer. Probably an average-sized
person can lie down inside with room to spare, so each dimension
is roughly 2 m, and the interior volume is
V 2 m2 m2 m 10 m
3
= 10
7
cm
3
. (1.3)
In this text, 222 is almost always 10. We are already working
with crude approximations, which we signal by using in N
V/v, so we do not waste eort in keeping track of a factor of
1.25 (from using 10 instead of 8). We converted the m
3
to cm
3
in anticipation of the dollar-bill-volume calculation: We want to
use units that match the volume of a dollar bill, which is certainly
much smaller than 1 m
3
.
Now we estimate the volume of a dollar bill (the volumes of us
denominations are roughly the same). You can lay a ruler next to
a dollar bill, or you can just guess that a bill measures 2 or 3 inches
by 6 inches, or 6 cm15 cm. To develop your feel for sizes, guess
rst; then, if you feel uneasy, check your answer with a ruler. As
your feel for sizes develops, you will need to bring out the ruler
less frequently. How thick is the dollar bill? Now we apply another
order-of-magnitude technique: guerrilla warfare. We take any
piece of information that we can get.
2
Whats a dollar bill? We lie
skillfully and say that a dollar bill is just ordinary paper. How thick
is paper? Next to the computer used to compose this textbook is
a laser printer; next to the printer is a ream of laser printer paper.
The ream (500 sheets) is roughly 5 cm thick, so a sheet of quality
paper has thickness 10
2
cm. Now we have the pieces to compute
the volume of the bill:
v 6 cm15 cm10
2
cm 1 cm
3
. (1.4)
The original point of computing the volume of the armored car
and the volume of the bill was to nd how many bills t into the
car: N V/v 10
7
cm
3
/1 cm
3
= 10
7
. If the money is in $20 bills,
then the car would contain $200 million.
The bills could also be $1 or $1000 bills, or any of the inter-
mediate sizes. We chose the intermediate size $20, because it lies
nearly halfway between $1 and $1000. You naturally object that
$500, not $20, lies halfway between $1 and $1000. We answer that
objection shortly. First, we pause to discuss a general method of
estimating: talking to your gut. You often have to estimate
quantities about which you have only meager knowledge. You can
then draw from your vast store of implicit knowledge about the
worldknowledge that you possess but cannot easily write down.
You extract this knowledge by conversing with your gut; you ask
that internal sensor concrete questions, and listen to the feelings
that it returns. You already carry on such conversations for other
aspects of life. In your native language, you have an implicit knowl-
edge of the grammar; an incorrect sentence sounds funny to you,
even if you do not know the rule being broken. Here, we have to
estimate the denomination of bill carried by the armored car (as-
suming that it carries mostly one denomination). We ask ourselves,
How does an armored car lled with one-dollar bills sound? Our
gut, which knows the grammar of the world, responds, It sounds
a bit ridiculous. One-dollar bills are not worth so much eort; plus,
every automated teller machine dispenses $20 bills, so a $20 bill is
a more likely denomination. We then ask ourselves, How about
a truck lled with thousand-dollar bills? and our gut responds,
no, sounds way too bignever even seen a thousand-dollar bill,
probably collectors items, not for general circulation. After this
edifying dialogue, we decide to guess a value intermediate between
$1 and $1000.
We interpret between using a logarithmic scale, so we choose
a value near the geometric mean,
1 1000 30. Interpolating

on a logarithmic scale is more appropriate and accurate than is
interpolating on a linear scale, because we are going to use the
number in a chain of multiplications and divisions. Lets check
whether 30 is reasonable, by asking our gut about nearby esti-
mates. It is noncommittal when asked about $10 or $100 bills;
both sound reasonable. So our estimate of 30 is probably rea-
sonable. Because there are no $30 bills, we use a nearby actual
denomination, $20.
Assuming $20 bills, we estimate that the car contains $200 mil-
lion, an amount much greater than our initial guess of $1 million.
Such a large discrepancy makes us suspicious of either the guess
or this new estimate. We therefore cross-check our answer, by
estimating the monetary value in another way. By nding another
method of solution, we learn more about the domain. If our new
estimate agrees with the previous one, then we gain condence
that the rst estimate was correct; if the new estimate does not
agree, it may help us to nd the error in the rst estimate.
We estimated the carrying capacity using the available space.
How else could we estimate it? The armored car, besides having
limited space, cannot carry innite mass. So we estimate the mass
of the bills, instead of their volume. What is the mass of a bill?
If we knew the density of a bill, we could determine the mass us-
ing the volume computed in (1.4). To nd the density, we use the
guerrilla method. Money is paper. What is paper? Its wood or fab-
3. It is unfortunate that mass is not a
transitive verb in the way that weigh
is. Otherwise, we could write that the
truck masses 10 tons. If you have more
courage than we have, use this construc-
tion anyway, and start a useful trend.
ric, except for many complex processing stages whose analysis is
beyond the scope of this book. Here, we just used another order-of-
magnitude technique, punt: When a process, such as papermak-
ing, looks formidable, forget about it, and hope that youll be okay
anyway. Ignorance is bliss. Its more important to get an estimate;
you can correct the egregiously inaccurate assumptions later. How
dense is wood? Once again, use the guerrilla method: Wood barely
oats, so its density is roughly that of water, 1 g cm
3
. A bill,
which has volume v 1 cm
3
, has mass m 1 g. And 10
7
cm
3
of
bills would have a mass of 10
7
g = 10 tons.
3
This cargo is large. [Metric tons are 10
6
g; English tons (may
that measure soon perish) are roughly 0.9 10
6
g, which, for our
purposes, is also 10
6
g.] What makes 10 tons large? Not the number
10 being large. To see why not, consider these extreme arguments:
In megatons, the cargo is 10
5
megatons, which is a tiny cargo
because 10
5
is a tiny number.
In grams, the cargo is 10
7
g, which is a gigantic cargo because
10
7
is a gigantic number.
You might object that these arguments are cheats, because nei-
ther grams nor megatons is a reasonable unit in which to measure
truck cargo, whereas tons is a reasonable unit. This objection is
correct; when you specify a reasonable unit, you implicitly choose
a standard of comparison. The moral is this: A quantity with
unitssuch as tonscannot be large intrinsically. It must be large
compared to a quantity with the same units. This argument fore-
shadows the topic of dimensional analysis, which is the subject of
Chapter 2.
So we must compare 10 tons to another mass. We could com-
pare it to the mass of a bacterium, and we would learn that 10 tons
is relatively large; but to learn about the cargo capacity of Brinks
armored cars, we should compare 10 tons to a mass related to
transport. We therefore compare it to the mass limits at railroad
crossings and on many bridges, which are typically 2 or 3 tons.
Compared to this mass, 10 tons is large. Such an armored car
could not drive many places. Perhaps 1 ton of cargo is a more
reasonable estimate for the mass, corresponding to 10
6
bills. We
can cross-check this cargo estimate using the size of the armored
cars engine (which presumably is related to the cargo mass); the
engine is roughly the same size as the engine of a medium-sized
pickup truck, which can carry 1 or 2 tons of cargo (roughly 20
or 30 book boxessee Example 3.1). If the money is in $20 bills,
then the car contains $20 million. Our original, pure-guess esti-
mate of $1 million is still much smaller than this estimate by
roughly an order of magnitude, but we have more condence in
this new estimate, which lies roughly halfway between $1 million
and $200 million (we nd the midpoint on a logarithmic scale).
The Reuters newswire of 18 September 1997 has a report on the
largest armored car heist in us history; the thieves took $18 mil-
Pasadena
A 100 km
2
10 km
10 km
a (50 m)
2
Figure 1.2. Map of Pasadena, Cali-
fornia drawn to order of magnitude.
The small shaded box is the area gov-
erned by one lamp; the box is not drawn
to scale, because if it were, it would be
only a few pixels wide. How many such
boxes can we t into the big square? It
takes 10 min to leave Pasadena by car,
so Pasadena has area A (10 km)
2
=
10
8
m
2
. While driving, we pass a lamp
every 3 s, so we estimate that theres a
lamp every 50 m; each lamp covers an
area a (50 m)
2
.
lion; so our estimate is accurate for a well-stocked car. (Typical
heists net between $1 million and $3 million.)
We answered this rst question in detail to illustrate a num-
ber of order-of-magnitude techniques. We saw the value of ly-
ing skillfullyapproximating dollar-bill paper as ordinary paper,
and ordinary paper as wood. We saw the value of waging guer-
rilla warfareusing knowledge that wood barely oats to esti-
mate the density of wood. We saw the value of cross-checking
estimating the mass and volume of the cargoto make sure that
we have not committed a gross blunder. And we saw the value of
divide and conquerbreaking volume estimations into products
of length, width, and thickness. Breaking problems into factors,
besides making the estimation possible, has another advantage: It
often reduces the error in the estimate. There probably is a general
rule about guessing, that the logarithm is in error by a reasonably
xed fraction. If we guess a number of the order of 1 billion in
one step, we might be in error by, say, a factor of 10. If we factor
the 1 billion into four pieces, the estimate of each piece will be in
error by a factor of = 10
1/4
. We then can hope that the errors
are uncorrelated, so that they combine as steps in a random walk.
Then, the error in the product is
4
= 10
1/2
, which is smaller
than the one-shot error of 10. So breaking an estimate into pieces
reduces the error, according to this order-of-magnitude analysis of
error.
1.1.2 Cost of lighting Pasadena, California
What is the annual cost of lighting the streets of Pasadena, Cali-
fornia?
Astronomers would like this cost to be huge, so that they could
argue that street lights should be turned o at night, the better
to gaze at heavenly bodies. As in Section 1.1.1, we guess a cost
right away, to train our intuition. So lets guess that lighting costs
$1 million annually. This number is unreliable; by talking to our
gut, we nd that $100,000 sounds okay too, as does $10 million
(although $100 million sounds too high).
The cost is a large number, out of the ordinary range of costs,
so it is dicult to estimate in one step (we just tried to guess it,
and were not sure within a factor of 10 what value is correct). So
we divide and conquer. First, we estimate the number of lamps;
Figure 1.3. Fraction of Pasadena that
is lighted. The streets (thick lines) are
spaced d 100 m apart. Each lamp,
spaced 50 m apart, lights a 50 m 50 m
area (the eight small, unshaded squares).
The only area not lighted is in the center
of the block (shaded square); it is one-
fourth of the area of the block. So, if
every street has lights, f = 0.75.
then, we estimate how much it costs to light each lamp.
To estimate the number of lamps (another large, hard-to-guess
number), we again divide and conquer: We estimate the area of
Pasadena, and divide it by the area that each lamp governs, as
shown in Figure 1.2. There is one more factor to consider: the
fraction of the land that is lighted (we call this fraction f). In the
desert, f is perhaps 0.01; in a typical city, such as Pasadena, f
is closer to 1.0. We rst assume that f = 1.0, to get an initial
estimate; then we estimate f, and correct the cost accordingly.
We now estimate the area of Pasadena. What is its shape? We
could look at a map, but, as lazy armchair theorists, we lie; we
assume that Pasadena is a square. It takes, say, 10 minutes to
leave Pasadena by car, perhaps traveling at 1 km/min; Pasadena
is roughly 10 km in length. Therefore, Pasadena has area A
10 km 10 km = 100 km
2
= 10
8
m
2
. (The true area is 23 mi
2
, or
60 km
2
.) How much area does each lamp govern? In a carsay,
at 1 km/min or 20 ms
1
it takes 2 or 3 s to go from lamppost
to lamppost, corresponding to a spacing of 50 m. Therefore,
a (50 m)
2
2.510
3
m
2
, and the number of lights is N A/a
10
8
m
2
/2.5 10
3
m
2
4 10
4
.
How much does each lamp cost to operate? We estimate the
cost by estimating the energy that they consume in a year and the
price per unit of energy (divide and conquer). Energy is power
time. We can estimate power reasonably accurately, because we
are familiar with lamps around the home. To estimate a quantity,
try to compare it to a related, familiar one. Street lamps shine
brighter than a household 100 W bulb, but they are probably more
ecient as well, so we guess that each lamp draws p 300 W. All
N lamps consume P Np 4 10
4
300 W 1.2 10
4
kW.
Lets say that the lights are on at night8 hours per dayor
3000 hours/year. Then, they consume 410
7
kWhour. An electric
bill will tell you that electricity costs $0.08 per kWhour (if you
live in Pasadena), so the annual cost for all the lamps is $3 million.
Now lets improve this result by estimating the fraction f.
What features of Pasadena determine the value of f? To answer
this question, consider two extreme cases: the desert and New
York city. In the desert, f is small, because the streets are widely
separated, and many streets have no lights. In New York city, f
is high, because the streets are densely packed, and most streets
are lled with street lights. So the relevant factors are the spacing
between streets (which we call d), and the fraction of streets that
are lighted (which we call f
l
). As all pedestrians in New York city
know, 10 northsouth blocks or 20 eastwest blocks make 1 mile
(or 1600 m); so d 100 m. In street layout, Pasadena is closer
to New York city than to the desert. So we use d 100 m for
Pasadena as well. If every street were lighted, what fraction of
Pasadena would be lighted? Figure 1.3 shows the computation;
the result is f 0.75In New York city, f
L
1; in Pasadena,
f
L
0.3 is more appropriate. So f 0.75 0.3 0.25. Our
estimate for the annual cost is then $1 million. Our initial guess
is unexpectedly accurate.
As you practice such estimations, you will be able to write
them down compactly, converting units stepwise until you get to
your goal (here, $/year). The cost is
cost 100 km
2
. .
A
10
6
m
2
1 km
2

1 lamp
2.5 10
3
m
2
. .
a
8 hrs
1 day
. .
night
365 days
1 year

$0.08
1 kWhour
. .
price
0.3 kW0.25
$1 million.
(1.5)
It is instructive to do the arithmetic without using a calculator.
Just as driving to the neighbors house atrophies your muscles,
using calculators for simple arithmetic dulls your mind. You do
not develop an innate sense of how large quantities should be, or
of when you have made a mistake; you learn only how to punch
keys. If you need an answer with 6-digit precision, use a calculator;
thats the task for which they are suited. In order-of-magnitude es-
timates, 1- or 2-digit precision is sucient; you can easily perform
these low-precision calculations mentally.
Will Pasadena astronomers rejoice because this cost is large?
A cost has units (here, dollars), so we must compare it to another,
relevant cost. In this case, that cost is the budget of Pasadena. If
lighting is a signicant fraction of the budget, then can we say
that the lighting cost is large.
1.1.3 Pasadenas budget
What fraction of Pasadenas budget is alloted to street lighting?
We just estimated the cost of lighting; now we need to estimate
Pasadenas budget. First, however, we make the initial guess. It
would be ridiculous if such a trivial service as street lighting con-
sumed as much as 10 percent of the citys budget. The city still
has road construction, police, city hall, and schools to support. 1
percent is a more reasonable guess. The budget should be roughly
$100 million.
Now that weve guessed the budget, how can we estimate it?
The budget is the amount spent. This money must come from
somewhere (or, at least, most of it must): Even the us government
is moderately subject to the rule that income spending. So we
can estimate spending by estimating income. Most us cities and
towns bring in income from property taxes. We estimate the citys
income by estimating the property tax per person, and multiplying
the tax by the citys population.
Each person pays property taxes either directly (if she owns
land) or indirectly (if she rents from someone who does own land).
A typical monthly rent per person (for a two-person apartment) is
$500 in Pasadena (the apartments-for-rent section of a local news-
paper will tell you the rent in your area), or $6000 per year. (Places
with ne weather and less smog, such as the San Francisco area,
have higher monthly rents, roughly $1500 per person.) According
to occasional articles that appear in newspapers when rent sky-
rockets and interest in the subject increases, roughly 20 percent of
rent goes toward landlords property taxes. We therefore estimate
that $1000 is the annual property tax per person.
Pasadena has roughly 210
5
people, as stated on the road signs
that grace the entries to Pasadena. So the annual tax collected is
$200 million. If we add federal subsidies to the budget, the total
budget is probably double that, or $400 million. A rule of thumb
in these nancial calculations is to double any estimate that you
make, to correct for costs or revenues that you forgot to include.
This rule of thumb is not infallible. We can check its validity in this
case by estimating the federal contribution. The federal budget
is roughly $2 trillion, or $6000 for every person in the United
States (any recent almanac tells us the federal budget and the
us population). One-half of the $6000 funds defense spending and
interest on the national debt; it would be surprising if fully one-
half of the remaining $3000 went to the cities. Perhaps $1000 per
person goes to cities, which is roughly the amount that the city
collects from property taxes. Our doubling rule is accurate in this
case.
For practice, we cross-check the local-tax estimate of $200 mil-
lion, by estimating the total land value in Pasadena, and guess-
ing the tax rate. The area of Pasadena is 100 km
2
36 mi
2
, and
1 mi
2
= 640 acres. You can look up this acresquare-mile conver-
sion, or remember that, under the Homestead Act, the us govern-
ment handed out land in 160-acre parcelsknown as quarter lots
because they were 0.25 mi
2
. Land prices are exorbitant in south-
ern California (sun, sand, surf, and mountains, all within a few
hours drive); the cost is roughly $1 million per acre (as you can
determine by looking at the homes-for-sale section of the newspa-
per). We guess that property tax is 1 percent of property value.
You can determine a more accurate value by asking anyone who
owns a home, or by asking City Hall. The total tax is
W 36 mi
2
. .
area
640 acres
1 mi
2

$1 million
1 acre
. .
land price
0.01
..
tax
$200 million.
(1.6)
This revenue is identical to our previous estimate of local revenue;
the equality increases our condence in the estimates. As a check
on our estimate, we looked up the budget of Pasadena. In 1990,
it was $350 million; this value is much closer to our estimate of
$400 million than we have a right to expect!
The cost of lighting, calculated in Section 1.1.2, consumes only
0.2 percent of the citys budget. Astronomers should not wait for
Pasadena to turn out the lights.
1.1.4 Diaper production
How many disposable diapers are manufactured in the United
States every year?
We begin with a guess. The number must be in the millions
say, 10 millionbecause of the huge outcry when environmental-
ists suggested banning disposable diapers to conserve landll space
and to reduce disposed plastic. To estimate such a large number,
we divide and conquer. We estimate the number of diaper users
babies, assuming that all babies use diapers, and that no one else
doesand the number of diapers that each baby uses in 1 year.
These assumptions are not particularly accurate, but they provide
a start for our estimation. How many babies are there? We hereby
dene a baby as a child under 2 years of age. What fraction of
the population are babies? To estimate this fraction, we begin by
assuming that everyone lives exactly 70 yearsroughly the life
expectancy in the United Statesand then abruptly dies. (The
life expectancy is more like 77 years, but an error of 10 percent is
not signicant given the inaccuracies in the remaining estimates.)
How could we have gured out the average age, if we did not
already know it? In the United States, government retirement (So-
cial Security) benets begin at age 65 years, the canonical retire-
ment age. If the life expectancy were less than 65 yearssay, 55
yearsthen so many people would complain about being short-
changed by Social Security that the system would probably be
changed. If the life expectancy were much longer than 65 years
say, if it were 90 yearsthen Social Security would cost much
more: It would have to pay retirement benets for 90 65 = 25
years instead of for 75 65 = 10 years, a factor of 2.5 increase. It
would have gone bankrupt long ago. So, the life expectancy must
be around 70 or 80 years; if it becomes signicantly longer, expect
to see the retirement age increased accordingly. For deniteness,
we choose one value: 70 years. Even if 80 years is a more accurate
estimate, we would be making an error of only 15 percent, which
is probably smaller than the error that we made in guessing the
cuto age for diaper use. It would hardly improve the accuracy of
Area
2
70
3 10
8
10
7
Order-of-magnitude age distribution
2 70
0
4
True age distribution
Age (years)
People (10
6
)
Figure 1.4. Number of people versus age
(in the United States). The true age dis-
tribution is irregular and messy; without
looking it up, we cannot know the area
between ages 0.0 years and 2.0 years
(to estimate the number of babies). The
rectangular graphwhich has the same
area and similar widthimmediately
makes clear what the fraction under
2 years is: It is roughly 2/70 0.03.
The population of the United States is
roughly 3 10
8
, so the number of babies is
0.03 3 10
8
10
7
.
the nal estimate to agonize over this 15 percent.
To compute how people are between the ages of 0 and 2.0 years,
consider an analogous problem. In a 4-year university (which grad-
uates everyone in 4 years and accepts no transfer students) with
1000 students, how many students graduate in each years class?
The answer is 250, because 1000/4 = 250. We can translate this
argument into the following mathematics. Let be lifetime of a
person. We assume that the population is steady: The birth and
death rates are equal. Let the rates be

N. Then the total pop-
ulation is N =

N, and the population between ages
1
and
2
is
N
=

N(
2
1
). (1.7)
So, if everyone lives for 70 years exactly, then the fraction of the
population whose age is between 0 and 2 years is 2/70 or 0.03
(Figure 1.4). There are roughly 310
8
people in the United States,
so
N
babies
3 10
8
0.03 10
7
babies. (1.8)
We have just seen another example of skillful lying. The jagged
curve in Figure 1.4 shows a cartoon version of the actual mortal-
ity curve for the United States. We simplied this curve into the
boxcar shape (the rectangle), because we know how to deal with
rectangles. Instead of integrating the complex, jagged curve, we
integrate a simple, civilized curve: a rectangle of the same area and
similar width. This procedure is order-of-magnitude integra-
tion. Similarly, when we studied the Brinks armored-car example
(Section 1.1.1), we pretended that the cargo space was a cube;
that procedure was order-of-magnitude geometry.
How many diapers does each baby use per year? This num-
ber is largemaybe 100, maybe 10,000so a wild guess is not
likely to be accurate. We divide and conquer, dividing 1 year into
365 days. Suppose that each baby uses 8 diapers per day; new-
borns use many more, and older toddlers use less; our estimate is a
reasonable compromise. Then, the annual use per baby is 3000,
and all 10
7
babies use 3 10
10
diapers. The actual number manu-
Reported meteor impact
a 10 m
2
Earths surface
A 5 10
14
m
2
Locations where impact would
be reported, N 10
9
Figure 1.5. Large-meteorite impacts on
the surface of the earth. Over the sur-
face of the earth, represented as a circle,
every year one meteorite impact (black
square) causes sucient damage to be
reported by Sky&Telescope. The gray
squares are areas where such a meteorite
impact would have been reportedfor
example, a house or car in an indus-
trial country; they have total area Na
10
10
m
2
. The gray squares cover only
a small fraction of the earths surface.
The expected number of large impacts
over the whole earth is 1A/Na 510
4
,
where A 5 10
14
m
2
is the surface area
of the earth.
factured is 1.6 10
10
per year, so our initial guess is low, and our
systematic estimate is high.
This example also illustrates how to deal with ows: People
move from one age to the next, leaving the ow (dying) at dierent
ages, on average at age 70 years. From that knowledge alone, it is
dicult to estimate the number of children under age 2 years; only
an actuarial table would give us precise information. Instead, we
invent a table that makes the calculation simple: Everyone lives
to the life expectancy, and then dies abruptly. The calculation
is simple, and the approximation is at least as accurate as the
approximation that every child uses diapers for exactly 2 years.
In a product, the error is dominated by the most uncertain factor;
you waste your time if you make the other factors more accurate
than the most uncertain factor.
1.1.5 Meteorite impacts
How many large meteorites hit the earth each year?
This question is not yet clearly dened: What does large mean?
When you explore a new eld, you often have to estimate such ill-
dened quantities. The real world is messy. You have to constrain
the question before you can answer it. After you answer it, even
with crude approximations, you will understand the domain more
clearly, will know which constraints were useful, and will know how
to improve them. If your candidate set of assumptions produce
a wildly inaccurate estimatesay, one that is o by a factor of
100,000then you can be sure that your assumptions contain a
fundamental aw. Solving such an inaccurate model exactly is a
waste of your time. An order-of-magnitude analysis can prevent
this waste, saving you time to create more realistic models. After
you are satised with your assumptions, you can invest the eort
to rene your model.
Sky&Telescope magazine reports approximately one meteorite
impact per year. However, we cannot simply conclude that only
one large meteorite falls each year, because Sky&Telescope pre-
sumably does not report meteorites that land in the ocean or in
the middle of corn elds. We must adjust this gure upward, by
a factor that accounts for the cross-section (eective area) that
Sky&Telescope reports cover (Figure 1.5). Most of the reports cite
impacts on large, expensive property such as cars or houses, and
are from industrial countries, which have N 10
9
people. How
m
F
R
Density
Mass M
M R
3
F =
GMm
R
2
Rm
Figure 1.6. Order-of-magnitude astro-
nomical body. The body, which we ap-
proximate as a sphere, has radius R and
uniform density . A block of mass m
sits on the surface and feels a gravi-
tational force F = GMm/R
2
, where
M R
3
is the mass of the astro-
nomical body. The resulting accelera-
tion is g = F/m = GR R; if is
the same for all astronomical bodies in
which were interested, then g R.
much target area does each persons car and living space occupy?
Her car may occupy 4 m
2
, and her living space (portion of a house
or apartment) may occupy 10 m
2
. [A country dweller living in a
ranch house presents a larger target than 10 m
2
, perhaps 30 m
2
.
A city dweller living in an apartment presents a smaller target
than 10 m
2
, as you can understand from the following argument.
Assume that a meteorite that lands in a city crashes through 10
stories. The target area is the area of the building roof, which
is one-tenth the total apartment area in the building. In a city,
perhaps 50 m
2
is a typical area for a two-person apartment, and
3 m
2
is a typical target area per person. Our estimate of 10 m
2
is
a compromise between the rural value of 30 m
2
and the city value
of 3 m
2
.]
Because each person presents a target area of a 10 m
2
, the
total area covered by the reports is Na 10
10
m
2
. The surface
area of the earth is A 4(610
6
m)
2
510
14
m
2
, so the reports
of one impact per year cover a fraction Na/A 2 10
5
of the
earths surface. We multiply our initial estimate of impacts by the
reciprocal, A/Na, and estimate 510
4
large-meteorite impacts per
year. In the solution, we dened large implicitly, by the criteria
that Sky&Telescope use.
1.2 Scaling analyses
In most of the previous examples, we used opportunistic tricks to
determine what numbers to multiply together. We now introduce
a new method, scaling, for problems where simple multiplication
is not sucient. Instead of explaining what a scaling argument is,
we rst make one, and then explain what we did. The fastest way
to learn a language is to hear and speak it. Physics is no exception;
you hear it in the examples, and you speak it in the exercises.
1.2.1 Gravity on the moon
What is acceleration due to gravity on the surface of the moon?
First, we guess. Should it be 1 cms
2
, or 10
6
cms
2
, or perhaps
10
3
cms
2
? They all sound reasonable, so we make the guess of
least resistancethat everywhere is like our local environment
and say that g
moon
g
earth
, which is 1000 cms
2
. Now we will
make a systematic estimate.
This method that we use eventually shows you how to make
estimates without knowing physical constants, such as the gravi-
tational constant G. First, we give the wrong solution, so that we
can contrast it with the rightand simplerorder-of-magnitude
solution. The acceleration due to gravity at the surface of the
moon is given by Newtons law of gravitation (Figure 1.6):
g =
F
m
=
GM
R
2
. (1.9)
In the wrong way, we look upperhaps in the thorough and useful
CRC Handbook of Chemistry and Physics [13]M and R for the
moon, and the fundamental constant G, and get
g
moon

6.7 10
8
cm
3
g
1
s
2
7.3 10
25
g
(1.7 10
8
cm)
2
160 cms
2
.
(1.10)
Here is another arithmetic calculation that you can do mentally,
perhaps saying to yourself, First, I count the powers of 10: There
are 17 (8 + 25) powers of 10 in the numerator, and 16 (8 +
8) in the denominator, leaving 1 power of 10 after the division.
Then, I account for the prefactors, ignoring the factors of 10. The
numerator contains 6.7 7.3, which is roughly 7 7 = 49. The
denominator contains 1.7
2
3. Therefore, the prefactors produce
49/3 16. When we include one power of 10, we get 160.
This brute-force methodlooking up every quantity and then
doing arithmeticis easy to understand, and is a reasonable way
to get an initial solution. However, it is not instructive. For exam-
ple, when you compare g
moon
160 cms
1
with g
earth
, you may
notice that g
moon
is smaller than g
earth
by a factor of only 6.
With the huge numbers that we multiplied and divided in (1.10),
g
moon
could easily have been 0.01 cms
2
or 10
6
cms
2
. Why are
g
moon
and g
earth
nearly the same, dierent by a mere factor of 6?
The brute-force method shows only that huge numbers among G,
M, and R
2
nearly canceled out to produce the moderate acceler-
ation 160 cms
2
.
So we try a more insightful method, which has the benet
that we do not have to know G; we have to know only g
earth
.
This method is not as accurate as the brute-force method, but it
will teach us more physics. It is an example of how approximate
answers can be more useful than exact answers.
We nd g
moon
for the moon by scaling it against g
earth
. [It
is worth memorizing g
earth
, because so many of our estimations
depend on its value.] We begin with (1.9). Instead of M and R,
we use density and radius R as the independent variables; we
lose no information, because we can compute density from mass
and radius (assuming, as usual, that the astronomical body has
the simplest shape: a sphere). We prefer density to mass, because
density and radius are more orthogonal than mass and radius. In a
thought experimentand order-of-magnitude analyses are largely
thought experimentswe might imagine a larger moon made out
of the same type of rock. Enlarging the moon changes both M
and R, but leaves alone. To keep M xed while changing R
requires a larger feat of imagination (we shatter the moon and
use scaolding to hold the fragments at the right distance apart).
For a sphere of constant density, M = (4/3)R
3
, so (1.9)
becomes
g R. (1.11)
This scaling relation tells us how g variesscaleswith density
and radius. We retain only those variables and factors that change
from the earth to the moon; the proportionality sign allows us
to eliminate constants such as G, and numerical factors such as
4/3.
If the earth and moon have the same radius and the same
average density of rock, then we can further simplify (1.11) by
eliminating and R to get g 1. These assumptions are not
accurate, but they simplify the scaling relation; we correct them
shortly. So, in this simple model, g
moon
and g
earth
are equal, which
partially explains the modest factor of 6 that separates g
moon
and
g
earth
. Now that we roughly understand the factor of 6, as a con-
stant near unity, we strive for more accuracy, and remove the most
inaccurate approximations. The rst approximation to correct is
the assumption that the earth and moon have the same radius. If
R can be dierent on the earth and moon, then (1.11) becomes
g R, whereupon g
earth
/g
moon
R
earth
/R
moon
.
What is R
moon
? Once again, we apply the guerrilla method.
When the moon is full, a thumb held at arms length will just
cover the moon perceived by a human eye. For a typical human-
arm length of 100 cm, and a typical thumb width of 1 cm, the angle
subtended is 0.01 rad. The moon is L 4 10
10
cm from the
earth, so its diameter is L 0.01L; therefore, R
moon
210
8
cm.
By contrast, R
earth
6 10
8
cm, so g
earth
/g
moon
3. We have al-
ready explained a large part of the factor of 6. Before we explain
the remainder, lets estimate L from familiar parameters of the
moons orbit. One of the goals of order-of-magnitude physics is to
show you that you can make many estimates with the knowledge
that you already have. Lets apply this philosophy to estimating
the radius of the moons orbit. One familiar parameter is the pe-
riod: T 30 days. The moon orbits in a circle because of the
earths gravitational eld. What is the eect of earths gravity at
distance L (from the center of the earth)? At distance R
earth
from
the center of the earth, the acceleration due to gravity is g; at L,
it is a = g(R
earth
/L)
2
, because gravitational force (and, therefore,
acceleration) are proportional to distance
2
. The acceleration re-
quired to move the moon in a circle is v
2
/L. In terms of the period,
which we know, this acceleration is a = (2L/T)
2
/L. So
g
_
R
earth
L
_
2
. .
a
gravity
=
_
2L
T
_
2
1
L
. .
a
required
. (1.12)
The orbit radius is
L =
_
gR
2
earth
T
2
4
2
_
1/3
_
1000 cms
2
(6 10
8
cm)
2
(3 10
6
s)
2
40
_
1/3
5 10
10
cm,
(1.13)
which closely matches the actual value of 4 10
10
cm.
Now we return to explaining the factor of 6. We have already
explained a factor of 3. (A factor of 3 is more than one-half of a
factor of 6. Why?) The remaining error (a factor of 2) must come
largely because we assumed that the earth and moon have the
same density. Allowing the density to vary, we recover the original
scaling relation (1.11). Then,
g
earth
g
moon

earth
moon
R
earth
R
moon
. (1.14)
Typically,
crust

moon
3 g cm
3
, whereas
earth
5 g cm
3
(here,
crust
is the density of the earths crust).
Although we did not show you how to deduce the density of
moon rock from well-known numbers, we repay the debt by pre-
senting a speculation that results from comparing the average den-
sities of the earth and the moon. Moon rock is lighter than earth
rock; rocks in the earths crust are also lighter than the average
earth rock (here rock is used to include all materials that make
up the earth, including the core, which is nickel and iron); when
the earth was young, the heavier, and therefore denser, elements
sank to the center of the earth. In fact, moon rock has density
close to that of the earths crustperhaps because the moon was
carved out of the earths crust. Even if this hypothesis is not true,
it is plausible, and it suggests experiments that might disprove
it. Its genesis shows an advantage of the scaling method over the
brute-force method: The scaling method forces us to compare the
properties of one system with the properties of another. In making
that comparison, we may nd an interesting hypothesis.
Whatever the early history of the moon, the density ratio con-
tributes a factor of 5/3 1.7 to the ratio (1.14), and we get
g
earth
/g
moon
3 1.7 5. We have explained most of the factor
of 6as much of it as we can expect, given the crude method that
we used to estimate the moons radius, and the one-digit accuracy
that we used for the densities.
The brute-force methodlooking up all the relevant numbers
in a tabledefeats the purpose of order-of-magnitude analysis.
Instead of approximating, you use precise values and get a precise
answer. You combine numerous physical eects into one equa-
tion, so you cannot easily discern which eects are important.
The scaling method, where we rst approximate the earth and
moon as having the same density and radius, and then correct
L
v
Brick wall
Figure 1.7. Order-of-magnitude car
about to hit a brick wall. It hits with
speed v, which provides kinetic energy
Mv
2
, where M is the mass of the car.
The energy required to distort a xed
fraction of the bonds is proportional to
the number of bonds. If toy and real cars
are made of the same metal, then the
number of atoms, and the total bond-
distortion energy, will be proportional
to M, the mass of the car. The avail-
able kinetic energy also is proportional
to M, so the necessary crash velocity is
the same at all masses, and, therefore,
at all sizes.
the most inaccurate assumptions, teaches us more. It explains
why g
moon
g
earth
: because the earth and moon are made of
similar material and are roughly the same size. It explains why
g
moon
/g
earth
1/6: because moon rock is lighter than earth rock,
and because the moon is smaller than the earth. We found a series
of successive approximations:
g
moon
g
earth
,
g
moon

R
moon
R
earth
g
earth
,
g
moon

moon
earth
R
moon
R
earth
g
earth
.
(1.15)
Since the approximations each introduce only one physical eect,
they are easy to understand. Another benet of the scaling method
is that it can suggest new theories or hypotheses. When we con-
sidered the density of moon rock and earth rock, we were led to
speculate on the moons origin from the earths crust. Order-of-
magnitude reasoning highlights the important factors, so that our
limited brains can digest them, draw conclusions from them, and
possibly extend them.
1.2.2 Collisions
Imagine that you work for a government safety agency testing how
safe various cars are in crashes. Your budget is slim, so you rst
crash small toy cars, not real cars, into brick walls. (Actually, you
might crash cars in computer simulation only, but, as the order-
of-magnitude analysis of computer programs is not the topic of
this example, we ignore this possibility.) At what speed does such
a crash produce mangled and twisted metal? Metal toy cars are
still available (although hard to nd), and we assume that you are
using them.
For our initial guess, lets estimate that the speed should be
roughly 50 mph or 80 kphroughly the same speed that would
badly mangle a real car (mangle the panels and the engine com-
partment, not just the fenders). Why does a crash make metal
bend? Because the kinetic energy from the crash distorts the
metallic bonds. We determine the necessary crash speed using a
scaling argument.
Figure 1.7 shows a car about to hit a brick wall. In an order-of-
magnitude world, all cars, toy or real, have the same proportions,
so the only variable that distinguishes them is their length, L.
(Because we are assuming that all cars have the same proportions,
we could use the width or height instead of the length.) The kinetic
energy available is
E
kinetic
Mv
2
. (1.16)
The energy required to distort the bonds is
E
required

M
m
atom
. .
no. of atoms
c
f, (1.17)
where
c
is the binding, or cohesive, energy per atom; and f is a
fractional fudge factor thrown in because the crash does not need
to break every bond. We discuss and estimate cohesive energies
in Section 3.2.2; for now, we need to know only that the cohesive
energy is an estimate of how strong the bonds in the substance
are. Lets assume that, to mangle metal, the collision must break
a xed fraction of the bonds, perhaps f 0.01. Equating the
available energy (1.16) and the required energy (1.17), we nd
that
Mv
2
M

c
m
atom
f. (1.18)
We assume (reasonably) that
c
, f, and m
atom
are the same for
all cars, toy or real, so once we cancel M, we have v 1. The
required speed is the same at all sizes, as we had guessed.
Now that we have a zeroth-order understanding of the prob-
lem, we can improve our analysis, which assumed that all cars have
the same shape. The metal in toy cars is proportionally thicker
than the metal in real cars, just as roads on maps are proportion-
ally wider than real roads. So a toy car has a larger mass, and is
therefore stronger than the simple scaling predicts. The metal in
full-size cars mangles in a 80 kph crash; the metal in toy cars may
survive an 80 kph crash, and may mangle only at a signicantly
higher speed, such as 200 kph.
Our solution shows the benet of optimism. We do not know
the fudge factor f, or the cohesive energy
c
, but if we assume
that they are the same for all cars, toy or real, then we can ig-
nore them. The moral is this: Use symbols for quantities that
you do not know; they might cancel at the end. Our example il-
lustrated another technique: successive approximation. We made
a reasonable analysisimplicitly assuming that all cars have the
same shapethen improved it. The initial analysis was simple,
and the correction was almost as simple. Doing the more accurate
analysis in one step would have been more dicult.
1.2.3 Jump heights
We next apply scaling methods to understand how high an animal
can jump, as a function of its size. We study a jump from stand-
ing (or from rest, for animals that do not stand); a running jump
m
m
h
E
jump
mgh
Figure 1.8. Jumping animal. An ani-
mal of mass m (the block) stores energy
in its muscles (the compressed, mass-
less spring). It uses the energy to jump
a height h o the ground. The energy
required is E
jump
mgh.
depends on dierent physics. This jump-height problem also looks
underspecied. The height depends on how much muscle an ani-
mal has, how ecient the muscles are, what the animals shape is,
and much else. So we invoke another order-of-magnitude method:
When the going gets tough, lower your standards. We cannot
easily gure out the absolute height; we estimate instead how the
height depends on size, leaving the constant of proportionality to
be determined by experiment. In Section 1.2.3.1, we develop a
simple model of jumping; in Section 1.2.3.2, we consider physical
eects that we neglected in the crude approximations.
1.2.3.1 Simple model for jump height. We want to determine only
how jump height scales (varies) with body mass. Even this prob-
lem looks dicult; the height still depends on muscle eciency,
and so on. Lets see how far we get by just plowing along, and using
symbols for the unknown quantities. Maybe all the unknowns can-
cel. We want an equation for the height h, such as h f(m), where
m is the animals mass. Jumping requires energy, which must be
provided by muscles. [Muscles get their energy from sugar, which
gets its energy from sunlight, but we are not concerned with the
ultimate origins of energy here.] If we can determine the required
energy, and compare it with the energy that all the muscles in
an animal can supply, then we have an equation for f. Figure 1.8
shows a cartoon version of the problem.
A jump of height h requires energy E
jump
mgh. So we can
write
E
jump
mh. (1.19)
The sign means that we do not have to worry about making
the units on both sides match. We exploited this freedom to get
rid of the irrelevant constant g (which is the same for all animals
on the earth, unless some animal has discovered antigravity). The
energy that the animal can produce depends on many factors.
We use symbols for each of these unknowns. First, the energy
depends on how much muscle an animal has. So we approximate
by assuming that a fraction, , of an animals mass is muscle, and
that all muscle tissue can store the same energy density, E (we are
optimists). Then, the energy that can be stored in muscles is
E
stored
mE m. (1.20)
Here we have derived a scaling relation, showing how energy stored
varies with mass; we used the freedom provided by to get rid
of and E, presumed to be the same for all animals. Equating
the required energy from (1.19) with the available energy from
(1.20), we nd that mh m, or that h 1; this proportionality
says that h is independent of mass. This result seems surprising.
Our intuition tells us that people should be able to jump higher
than locusts. Table 1.1 shows measured jump heights for animals
of various sizes and shapes; the data are also plotted in Figure 1.9.
Flea
Click beetle
Locust
Human
10
3
10
1
10
5
10
30
60
Mass (g)
Height (cm)
Figure 1.9. Jump height versus body
mass. This graph plots the data in Ta-
ble 1.1. Notice the small range of vari-
ation in height, compared to the range
of variations in mass. The mass varies
more than 8 orders of magnitude (a fac-
tor of 10
8
), yet the jump height varies
only by a factor of 3. The predicted scal-
ing of constant h (h 1) is surprisingly
accurate. The largest error shows up at
the light end; eas and beetles do not
jump as high as larger animals, due to
air resistance.
Animal Mass (g) Height (cm)
Flea 0.5 10
3
20
Click beetle 0.04 30
Locust 3 59
Human 7 10
4
60
Table 1.1. Jump height as a function of
mass. Source: Scaling: Why Animal Size
is So Important [17, page 178].
Surprising or not, our result is roughly correct.
1.2.3.2 Extensions of the simple model. Now that we have a crude
understanding of the situationthat jump height is constant
we try to explain more subtle eects. For example, the scaling
breaks down for tiny animals such as eas; they do not jump as
high as we expect. What could limit the jump heights for tiny
animals? Smaller animals have a larger surface-to-volume ratio
than do large animals, so any eect that depends on the surface
area is more important for a small animal. One such eect is air
resistance; the drag force F on an animal of size L is F L
2
, as we
show in Section 2.3.3. The resulting deceleration is F/m L
1
,
so small animals (small L) get decelerated more than big animals.
We would have to include the constants of proportionality to check
whether the eect is suciently large to make a dierence; for
example, it could be a negligible eect for large animals, and 10
times as large for small animals, but still be negligible. If we made
the estimate, we would nd that the eect of air resistance is
important, and can partially explain why eas do not jump as high
as humans. The constant jump height also fails for large animals
such as elephants, who would break their bones when they landed
if they jumped as high as humans.
You might object that treating muscle simply as an energy
storage medium ignores a lot of physics. This criticism is valid, but
if the basic model is correct, its simpler to improve the model one
step at a time, instead of including every eect in the early stages.
As an example of successive renement, lets consider the power
requirements for jumping. How does the power required scale with
animal size, and do limitations on power prevent animals from
attaining their theoretical jump height?
Power is energy per time; in this case, it is energy required
for the jump divided by time during which the energy is released.
In (1.19) we found that E mh; because h is constant, E m.
[Successive renement, which we are doing here, depends on an
at least rudimentary understanding of the problem. If we had not
already solved the problem crudely, we would not know that E
m or that h 1.]
We now need to estimate the time required to release the en-
ergy, which is roughly the time during which the animal touches
the ground while launching. Suppose that the animal blasts o
with velocity v. The animal squats to zero height, the clock starts
ticking, and the animals starts to push. At the end of the push,
when the clock stops ticking, the animal is moving with speed v; we
assume that it moves with the same speed throughout its launch
(the rectangle assumption). The clock, therefore, stops ticking at
time L/v. The takeo speed v is roughly the same for all
animals, because v
gh
h, and h is roughly constant. So

L.
How does the energy vary with L? We make the simplest
assumptionthat all animals have the same density and the same
cubical shape. Then, E m, and m L
3
, so E L
3
.
From our estimates for the energy and the time, we estimate
that the power required is P E/ L
2
. Per unit volume, the
power required is P
req
L
1
. If there is a maximum power per
unit volume, P
max
, that an animal can generate, then suciently
tiny animalsfor whom P
req
is largemight not be able to gener-
ate sucient power. Click beetles overcome this problem by stor-
ing energy in their exoskeleton, and jumping only after they have
stored sucient energy: They increase the eective , and thus
decrease P
req
.
The analysis of this extreme casetiny animalsand the anal-
ysis of the power requirements show the value of making a simple
analysis, and then rening it. To complete the more detailed anal-
ysis, we required results from the simple analysis. If we had tried
to include all factorssuch as air resistance, bone breakage, power
consumption, and energy storagefrom the beginning, we would
have cooked up a conceptual goulash, and would have had trouble
digesting the mess. The approximate model provides a structure
on which we can build the more detailed analyses.
1.3 What you have learned
You now know a basic repertoire of order-of-magnitude techniques:
Divide and conquer: Split a complicated problem into manage-
able chunks, especially when you must deal with tiny or huge
numbers, or when a formula naturally factors into parts (such
as V l w h).
Guess: Make a guess before solving a problem. The guess may
suggest a method of attack. For example, if the guess results in
a tiny or huge number, consider using divide and conquer. The
guess may provide a rough estimate; then you can remember
the nal estimate as a correction to the guess. Furthermore,
guessingand checking and modifying your guessimproves
your intuition and guesses for future problems.
Talk to your gut: When you make a guess, ask your gut how
it feels. Is it too high? Too low? If the guess is both, then its
probably reliable.
Lie skillfully: Simplify a complicated situation by assuming
what you need to know to solve it. For example, when you do
not know what shape an object has, assume that it is a sphere
or a cube.
Cross-check: Solve a problem in more than one way, to check
whether your answers correspond.
Use guerrilla warfare: Dredge up related facts to help you
make an estimate.
Punt: If youre worried about a physical eect, do not worry
about it in your rst attempt at a solution. The productive
strategy is to start estimating, to explore the problem, and
then to handle the exceptions once you understand the domain.
Be an optimist: This method is related to punt. If an assump-
tion allows a solution, make it, and worry about the damage
afterward.
Lower your standards: If you cannot solve the entire problem
as asked, solve those parts of it that you can, because the sub-
problem might still be interesting. Solving the subproblem also
claries what you need to know to solve the original problem.
Use symbols: Even if you do not know a certain valuefor
example, the energy density stored in muscledene a symbol
for it. It may cancel later. If it does not, and the problem is
still too complex, then lower your standards.
We apply these techniques, and introduce a few more, in the chap-
ters to come. With a little knowledge and a repertoire of tech-
niques, you can estimate many quantities.
23
2 Dimensional Analysis
In this chapter, you learn a general method of coping with com-
plicated equations without actually solving them: dimensional
analysis. Because the method is fastmuch faster than nding
an honest solution would beyou can use it to discard unpromis-
ing approaches early, sparing you time to think of new approaches.
Rather than bore you with a theoretical discussion of dimensional
analysis, we illustrate the method with examples.
The idea of dimensional analysis is that unitsbe they meters,
feet, seconds, pints, fortnights, or furlongsare articial. The uni-
verse cares not for our choice of units. Valid physical laws must
have the same form in any system of units. Only dimensionless
quantitiespure numbersare the same in every unit system, so
we write equations in a universe-friendly, dimensionless form. Of-
ten, there is only one such form. Then, without doing any work,
we have solved the problem.
2.1 Newtons law
For example, suppose that we live in Newtons time. Lacking his
insight, we do not realize that force (F), mass (m), and accelera-
tion (a) are related by F = ma. We know, perhaps from experi-
ments, only that F, m, and a are related. Our problem is to nd
that way. The most general relation between the three variables
is
h(F, m, a) = 0, (2.1)
where h is a function that we must nd.
We use dimensional analysis to nd h. Dimensional analysis
tells us that we can rewrite the general form (2.1) using only
quantities without units. With luck, there is only one such form:
If we nd it, we obtain a law of nature.
2.1.1 Finding dimensionless groups
To get rid of the units, we need rst to know them. In cgs units, the
units of force are g cms
2
; in a general masslengthtime system
of units, the units are
[F] = [M][L][T]
2
. (2.2)
We introduce the useful notation that [M] stands for a unit of
mass, [L] for a unit of length, and [T] for a unit of time. Also,
[x] stands for the units of the enclosed variable x; context distin-
guishes the two usages of []. Similarly, the units of acceleration
are
[a] = [L][T]
2
, (2.3)
2. Dimensional Analysis 24
0 2 4 6
3
2
1
0
1
2
3
x = F/ma
f(x)
Figure 2.1. Zero-crossings of a random
function f. They are located at 4.894,
5.61392, 1.01346, 1.9288, 2.4041, and
4.12762 and are marked with dots.
1. A good table of functions of one
variable may require a page; that of a
function of two variables a volume; that
of a function of three variables a book-
case; and that of a function of four vari-
ables a library.
H. Jeries [14, p. 82]
and the units of mass are
[m] = [M]. (2.4)
We can combine these three variables into a quantity with no
unitsa dimensionless group. An example of an illegal group
is F/a, which has units of mass. Legal groups include
F
ma
,
_
F
ma
_
2
, sin
_
F
ma
_
, and
ma
F
. (2.5)
These legal groupsall legal groupsare functions of F/ma. They
are also functions of ma/F or 2F/ma. It does not matter which
base form we choose, so we choose F/ma. This form does not
contain an arbitrary constant (a weakness of 2F/ma), which is
superuous in a dimensional analysis. This form also makes later
manipulations most familiar. The general form of a dimensionless
group is therefore
f
_
F
ma
_
, (2.6)
where f is a dimensionless function: It takes in dimensionless num-
bers and spits out a dimensionless number.
2.1.2 Finding the dimensionless relation
The most general relation using (2.6) is
f
_
F
ma
_
= 0. (2.7)
This form is the dimensionless version of (2.1). To make sure that
we have not gone astray, lets check that (2.7) does not exclude
what we know to be the right answer: F = ma. If we choose
f(x) = x 1, then (2.7) becomes F/ma = 1, or F = ma.
We have made progress. The function h has three parameters;
f has only one.
1
We can further simplify (2.7), because we need
to nd only the zeros of f; we do not need to nd f itself. We
have sketched a dimensionless function in Figure 2.1. Its six zero-
crossings determine the possible values of F/ma.
Which zero is correct? In some problems, more than one zero
is possible; for example, an electron in a hydrogen atom can have
an innity of possible energies. In this problem, only one zero is
possiblea fact that we could discover from experiment; dimen-
sional analysis cannot decide this question for us. We call the
unknown zero , in honor of the Buckingham Pi theorem (Theo-
rem 2.1). The solution of (2.7) is F/ma = , or
F = ma. (2.8)
This result is Newtons second law. We deduced it without know-
ing much physics. [This example is unfair to Newton, because it
implies that he was dense for not instantly realizing that F = ma.
However, we took for granted much knowledge that was unknown
in Newtons time. At that time, the concept of forcelet alone its
unitswas not clearly understood, and it was the great triumph
of Newton that he gave an implicit denition of force in F = ma.]
Even with our extra knowledge, we have not matched Newton.
Dimensional analysis does not tell us the value of . It cannot,
because numbers, such as , have no units; they are therefore in-
visible to dimensional analysis, which cares about only the units
of a quantity, not its magnitude. We could do an experiment to
determine ; if we did, we would nd that = 1. Much of this
text assumes (or hopes) that is near 1. We often use dimensional
analysis to solve problems for which we cannot easily determine
; at the end, we pretend that = 1. The closer is to unity,
the more accurate is the resulting estimate.
2.1.3 Summarizing the argument
Lets extract the general style of argument that we used in this
example; we use a similar pattern in many order-of-magnitude
analyses. What did we have to know before we could begin? We
had to know which variables were important: F, m, and a. In this
example, the variables were given by at, as part of the thought
experiment. In more realistic examples, we have to think care-
fully to choose the variables: Choosing them is the hardest part
in dimensional analysis. Aristotle thought that force, mass, and
velocity were the relevant variables; it took almost 2000 years be-
fore scientists realized that acceleration, rather than velocity, was
the relevant variable.
If we leave out a necessary variable, we invite trouble. For ex-
ample, suppose that we use only F and a. No combination of F
and a is dimensionless; this failure is a clue that weve left out
a necessary variablein this case, the mass. We can also make
more subtle mistakes. Suppose that we correctly realize that mass
should be a variable, but we incorrectly think that the relevant
mass is the mass of the earth m
, because of a vague hunch about

gravity. We form the dimensionless group F/m
a, and nd that
F = m
a. Although dimensionally correct by construction, this

equation is empirically bogus. Newtons law is accurate in far re-
gions of the solar system, where the earths gravity, and m
, are
?
Grouper
Relation
nder
Simplier
F
ma
f
_
F
ma
_
= 0
F, [M] [L] [T]
2
m, [M]
a, [L] [T]
2
F = ma
Figure 2.2. The Pi machine. After
lucky guessing, we decide what vari-
ables to feed the grouper. We feed them
in, along with their units. The grouper
hunts for combinations that have no
units (for dimensionless groups), and
spits them out (here, there is only one
group). That group enters the relation
nder, which produces the most general
dimensionless relation from its inputs.
The simplier simplies this equation,
presenting the answer F = ma.
irrelevant. You must think physically, appeal to experiment, and
make lucky guesses if you are to include all relevant variables.
We do not want to include every possible variable, however,
because irrelevant variables multiply the possibilities for dimen-
sionless relations. For example, suppose that, to hedge our bets,
we include both m and m
. Then, there are two dimensionless

groups, F/ma and F/m
a. The most general relation using these

groups is
f
_
F
ma
,
F
m
a
_
= 0, (2.9)
where f is a dimensionless function of two variables. We now have
to use physics knowledge to restrict the form of f. Here, we add
the knowledge that the mass of the earth is irrelevant, because
Newtons law works in the far solar system; then, we recover the
simpler relation f(F/ma) = 0 that we derived before, but, this
time, we had to work harder. Section 2.3.2 contains another ex-
ample of the excess-variable problem; there, we use knowledge of
uid mechanics to simplify the resulting relations. We know of no
recipe for choosing the right set of variables, except to practice.
[We sloppily used the same letter f to denote two dierent
functions: In (2.7), it is a function of one variable, yet in (2.9), it
is a function of two variables. The number of parameters disam-
biguates the two uses. For maximum clarity, we should call the two
functions by dierent names, such as f
1
and f
2
. In dimensional-
analysis problems, however, we often start with, say, a three-
variable function; then use physics knowledge to reduce it to a
two-variable function; then use more knowledge to reduce it to a
one-variable function. The sloppy notation is therefore convenient.
We avoid redundant subscripts, and we do not need to invent ar-
bitrary names such as f, g, and h.]
Once we chose the variables, we found the only dimensionless
group (apart from transformations) F/ma. The general relation,
f(F/ma) = 0 simplied to F = ma. The unknown constant we
can determine by experiment, or from other knowledge. Figure 2.2
illustrates the steps of the argument.
2.1.4 Generalizing the argument: the Buckingham Pi theorem
The art of dimensional analysis lies in choosing the right set of rel-
evant variables; a computer would nd this stage dicult, if not
impossible. The part that we could programnding the dimen-
2. You may object that we have person-
ied the pendulum bob, that we have en-
dowed it with the capacity for feeling, at
least, for feeling forces. Even worse, we
have become a pendulum bob. We plead
guilty, with an explanation. We person-
ify the bob, because picturing an active
bob enhances our intuition for how it
behaves; to make a vivid picture of an
active bob, we even pretend to be a bob.
Similarly, looking for a lost marble, we
often ask ourselves, If I were a mar-
ble and somebody dropped me, where
would I roll? This style may not en-
hance your intuition; try it for month as
an experiment.
sionless groupswe normally do by guessing. In this example, we
started with three relevant variables, composed of two basic units,
[M], and [L][T]
2
. It looks like the variables contain units[M],
[L], and [T]but the single combination [L][T]
2
suces to re-
place [L] and [T]. As an analogy, imagine three functionsa, b,
and cwith the forms
a(x, y, z) = xyz
2
,
b(x, y, z) = x,
c(x, y, z) = yz
2
.
(2.10)
Here, a, b, and c play the roles of [F], [m], and [a], respectively;
and x, y, and z play the roles of [M], [L], and [T], respectively. The
variables y and z occur in only the combination yz
2
. So there
are only two independent variables: x and yz
2
. In the F = ma
example, there are only two independent units: [M] and [L][T]
2
.
From n = 3 variables, composed of r = 2 units, we found n r =
1 dimensionless group. This pattern is general, and is stated in
Theorem 2.1.
Theorem 2.1 (Buckingham Pi theorem) A system described by
n variables, built from r independent dimensions, is also described
by n r independent dimensionless groups.
We also call the groups variables. Now we apply this method
to examples with more than one dimensionless group, where the
argument is more complex.
2.2 Pendula
The next example is familiar to many of you from freshman-
physics problem sets. A pendulum bob hangs at the end of a
massless rope, and, from a resting state, the bob starts oscillating
(Figure 2.3). What is the oscillation period?
2.2.1 Finding dimensionless groups
Following the pattern of Section 2.1, we rst choose the relevant
variables. We make sure to include the variable for which were
trying to solve, the period ; a common mistake is to forget the
goal variable, and thus to end up one variable short. The period
may depend on the initial angle, so we include
0
. To decide which
other variables to include, we imagine life as a pendulum bob. Why
do we move at all? We move because of gravity, so g belongs on
our list. A more massive bob feels a stronger gravitational force,
so we include m in our list.
2
From , g, and m,
0
four variables with three unitswe
can form one dimensionless group,
0
. We cannot form one that
includes (try to do so); so we really have only three variables,
and zero dimensionless groups. We must have left out a variable. A
parameter of the system that we have not mentioned is the length
of the rope. Perhaps that variable will rescue us, so we include it
in our list. Table 2.1 contains the variables that we have collected.
Var. Units Description
0
angle of release
m [M] mass of bob
[T] period
g [L][T]
2
acceleration due to gravity
l [L] length of rope
Table 2.1. Variables that may deter-
mine a pendulums oscillation period.
m
l
0
F = mg sin
0
Figure 2.3. A pendulum bob of mass m
hangs from a massless rope of length l.
The bob is released from rest at an angle
0
.
3. Any club that would admit me as a
member, I wouldnt want to join.
Groucho Marx
(From previous experience with this problem, you know that m
does not matter, and that l does; we will use those facts as a sanity
check on our solution.)
We now form dimensionless groups. Finding these groups re-
quires that we have handy the units of the variables, which is why
we listed the variables in a table. These n = 5 variables contain
r = 3 basic dimensions: length, mass, and time. The Bucking-
ham Pi theorem says that we can form n r = 2 independent
dimensionless groups from this set. Guesswork is the easiest way
to nd them. One original variable,
0
, is already dimensionless
(why are angles dimensionless?), so we start with
1

0
. Only
one original variable, m, contains a unit of mass. Therefore, no di-
mensionless group can contain m. [To be dimensionless, the group
has to contain another variable that cancels the unit of mass con-
tributed by m; however, m is the only mass in this problem, so m
must cancel itself from any group that it tries to join.
3
] We form
the second independent dimensionless variable using g, l, and .
[If we did not require that the n r variables be independent,
then we could choose, for example, the useless pair
0
and
2
0
.]
The units of g show us how to form the group: Cancel the length
in g with l
1
, and cancel the time with
2
, to make
2
g
2
/l.
The two groups are then
1
=
0
,
2
=
g
2
l
.
(2.11)
2.2.2 Finding dimensionless relations
The most general relation using the groups in (2.11) is
f(
1
,
2
) = 0, (2.12)
where f is a dimensionless function. What is f? Figure 2.4 outlines
the argument that we use nd it.
In Section 2.1.2, we rewrote f(
1
) = 0 as
1
= , where
1
= F/ma was the dimensionless group, and was an unknown
constantthe zero of f that is experimentally correct. With the
more complicated relation (2.12), we can make the same transfor-
mation. The two-parameter function f is equivalent to a family
of one-parameter functions, f
1
, indexed by
1
(Figure 2.5). The
general form (2.12) is then
f
1
(
2
) = 0. (2.13)
?
Grouper
Relation
nder
Simplier
, [T]
0
, -
m, [M]
g, [L] [T]
2
l, [L]
0
g
2
l
f
_
0
,
g
2
l
_
= 0
g
2
l
=

f(
0
)
Figure 2.4. Dimensional analysis for
the pendulum period. After thinking,
we choose ve relevant variables. They
enter the grouper. It counts the inputs
(ve) and the number of independent
units in the inputs (three), and spits out
two dimensionless groups. The relation
nder produces the most general rela-
tion using the two groups. The simpli-
er transforms the relation into a form
more convenient for solving for .
In Section 2.1.2, where f was a one-parameter function, we trans-
formed f(
1
) = 0 to
1
= . Here, we transform f
1
(
2
) = 0
to
2
=

f(
1
), (2.14)
where

f(
1
) is the zero of f
1
(Figure 2.6). (This transformation
is valid when f
1
has only one physically valid zero, which we
assume for simplicity.) We have extracted
2
from its hiding place
in f. Alternatively, we could have written f(
1
,
2
) = f
2
(
1
),
and have dened

f(
2
) as the zero of f
2
. Then, we could have
extracted
1
:
1
=

f(
2
). (2.15)
We choose which form to use by the location of the goal variable. If
it belongs to
2
, we use (2.14). If it belongs to
1
, we use (2.15). If
it belongs to both, then we have to think. If it belongs to neither,
then we have too few starting variables and dimensionless groups.
We wish to solve for the period, which is contained in
2
=
g
2
/l, so we use the form (2.14). In terms of the original variables,
(2.14) is g
2
/l =

f(
0
), or
=

f(
0
)
l
g
. (2.16)
So depends on l but not on m (the promised sanity check). We
do not know the function

f(
0
). When we derived Newtons law in
Section 2.1.2, we could not, using dimensional analysis, determine
the constant (or zero-parameter function) . Here, we cannot de-
termine the one-parameter function

f. We can determine it by
experiment: We release a pendulum at various
0
, and measure
(
0
). Then,

f is
f(
0
) = (
0
)
_
g
l
. (2.17)
We do not have to repeat the experiments for dierent l (say,
for another pendulum) or g (say, on another planet), because

f
is universal. All pendulumslong or short, on the earth or on
Marsobey the same

f.
0 3 6 9
1 0 1 2
10
5
0
5
10
15
2
f
1
(x)
Figure 2.5. The two-parameter function
f(
1
,
2
) = e
2
+
2

1
, plotted
for various
1
. The result is a family of
functions, f
1
(
2
). Their zero-crossings
are marked with dots, and labeled with
the values of
1
.
In the small-amplitude limit (
0
0), we can simplify (2.16):
=

f(0)
l
g
, (2.18)
where

f(0) is a constant. In speaking of

f(0), we are tacitly as-
suming that lim
x0

f(x) exists. Dimensional analysis makes no
promises that it does; we have to appeal to physics. Here is a scal-
ing argument that makes the existence of the limit plausible. We
estimate the oscillation time by estimating the acceleration and
the oscillation distance. The pendulum feels a force F mg sin
0
(see Figure 2.3), which makes it accelerate at a g sin
0
. For
small
0
, the acceleration is a g
0
, because sin
0

0
for small
0
. In time , it travels a distance a
2
g
0
2
. It needs to travel
a distance d l
0
(neglecting constants) to complete a cycle, so
g
0
2
l
0
. (2.19)
The amplitude
0
cancels, which is the physical equivalent of the
mathematical statement that

f(0) exists. We even get an estimate
for :

l
g
, (2.20)
which is a physical derivation of (2.18). Alternatively, we can es-
timate a typical velocity for the bob, and from the velocity, esti-
mate the period. The maximum potential energy is PE mgh,
where the change in height is h = l(1 cos
0
) l
2
0
. The max-
imum kinetic energy is equal to the maximum potential energy.
So the maximum velocity is given by mv
2
mgl
2
0
. (This rela-
tion equates two energies, potential and kinetic. This method is
an example of balancing; we discuss balancing more fully in Sec-
tion 3.1.2.) The maximum velocity is then v
0
gl; we use this

value as an estimate for a typical velocity. The time to complete
a cycle is d/v
_
l/g, as we found using the acceleration
method.
0 2 4 6 8 10
0.5
0
0.5
1
1.5
2
1
Zero of
f
1
(x)
Figure 2.6. Zero of f
1
(
2
), as a func-
tion of
1
. The function f(
1
,
2
) is
from Figure 2.5.
Density
Viscosity
sp
R
v
Figure 2.7. Sphere falling in a liquid.
Its terminal velocity is v.
We call

f(0) the constant , and (2.18) becomes
=
l
g
. (2.21)
We cannot evaluate with dimensional analysis. It turns out to
be 2, but we would either have to solve an equation to determine
itand solving equations is beyond the scope of the textor have
to determine it experimentally, by timing one pendulum swing.
We release the pendulum from a small anglesay,
0
= 0.1 rad
and measure the period, . Knowing , l, and g, we use (2.21) to
determine .
2.3 Drag in uids
Every freshman-physics course solves the pendulum dierential
equation to nd (2.21). The freshman course usually continues
beyond where dimensional analysis can compete: The solution to
the dierential equation contains the value of . The pendulum
problem does not show the benet of dimensional analysis. So we
try our hand at uid mechanicsa subject notorious for its math-
ematical and physical complexity; Chandrasekhars books [3, 4]
show how complex the mathematics can become. As usual, we
study examples. The examples are a marble falling through corn
syrup (Section 2.3.2), and a person falling from the sky after being
thrown from an airplane (Section 2.3.3). We calculate the termi-
nal velocity: The velocity that the object reaches after falling
for suciently long. The examples illustrate two extremes of uid
ow: oozing and turbulent.
We begin by solving the partial-dierential equations of uid
mechanics (for the incompressible ow of a Newtonian uid):
v
t
+ (v)v =
1
p +
2
v,
v = 0.
(2.22)
Here, v is the uid velocity eld, is the uid density, is the
kinematic viscosity, and p is the pressure. We have to solve six
4. Viscosity is only skin deep.
coupled, nonlinear partial-dierential equations, and to nd the
velocity of the marble as t . A miserable task.
Do not worrywe will not solve these equations. We put them
there to scare you into using dimensional analysis, to show you
that it is easier than solving equations.
So lets use dimensional analysis to nd the terminal velocity,
v. We could start with our usual method: Choose a group of vari-
ables that includes v, form dimensionless groups, and then solve
for the velocity. This direct approach would gives us too many
dimensionless groups; we would have to add physics knowledge at
the end, to simplify the resulting dimensionless relations. Instead,
we use physics knowledge at the beginning, to simplify the deriva-
tion early on. You save eort if you simplify early. If you simplify
later, you carry around baggage that you eventually discard. If
youre going on holiday to the Caribbean, why pack snow shoes?
The adjective terminal in the phrase terminal velocity hints
at the physics that determines the velocity. It indicates that the
velocity is constant, which can happen only if there is no net force
on the marble. What forces act on the marble? There is gravity,
which speeds it up. There is drag, which slows it down. To nd the
terminal velocity, then, we nd the drag and gravitational forces,
and equate them. We have split the terminal-velocity problem
into two simpler problems (divide and conquer). The gravitational
force is mg (or close to that, as we will see), where m is the mass
of the marble. What is the drag force?
2.3.1 Dimensional analysis for the drag force
We use dimensional analysis to nd the drag force, F
d
. The rst
step is to choose the relevant variables. We begin by including the
variable for which were solving, F
d
. What characteristics of the
sphere are relevant to the drag force? The drag force has no idea
what is inside the sphere. Picture the uid as a huge computer
that implements the laws of uid dynamics. How can it know how
dense the sphere or person is? It cannot see inside the object. To
the drag force, the parameters v and R are the only relevant at-
tributes of a sphere moving through the atmosphere. What lies
underneath the persons surface does not aect the uid ow.
4
So
the computer can determine the ow (if it has tremendous pro-
cessing power) knowing only v and R. Therefore,
sp
is irrelevant.
What characteristics of the uid are relevant? The uid supercom-
puter needs to know the density of the uid, to determine how fast
the pieces of uid move because of the object. So we include
.
What about viscosity? Viscosity is a measure of the tendency
of a uid to smear velocity dierences in the ow. You can observe
an analogue of viscosity in trac ow on a multilane highway. If
one lane moves much faster than another, drivers switch from
the slower to the faster lane, eventually slowing down the faster
lane. Local decisions of the drivers reduce the velocity gradient.
Similarly, in a uid, molecular motion transports speed (really,
5. Sadly, we could not use the more
melliuous term uid mechanics to
signify a host of physicists agonizing
over the equations of uid mechanics;
it would not distinguish the toilers from
their toil.
F
d
[M][L][T]
2
drag force
R [L] marble radius
[L]
2
[T]
1
kinematic viscosity
[M][L]
3
uid density
v [L][T]
1
terminal velocity
Table 2.2. Variables that determine the
drag force.
momentum) from fast- to slow-owing regions. This transport re-
duces the velocity dierence between the regions. Thicker, oozier
(more viscous) uids probably cause more drag than thin uids
do. So we include viscosity in our list.
Fluid mechanicians
5
have dened two viscosities, dynamic vis-
cosity , and kinematic viscosity . They are related by =
v.
Life in Moving Fluids [18, pp. 2325] discusses the two types of
viscosity in detail. In Section 4.5, we estimate the viscosity of
gases by examining the molecular origin of viscosity. For the anal-
ysis of drag force, we need to know only that viscous forces are
proportional to viscosity. Which viscosity should we use? Dynamic
viscosity hides
inside the product
; equations that contain
sp
and then look less dimensionless than they are, because
sp
has no partner. So we use the kinematic viscosity, .
Table 2.2 lists the variables that we have collected. The drag
force does not depend on
sp
or g; it is simpler to estimate than
is the terminal velocity (which does depend on g and
sp
). The
ve variables in the list are composed of three basic dimensions;
we look for two dimensionless groups. We nd one group by the
method of divide and conquer. The list already includes a velocity,
v. If we can concoct another quantity, V , also with units of veloc-
ity, then we can form the dimensionless group v/V . The viscosity
almost works. It has an extra length, which R can eliminate:
V /R. The dimensionless group is
1

v
V
=
vR
, (2.23)
which is the Reynolds number, Re. It is a dimensionless mea-
sure of the ow speed. [In Section 2.3.3, we give a more detailed
physical interpretation of Re.] The velocity alone cannot distin-
guish fast from slow ow, because v is not dimensionless (see the
discussion of large cargos in Section 1.1.1). If you hear that a
quantity is small, or fast, or large, or whatever, your rst reaction
should be, compared to what? To distinguish fast from slow
ow, we have to use a dimensionless quantity related to vthe
Reynolds number. It compares v to V . Low values of Re indicate
slow, viscous ow (cold honey oozing out of a jar); high values
indicate turbulent ow (a jet ying at 600 mph). We once again
refer you to the excellent Life in Moving Fluids [18], this time for
more details on dimensionless ratios in uid mechanics. We intro-
duce another dimensionless ratiothe Froude numberwhen we
discuss speedboating in Section 5.2.1.2.
What is the second dimensionless group? The drag force is
absent from
1
, so it has to be part of
2
(else how can we solve
for F
d
?). Instead of dreaming up the dimensionless group in one
lucky guess, we construct two quantities with the same units, and
divide them to get a dimensionless quantity. Notice that F
d
R is
an energy. We construct another energy (call it E), then form the
dimensionless ratio F
d
R/E. Energy contains one power of mass;
the only variable other than F
d
that contains mass is
. So E
contains one power of
. How can we make an energy from
?
We start with
R
3
, which is a mass. Kinetic energy is mv
2
/2, so it
has units of [M] [V]
2
. [A formula is a reliable and simple method
for determining the units of a quantity. Even if the formula is
approximatemv
2
/2 is valid at speeds only much less than the
speed of lightit still tells you the correct units.] The second
energy is therefore
E
R
3
. .
mass
v
2
. (2.24)
The group
2
is the ratio of the two energies:
2

F
d
R
E
=
F
d
R
2
v
2
. (2.25)
The most general relation is f(
1
,
2
) = 0. We want to solve
for F
d
, which is contained in
2
, so we use the (2.14):
2
=

f(
1
).
In terms of the original variables, the drag force is
F
d
=
R
2
v
2

f
_
vR
_
. (2.26)
Now we are stuck; we can go no farther using only dimensional
analysis. To learn the form of

f, we specialize our problem to
extreme conditions: to turbulent, high-speed ow (Re 1), or to
viscous, low-speed ow (Re 1).
2.3.2 Viscous limit
We rst analyze the low-speed limit: a marble falling in corn syrup.
You may wonder how often marbles fall in corn syrup, and why we
bother with this example. The short answer to the rst question
is, not often. However, the same physics that determines the
behavior of marbles in syrup also determines, for example, the
behavior of fog droplets in air, of bacteria swimming in water,
or of oil drops in the Millikan oil-drop experiment. We study the
marble problem because it illustrates the physical principles, and
because we can check our estimates with a home experiment.
In slow, viscous ows, the drag force comes from viscous forces,
which are proportional to . Therefore, F
d
. The viscosity ap-
pears exactly once: in the argument of

f, and in the denominator.
So that ips into the numerator, we choose

f(x) 1/x. [We
used to avoid writing the constant repeatedly. The symbol
means that the two sides have the same units (none in this case),
but that they may dier by a dimensionless constant.] With this
f(x), the drag force (2.26) becomes
F
d
=
R
2
v
2

vR
=
vR, (2.27)
where is a dimensionless constant.
Sadly, we cannot compute the magic factor using dimen-
sional analysis. We must call a uid mechanic to do the messy
work of calculating it; but we can hope that her burden is light,
because we have worked out the solution except for one number.
Stokes rst derived the value of , and found that = 6 (for a
sphere). With this value, (2.27) becomes
F
d
= 6
vR. (2.28)
Lets pause to sanity check this result. Larger or faster marbles
should feel more drag; and a denser or more viscous uid should
produce more drag. The drag force (2.28) passes these tests.
We can make a more detailed check by explaining the result
physically. The argument is similar in style to the argument that
we made in Section 2.2.2 to justify
_
l/g for a pendulum.
In the viscous limit, drag is caused by viscosity only: The drag
force is the viscous force. To estimate the viscous force, we have
to introduce more knowledge about viscosity. We also work in the
reference frame of the sphere. Viscosity is a consequence of the no-
slip boundary condition: At the surface of the sphere, the uid
is stationary (relative to the sphere). Far away from the sphere,
the uid does not know about the sphere: The uid moves as if
the sphere were not there, with velocity v (relative to the sphere).
Viscosity reduces this velocity dierence by transporting momen-
tum from the sphere to the distant uid: It slows the sphere. The
momentum transport produces a stress

(roughly, a force per
area) that is proportional to dynamic viscosity and to velocity
gradient:
u, (2.29)
where u is the uid velocity as a function of distance from the
spheres center. The velocity gradient u is a complicated object:
a tensor. We can pretend that it is an ordinary number (a scalar),
and approximate its magnitude using two data points: u = 0 near
the sphere, and u = v far away from the sphere. How far is far
away? Equivalently, we can ask: Far relative to what length? The
natural length to use as a standard of comparison is R, the radius
of the sphere. So lets say that u(R) = 0 and u(2R) = v. We could
also use u(10R) = v or u(8R) = v; the constant that exactly
characterizes far away is not important, because we are going
to ignore it anyway. Because u changes by v in a length R, its
gradient is approximately v/R. This result is a consequence of a
general order-of-magnitude rule of thumb:
f
typical value of f
length over which f changes signicantly
. (2.30)
This rule of thumb is the method of order-of-magnitude dif-
ferentiation. [One way to remember it is to write
df
dx

f
x
, (2.31)
once we cancel the ds.] The viscous stress is therefore
v/R.
The viscous force is the stress times the surface area over which
Higher p
Buoyant force
Lower p
Sphere
Figure 2.8. Gravity causes buoyancy.
Because of gravity, the pressure at the
bottom of the sphere (submerged in a
uid that is not shown) is greater than
the pressure at the top. The result is an
upward force: buoyancy.
it acts:
F
viscous

v/R
. .
R
2
..
area

vR. (2.32)
This result agrees with the estimate result (2.27), which we got
from dimensional analysis. We neglected constants throughout
this physical argument, which is therefore incapable of justifying
the factor of 6 in (2.28).
This magic factor of 6 in (2.28) comes from doing honest
physics, here, from solving the NavierStokes equations (2.22). In
this book, we wish to teach you how to avoid suering, so we do
not solve such equations. Occasionally, we quote the factor that
honest physics produces, to show you how accurate (or sloppy)
the order-of-magnitude approximations are. The honest-physics
factor is usually of order unity. Such a number suits our neural
hardware: It is easy to remember and to use. If we know the order-
of-magnitude derivation and remember this one number, we can
reconstruct the exact result without solving dicult equations.
We now return to the original problem: nding the terminal
velocity, which we estimate by balancing drag against gravity. The
gravitational force is F
g
= mg, where m is the mass of the marble.
Instead of using m, we use
sp
the density of the marble. We can
recover the mass from
sp
and R if we need it. We prefer to use
density, because the drag force contains a density,
. We therefore
write the gravitational force as
F
g

sp
gR
3
, (2.33)
where we have neglected the 4/3 factor. We equate the drag force
from (2.27) with the gravitational force:
vR
sp
gR
3
. (2.34)
The terminal velocity is
v =
gR
2
sp
. (2.35)
We can increase our condence in this expression by checking
whether the correct variables are upstairs (a pictorial expression
meaning in the numerator) and downstairs (in the denominator).
Denser marbles should fall faster than less dense marbles, so
sp
should live upstairs. Gravity accelerates marbles, so g should also
live upstairs. Viscosity slows marbles, so should live downstairs.
The terminal velocity (2.35) passes these tests. We therefore have
more condence in our result (although the tests did not check
the exponents, or the location of R).
Our result is correct, except that we neglected an important
physical eect: buoyancy. Equivalently, we forgot one dimension-
less group:
3

/
sp
. Buoyancy reduces the eective weight of
the marble. Equivalently, it reduces the value of g. If
sp
=
,
then the eective g is 0; if
= 0, then gravity retains its full

eect. Between these two limits, the eective g varies linearly
with
. Why must the variation be linear? Gravity makes pres-

sure in a uid increase with depth; this increase causes buoyancy
(Figure 2.8). So gravity is responsible for buoyancy. Gravitational
forceand therefore buoyant forceis proportional to
. So the
eective g should vary linearly with
. Therefore, we make the

transformation
g g
_
1

sp
_
. (2.36)
The gravitational force (2.33) becomes
F
g
(
sp
)gR
3
, (2.37)
and the terminal velocity becomes
v
gR
2
sp
1
_
. (2.38)
If we had carried through the constants of proportionality, starting
with the magic 6 from the drag force (2.28) and including the
4/3 that belongs in the gravitational force (2.33), then we would
have found that the missing constant of proportionality in the
terminal velocity (2.38) is 2/9.
Using (2.38), we tried a home experiment to determine the
viscosity of corn syrup. We dropped a glass marble with (R =
0.8 cm) in a tube of corn syrup, and measured v 2 cms
1
. With
the magic 2/9 factor, the viscosity is
=
2
9
gR
2
v
_
sp
1
_
. (2.39)
The density of glass is
sp
2 g cm
3
, and that of corn syrup is
sp
1 g cm
3
. With all the numbers, the viscosity estimate is
0.2
1000 cms
2
(0.8 cm)
2
2 cms
2
1 60 cm
2
s
1
. (2.40)
Many tables list , for which we predict 60 g cm
1
s
1
(because
sp
1 g cm
3
).
A World Wide Web search for the viscosity of corn syrup
yielded 60 g cm
1
s
1
for deionized corn syrup, and 25 g cm
1
s
1
for regular corn syrup. Presumably, the corn syrup that we pur-
chased from the grocery store was merely regular corn syrup. Our
estimate is therefore in error by roughly a factor of 2. The modest
size of the factor indicates that we have included much of the rel-
evant physics in our model. In this case, we have captured much
of the physics (although we did insert the 2/9 by at); the major
uncertainty is in the viscosity of our corn syrup. (Another source
of error is the eect of the tube wall; this order-of-magnitude anal-
ysis assumed that the tube is innitely wide, so that the marble
feels no eects from the wall.)
In deriving the drag force (2.28), we assumed that the drag
force was proportional to which it is indeed in slow ows, or in
highly viscous liquids. Is our ow suciently slow? The Reynolds
numberdened in (2.23)is the correct measure of ow speed.
Lets make sure that Re is small:
Re
vR

2 cms
1
0.8 cm
60 cm
2
s
1
0.03, (2.41)
which is 1. The ow oozes; so our assumption is self-consistent.
2.3.3 Turbulent limit
We now compute drag in the other ow extreme: high-speed, or
turbulent, ow. We estimate the terminal velocity for a person
thrown from a high-ying airplane. How high is the airplane y-
ing? High enough so that the person reaches terminal velocity. You
may reasonably question whether people frequently get thrown
from airplanes. Fortunately, they do not. However, our results are
generic. For example, even when a child rises from a chair, the
airow around her is high-speed ow. Say that the child is 20 cm
wide, and that she rises with velocity 50 cms
1
. Then,
Re
vR
air
50 cms
1
20 cm
0.2 cm
2
s
1
5000, (2.42)
which is 1. The ow is turbulent. Larger objects, such as adults
and cars, create turbulence even when they travel only 10 cms
1
.
Most uid owlarge raindrops falling in air, ships traveling in
water, cyclists racing in airis turbulent.
We now specialize the results of Section 2.3.1 to high-speed
ow. To begin our analysis, we lie: We assume that a person is a
sphere. It is a convenient approximation. Even if people are cylin-
drical in everyday life, a person thrown out of an airplane might,
from a sensible fear reex, tuck into fetal position, and vindicate
our approximation. At high speeds (more precisely, at high Re),
the ow is turbulent. Viscositywhich aects only slow ows, but
does not inuence the shearing and whirling of turbulent ows
becomes irrelevant. Lets see how much we can understand about
turbulent drag, knowing only that drag is nearly independent of
viscosity. The viscosity appears in only
1
. If viscosity disappears,
so does
1
. This argument is slightly glib. More precisely, we re-
move from the list in Table 2.2, and search for dimensionless
groups. From four variables, we can nd one dimensionless group:
the
2
from the marble example.
Why is drag independent of Re at high speeds? Equivalently,
why can we remove from the list of variables and still get the
correct form for the drag force? The answer is not obvious. Our
construction of Re in (2.23)as a ratio of v and V provides a
partial answer. A natural length in this problem is R, so from each
velocity, we can form a quantity with units of time:
v

R
v
,
V

R
V

R
2
.
(2.43)
Object c
d
Sphere 0.5
Cylinder 1.0
Flat plate 2.0
Car 0.4
Table 2.3. Approximate drag coe-
cients (c
d
) of objects in high-speed ow
(Re 1). The cylinder moves perpen-
dicular to its axis; the at plate moves
perpendicular to its surface.
Note that Re
V
/
v
. The quantity
v
is the time that uid takes
to travel around the sphere (apart from constants). As we discuss
in Section 4.5, kinematic viscosity is the diusion coecient for
momentum, and the time for momentum to diuse a distance x is
x
2
/. So
V
is the time that momentum takes to diuse around
the sphere (apart from constants). If
V

v
in which case,
Re 1then momentum diuses before uid travels around the
sphere. Momentum diusion equalizes velocities, if it has time.
In this low-Re limit, it does. It prevents ow at the front from
being radically dierent from the ow at the back. There is no
turbulence. In the other limit
V

v
, or Re 1momentum
diusion gets soundly beaten by uid ow, so the uid is free to
do whatever it wants: It is turbulent. The degree to which uid
ow beats momentum diusion does not aect the ow; the ow
is already unconstrained by viscosity. (This statement is a lie, but
its reasonably accurate.)
The dimensionless group that remains is
2
=
F
d
R
2
v
2
. (2.44)
The most general relation is f(
2
) = 0, so
2
= , and
F
d

R
2
v
2
. (2.45)
We computed the drag force for a sphere; what about other
shapes? So that the drag force generalizes to more complex shapes,
we express it using the cross-sectional area of the objecthere,
A = R
2
, so F
d

Av
2
. The constant of proportionality is called
c
d
/2, where c
d
is the drag coecient. The drag coecient is a
dimensionless measure of the drag force, and it depends on the
shape of the object (on how streamlined the object is). Table 2.3
lists c
d
for various shapes. (The drag coecient is proportional to
our function

f(Re).) With the drag coecient, the drag force is
F
d

1
2
c
d
v
2
A. (2.46)
We For most purposes, we can assume that c
d
1. The drag
coecient varies slightly with Re, so we lied a little when we said
that, at high Re, drag is independent of viscosity (which appears
in the Reynolds number). The lie is suciently accurate for most
order-of-magnitude purposes.
Now that we have an expression for the drag force, we estimate
the gravitational force; computing it requires far less work. The
gravitational forceincluding the buoyant forcedoes not care
whether the ow is turbulent. So the result for viscous ow (2.37)
applies here. For your convenience, we restate that result:
F
g
(
sp
)gR
3
. (2.47)
To nd the terminal velocity, we equate F
g
with the drag force
(2.45):
R
2
v
2
g(
sp
)R
3
. (2.48)
6. Gravity partially determines at-
mospheric pressure and density.
Holding the atmospheric density
constant while increasing gravity
might be impossible in real life, but
we can do it easily in a thought ex-
periment.
The terminal velocity is
v
_
gR
sp
. (2.49)
We pause to sanity check this result. Are the right variables
upstairs and downstairs? We consider each variable in turn.
: The terminal velocity is smaller in a denser uid (try run-

ning in a swimming pool), so
should be in the denominator.

g: Imagine a person falling on a planet that has a gravitational
force stronger than that of the earth.
6
The drag force does not
depend on g, so gravity wins the tug of war, and v increases:
g should be upstairs.
sp
: Imagine a person who has twice normal density (but has
normal size). The gravitational force doubles. The drag force
stays the same, as we saw using the uid-is-a-computer argu-
ment in Section 2.3.1. So
sp
should be upstairs in (2.49).
R: To determine where the radius lives, we require a more
subtle argument, because changing R changes the gravitational
force and the drag force. The gravitational force increases as
R
3
; the drag force increases as R
2
. At larger R, gravity wins;
greater gravity means greater v, so R should be live upstairs.
: At high speeds, viscosity does not aect drag, so should
appear neither upstairs nor downstairs.
The velocity (2.49) passes all tests.
At last, we can compute the terminal velocity for the (in-
voluntary) skydiver. We assume that he tucks into a fetal po-
sition, making him roughly spherical, with, say, R 40 cm. A
person is mostly water, so
sp
1 g cm
3
. The density of air is
10
3
g cm
3
, so

sp
; buoyancy is not an important ef-
fect. In (2.49), we can neglect
in
sp
. We put these numbers

into the terminal velocity (2.49):
v
_
1000 cms
2
40 cm
1 g cm
3
10
3
g cm
3
_
1/2
6 10
3
cms
1
,
(2.50)
or 120 mph (200 kph).
This calculation assumed that Re 1. We now check that
assumption. We do not need to calculate Re from scratch, because,
as we worked out in (2.42), a child (R 20 cm) rising from her
chair (v 50 cms
1
) creates a ow with Re 5000; this ow is
turbulent. The ow created by the skydiver is faster by a factor
of 100, and the skydiver is larger by a factor of 2, so the ow is
certainly turbulent. Scaling the Reynolds number from (2.42), we
get
Re Re
child
v
skydiver
v
rising
R
skydiver
R
child
5000 100 2 = 10
6
,
(2.51)
which is 1. The ow is extremely turbulent.
Re 1
Re
vR
F
d
v
2
R
2
v
_
gR
sp
1
3
4 2
Assume Derive
Calculate Check
Figure 2.9. Correct solution order for
terminal velocity in turbulent ow. By
starting in the assume box, we sim-
plied the solution procedure. Step 1:
On that assumption, we estimated the
drag force (the derive box). Step 2: From
the drag force, we estimated the termi-
nal velocity (the calculate box). Step
3: From the terminal velocity, we esti-
mated the Reynolds number (the check
box). Step 4: To close the loop, we ver-
ied the starting condition. Note: For
compactness, the terminal-velocity for-
mula ignores the normally small eect
of buoyancy.
Re 1
Re
vR
F
d

Rv
v
gR
2
sp
1
3
4 2
Assume Derive
Calculate Check
Figure 2.10. Correct solution order for
terminal velocity in viscous ow. Note
the similarities with Figure 2.9.
7. We agree with the Fowlers [10] and
Dupre [6]: that is restrictive, which
is not. If we had said . . .from honest
physics that we did not do, we would
imply that, somewhere in this book, we
did do honest physics. We plead not
guilty.
Now that we have found the terminal velocity, lets extract the
pattern of the solution. The order that we followed was assume,
derive, calculate, and check (Figure 2.9). This order is more power-
ful than the usual order of derive and calculate. Without knowing
whether the ow is fast or slow, we cannot derive a closed-form
expression for F
d
(such a derivation is probably beyond present-
day understanding of turbulence). Blocked by this mathematical
Everest, we would remain trapped in the derive box; we would
never determine F
d
; and we would never realize that Re is large
(the assume box), which is the assumption that makes it possible
to estimate F
d
. We used the same, correct order when we solved
for the terminal velocity in corn syrup; Figure 2.10 shows how
similar the logic is.
2.3.4 Combining solutions from the two limits
We computed the drag force in two limits. Lets compare the re-
sults, to see whether we can combine them. We can write the
solution for viscous ow (2.28) in terms of the Reynolds number
and cross-sectional area:
F
d
=
1
Av
2

f(Re). (2.52)
We showed that

f(x) = /x, where the magic constant = 6
comes from honest physics, which we did not do.
7
So
F
d
=
6
Re
Av
2
. (2.53)
Compare this viscous drag force to the turbulent drag force (2.46).
If we substitute c
d
= 12/Re, then (2.46) becomes (2.53). So we
can use the drag formula (2.46) for both high- and low-speed ow,
provided that we let c
d
vary with Re. For high Re, we quoted
c
d
1/2. If the low-Re approximation were valid at all Reynolds
numbers, then c
d
would cross 1/2 near Re 24, at which point
we would expect the high-Re approximation to take over. The
crossing point is a reasonable estimate for the transition between
low- and high-speed ow. Experimental data place the crossover
nearer to Re 5, at which point c
d
2. If a dimensionless vari-
able, such as Re, is close to unity, calculations become dicult.
Approximations that depend on a quantity being either huge or
tiny are no longer valid. When all quantities are roughly the same
size, you cannot get rid of many terms in your equations. To get
results in these situations, you have to do honest work: You must
solve the nonlinear equations (2.22) numerically or analytically.
Every valid physical equation can be written in a form without
units. To nd such forms, you follow these steps:
1. Write downby magic, intuition, or luckthe physically
relevant variables. For illustration, lets say that there are
n of them.
2. Determine the units of each variable. Count how many in-
dependent dimensions these variables comprise. Call this
number r. Usually, length, mass, and time are all that you
need, so r = 3.
3. By playing around, or by guessing with inspiration, nd
n r independent dimensionless combinations of the vari-
ables. These combinations are the dimensionless groups, or
Pi variablesnamed for the Buckingham Pi theorem.
4. Write down the most general relation using these groups.
Then try to eliminate dimensionless groups, or to restrict
the form of the relation, using physical arguments.
5. Using physical arguments, simplify the dimensionless rela-
tion by eliminating dimensionless groups, or by otherwise
constraining the form of the relation.
In dimensionless form, even uid mechanics is simple.
The Reynolds number is a dimensionless measure for the ow
speed; it distinguishes viscous from turbulent ow, and com-
pares the momentum transport time to the uid transport
time.
When you solve a dicult problemsuch as computing the
drag forcesimplify by assuming an extreme case: Assume
that one or more of the dimensionless variables are nearly zero
or nearly .
Using your solution, check your assumption!
In the next two chapters, we use these dishonest methods, along
with physical reasoning, to estimate the properties of materials.
43
3 Materials I
+
e
a
0
Figure 3.1. Hydrogen. A proton (the
+) holds an electron in jail. The jail
radius is a
0
, the Bohr radius.
Why are there are no 100 km mountains on earth (Section 3.3.2)?
Why must a Thanksgiving turkey cook all morning (Example 4.2)?
How large are white dwarfs (Section 3.4)? The answers to these
questions depend on the mechanical and thermal properties of
materials. We estimate these properties by using the techniques of
the previous chapters, and introduce a new technique, balancing.
In this chapter, we study mechanical properties of materials: their
density, strength, and surface tension. In Chapter 4, we study
thermal properties of materials.
Our estimates of material properties depend on the atomic
theory. In Feynmans words:
If, in some cataclysm, all of scientic knowledge were to be de-
stroyed, and only one sentence passed on to the next generations
of creatures, what statement would contain the most informa-
tion in the fewest words? I believe it is the atomic hypothesis
(or the atomic fact, or whatever you wish to call it) that all
things are made of atomslittle particles that move around in
perpetual motion, attracting each other when they are a little
distance apart, but repelling upon being squeezed into one an-
other. In that one sentence, you will see, there is an enormous
amount of information about the world. . . [8, p. 1-2]
The atomic theory was rst stated by Democritus. Bertrand Rus-
sell, himself a scientist and philosopher, describes early Greek sci-
ence and philosophy with wit, sympathy, and insight in his His-
tory of Western Philosophy [16]. Democritus could not say much
about the properties of atoms. With our modern knowledge of
classical and quantum mechanics, we can go farther. We begin
by determining how large atoms are. After we understand what
forces determine the sizes of atoms, we extend our understanding
to molecules: We divide and conquer.
3.1 Sizes
We begin our study of atoms with a simple picturealways draw
a simple pictureof the simplest atom, hydrogen (Figure 3.1).
What is the radius of the orbit?
3.1.1 Dimensional analysis
We estimate the radius rst using dimensional analysis. We make
a list of relevant variables by considering what physics determines
the properties of hydrogen. The electrostatic force holds the elec-
tron in orbit, so e, the charge on the electron, is certainly on the
list. We use e
2
instead of e, because, in cgs units, e
2
has familiar
3. Materials I 44
a
0
[L] radius
e
2
[E][L] charge
2
[E][T] quantum mechanics
m
p
[M] proton mass
m
e
[M] electron mass
size of hydrogen. The notation [E] is a
shorthand for [M][L]
2
[T]
2
.
dimensions: [E][L]. If we had used e, then we would wonder how
to combine esu with more familiar dimensions. [In mks units, we
would have to introduce the permittivity of free space,
0
, along
with the charge. Symbolic results in electrostatics often easier to
understand in cgs units than in mks units. Numerical estimates are
often simpler in mks units, because tables quote values in ohms,
or Coulombs, rather than in statohms or esu. If you want both to
think and to calculate, become comfortable with both systems.]
The mass of the electron m
e
determines the acceleration that the
electrostatic force produces, so we include m
e
. The mass of the
proton, m
p
, might also be relevant. From these four variables
a
0
, e
2
, m
e
, and m
p
we can form one dimensionless group,
1

m
e
m
p
. (3.1)
That group does not contain a
0
; we are stuck. Our failure to nd
a suitable dimensionless group tells us that weve left out physics
that determines a
0
.
We get a clue to this missing physics, because the picture of
hydrogen in Figure 3.1 cannot be classical. A classical electron,
moving in a circle, would radiate, lose energy, and spiral into its
doom as the proton swallowed it; classical atoms cannot exist. For-
tunately, quantum mechanics comes to the rescue. Unfortunately,
quantum mechanics is a large, fearsome subject. Thanks to dimen-
sional analysis, we do not have to understand quantum mechanics
to compute results based on it. For the purposes of dimensional
analysis, the whole content of quantum mechanics is contained in
another constant of nature: , whose value is 1.05 10
27
erg s. By
adding to our four variables, we get the list in Table 3.1.
One more variable produces one more dimensionless group. We
are sunk if this group does not contain a
0
, so lets assume that
it does. How can we make a length from the other four variables,
to cancel the length in a
0
? We divide and conquer. Because
_
e
2
is [E][L], we make an energy from e

2
, , and m
e
(we could also
have used m
p
, but using both m
p
and m
e
allows us no additional
units), and use the constructed energy to cancel the energy in
[E][L]. Notice that e
2
/ is a velocity, so m
e
(e
2
/)
2
is an energy,
with which we can cancel energy from e
2
. We do so, and construct
the length
e
2
m
e
(e
2
/)
2
, (3.2)
which simplies to
2
/m
e
e
2
. The new dimensionless group is a
0
divided by this length:
2
=
a
0
2
/m
e
e
2
. (3.3)
The other group is
1
= m
e
/m
p
, so we write
2
= f(
1
), where
f is a dimensionless function, or, equivalently,
a
0
=

2
m
e
e
2
f
_
m
e
m
p
_
. (3.4)
3. Materials I 45
The proton is much more massive than the electron, so we exagger-
ate and pretend that m
e
/m
p
= 0; we simplify by exaggerating.
If f(x) has a limit as x 0, then (3.4) becomes
a
0
=

2
m
e
e
2
, (3.5)
where the constant is f(0). This important result deserves a
cross-check, which we perform in Section 3.1.2.
Before we do the cross-check, we discuss the validity of assum-
ing that f(0) exists. If we had constructed
2
from m
p
instead of
from m
e
, then f(x) would be 1/x, which does not have a nite
limit. How can you decide which mass avoids this problem, short
of trying both, or getting lucky? There is no foolproof procedure,
but we can formulate rules of thumb.
First, choose dimensionless groups with minimal overlap. In
the denition of
2
, we could have used m
p
+ m
e
instead of m
e
.
This choice would make m
p
and m
e
appear in two dimensionless
groups; the choice that we madeto use m
e
in
2
, and m
e
and m
p
in
1
has less overlap among the original quantities. When we
consider the limit of f, we usually hope to eliminate one dimen-
sionless group. This elimination is more likely if there are fewer
overlaps.
Second, decide which variable is likely to vanish, and segre-
gate it in one dimensionless group. In this example, the proton
and electron mass are so disparate (their ratio is 2000) that
we immediately imagine a world where the disparity is more pro-
nounced. There are many such imaginary worlds; m
p
1000 kg
and m
e
1 g, or perhaps m
p
10
50
g and m
e
10
60
g. These
examples are unreasonable. If possible, we vary only one quantity,
so as to tell the fewest lies about the actual world. In our imag-
inary world, should we reduce m
e
or increase m
p
? To decide, we
use physical reasoning. The momenta of the proton and the elec-
tron are equal (and opposite) with, say, magnitude P. A typical
kinetic energy is E P
2
/m. The mass is downstairs, so the heav-
ier particle (the proton) contributes much less energy than the
lighter particle contributes. We therefore imagine that m
p
gets
larger; that choice eliminates only a minor contributor to the to-
tal energy. This argument shows that hydrogen has a physical
limit when m
p
and m
e
stays xed. Therefore, f(m
e
/m
p
)
has a limit at 0. What would happen if we had not segregated
m
p
? Suppose, for example, that we chose
1
= a
0
m
p
e
2
/
2
and
2
= m
e
/m
p
. As with our previous pair of dimensionless groups,
we would have
1
= g(
2
), where g is a dimensionless function,
or
a
0
=

2
m
p
e
2
g(m
e
/m
p
). (3.6)
Now g behaves badly. As m
p
, the factor in front goes to 0; so
g must go to innity. To make progress in spite of gs intransigence,
we import the knowledge that a
0
is independent of m
p
as m
p

3. Materials I 46
1. In an order-of-magnitude estimate,
we ignore details, such as the precise
path that an object follows, and replace
them by a typical size and, perhaps,
speed. Quantum mechanics justies this
slothfulness as a principle of physics.
E = 0
e
2
/2r
e
2
/r
KE e
2
/2r
x r
Figure 3.2. Order-of-magnitude hydro-
gen. We replace the Coulomb potential
(dashed curve) by a box potential (solid
line) with the same characteristic width
and height. This rectangle trick is famil-
iar from Section 1.1.4. The total energy
of the electron (thick line) is the sum
of the potential energy, e
2
/r, and the
kinetic energy, e
2
/2r.
. Therefore, we must have g(x) 1/x, so that g(x) can cancel
the m
p
that sits in front. If we segregate m
p
to begin withas
we did in (3.1) and (3.3)then we can simply discard
1
without
going through this more involved argument.
3.1.2 Physical meaning
For the second computation of a
0
, we use a method that substi-
tutes physical reasoning for dimensional analysis. We get a phys-
ical interpretation of the result (3.5). We want to nd the orbit
radius with the lowest energy (the ground state) of hydrogen. The
energy is a sum of kinetic and potential energy: kinetic from mo-
tion of the electron, and potential from electrostatic attraction.
What is the origin of the kinetic energy? The electron does not
orbit in any classical sense. If it did, it would radiate its energy and
spiral into the nucleus. Instead, according to quantum mechanics,
the proton connes the electron to some region of size r.
1
Con-
nement gives the electron energy according to the uncertainty
principle, which says that xp , where x is the position
uncertainty, and p is the momentum uncertainty of the electron.
In hydrogen, x r, as shown in Figure 3.2, so p /r. The
kinetic energy of the electron is
KE
(p)
2
m
e

2
m
e
r
2
. (3.7)
This energy is the connement energy, or the uncertainty
energy. We use this idea many times in this book; our next ap-
plication for it is in Section 3.4.
The potential energy is the classical expression
PE
e
2
r
. (3.8)
Here the sign indicates the electron is not precisely at a radius
r; rather, its typical, or characteristic, distance is r.
The total energy is the combination
E = PE + KE
e
2
r
+

2
m
e
r
2
. (3.9)
The two energies compete. At small r, kinetic energy wins, because
of the 1/r
2
; at large r, potential energy wins, because it goes to
zero less rapidly. Is there a minimum combined energy at some
intermediate value of r? There has to be. At small r, E/r is
negative; at large r, it is positive; at an intermediate r, the E/r
must be zero. The potential has a minimum at this r. The location
is easy to estimate if we write (3.9) in dimensionless form; such a
rewriting is not necessary in this example, but it is helpful in more
complicated examples. In deriving (3.3), we constructed a length,
l =
2
/m
e
e
2
. So lets dene scaled length as r r/l. Furthermore,
e
2
/l is an energy, so lets dene scaled energy as

E E/(e
2
/l).
Then (3.9) becomes
E
1
r
+
1
r
2
. (3.10)
3. Materials I 47
We can nd the minimum-energy r by calculus, which gives r
min
=
2. Or, we can use order-of-magnitude minimization: When
two terms compete, the minimum occurs when the terms are
roughly equal (Figure 3.3). We therefore equate r
1
and r
2
;
we get r
min
1. In normal units, the minimum-energy radius
is r
min
= l r
min
=
2
/m
e
e
2
, which is exactly the Bohr radius; in
this case, the sloppiness in denition cancels the error in order-of-
magnitude minimization.
To justify this method, consider a reasonable general form for
E:
E(r) =
A
r
n

B
r
m
. (3.11)
This form captures the important feature of (3.9): The two parts
compete. To nd the minimum, we solve E
(r
min
) = 0, or
n
A
r
n+1
min
+m
B
r
m+1
min
= 0. (3.12)
The solution is
A
r
n
min
=
n
m
B
r
m
min
. (3.13)
The order-of-magnitude method minimizes (3.11) by equating the
two terms A/r
n
and B/r
m
, so this method cannot derive the n/m
factor in (3.13). The ratio of the two estimates for r
min
is
order-of-magnitude estimate
calculus estimate

_
n
m
_
1/(mn)
, (3.14)
which is < 1 if n > m. So the order-of-magnitude method underes-
timates minima and overestimates maxima (for maxima, the same
argument carries through, except that m and n change places).
The potential between nonpolar atoms and molecules is well
approximated with m = 6 and n = 12:
U(r)
A
r
12

B
r
6
. (3.15)
This potential is known as the LennardJones potential. It de-
scribes the interaction of two nonpolar atoms (or molecules). Such
atoms (for example, helium or xenon) have no permanent dipole
moment. However, the charge distribution around the nucleus
(and therefore the dipole moment) uctuates. This uctuation fa-
vors opposite uctuations in the other atom. The induced dipole
moments have opposite directions; therefore, the atoms attract
each other, no matter what the sign of the uctuation. Lets use
this physical picture to explain the r
6
dependence in the attrac-
tive part of the potential. We begin by estimating the force on a
dipole of moment sitting in an electric eld E. (In this para-
graph, E is an electric eld, not an energy. Sorry!) A dipole is
an ideal version of a positive and a negative charge (+q and q)
separated by a small distance d; the dipole moment is the product
= qd. The force on a dipole is therefore a sum of the force on
the positive charge and on the negative charge. So
F q (E(r +d) E(r)) . (3.16)
3. Materials I 48
Estimate
Actual
Total
KE
PE
1 2 3 4 5
0
0.5
1
1.5
2
r
E Figure 3.3. Order-of-magnitude calcu-

lus: minimizing scaled energy

E versus
scaled bond length r. The scaled energy
is the sum of potential and kinetic en-
ergy. In absolute value, potential and
kinetic energy are equal at r
min
= 1,
so the order-of-magnitude minimum is
at r
min
= 1. Honest calculus produces
r
min
= 2.
Here, E(r) is the eld at a distance r from the other atom. Because
d is small, this force is approximately qdE
(r), or E
(r), where
E
(r) = E/r. We assume that E r

b
, for some constant b;
then E
(r) E(r)/r and F E/r. What is the dipole moment?

The dipole is induced by the eld of the other atom; that eld
polarizes this atom. So = aE, where a is the polarizability of
the atom (a characteristic of the atom that is independent of eld,
as long as the eld is not too strong). Then F aE
2
/r. What
is the eld E? The eld from a dipole is proportional to r
3
, so
F r
7
. Therefore, U r
6
, which explains the attractive part
of the LennardJones potential. The r
12
term represents short-
range repulsion; it is a t to experimental data, and cannot be
estimated with these simple arguments.
For the LennardJones potential, the ratio n/m is 2, so the
order-of-magnitude estimate of r
min
is in error by a factor of 2
1/6
1.1. Here is another example: Minimize the function f(r) = e

r
+
1/r
2
. The minimum is at r
0
0.925 (the solution of e
r
= 2/r
3
),
whereas the order-of-magnitude method predicts that r
0
0.7
(the solution of e
r
= 1/r
2
). So even when the two contributors
to f are not power laws, the order-of-magnitude method gives a
reasonable answer. [To understand the range of validity of this
method, construct functions for which it is grossly inaccurate.]
Now we return to the original problem: determining the Bohr
radius, for which order-of-magnitude minimization predicts the
correct value. Even if the minimization were not so charmed, there
would be no point in doing the a calculus minimization; the cal-
culus method is too accurate. The kinetic-energy estimate uses
a crude form of the uncertainty principle, px , whereas
the true statement is that px /2. It uses a crude def-
inition of x, that x r. It uses the approximate formula
KE (p)
2
/m. This formula not only contains m instead of
2m in the denominator, but also assumes that we can convert p
into an energy as though it were a true momentum, rather than
merely a crude estimate of the root-mean-square momentum. Af-
ter making these approximations, its pointless to minimize using
the elephant gun of dierential calculus. The order-of-magnitude
3. Materials I 49
method is as accurate as the approximations that we used to de-
rive the energy.
Our name for this method of equating competing terms is bal-
ancing; we balanced the kinetic energy against the potential en-
ergy and found that a
0

2
/m
e
e
2
. Nature could have been un-
kind: The potential and kinetic energies could have been dierent
by a factor of 10 or 100. But she is kind: The two energies are
roughly equal, up to a constant that is nearly 1. Nearly 1 is
also called of order unity. This rough equality occurs in many
examples that we examine; we often get a reasonable answer if we
pretend that two energies (or, in general, two quantities with the
same units) are equal. When the quantities are potential and ki-
netic energy, the virial theorem, which we discuss in Section 3.2.1,
protects us against large errors.
3.1.3 Numerical evaluation
Now that we have derived
a
0

2
m
e
e
2
(3.17)
in two ways, we evaluate a
0
: We put in numbers. You might think
that we need to know or look up , e, and m
e
. Not so. We can
avoid looking up , m
e
, and e, by using three useful combinations
of fundamental constants. The rst two are round values that use
eV as the unit of energy:
c 2000 eV
A, (3.18)
m
e
c
2
500 keV. (3.19)
The third combination is

e
2
c

1
137
0.01. (3.20)
This combination is the ne-structure constant, a dimension-
less measure of the strength of the electrostatic force. Dimension-
less numbers, such as , are worth memorizing, because their value
is the same in any unit system.
The Bohr-radius estimate (3.17) contains e
2
downstairs, so we
multiply by c/c to manufacture
1
:
a
0

2
m
e
c
c
e
2
. (3.21)
This expression contains m
e
, so we multiply by c
2
/c
2
. After can-
celing one c in the numerator and in the denominator, we nd
that
a
0

c
m
e
c
2
c
e
2
. (3.22)
We could have gone directly from (3.17) to (3.22) by multiply-
ing by c
2
/c
2
. We wrote out the individual steps to distinguish two
3. Materials I 50
2. See any textbook on quantum me-
chanics for an explanation of the Pauli
principle; Gasiorowiczs text [11] is par-
ticularly compact and clear.
a
a
Figure 3.4. Packing of atoms in a solid
or liquid. For clarity, we draw only a
two-dimensional slice of the substance.
In this crude picture, each atom occu-
pies a cube of side length a 3
A, or a
volume of 30
A
3
or 3 10
23
cm
3
.
tricks that often go together. The rst trick is to insert c/c. The
second trick is to insert c
2
/c
2
to make m
e
c
2
, which has the round
value 500 keV. Often, if you add enough powers of c to convert ev-
ery into c, you also convert every m
e
into m
e
c
2
. In such cases,
the two tricks combine into one, which here is multiplication by
c
2
/c
2
.
Now we can put easy-to-remember numbers into (3.22) and
determine the Bohr radius:
a
0

(2000 eV
A)
5 10
5
keV
100 0.5
A. (3.23)
Knowing the size of hydrogen, we can understand the size of
more complex atoms. Hydrogen is the simplest atom; it has one
electron, and therefore one energy shell. The second row of the
periodic table contains elements with two shells; the third row
contains elements with three shells. The most abundant elements
on earth (oxygen, carbon, silicon) come from the second and third
rows. As a rule of thumb, the diameter of an atom with n shells
is n
A, for n 3; for n > 3, the diameter is still 3
A, because the
extra nuclear charge in those atoms drags the electrons closer, and
makes up for the increased number of shells.
The following argument expands on this statement about ex-
tra nuclear charge. Consider an atom with atomic number Z; it
has Z protons and Z electrons. The outermost electron moves in a
potential created by the Z protons and the Z 1 other electrons.
We cannot easily calculate the charge distribution, so we need to
simplify. Imagine that the other electrons orbit inside the outer-
most electron. This assumption is a crude approximation; it does
not account for important facts from quantum mechanics, such
as the Pauli exclusion principle,
2
nor does it accurately represent
atoms in which two or more electrons are in the outermost shell.
However, it is a simple assumption, and has truth in it, so lets
use it. What charge distribution does the outermost electron see?
It sees a nucleus with charge e (eectively, a single proton): The
Z protons and the Z 1 electrons almost cancel. An outermost
electron orbits a single protonthis conguration is the descrip-
tion of hydrogen. So the environment of the outermost charge is
independent of Z, and every large-Z atom is the size of hydrogen.
Part of this conclusion is reasonably accurate: that every large-Z
atom is a similar size. Part of the conclusion is not correct: that
the size is the size of hydrogen. It is incorrect because of the ex-
treme approximation in assuming that every other electron orbits
inside the outermost electron, and because it neglects the Pauli
exclusion principle. We retain the reasonably correct part, and use
a 3
A for a typical atomic diameter.

3.1.4 Densities
From atomic sizes, we can estimate densities. An atom is a positive
nucleus surrounded by a cloud of negative charge. A solid or liquid
contains atoms jammed together. The electron cloud limits how
3. Materials I 51
3. This density estimate also shows why,
for materials physics, cgs units are more
convenient than mks units. A typical cgs
density is 3, which is a modest, easy-to-
work with number. A typical mks density
is 3000, which is unwieldy.
(g cm
3
)
Element Estimated Actual
Li 0.47 0.54
H
2
O 1.2 1.0
Si 1.87 2.4
Fe 3.73 7.9
Hg 13.4 13.5
Au 13.1 19.3
U 15.9 18.7
Table 3.2. Actual and estimated densi-
ties of solids and liquids. The estimate
is from (3.25).
closely the atoms can approach each other, because the repulsive
force between the electron clouds is large when the clouds overlap.
At large distances (relative to the Bohr radius), two atoms hardly
interact. Between these extremes, there is a minimum-energy dis-
tance: a, the diameter of the electron cloud. So a 3
A is also a
typical interatomic spacing in a solid or liquid.
Let A be the atomic mass of the atom; A is roughly the number
of protons and neutrons in the nucleus. Although it is called a
mass, A is dimensionless. Each atom occupies a cube of side length
a 3
A (Figure 3.4), and has mass Am

p
. The density of the
substance is
=
mass
volume

Am
p
(3
A)
3
, (3.24)
or, with m
p
2000m
e
2 10
24
g,

A 2 10
24
g
3 10
23
cm
3

A
15
g cm
3
. (3.25)
In Table 3.2, we compare this estimate against reality. This es-
timate explains why most densities lie between 1 and 10 g cm
3
:
Most everyday elements have atomic masses between 15 and 150.
3
We even threw into the pack a jokerwater, which is not an
elementand our estimate was accurate to 20 percent. In Ex-
ample 3.1, we answer a question that may be painfully familiar if
you have moved house or apartment.
Example 3.1 How heavy is a small box lled with books?
Books are mostly paper; as we reasoned in Section 1.1.1,
paper has the same density as water, so
book
1 g cm
3
. In
the United States, the canonical book box is the small United
Parcel Service box. Its volume is 60 cm 30 cm 30 cm
5 10
4
cm
3
, so its mass is m 50 kgapproximately the mass
of a person. It is no wonder that these boxes are so heavy, and
no wonder that they are no larger.
3.2 Energies
Part of the order-of-magnitude picture of materials is the spacing
between atoms (bond size); another part is the force or interaction
energy between atoms (bond energy). Following the pattern of
Section 3.1, we rst estimate the binding energy of hydrogen,
and then extend our understanding to more common materials.
3.2.1 Binding energy of hydrogen
In hydrogen, the binding energy is the energy required to drag the
electron innitely far from the proton. In more complex atoms, it is
the energy to remove all the electrons. We can estimate the binding
energy by rst estimating the potential energy in hydrogen. The
potential energy of an electron and a proton separated by the Bohr
radius is
PE
e
2
a
0

m
e
e
4
2
. (3.26)
3. Materials I 52
The binding energy is E
total
, where E
total
is the total energy
it includes kinetic as well as potential energy. What is the kinetic
energy? The virial theorem says that, in a 1/r
n
potential (n = 2
for hydrogen),
PE = nKE. (3.27)
Therefore, E
total
= PE + KE = PE/2, and the binding energy is
E
0
= E
total
=
1
2
m
e
e
4
2
. (3.28)
To evaluate this energy, we create (c)
2
and m
e
c
2
by multiplying
E
0
by c
2
/c
2
:
E
0

1
2
m
e
c
2
e
4
(c)
2
=
1
2
m
e
c
2
1
2
5.1 10
5
eV
1
137
2
= 13.6 eV.
(3.29)
By luck, the errors in our approximations have canceled: The
honest-physics factor is 1. Equation (3.29) is the correct ground-
state energy of hydrogen (neglecting relativistic eects, such as
spin). We can rewrite the binding energy as m
e
(c)
2
/2, which is
the kinetic energy of an electron with v = c: the ne-structure
constant is the velocity of an electron in atomic units, where
charge is measured in units of e, length in units of a
0
, and veloc-
ity in units of c.
For future reference, we quote useful energy conversions:
1 eV 1.6 10
12
erg
1.6 10
19
J,
(3.30)
1 cal 4 J, (3.31)
and
1 eV/molecule 25 kcal mole
1
100 kJ mole
1
.
(3.32)
3.2.2 Covalent- and ionic-bond energies
Covalent and ionic bonds are formed by attractions between elec-
trons and protons; the hydrogen atom is a crude model of this
interaction. The main defect of this model is that the electron
proton distance in a hydrogen atom is much smaller than it is in
most materials. In most materials, the distance is roughly a 3
A,
rather than a
0
0.5
A. For covalent and ionic bondstheir bind-

ing energy is from the electrostatic attraction of monopolesthe
binding energy is smaller than E
0
by a factor of 6: E
bond
2 eV.
The factor of 6 occurs because a 6a
0
, and electrostatic energy
scales as E 1/r. [Scaling E is more direct than is evaluating e
2
/a
from scratch. It does not clutter the derivation, or our thinking,
with irrelevant information, such as the value of or .] Table 3.3
lists bond energies. For bonds that include carbon, oxygen, or
3. Materials I 53
Bond Energy (eV)
CC 6.3
HH 4.5
NaCl 4.3
HO 4.4
FeO 4.0
CH 3.5
SiSi 3.3
HgS 2.2
CdS 2.1
ZnS 2.1
Table 3.3. Approximate covalent and
ionic bond energies. Source: CRC Hand-
book of Chemistry and Physics [13,
pages 9-1239-128].
hydrogenwhich have a 1.5
Awe expect E
bond
4 eV. This
expectation is conrmed by the tabulated values. Van der Waals
bonds are much weaker than covalent or ionic bonds, as you would
expect from the high reciprocal powers of r in the LennardJones
potential (3.15).
The cohesive energy is the energy required to break the
weakest type of bond in an object. In water, it is the energy re-
quired to remove a water molecule from the liquid into the vapor.
In hydrogen, it is the energy required to break the electronproton
bond (given by (3.29)). We denote the cohesive energy by
c
. The
typical magnitude of
c
indicates why the electronVolt is a con-
venient unit in materials physics: Cohesive energies measured in
eV are of order unity (for strongly bonded substances).
3.3 Elastic properties
The size of white dwarfs or of raindrops, the speed of sound
these quantities we can estimate using our knowledge of atomic
sizes and energies, and the methods that we have developed. We
begin with the speed of sound. It illustrates a method that we use
frequently: approximation by springs.
3.3.1 Speed of sound
In a gas at pressure P, the speed of sound is roughly
c
s

_
P/, (3.33)
where P is the pressure and is the density. We pretend that the
gas is an ideal gas, even though this pretense introduces a theoret-
ical diculty: Sound cannot travel in an ideal gas, because sound
waves require molecular collisions to propagate, and in an ideal
gas, molecules do not collide. A more detailed analysis shows that
sound waves with frequency can propagate provided that each
molecule has a collision frequency f . In air, which for most
purposes can be treated as an ideal gas, the collision frequency
is roughly 10
9
Hz. This frequency is high enough to include most
sound waves that we are interested in. The slight deviation from
ideal-gas behavior is sucient to make sound possible, so we will
not worry about this theoretical incompatibility between sound
propagation and the ideal-gas approximation.
The sound-speed formula (3.33) is the only dimensionally cor-
rect combination of c
s
, P, and . For a solid or liquid, we expect a
similar relation, but we need to replace pressure with an analogous
quantity. The units of pressure are force per area, or energy per
volume: Pressure is energy density. We already have a volume: the
interatomic volume a
3
. We have an energy: the cohesive energy
c
.
An analogue of pressure could be
M
energy
volume

c
a
3
. (3.34)
The speed of sound is then
c
s

_

c
a
3

_
c
m
(3.35)
3. Materials I 54
A
l
l
F
Figure 3.5. Wire stretched by a force.
The force F stretches the wire by l
from its relaxed length l; it has cross-
sectional area A.
1
1
2 3
2 3
d
m
k
Figure 3.6. Small piece of a solid. The
atoms, with mass m, are connected by
bonds approximated as ideal springs of
spring constant k. At t = 0, atom 1
gets a kick, and moves a distance d (sec-
ond row). When does atom 2 follows in
atom 1s footsteps?
where a
3
m is the mass of an atom.
The quantity M is the elastic modulus. It is analogous to
the spring constant, but it is more useful. Imagine a wire (Fig-
ure 3.5) stretched by a force. Its spring constant determines how
far it stretches: F = k
w
l. A thicker wire (of the same material)
with, say, area 4Ahas a correspondingly larger spring constant:
k
1
= 4k
w
. A longer wirewith, say, length 2l (and the same
area)has a correspondingly smaller spring constant: k
w
/2. The
spring constant is not a property of the wire alone; it depends on
the wires length and area as well. The elastic modulus xes this
problem; it depends on only the substance.
How can we dene such a quantity macroscopically, in terms
of lengths, areas, and forces? (Equation (3.34) is a microscopic
denition.) One dependence is on area: k/A is independent of
area. The other is on length: k
w
l is independent of length. The
combined quantity k
w
l/A is independent of area and length. In
terms of this quantity, F = k
w
l becomes
F =
_
k
w
l
A
_
A
l
l, (3.36)
or
F
A
..
=
k
w
l
A
..
M
l
l
..
, (3.37)
where is stress (or pressure), Mis elastic modulus, and is frac-
tional change in length, or strain. The strain is the dimensionless
measure of extension. We show shortly that
M= k
w
l/A (3.38)
is the macroscopic equivalent of (3.34).
We just imagined a wire as a spring. We can apply to the spring
picture to the atomic scale as well. On the atomic scale, the bonds
are springs. Imagine a line of atoms, each connected to its neigh-
bors by complex forces: electrostatics and kinetic energy in deli-
cate balance. We replace the complexity of a bond with the sim-
plicity of a spring (Figure 3.6). We can use this picture to estimate
the sound speed, and then compare the microscopic estimate with
the dimensional guess (3.35). A sound wave transmits atomic mo-
tion from one location to another. To idealize the complex motion
of the atoms, imagine that atom 1 gets a swift kick (an impulse)
that moves it a distance d to the right. When does atom 2 hear the
news, and move a distance d? Before the kick, the atoms rested
peacefully in their equilibrium positions. Right after atom 1 moves,
the force on atom 2 is F kd, where k is the spring constant, so
atom 1s acceleration is a = F/m kd/m. After time t, atom 2
has traveled a distance at
2
kt
2
d/m. This distance reaches d at

_
m/k. The sound speed is then c
s
a/
_
ka
2
/m. To
estimate the spring constant, k, we use
c
ka
2
. This approxima-
tion is equivalent to assuming that a bond breaks when its length
3. Materials I 55
E
r
a
Figure 3.7. Bond energy versus separa-
tion. As two atoms approach, electron
repulsion contributes a large positive
energy. As they separate, electrostatic
attraction weakens (becomes less neg-
ative). In between is a happy compro-
mise, the energy minimum, a. Around
that minimum, we t a parabola: We
assume that the bond is spring (we dis-
cuss the near-universal value of mod-
eling physical processes with springs in
Section 5.1).
M
Substance
_
10
11
erg cm
3
_
Steel 20
Cu 11
Ice (5
C) 0.9
Al 7.9
Pb 1.8
C (diamond) 44
C (graphite)
to planes 3.5
to planes 100
Ash (white) 12
Glass 60.5
Table 3.4. Elastic moduli. There are
numerous elastic moduli. We list for
each substance the Youngs modulus,
which is a combination of the shear
modulus and the bulk modulus. Note
how strong M for graphite depends on
direction, because result of graphites
layered structure. If M
were as high as
M
, then lead pencils would be diamond

styluses. Source: Smithsonian Physical
Tables [9].
is doubled (Figure 3.7). With this approximation, c
s

_
c
/m,
as we derived in (3.35) by analogy with gases.
We can count bond springs to justify the macroscopic expres-
sion M = k
w
l/A from (3.38). The wire of Figure 3.5 is a bundle
of laments, where each lament is a line of atoms. Each lament
has N
l
l/a springs; the spring constant of a lament is therefore
reduced by l/a compared to the spring constant of a bond. The
wire contains N
f
A/a
2
laments; this factor increases the spring
constant of the wire compared to the spring constant of a lament.
The spring constant of the wire is then k
w
= kN
f
/N
l
kA/la.
With this expression, the macroscopic denition of elastic modu-
lus (3.38) becomes M k/a. We need to estimate k, for which we
use ka
2

c
. Then we have M
c
/a
3
, which is the microscopic
estimate (3.34).
To estimate a typical sound speed, we rst evaluate a typical
elastic modulus using (3.34):
M

c
a
3

2 eV
(3
A)
3

2 1.6 10
12
erg
3 10
23
cm
3
10
11
erg cm
3
.
(3.39)
This estimate for Mis reasonable for many materials (Table 3.4).
A typical solid has atomic mass, say, 40; from (3.25), it has
density 3 g cm
3
, so
c
s

10
11
erg cm
3
3 g cm
3
1.7 kms
1
. (3.40)
This estimate is reasonably accurate (Table 3.5), indicating that
we have included the important physics.
3.3.2 Yield strength
How strong are materials? To break a perfect material (for exam-
ple, diamond or carbon lament with no aws), we would have to
apply a stress M. Most materials break long before the stress
reaches M, because aws in their crystal structure concentrate
stress, so locally the stress may reach M even if the global stress
is much less than M. A typical breaking stress (or yield stress) is
between 0.001Mand 0.01M. The dimensionless factor is the yield
3. Materials I 56
mg
h
Shear
Shear
A h
2
Figure 3.8. Mountain. This approx-
imate mountain is conical, with 45
slopes. The shaded top block, which has

mass m, squeezes the hatched chunk
(whose footprint is the lightly shaded
ring). When the mountain gets too high,
the squeeze becomes greater than the
mountain rock can withstand.
Substance v (kms
1
)
Glass 5.5
Steel 5.0
Brick 3.7
Pine 3.3
Water 1.5
Hg 1.5
Cu 3.6
Cork 0.5
Granite 4.0
Table 3.5. Speed of sound at room tem-
perature. Our generic estimate is v
1.7 kms
1
. It is a useful rule of thumb;
but the rule of thumb underestimates
the speed in metals (and sti substances
such as glass), and overestimates the
speed in soft substances such as cork.
Source: Smithsonian Physical Tables [9,
p. 306].
Substance
y
Steel 0.005
Cu 0.002
Al 0.001
Rock 0.001
Table 3.6. Approximate breaking strain,
y
.
or breaking strain,
y
, which we list in Table 3.6. A typical yield
stress is 10
9
erg cm
3
. We now apply these estimates to mountain
climbing.
3.3.2.1 Mountain heights on the earth. How much energy does it
take to climb the tallest mountain on a planet? We rst determine
the height of such a mountain (Figure 3.8). The strength of the
rock limits its height. The mass of the top conical block is h
3
,
where is the density of rock; it produces a force F gh
3
. The
resulting stress is

F
A

gh
3
h
2
= gh. (3.41)
The strain is
=

M

gh
M
. (3.42)
For rock, M 10
12
erg cm
3
. When
y
, the rock yields, and
the mountain shrinks until falls below
y
. Therefore, the maxi-
mum height for a mountain on the earth is
h
max

M
y
g
. (3.43)
For a typical rock,
y
10
3
and 3 g cm
3
, so
h
max

10
12
erg cm
3
10
3
3 g cm
3
1000 cms
2
3 km. (3.44)
In this estimate, we neglected many numerical factors; not surpris-
ingly, some mountains, such as Mount Everest (where h 10 km),
are larger than our estimated h
max
. (Perhaps Everest is made of
extra-hard rock, such as granite.)
3.3.2.2 Mountain heights on other planets. Is it more dicult to
climb the highest mountains on Mars, or Mercury, or the as-
teroid Ceres, compared to the highest mountains on the earth?
We assume that all astronomical objects are made of the same
type of rock. The energy required to climb to the top is E
3. Materials I 57
mgh
max
mM
y
/, where m is the mass of the climber. For a
typical climber, m 50 kg, so
E
5 10
4
g 10
12
erg cm
3
10
3
3 g cm
3
1.7 10
13
erg
400 kcal,
(3.45)
or one chocolate bar (at perfect eciency). This energy is inde-
pendent of g. People do not climb with perfect eciency; no one
could climb Everest on 1 or even 10 chocolate bars.
How high are these mountains? The energy to the climb the
mountains does not depend on g, but the maximum height does.
As we found in (1.2.1), g R. So h
max
R
1
, and the fractional
size of mountains, h
max
/R is R
2
. For the earth, h
max
3 km
and R
6000 km, so h/R
10
3
. For an asteroid of radius
R R
/30 (roughly 200 km), the fractional height is roughly 1.

This estimate is reasonably accurate. Asteroids with R 200 km
have mountains roughly their own size; these mountains make the
asteroids look oblate. In Section 3.4, we estimate mountain heights
on white dwarfs.
3.3.3 Surface tension
Water droplets are spherical, because spheres have the smallest
surface area for a given volume. This mathematical explanation
hides a physical question: Why do water droplets minimize their
surface area? The reason is that water has surface tension. Sur-
face tension is the energy required to create one unit of new surface
(surface tension has units of energy per area). On our picture of a
solid, every atom in the interior is bonded to six neighbors. Each
atom at the surface has only ve neighbors. In a surface of area
A, there are N = A/a
2
atoms. To make such a surface, we must
break N bonds, one for each atom. The cohesive energy is the
energy to break all six bonds, so E
bond

c
/3. To avoid counting
each bond twice, we divide by 3 instead of by 6. So the surface
energy is E AE
bond
/a
2
A
c
/3a
2
. The surface tension is the
ratio E/A:

E
A

c
3a
2
. (3.46)
We can estimate cohesive energies from heats of vaporization, as
we discuss in Section 4.2. For water,
c
0.5 eV. Using our stan-
dard spacing a 3
A, we get
water

0.5 eV 1.6 10
12
erg eV
1
3 (3 10
8
cm)
2
300 erg cm
2
.
(3.47)
Although our basic model is sound, this estimate is large by a fac-
tor of 4, partly because of the crudity of our six-neighbor picture
3. Materials I 58
Surface-tension forces
r
Circumference 2r
Drag force F
d
Drag force F
d
Velocity v
Figure 3.9. Falling raindrop of density
w
and radius r. The drop has reached
terminal velocity v. Surface tension (the
facing arrows) binds the two halves of
the drop along the circumference. The
shaded area shows a cross section of the
drop.
of solids.
Surface tension determines the size of raindrops. Falling rain-
drops feel a drag force, which tries to split the drop into smaller
droplets. Many small raindrops cost extra surface energy, which
ghts the splitting. There is a happy, intermediate size, which we
estimate by balancing the surface tension (as a force, F
) and the
drag force, F
drag
(Figure 3.9).
We rst estimate the surface-tension force. Surface tension is
energy per area, which is also force per length. To obtain a force,
we multiply by the only length involved: the circumference,
which is roughly r. So F
r. The drop moves at constant

velocity (terminal velocity), so we equate the drag force and the
weight: F
drag
= r
3
g, where r
3
is the mass of the drop. Equating
the drag and surface-tension forces, we get r
3
g r, or
r
max

_

w
g
. (3.48)
For water, 70 dyne cm
1
, so
r
max

70 dyne cm
1
1 g cm
3
10
3
cms
2
0.25 cm. (3.49)
The terminal velocity is given by the high-speed result (2.49):
v
_
gr
max
water
air
. (3.50)
Because
water
/
air
1000, we get v 10 m/s, which is the speed
of slow driving, or fast sprinting. This velocity seems about right;
if you drive slowly in a rainstorm, the drops strike at roughly a
45
angle.
3.4 Application to white dwarfs
White dwarfs are small, dense stars in which the pressure is high
enough to completely ionize the atoms. They are dead stars; there
is no nuclear fusion in their interiors. We apply our estimation
methods to this bizarre state of matter, to show you that the
principles have wide use.
3. Materials I 59
3.4.1 Size of a white dwarf
How large is a white dwarf? In a white dwarf, the atoms live in a
sea of electrons. What prevents gravity from squeezing the white
dwarf to zero size? You could ask a similar question about hydro-
gen: What prevents the electrostatic force from squeezing the ra-
dius to zero? In hydrogen, and in white dwarfs, connement energy
prevents collapse. For a white dwarf, we can nd the minimum-
energy radius by balancing gravitational and connement energy.
[The concept of connement energy is a simple version of a more
complicated argument that uses the Pauli exclusion principle; see
Section 4.6.2.]
We rst work out the connement energy. Imagine a star of
radius R and mass M, composed of atoms with atomic number Z
and atomic mass A. We dene the dimensionless ratio Z/A,
because this factor shows up often in the derivation. The star
contains M/Am
p
atoms, and N
e
= (M/m
p
)(Z/A) = M/m
p
electrons. The number density of electrons is n
e
, so each elec-
tron is conned to a region of size x n
1/3
e
. (We shamelessly
neglect numerical factors such as 4/3.) The electrons momen-
tum is p n
1/3
e
, and its connement energy is (p)
2
/m
e

2
n
2/3
e
/m
e
. This result is valid for nonrelativistic electrons. The
total connement (or kinetic) energy is this value times N
e
:
KE

2
n
2/3
e
m
e
N
e

2
N
5/3
e
m
e
R
2
, (3.51)
because n
2/3
e
= N
2/3
e
/R
2
. In terms of M and m
p
, this energy is
KE

2
M
5/3
m
e
m
5/3
p
R
2
5/3
. (3.52)
We next work out the gravitational (potential) energy, which
is the sum of all pairwise interactions between all particles in the
star (protons, electrons, and neutrons). What a mess! We rst
try a dimensional argument. We construct an energy using the
variables m
e
, m
p
, M, and R. Any expression of the form
PE
Gmass mass
R
, (3.53)
where G is the gravitational constant, is an energy. Unfortunately,
there are three masses from which to choose, and two slots in which
to put them. The possibilities are numerous. Fortunately, we can
restrict the freedom. Gravity does not care about the composi-
tion of the star; muons, quarks, electrons, and protons all obey
universal gravitation. So m
e
and m
p
are irrelevant, and
PE
GM
2
R
. (3.54)
A physical argument also gives us this result. Two random points
in the white dwarf have a separation of R. Imagine one-half of
3. Materials I 60
the star clumped at one point, and one-half clumped at the other
point. The gravitational potential energy between the two clumps
is G(M/2)
2
/R. Once we neglect the numerical factors, we get
PE GM
2
/R, in agreement with (3.54), but with the correct
sign.
Now we equate PE and KE (in absolute value) to solve for R:
GM
2
R

2
5/3
M
5/3
m
e
m
5/3
p
R
2
, (3.55)
so
R M
1/3

2
5/3
Gm
e
m
5/3
p
. (3.56)
This expression contains two factors of , which suggests that
we multiply it by c
2
/c
2
to manufacture two factors of c. After
regrouping, we nd that the radius is
R
c
m
e
c
2
c
Gm
2
p
_
m
p
M
_
1/3
5/3
. (3.57)
This expression contains factors familiar from our computation
of the Bohr radius. The factor c/m
e
c
2
is also in (3.22). This
parallel suggests that we compare the second factor, c/Gm
2
p
, to
the second in (3.22), which is c/e
2
(the reciprocal of the ne-
structure constant). We therefore call Gm
2
p
/c the gravitational
ne-structure constant. Its value is 6 10
39
: Gravity is much
weaker than electrostatics. The remaining factors in (3.57), which
are dimensionless, have no analogue in the hydrogen calculation.
If our sun were a white dwarf, how large would it be? We
use (3.57) and scale the result against M
. For most atoms, =

Z/A 0.5. The radius is
R
c
m
e
c
2
c
Gm
2
p
_
m
p
M
_
1/3
5/3
_
M
M
_
1/3
. (3.58)
The numerical factors combine to give
R 2 10
3
km
_
M
M
_
1/3
. (3.59)
When M = M
, R is roughly 0.3 earth radii. A more exact cal-

culation (honest physics) produces a factor of 4.5 in front: The
actual radius is roughly one earth radii. White dwarfs are dense.
Our computation of the massradius relation followed the same
procedure as did the computation of the Bohr radius. Both sys-
tems have a potential energy source: gravity in the star, elec-
trostatics in hydrogen. Without opposition, the potential energy
would be a minimum at zero size (R = 0 or a
0
= 0). In both
systems, connement energy ghts collapse.
Note that the radius of a white dwarf is M
1/3
, so the
volume is inversely proportional to the mass. For xed Z and
3. Materials I 61
A, the electron density is therefore M
2
. At a certain mass, n
e
becomes large enough that the electrons become relativistic (this
derivation assumed that the electrons are nonrelativistic). We can
estimate this critical mass equating the kinetic energy per electron
to m
e
c
2
, and solving for the white-dwarf mass. The kinetic energy
per electron is KE/N
e
, where KE is given in (3.51):
E
e

2
N
2/3
e
m
e
R
2
. (3.60)
We equate this energy to m
e
c
2
:
2
N
2/3
e
m
e
R
2
m
e
c
2
. (3.61)
We now use the expression for R in (3.56), and the expression N
e
=
M/m
p
for the number of electrons. With these substitutions,
(3.61) becomes
2/3
M
4/3
m
2/3
p
m
e
m
e
c
2
M
2/3
10/3
G
2
m
2
e
m
10/3
p
. .
R
2
. (3.62)
After canceling common factors and rearranging, we nd that the
critical mass is
M
critical

2
_
c
Gm
2
p
_
3/2
m
p
. (3.63)
With an honest-physics factor of 3.15 in front, this mass is the
Chandrasekhar mass, which he derived in 1931; it is the largest
mass that a white dwarf can have, and its value is roughly 1.4M
.
We did not try to calculate this masswe tried to calculate only
the transition mass between nonrelativistic and relativistic re-
gimesbut we found it anyway. The reason that the transition
mass and the maximum mass are related is that a relativistic
white dwarf is less sti than a nonrelativistic one (it does not
withstand compression as well). Imagine a nonrelativistic white
dwarf to which we slowly add mass. As the mass increases, the
radius decreases (R M
1
), and so does the electron velocity.
As the electron velocity approaches c, however, the less-sti rel-
ativistic behavior begins to dominate, and the radius decreases
faster than M
1
. The electron velocity then increases even more
than we would expect from the nonrelativistic relation, and the
less-sti relativistic behavior dominates even more. You can see
positive feedback in this description. At a large enough mass, this
feedback makes the radius go to zero. The mass at which this col-
lapse occurs is the Chandrasekhar mass. (The radius does not ac-
tually go to zerothe white dwarf becomes instead a much denser
object, a neutron star.) Because the same physics determines the
Chandrasekhar mass and the transition mass, our estimate is also
an estimate of the Chandrasekhar mass.
3. Materials I 62
3.4.2 Mountains on white dwarfs
How high are the tallest mountains on white dwarfs? The energy
density is the elastic modulus: M PE/R
3
GM
2
/R
4
. The
mass density is M/R
3
. And the acceleration due to gravity is
g GM/R
2
. The height estimate (3.43) becomes
h
max

y
GM
2
/R
4
(M/R
3
) (GM/R
2
)

y
R. (3.64)
The morass of proton and electron masses, , and so on, have
canceled. All that remains is the yield strain and the white-dwarf
radius. Perhaps white-dwarf matter has
y
0.01 (because the
high pressures squash imperfections in the lattice). We use (3.59)
for R. The mountain height is then
h
max
20 km
_
M
M
_
1/3
. (3.65)
If M = M
, the mountain is 20 km tall. It would be a many-

chocolate-bar climb.
Quantum mechanics: We can understand many of the conse-
quences of quantum mechanics by simply adding to the list of
relevant variables for dimensional analysis. Quantum mechan-
ics, through , introduces a new momentum scale, the uncer-
tainty momentum p, and indirectly, the uncertainty energy.
This energy is also called the connement energy.
Balancing: Many physical systems contain two competing pro-
cesses. For example, one energy (perhaps gravity) competes
with another energy (perhaps the uncertainty energy). The en-
ergies usually are equal near the minimum-energy state. Look
for these competitions!
Atomic sizes: Atoms have a diameter of a few

A.
Binding energies: Typical covalent and ionic bond energies are
a few eV, as a consequence of electrostatic attraction.
63
4 Materials II
How fast does a turkey cook? How quickly do the ground and
the air cool on a cloudless night? How high a fever do you run
after 30 minutes in a hot tub? At what temperature does water
boil? We can answer such questions by estimating the thermal
properties of materials. When we estimate boiling temperatures,
we introduce an extended example of successive approximation
the main technique of order-of-magnitude physics. As a warmup,
we begin with thermal expansion.
4.1 Thermal expansion
Most materials expand when they get hotter, and contract when
they get cooler. If bridge designers forget this fact, and design
a bridge so that its joints exactly mate in the summer, then it
will fall down in the winter. We would like to understand why
materials expand, and by how much.
Lets look at the potential of an atomic bond, and see what
it implies about thermal expansion. In Section 3.3.1, we approxi-
mated a bond as a spring. We can use this approximation to esti-
mate a typical thermal-expansion coecientthe fractional
length change per unit change in temperature. At a temperature
T, the bond spring vibrates with average energy kT (actually,
kT/2, but we ignore the 2). How far does that energy stretch the
bond? The bond energy is, in this approximation,
E(r) E
0
+
1
2
k
s
(r a)
2
, (4.1)
where E
0
is the bond energy, k
s
is the spring constant, a is the
equilibrium bond length, and r is the bond length. We can factor
out E
0
:
E(r)
E
0
1 +
k
s
(r a)
2
/2
E
0
. (4.2)
We can write E
0
= fk
s
a
2
/2, where f is a dimensionless constant
of order unity. Then
E(r)
E
0
1 +
_
r
a
1
_
2
f. (4.3)
To simplify the algebra, lets do the calculation in a dimensionless
unit system. We measure energy in units of E
0
and distance in
units of a. Then
E(r) 1 + (r 1)
2
f. (4.4)
For now, lets simplify by assuming that f = 1. Then
E(r) 1 + (r 1)
2
. (4.5)
4. Materials II 64
E
r
a
Figure 4.1. Potential energy versus
separation. The potential energy, E,
is plotted on the vertical axis (in arbi-
trary units); the equilibrium spacing,
a, is marked on the horizontal axis. At
higher energies (higher temperatures),
the curve skews toward positive r, mak-
ing the material expand as temperature
rises. The parabola, although an ade-
quate approximation to E(r) for small
deviations from a, gives a totally incor-
rect thermal-expansion coecient: zero.
If we dene x r 1, so x is the change in length (which is what
we are interested in), then the change in energy is E(x) x
2
. To
nd x, we equate E(x) to T (we use also units in which k 1,
so temperature is energy). We nd that x
2
T, or x
T.
[In normal units, this distance is x
normal
a
_
kT/E
0
, but that
value is irrelevant for the coming surprise.] How far does the bond
shrink? It shrinks exactly the same distance as it expands, because
a spring potential is symmetric. (Even if we had retained f, this
conclusion would still hold.) So the average separation in a spring
potential is independent of temperature: The thermal-expansion
coecient is zero!
We know that this result is not correct. We made too drastic an
approximation when we replaced the bond with a spring. An ac-
tual interatomic potential is asymmetric, as shown in the cartoon
potential-energy curve of Figure 4.1. The asymmetry produces a
nonzero thermal-expansion coecient.
A more accurate approximation to E includes terms up to x
3
:
E(x) x
2
+x
3
, (4.6)
where is a constant that measures the asymmetry of the poten-
tial. (We do not have to include the x
4
contribution, because it is
symmetric, and therefore contributes no average length change).
With the more rened bond energy (4.6), how much does the bond
stretch or shrink? The change in energy is
E(x) x
2
+x
3
. (4.7)
We equate this energy to T:
x
2
+x
3
= T. (4.8)
If x is small, we can solve this equation by successive approxima-
tion. Can we assume that x is small? Although the analysis using
a spring potential does not explain thermal expansion, it does give
us the estimate that x
T (in dimensionless units). A typical T

is 0.025 eV; a typical covalent bond energy is E
0
2.5 eV, so in di-
mensionless units, T 0.01 and x 0.1. For order-of-magnitude
calculations, x 0.1 is small, and we can solve (4.8) by successive
approximations. For the rst approximation, we ignore the x
3
4. Materials II 65
Substance (10
5
K
1
)
Cu 1.6
Si 0.5
Quartz 0.05
Diamond 0.1
Wood
along gr. 0.5
against gr 5
Glass 0.9
Pyrex 0.3
Vycor 0.08
Table 4.1. Thermal-expansion coe-
cients (at room temperature).
term, and recover the old result: x
1

T; the plus sign repre-

sents bond stretching, the minus sign represents bond shrinking.
For the next approximation, we replace x
3
by x
3
1
:
x
2
T
3/2
= T. (4.9)
The solution is
x =
_
T T
3/2
. (4.10)
(We have to carefully distinguish from , by carefully track-
ing bond stretching and bond shrinking. For expansion, take the
top sign; for contraction, take the bottom sign. Remember that
will be negative for most potentials, such as for the potential in
Figure 4.1.) The average x is (if we neglect a factor of 2)
x
avg
=
_
T T
3/2
_
T +T
3/2
. (4.11)
We factor out
T and use the approximation that
1 +h 1 +
h/2:
x
avg
=
T
_
_
1 T
1/2
_
1 +T
1/2
_
T((1 T
1/2
/2) (1 +T
1/2
/2))
= T
. (4.12)
The thermal-expansion coecient is dx
avg
/dT, which is . What
is the value in a normal unit system? Lets restore normal units to
x
avg
= T. The quantity x
avg
is really x
avg
/a. The temperature
T is really the energy kT, and the energy kT is really kT/E
0
.
So (4.12) becomes x
avg
/a kT/E
0
. If we had carried the f
through, then we would nd
x
avg
a

f
kT
E
0
. (4.13)
The thermal-expansion coecient is the fractional length change
per unit change in temperature:

d(x
avg
/a)
dT

f
k
E
0
. (4.14)
The Boltzmann constant is roughly k 10
4
eVK
1
; a typical
bond energy is E
0
3 eV. For many interatomic potentials, 1.
For example, for the potential E(r) = 4r
4
4r
2
(in dimen-
sionless units), 2. With 1, a typical thermal expansion
coecient is 3 10
5
K
1
.
Table 4.1 contains thermal-expansion data for various materi-
als. Our estimate seems to be an overestimate, although we can
partially explain the trends. For example, sti materials such as
quartz and diamond have a particularly low . Sti materials are
sti, because their bonds are strong, so , which is inversely pro-
portional to bond strength, should be low. (Quartz has silicon
oxygen bonds, and diamond has carboncarbon bonds; both bonds
are strong.) Perhaps quartz and diamond also have particularly
symmetric potentials (low ), which would also lower the expan-
sion coecient.
4. Materials II 66
T
vap
(K)
Substance
c
(eV) actual A B
Water 0.50 373 577 533
NH
3
0.31 240 374 347
HCl 0.21 188 263 244
O
2
0.071 90 99 93
Au 3.35 3130 3293 3022
Xe 0.13 165 171 159
He 0.00086 4.2 2.2 2.5
Hg 0.61 630 691 638
N
2
0.058 77 83 78
Table 4.2. Cohesive energy,
c
, per
atom or molecule; actual and predicted
boiling temperatures (at 1 atm). The co-
hesive energy is L
vap
/N
A
, where L
vap
the latent heat of vaporizationcomes

from experimental data. We use the co-
hesive energy to predict the boiling tem-
perature. Column A is the prediction
from (4.24). Column B is the prediction
from (4.27). Source: [13, 6-1036-106]
4.2 Phase changes
4.2.1 Boiling
To vaporize, or boil, a liquid, we must supply energy to move
molecules from the liquid into the gas phase. How much energy is
required? Per molecule, the energy required is roughly the cohesive
energy,
c
. The molar heat of vaporization (or enthalpy of vapor-
ization) is the energy required to boil one mole of a substance,
and we can estimate it as
L
vap

c
N
A
23 kcal mole
1
_

c
1 eV
_
. (4.15)
For water, for example,
c
0.5 eV, so L
vap
10 kcal mole
1
.
This estimate of the enthalpy leaves out a small contribution: the
energy to make room in the vapor for the evaporating molecules.
This energy is PV in the expression H = E+PV for the enthalpy.
The heat of vaporization is the change in H:
L
vap
H = E + (PV ). (4.16)
As we see shortly, after we estimate the boiling temperature,
(PV ) is small compared to E.
Predicting L
vap
from
c
is straightforward. Lets instead use
L
vap
for which accurate experimental data are availableto de-
termine
c
(the second column of Table 4.2). Now we have a mi-
croscopic quantity, for which we can make order-of-magnitude pic-
tures. What can we do with it? An essential feature of boiling is
the boiling temperature. So lets try to predict the boiling tem-
perature using the cohesive energy. We rst dene boiling temper-
ature. As a substance heats up, more molecules leave the liquid
and become vapor; the pressure of the vapor increases. The boil-
ing temperature is the temperature at which the vapor pressure
reaches atmospheric pressure (at sea level, roughly 10
6
dyne cm
2
,
or 1 atm). Atmospheric pressure is arbitrary; it is less on Mount
Everest, for example, and far less on Mars. It is unrelated to the
properties of the substance being boiledalthough it helps to de-
termine the boiling temperature. A more general question is how
boiling temperature depends on atmospheric pressure. You can
generalize the methods that we introduce to answer this question.
4.2.1.1 Simplest model. For our rst guess for the boiling tempera-
ture, we convert the cohesive energy into a temperature using
Boltzmanns constant: T
vap
=
c
/k, where is a dimensionless
constant. The conversion factor k is 1 eV 10
4
K (accurate to 20
percent). The data in Table 4.2 shows how inaccurate this guess is.
For example, for water, the predicted boiling temperature would
be 5000 K, instead of 373 K: Even on Mercury, oceans would freeze
solid. Or, for gold, the predicted temperature would be 30, 000 K
instead of 3000 K. If we by at insert a factor of 10, and assert
4. Materials II 67
that
T
vap

c
10k
, (4.17)
then our prediction would be reasonably accurate for both gold
and water. If T
vap
,
c
, and k are the only relevant variables in this
situation, then there is some constant such that T
vap
=
c
/k.
A few minutes with the data in Table 4.2 will convince you that
there is no such constant, although (4.17) provides a reasonable
approximation. Using this approximation, we can check that we
were justied in neglecting the PV term in the heat of vaporization
(4.16). From the ideal gas law, for one molecule in a gas at the
boiling temperature, PV = kT
vap
. Using (4.17), we nd that PV
is
c
/10. For order-of-magnitude purposes, PV is small compared
to
c
.
4.2.1.2 Volume contribution to the entropy: Part 1. Now lets
explain the approximate factor of 10. The factor arises because we
left out entropy, which is a fancy word for counting; the meaning
of entropy will become clear as we calculate it. At the boiling point,
vapor is in equilibrium with liquid: A molecule is equally happy in
either phase. In thermodynamic language, the Gibbs free energy
is the same in the two phases. Instead of analyzing the problem
using thermodynamic variables, which obscure the simplicity of
the mechanism, we analyze the feelings of one molecule deciding
where to go (the statistical-mechanics approach). If we consider
only energy, the molecule is more likely to be in the liquid than in
the vapor, by the Boltzmann factor e
c
/kT
vap
. So we have deduced
that the liquid does not boil at any temperature. However, we
must also include the entropy; that is, we must count the number
of accessible states in the vapor and in the liquid. We could, using
quantum mechanics, compute the number of accessible states in
the liquid; compute it in the solid; and then take the ratio. Instead,
we estimate the ratio directly, without stopping at the waystation
of quantum mechanics.
We assume that the vapor is an ideal gas, so that PV = NkT.
At atmospheric pressure, and at the boiling temperature, one
molecule is free to wander in a volume V
gas
= kT
vap
/P
0
, where
P
0
= 1 atm. In the liquid, the free volume for one molecule is a
factor times a
3
, but the factor is dicult to determine, because
liquids are poorly understood. One simple model of a liquid is a
solid with some atoms replaced by holes. How many holes should
there be? If no atom is surrounded by holes, then the substance is
a solid, not a liquid. If each atom is surrounded by six holes (all
its neighbors are holes), then the substance is a gas. A reasonable
compromise is that, on average, one or two holes surround each
atom. Each hole is shared among its six neighboring atoms (if
none of them are replaced by holes), so the free volume per atom
is perhaps a
3
/6 or a
3
/3. Lets be more general and say that the
free volume is V
liquid
(fa)
3
, where 0 < f < 1.
4. Materials II 68
Var. Name Value
E internal energy
H enthalpy E +PV
F Helmholtz free energy E TS
G Gibbs free energy H TS
Table 4.3. Thermodynamic energies.
The ratio of allowed volumes is
r
v

V
gas
V
liquid
kT
vap
P
0
a
3
f
3
. (4.18)
This ratio is 1, as we will nd out in a moment; therefore
this entropy factor encourages the molecule to move to the vapor.
This factor is independent of temperature, whereas the Boltzmann
factor depends strongly on temperature. The factors balance at
the boiling temperature:
e
c
/kT
vap
= r
v
, (4.19)
or
c
kT
vap
= log r
v
= log
_
kT
vap
P
0
a
3
f
3
_
. (4.20)
[All logs are natural logs.] This equation is transcendental; we can
solve it by successive approximation, or by drawing a graph.
Lets rst connect the result (4.20) to the (perhaps more fa-
miliar) thermodynamic quantities of Table 4.3. We are neglecting
the PV term, so the internal energy and the enthalpy are, in
this approximation, identical, as are the Gibbs and the Helmholtz
free energies. With T constant, which is what equilibrium means,
F = ETS. When the liquid and gas are in equilibrium, the
free energy change going from liquid to gas is zero, so E = TS.
Matching this equation against (4.20), we nd that
c
corresponds
to E i S = k log r
v
. Here we see the value of the free energy.
It is the energy with an adjustment TS; the adjustment includes
the counting of states.
Before solving by successive approximations, we should get a
qualitative feel for the sizes of the terms in (4.20), and see whether
the temperatures that it predicts are reasonable. According to
(4.20), the boiling temperature is given by
T
vap

c
k log r
v
. (4.21)
So lets evaluate r
v
for a couple substances using (4.18) and then
compute log r
v
. For water (with f = 1), we get
r
v

1.4 10
16
erg K
1
374 K
10
6
dyne cm
2
(3 10
8
cm)
3
200, (4.22)
and log r
v
5. For gold, we get log r
v
7.4. So we can explain
most of the factor of 10 in our rst guess (4.17): It comes from
entropy.
The rst step in solving a transcendental equation with a (nat-
ural) logarithm is to assume that any logarithm is 10. This rule
is due to Fermi, who was a master of order-of-magnitude physics.
A log could also be 10, although not in this case because
c
/kT
vap
is positive. At rst glance, this rule of 10 seems ridiculous, but it
has the following justication. Initially, most order-of-magnitude
4. Materials II 69
arguments assume power laws for the unknown functional forms;
this assumption excludes logarithms as solutions. If there is a loga-
rithm in the solution, it often arises in a later, more rened stage of
the analysis. In what circumstances would we rene the analysis?
Only when the original analysis was at least moderately accurate.
If it is wildly inaccurate, then we would throw out the whole anal-
ysis and start again. If it is extremely accurate, then a renement
is not likely to introduce a new function, such as a logarithm. So, if
a logarithm arises, we can be reasonably certain that the original
estimate was only moderately accurate, and that the logarithm in-
troduces a correction factor of at least 5, perhaps even as high as
20 or 30. We can further narrow this range. The logarithm takes a
dimensionless argument. How large can a dimensionless argument
become? For example, Avogadros number, N
A
6 10
23
, is huge,
and log N
A
55. However, this example is not relevant, because
Avogadros number is not dimensionless. Its units are mole
1
,
and the mole is an arbitrary unit whose size depends on the size
of the gram. A dimensionless argument must be the ratio of two
quantities with the same units; in this boiling-temperature prob-
lem, the two quantities are V
liquid
and V
gas
. In other problems,
they might be two characteristic lengths or energies. The ratio of
two characteristic quantities is almost never larger than r
v
= 10
8
,
in which case log r
v
20. So the logarithm typically lies between
5 and 20; a reasonable starting estimate is 10. With this starting
estimate, we recover (4.17).
Starting with log() 10, lets predict T
vap
for water. The
argument of the logarithm is
kT
vap
c
P
0
a
3
f
3
. (4.23)
To continue, we need a value for the fudge factor f. Should it
be 0 (the substance is virtually a solid) or 1 (the substance is
virtually a gas)? Its dicult to decide; both values are extreme,
but we have no clear argument favoring one extreme or the other.
In such situations of ignorance, we choose the middle value f =
0.5; this rule is the rule of one-half. We can estimate the spacing
a from the density and molar mass, but for simplicity we use
our usual estimate a 3
A. Then P
0
a
3
f
3
3.3 10
6
eV. For
convenience, we write F for log r
v
; so F =
c
/kT
vap
. In terms of
F, the transcendental equation is
F = log
_
F
1

c
3.3 10
6
eV
_
. (4.24)
For water,
c
0.5 eV. We start with F
0
= 10. The next iteration
produces
F
1
= log
_
1.5 10
4
_
9.62. (4.25)
Because F
1
F
0
, we do not need to compute F
2
; we might as
well use F = 10. Then T
vap

c
/10k 576 K. On the Celsius
scale, the error in this estimate is over 100 percent. However, the
4. Materials II 70
Celsius scale is irrelevant for physics. It has an arbitrary oset;
why should the universe care about the boiling temperature of
water? On the Kelvin scale, which has a natural denition for the
zero point, the error is 50 percent.
4.2.1.3 Volume contribution to the entropy: Part 2. Can we re-
duce it? The most arbitrary input was the factor f = 0.5. If we
can understand its origin, we can choose it more accurately. It
is a dimensionless measure of the molecular vibration amplitude.
Lets estimate this amplitude. The vibrations are thermal, with
a typical energy at the boiling temperature of kT
vap
/2. In the
model for surface tension in Section 3.3.3, each molecule had six
neighbors, and each bond had energy E
bond

c
/3. When we esti-
mated thermal expansion in Section 4.1, we made a spring model
in which the vibration energy (the change in bond energy relative
to the equilibrium bond energy) is E
bond
(x/a)
2
, where x is the
vibration amplitude. Lets solve for x as a function of kT
vap
. The
bond vibration gets kT
vap
/2 from thermal motion, so
kT
vap
2

c
3
_
x
a
_
2
. (4.26)
The solution is x a
_
1.5kT
vap
/
c
. If we take x as an estimate for
fa, then we nd that f
_
1.5kT
vap
/
c
. The factor kT
vap
/
c
is
familiar: It is F
1
, which is roughly 0.1. Therefore, f 0.4. If we
use this f in the successive-approximation procedure for T
vap
, we
get T
vap
540 K for water. This new T
vap
gives us a new estimate
for F, and thus a new estimate for f, which changes T
vap
. We are
solving the equation
F = log
_
F
1

c
2.6 10
5
eV f
3
_
, (4.27)
with f itself given in terms of F:
f =
1.5F
1
. (4.28).(4.29)
We can continue with further iterations; the succeeding iterations
converge to 533 K. This value is in error by 45 percent; our eort
has not reduced the error signicantly.
4.2.1.4 Rotational contribution to the entropy. Before despair-
ing, we should check our predictions on other, less peculiar sub-
stances; water has many strange properties. The prediction is ac-
curate for many substances, as you can see from column B of
Table 4.2. The exceptions are water, ammonia, and hydrochloric
acid. They share a feature, hydrogen bonds, and a common dis-
crepancy, a lower boiling temperature than we expected. What
eect do hydrogen bonds have? Lets see what eect they should
have to correct our predictions, and see whether we can nd a
physically justify the eect using hydrogen bonds. We predicted
the boiling temperature by calculating, or estimating, the entropy
4. Materials II 71
change. If our prediction for T
vap
is too high, then we must have
underestimated the entropy increase from liquid to gas. What
physical eect could add to the entropy increase? Because our
prediction is accurate for many substances, with widely diering
bond types, our volume estimate in the liquid is probably reliable,
even for hydrogen-bonded substances. So hydrogen bonds must
restrict the freedom in a way that we have not yet included. They
can do so if they have a directionif they are dipolar. And they
are; hydrogen bonds are like weak covalent bonds, so they have a
preferred direction. In a water molecule, the electron around hy-
drogen is dragged toward the oxygen, so the oxygen end of the
molecule is negatively charged, and the hydrogen end is positively
charged. Thus hydrogen ends tend to attract oxygen ends. The
bond is not as strong as a covalent bond, because the electron is
not dragged all the way to the oxygen.
In neglecting the rotational contribution to the entropy, we
implicitly, and incorrectly, assumed that a molecule in the liquid
could rotate as freely as a molecule in the gas; hydrogen bonds
prevent this easy rotation in the liquid. We need to compute a
new ratio,
r
rot

allowed solid angle of rotation in gas
allowed solid angle of rotation in liquid
, (4.30)
and include k log r
rot
in the entropy. The solid angle in the gas
is 4. To compute the liquid solid angle, we estimate the vibra-
tion angle of a water molecule, using the following crude model
of hydrogen bonding. We concentrate the cohesive energy into a
single hydrogen bond (the energy is actually spread over many hy-
drogen bonds). The resulting bond energy is E
0
2
c
. Lets also
assume that the bond is well approximated as the interaction of
two dipoles. Keep one dipole (one molecule) xed. Through what
angle, , does the unxed molecule rotate, because of thermal mo-
tion? For small , the dipoledipole potential can be written
U() E
0
(1
2
). (4.31)
To nd , we equate U() to kT
vap
, and nd E
0
2
kT
vap
, or
2
kT
vap
/2
c
, which is 0.5F
1
. The solid angle is roughly
2
,
so the ratio of solid angles is
r
rot

4
0.5F
1
8F. (4.32)
This result has a strange feature: The hydrogen-bond strength ap-
pears only indirectly, through F. The quantity F remains roughly
10, so the hydrogen bond strength hardly alters the rotational cor-
rection factor. This invariance has to be wrong. To see why, con-
sider a nonpolar substance; we can say that it has hydrogen bonds,
but that they are extremely weak. Should we then add k log r
rot
to the entropy? If we do, we spoil the excellent agreement between
4. Materials II 72
theory and experiment for those substances. Theres a way out of
our dilemma: This estimate of r
rot
is valid for only small . If the
hydrogen bond is too weak, then we cannot rely on it; in such a
case, the available solid angle in the liquid is 4, and r
rot
1.
Reassured that our result is not nonsense, we use it to estimate
the boiling temperature of water. For water, the preceding theories
gave F 10; more precisely, successive approximation applied to
(4.27) gave F 10.9. So log r
rot
4.46, and we should adjust F to
10.9+4.46 = 15.36 (because F is the entropy without Boltzmanns
constant). Then we recalculate r
rot
, and from that, recalculate
F, and so on; the procedure quickly converges to F = 15.7; the
predicted boiling temperature is 368 K. Our prediction is in error
by only 1 percent! You are invited to make the calculations for
the other hydrogen-bonded substances.
4.2.2 Melting
The heat of fusionthe energy released in freezingis more
dicult to estimate than the heat of vaporization is. We have a
clear picture of what happens when a liquid boils: Molecules leave
the liquid, and join the vapor; boiling destroys the short-range
order in the liquid. What happens when a solid melts? The short-
range order does not change signicantly; atoms in a solid are
closely packed, as are atoms in a liquid. Only the weaker long-
range order changes; the structured lattice of the solid becomes
the chaos of the liquid. This change is dicult to quantify. We can
make an attempt by using the hole model to estimate the entropy
change.
4.2.2.1 Entropy of fusion or melting. We assume that, at the melt-
ing point, every atom is surrounded by 1 hole: We replace one-
sixth of the atoms with holes. Let there be N atoms. What is the
entropy of this conguration? Equivalently, how many arrange-
ments of (5/6)N atoms and N/6 holes are there? The entropy is
the logarithm of this number (times Boltzmanns constant). For
generality, let be the fraction 1/6. There are
W =
_
N
N
_
=
N!
(N)!((1 )N)!
possible arrangements. We have used the symbol W for the num-
ber of states in honor of Boltzmann, whose epitaph reads S =
k log W. To compute S, we need log W:
log W = log N! log{(N)!} log{((1 )N)!}. (4.33)
For large N, we can use Stirlings formula, that log x! xlog x. A
more accurate statement is that log x! xlog x x, but the x
term always cancels in the log of a binomial coecient. Then
log W N log NN log(N)N(1) log{(1)N}. (4.34)
Many terms cancel after we expand the logarithms. What remains
is
log W N( log + (1 ) log(1 )). (4.35)
4. Materials II 73
L
vap
L
fus
Property (10
4
cal mole
1
) (10
3
cal mole
1
)
Water 1.0 1.5
Hg 1.5 0.6
Au 8 10
Table 4.4. Comparison of heats of fu-
sion and of vaporization. The estimate
in (4.40) is reasonably accurate. That
rule compares the second and the third
columns. We can also compare values
within a column, and try to understand
the variation. Water and mercury are
liquids at room temperature, so their
bonds are weak compared to the bonds in
gold, which is a solid at room tempera-
ture. Not surprisingly, gold has higher
heats of vaporization and of fusion.
The entropy per atom is
s =
k log W
N
log (1 ) log(1 ). (4.36)
For = 1/6, the entropy of liquid is s
liquid
0.5k (per atom).
In this model, a solid is a liquid with = 0, so s
solid
= 0. Then
s 0.5k. Experimentally, s k is more accurate; the hole
model accounts for only one-half of the entropy change. As we
noted before, liquids are poorly understood. In what follows, we
use s k, or, per mole, S R.
4.2.2.2 Melting temperature. The entropy is dicult to calculate, and
the melting temperature even harder. T
melt
must be less than the
boiling temperature, but how much less? We make the simplest
assumption, that T
melt
T
vap
. From the entropy and the melting
temperature, we can estimate the heat of fusion:
L
fus
T
melt
S RT
melt
RT
vap
. (4.37)
Equation (4.17) gives us an estimate for L
vap
per molecule: L
vap

10kT
vap
. Per mole,
L
vap
10RT
vap
. (4.38)
When we substitute this estimate into (4.37), we get the estimate
L
fus

L
vap
10
. (4.39)
Then, using (4.15) with
c
0.5 eV (for water), we nd the esti-
mates
L
fus
10
3
cal mole
1
,
L
vap
10
4
cal mole
1
.
(4.40)
Table 4.4 compares heats of fusion and vaporization.
4.3 Specic heat
How much energy does it take to heat water to bath temperature?
How many days of solar heating can the oceans store? The answers
to these questions depend on the energy that the substance stores
per unit temperature change: the specic heat. We approximate
this quantity for metals and for insulators (dielectrics).
Before thinking about the physics of specic heats, we make
the usual dimensional estimate. The units of specic heat are
[specic heat] =
energy
temperature amount of substance
. (4.41)
The amount can be whatever size is convenient: one mole, one
molecule, one gram, and so on. For our dimensional estimate, we
4. Materials II 74
Substance c
p
/k
I
2
4.4
Cl
2
4.1
O
2
3.5
N
2
3.5
Ni 3.1
Au 3.1
Zn 3.1
Fe 3.0
Xe 2.5
He 2.5
C (diamond)
0
C 0.6
223
C 1.6
823
C 2.6
Table 4.5. Specic heats at constant
pressure. All data are for room temper-
ature (unless otherwise noted) and at-
mospheric pressure. We have listed the
specic heats in dimensionless form
in units of k per atom or molecule
because we already know that the specic
heat must contain a factor of k. Source:
[9, p. 155].
choose the molecule, because it is a natural size: It involves the
fewest arbitrary parameters. A mole, or a gram, for example, de-
pends on human-chosen sizes. We already know one quantity with
units of energy per temperature: the Boltzmann constant. So our
rst estimates are
specic heat
molecule
k,
specic heat
mole
kN
A
= R.
(4.42)
Table 4.5 lists the specic heat of various substances.
We now estimate the missing constant of proportionality using
physical reasoning.
4.3.1 Ideal gases
We rst study ideal gases, because they illustrate the basic ideas
of thermal modes, the importance of quantum mechanics, and the
dierence between constant-pressure and constant-volume calcu-
lations. We choose the simplest ideal-gas atom: helium. Hydro-
gen, though containing fewer protons and electrons per atom, is
diatomic: It exists as H
2
. Helium, an inert gas, is monoatomic,
and so more suited for beginning our study.
4.3.1.1 Helium. What is the specic heat of helium? Imagine a collection
of helium atoms. In three dimensions, it has three degrees of free-
dom, one for each dimension. A degree of freedom corresponds to a
quadratic contribution to the energy of a molecule (here, quadratic
means quadratic in a position or velocity coordinate). The trans-
lational energy is m(v
2
x
+ v
2
y
+ v
2
z
)/2; each term corresponds to a
degree of freedom. At temperature T, each degree of freedom con-
tributes energy kT/2 to the internal energy of the atom; so the
internal energy per atom is u = 3kT/2, and the specic heat per
atom is u/T = 3k/2. The kT/2-per-degree-of-freedom rule holds
for only classical degrees of freedom; the meaning of this distinc-
tion will become clear when we discuss nitrogen in (4.3.1.2). The
specic heat should therefore be 1.5k. However, the value given in
Table 4.5 is roughly 2.5k. We somehow neglected a contribution
of kT per atom. What went wrong?
Specic heat measurements are of two types: constant pres-
sure (the more common) and constant volume, depending which
quantity is held xed as the temperature changes. If the volume
of the gas is held constant, then the specic heat is indeed 1.5k;
this specic heat is denoted c
v
. However, if the pressure is con-
stant, we must include another contribution. Consider N atoms of
the gas in a container of volume V . To measure the specic heat,
we increase the temperature by T. If we leave the volume un-
changed, the pressure increases by P = nkT, where n = N/V
is the number density of atoms. To keep the pressure constant, we
must increase the volume by the same fraction that T increased:
by T/T. So V = T(V/T). Let A be the surface area of the
4. Materials II 75
box. Then Adx = V , where dx is how far the walls move. As the
walls recede, the gas does work Fdx = PAdx = PV . Thus, the
heat required to raise the temperature must include PV (T/T),
which, for an ideal gas, is NkT. This energy contributes k to
the specic heat per atom. The specic heat at constant pressure
is therefore c
p
= c
v
+ k, which explains the formerly mysterious
c
p
= 2.5k. From the two specic heats, we can form a dimension-
less group:

c
p
c
v
, (4.43)
which is 5/3 for helium (and any other monoatomic ideal gas).
The value of determines how the temperature changes in
an adiabatic change of state. An adiabatic change of state is
a transformation that is rapid enough so that no heat ows to
or from the substance. Such changes include the change that a
parcel of air owing up the side of a mountain experiences. As
the parcel rises, it moves into regions with lower pressure, so it
expands. This expansion is adiabatic; in an adiabatic expansion,
the relation between pressure and volume is PV

= constant. So
the value of determines how P and V change, which determines,
via the ideal-gas law, how T changes. For air, 7/5, as we see
when we estimate c
p
for nitrogen (the main constituent of air).
4.3.1.2 Nitrogen. Nitrogen is more complicated than helium, because
it has two atoms in a molecule. How many degrees of freedom
does the molecule have? Its center of mass can translate in three
dimensions, which gives three degrees of freedom. The bond can
vibrate, which could give two degrees of freedom, one for the po-
tential energy in the bond, one for the relative motion of the atoms;
it actually gives none, for reasons that will become clear shortly.
The molecule can rotate about any of three axes; three axes could
contribute three degrees of freedom (but actually contribute only
two, as we see shortly). The total is ve degrees of freedom (if we
count none for vibration and two for rotation). Then u = 5kT/2,
so c
v
= 5k/2 and c
p
= 7k/2, which agrees with the tabulated
value.
What justies neglecting the two vibrational degrees of free-
dom, and one of the rotational degrees of freedom? Each evidently
contributes much less energy than kT/2. The reason why is quan-
tum mechanics. The kT/2 rule is valid only for classical degrees
of freedom. Consider, for example, a rotational degree of freedom
(or mode). What makes it classical? In an exact analysis, there
are only certain allowed rotations: Rotation is quantized. When
does the quantization become irrelevant? Suppose that the spac-
ing of the rotational energy levels is ; then mode is classical if
kT . In that case, the energy levels are eectively continuous,
and we can treat the mode with classical statistical mechanics.
What happens if kT ? Then the mode is mostly in the lowest
energy level; the probability that the mode is in a higher level is
roughly the Boltzmann factor e
/kT
, which is minute. The av-
4. Materials II 76
erage energy in the mode is roughly e
/kT
, which is kT/2
when kT . Lets estimate for bond vibrations in nitrogen.
We rst estimate the bond-vibration frequency . The spring con-
stant of the bond is roughly given by k
s
a
2
/2 E
0
. The vibration
frequency is
_
2k/M, where M is the mass of a nitrogen
atom. (We are being careful with factors of 2 in this derivation,
because the result is equivocal if we are not careful, as you can
check by neglecting the factors of 2.) So
2
_
E
0
Ma
2
. (4.44)
In a spring, the energy-level spacing is = , so
2
_
E
0
Ma
2
. (4.45)
To evaluate the spacing, we introduce c
2
/c
2
inside the square root:
2
(c)
2
E
0
(Mc
2
)a
2
. (4.46)
The nitrogen molecule contains a triple bond, so E
0
is substantial,
perhaps 7 eV; and a is short, perhaps 1
A (triple bonds are shorter

than double bonds, which are shorter than single bonds). The mass
of a nitrogen atom is M 14m
p
. So
2
4 10
6
eV
2
A
2
7 eV
14 10
9
eV 1
A
2
, (4.47)
because c 2000 eV
A and m
p
c
2
10
9
eV. After the dust clears,
we nd 0.1 eV. The thermal energy is kT 0.025 eV, so
kT : The vibration mode is not classical. We say that the
vibrations are frozen out: Room temperature is cold compared to
their intrinsic energy (or temperature) scale. A similar calculation
shows that rotation about the interatomic axis is also frozen out.
So quantum mechanics explains why c
v
5k/2. Then c
p
should
be 7k/2, and it is.
4.3.1.3 Iodine. Iodine is also diatomic, but it is more massive than ni-
trogen (larger M); and its bond is weaker (lower E
0
) and longer
(larger a). We estimate
I
by scaling the result for nitrogen. Rela-
tive to nitrogen, the mass of an iodine atom is larger by roughly a
factor of 10; lets say that the bond is weaker by a factor of 2, and
longer by a factor of 2. Then
I

N
/6 0.015 eV. Now kT > ,
so the two bond degrees of freedom unfreeze. (The interatomic-
axis rotational mode is still frozen out.) We therefore expect that,
for iodine, c
v
7k/2, and c
p
9k/2. We are not disappointed:
The tabulated value is c
p
4.4k.
4.3.2 Liquids and gases
4. Materials II 77
We now estimate specic heats of liquids and solids. Liquids and
solids hardly change volume, even if the volume is not held con-
stant, so c
p
c
v
. If each molecule sits in a three-dimensional har-
monic potential produced by the rest of the lattice, each molecule
has 3 potential-energy degrees of freedom. Combining them with
the three translational degrees of freedom produces 6 total de-
grees of freedom. The energy per molecule is therefore 3kT, and
c
p
3k. The specic heat per mole is
C
p
3R 24
J
mole K
6
cal
mole K
, (4.48)
which is a useful number to remember. This value is the lattice
specic heat. The prediction is quite accurate for most of the
solids and liquids (Ni, Au, Zn, and Fe) in Table 4.5. For dia-
mond, the prediction is accurate at only high temperature. As in
Section 4.3.1.2, we have an example of frozen modes. The bonds
in diamond are extremely strong, and therefore have a high vibra-
tion frequency, and large level spacing. At room temperature, the
thermal energy is not large compared to the energy-level spacing,
so most bond vibration is frozen out. Only at T 800
C is kT
large enough to make bond vibration a classical mode; at that tem-
perature, c
p
is 2.6k, which is close to our prediction of 3k. The
other curious fact is that nickel, gold, and zinc have c
p
slightly
greater than 3k! These substances are all metals; we neglected a
contribution to the specic heat that occurs in metals: the specic
heat of the electrons. In metals, the electrons are not bound to an
atom, but are free to move through the lattice; a small fraction of
the electrons can store thermal energy. Electrons therefore make a
small contribution to the specic heat. In Section 4.6.2, we explain
why the fraction of contributing electrons is small.
4.4 Thermal diusivity of liquids and solids
We would like to estimate the rate of heat transportthe thermal
conductivity. One piece in that calculation is the specic heat: how
much heat one molecule, or one mole, stores. The other piece,
which is the topic of this section, is the thermal diusivity.
The thermal diusivity determines how rapidly heat spreads
say, into the center of a turkey, where it denatures (cooks) the
proteins. Heat is the vibration of atoms. In a solid, the atoms
are conned in a lattice, and the vibrations can be represented as
combinations of many sound waves. More precisely, the waves are
phonons, which are sound waves that can have polarization (just
as light waves have polarization). Heat diusion is the diusion of
phonons.
Lets estimate how long it takes for phonons (or heat) to dif-
fuse a macroscopic distance L. Phonons act like particles: They
travel through lattice, bounce o impurities, and bounce of other
phonons. The phonon mean free path measures how far a phonon
travels before bouncing (or scattering), and then heading o in a
4. Materials II 78
random direction. In a solid without too many defects, at room
temperature, typically 10
A. After a scattering, the phonon

heads o in a random direction; so at each scattering, the phonon
takes one step of a random walk with step size . After N steps, it
has gone an average distance
N; this square-root dependence

is characteristic of random walks. See the classic book Random
Walks in Biology [1], which has an excellent treatment of random
walks, illustrated with fascinating examples. So it needs L
2
/
2
steps to travel a distance L. How long does each step take? A
typical phonon velocity is given by the velocity of sound c
e
(the e
indicates elastic wave); for most solids or liquids, c
e
3 kms
1
,
as we found in Section 3.3.1. (We treat a liquid as a disordered
solidas a solid with low .) The time between scatterings is /c
e
.
So the time to travel a distance L is

L
2
c
e
=
L
2
c
e
. (4.49)
The factor in the denominator is a characteristic of the substance,
andwith a magic factor of one-thirdis the thermal diusivity:

1
3
c
e

1
3
10
7
cm3 10
5
cms
1
10
2
cm
2
s
1
. (4.50)
The magic one-third comes from doing honest physicsfrom solv-
ing messy dierential equations, from worrying about boundary
conditions, from not treating every object as a sphere, and so on.
In terms of our new constant , the diusion time is
L
2
/. (4.51)
Lets look at two examples of (4.51).
Example 4.2 Cooking a turkey
We apply our knowledge to Thanksgiving, an American
holiday in which large turkeys are cooked, and many people
give thanks once all the turkey leftovers are nished. How long
does it take to cook a R 20 cm turkey?
turkey

(20 cm)
2
10
2
cm
2
s
1
10
5
s 0.5 days. (4.52)
Start cooking early in the morning! A similar example that
you can try is to predict the cooking time for an egg.
This estimate ignored an important parameter: the oven
temperature. We showed that diusion equalizes the central
temperature to the oven temperature after roughly 0.5 days.
However, the inside must still cook (the proteins must dena-
ture) after reaching this temperature. So now we have the
explanation of why the oven must be hot: A cold oven would
not cook the meat, even after the meat reached oven temper-
ature. What oven temperature is hot enough? Protein physics
4. Materials II 79
is complicated; we cannot expect to estimate the tempera-
ture from theory, so we rely on experiment. We know from
cooking meals that a thin piece of meat next to a hot skil-
let (perhaps at 200
C) cooks in a few minutes; we also know

that if the skillet is at 50
C, the meat will never cook (50
C
is not much hotter than body temperature). So if we set the
oven to 200
C, the turkey will cook in a few minutes after

attaining oven temperature; most of the cooking time is spent
attaining oven temperature, which is the assumption that we
made when we estimated the cooking time as L
2
/.
Example 4.3 Cooling the moon
How long does it take for the moon (R 1.7 10
8
cm) to
cool? For the moon, we nd that
moon

(1.7 10
8
cm)
2
10
2
cm
2
s
1

3 10
16
10
2
s 3 10
18
s 10
11
yr.
(4.53)
[Another useful number: 1 yr 3 10
7
s.] The universe has
been around for only 10
10
yr; so why is the moon cold already,
especially if heat is generated at its core by radioactive heat-
ing? Because of solid-state convection: Rock ows slowly (on
a billion-year timescale), and transports heat to the surface
much faster than molecular motions could.
This moon example shows one merit of order-of-magnitude
physics: eciency. We couldwith lots of computer time
solve the diusion equation for a mostly-spherical body like
the moon. We might nd that the more accurate cooling time
is 6.74 10
10
yr. What would that result gain us? The more
accurate time is still far longer than the age of the universe;
we would still conclude that convection must be responsible.
Order-of-magnitude analysis allow you to determine quickly
which approaches are worth pursuing; you can then spend
more time rening productive approaches, and less time chas-
ing unproductive ones.
4.5 Diusivity and viscosity of gases
Before estimating the thermal conductivity of liquids and solids,
we pause to harvest results that follow easily from our understand-
ing of the mechanism of diusivity: We estimate the thermal dif-
fusivity and the viscosity of a gas. In a solid or liquid, the atoms
themselves do not transport the heat that they store; phonons,
which are much more mobile than the atoms, do. In a gas, the
analogue of phonons is sound waves, but sound waves do not carry
the heat in a gas; the atoms (or molecules) themselves carry the
heat. We can estimate the thermal diusivity using (4.50), but
with parameters for a gas:
gas

1
3
v. (4.54)
4. Materials II 80
Now, is the mean free path of a gas molecule, and v is the
velocity of the molecule. The velocity we can estimate easily
it is the thermal velocity. The mean free path is given by
(n)
1
, where n is the number density of molecules, and is
the cross-sectional area of a molecule. This result is dimensionally
correct (a useful sanity check), and it has the right quantities
downstairs: Larger molecules scatter after a shorter distance than
smaller molecules do, and denser gases cause a molecule to scatter
sooner than a less dense gas does. So
gas

1
3
v
n
. (4.55)
For air at room temperature and atmospheric pressure, the cross
section is a
2
, where a 3
A; the number density is n

310
19
cm
3
; and the thermal velocity is v 310
4
cms
1
(roughly
the speed of sound). Then
gas

1
3
3 10
4
cms
1
3 10
19
cm
3
10
15
cm
2
0.3 cm
2
s
1
.
(4.56)
This result is too large by a factor of only 1.5: the measured value
is 0.2 cm
2
s
1
. Most of the error is in our estimate of .
What about kinematic viscosity, ? Thermal diusivity is the
heat- or energy-diusion coecient, and viscosity is the momen-
tum-diusion coecient. In a gas, momentum and energy are
transported by the same mechanism: molecular motion. So we
estimate that , and we are rewarded: 0.15 cm
2
s
1
.
Kinematic viscosity and thermal diusivity have the same units.
A natural dimensionless group is therefore
Pr

, (4.57)
which is the Prandtl number. It is close to unity in gases, and even
in many solids and liquids (for water, it is 6).
4.6 Thermal conductivity
Why can people (who are not too condent) walk on hot coals
without getting burned? How long does the 3-inch layer of frost in
the freezer take to melt? The answers to these questions depend
on the thermal conductivityhow fast a material can trans-
port heat. Now that we have estimated the specic heat and the
thermal diusivity, we can estimate the thermal conductivity.
4.6.1 Dielectrics
We rst study the simpler case of dielectrics, where we can neglect
the electronic contribution to thermal conductivity. The thermal
conductivity is dened as the constant K in the heat-ow equation
F = K
dT
dx
, (4.58)
4. Materials II 81
K
Substance
_
10
3
cal
cms K
_
Water 5
Glass 3
Rock 2
Wood 0.3
Asbestos 0.2
Table 4.6. Thermal conductivities for
common dielectrics. Note that we chose
the typical dielectric thermal conduc-
tivity as the unit size: We scaled the
value relative to a reasonable estimate.
Scaled means that we measure the phys-
ical quantities in units that we expect to
be reasonable: in units that should yield
a number near unity.
where F is the heat ux (power per area), and T is the tem-
perature. We can estimate the value of the constant using our
understanding of specic heat and thermal diusivity. Consider a
rectangle of material, with length x and cross-sectional area A.
Heat one half (measuring along the length) to temperature T,
and start the other at T = 0. One half contains a quantity of
heat H C
p
TxA, where C
p
is the heat capacity per volume.
After a time x
2
/, the temperature of the two halves equal-
izes (because of diusion); thus, heat must ow from the hot to
the cold half at a rate dotH H/ C
p
AT/x. The heat ux is
F

H/A, or F C
p
T/x. Now we match terms with the dening
equation (4.58); dT/dx corresponds to T/x, so C
p
corresponds
to K. In fact, this relation is exact: K = C
The specic heat per volume is C

p
/V
mol
, where V
mol
is the
molar volume. Therefore,
K =
C
p
V
mol
. (4.59)
In our order-of-magnitude world, the volume of one molecule is
a
3
, where a 3
A. So V
mol
N
A
a
3
, which is 16 cm
3
mole
1
. We
also substitute from (4.50), and C
p
from (4.48), and we nd
K
6 cal
mole K
10
2
cm
2
s
1
mole
16 cm
3
10
2
cal
cms K
. (4.60)
We neglected a factor of 2.5 so that the answer comes out in
a round number. This value includes only the lattice transport
of heat; it applies to dielectrics, but not to metals, where the
electrons transport most of the heat. Table 4.6 contains thermal
conductivities for dielectrics.
4.6.2 Metals
Metals feel colder to the touch than, say, concrete does because
metals transport heat faster than concrete does. If lattice conduc-
tion is the only factor responsible for heat conduction, then metals
and concrete should not have such a great disparity in conductiv-
ity. That they do is a consequence of the large contribution to
the thermal conductivity from free electrons in a metal. [A re-
lated example: A slab of granite feels much colder than a slab of
cork does. Why does granite carry heat more eciently than cork,
even though both substances are dielectrics, and therefore carry
heat via lattice conduction? The answer is that granite is much
denser than cork, so its specic heat per volume is much larger
than corks.]
Lets estimate the ratio of electronic to lattice thermal con-
ductivity. Because the conductivity factors into
diusivity specic heat, (4.61)
we rst estimate the ratio of electronic to lattice diusivities. Dif-
fusivity is
velocity mean free path, (4.62)
4. Materials II 82
so we subdivide the estimate of the diusivity ratio into estimates
of the velocity ratio and of the mean-free-path ratio.
Phonons move at speeds similar to the speed of sounda few
kilometers per second. Electrons move at the Fermi velocity
v
F
. We have seen this concept in disguise when we studied white
dwarfs in Section 3.4. In a white dwarf, the electrons are conned
by gravity; in a metal, they are conned by electrostatics (a metal
is like a giant hydrogen atom). In both systems, the electrons are
conned, which gives them a velocity. We could modify the white-
dwarf analysis to account for these changes; instead, we just redo
the relevant part. Let n
e
be the number density of free electrons
in the metal. Then p /x n
1/3
e
, where p is the momen-
tum uncertainty produced by connement, and x is the position
uncertainty. The velocity is the Fermi velocity: v
F
n
1/3
e
/m
e
.
A typical metal has two free electrons per atom, and an inter-
atomic spacing of a 3
A. So n
e
2/a
3
10
23
cm
3
. Then
v
F
8 cms
1
= 1000 kms
1
. This velocity is much faster than a
typical elastic wave So the velocity ratio is 300 (electrons move
faster).
Electrons also have a much longer mean free path than do
phonons; in copper,
e
100a, where a 3
A is a typical lattice
spacing. The phonon mean free path is
p
10
A. So the mean-
free-path ratio is 30, and the diusivity ratio is 300 30 10
4
in favor of the electrons.
We now have to compute the specic-heat ratio. Here, the
phonons win, because only a small fraction of the electrons carry
thermal energy. To understand why, we have to improve the Fermi-
sphere argument that we used to compute v
F
. A more accurate
picture is that the metal is a three-dimensional potential well (the
electrons are conned to the metal). There are many energy levels
in the well, which we can label by the momentum of the elec-
trons that occupy it. The free electrons ll the lowest levels rst
(two per level); the Pauli principle forbids more than two electrons
from being in the same energy level. The electrons in the highest
levels have velocity v
F
(that part of the previous argument was
correct), or momentum m
e
v
F
. As vectors, the momenta of the
highest-energy electrons are uniformly distributed over the sur-
face of a sphere in momentum space; the sphere has radius m
e
v
F
,
and is called the Fermi sphere. How is this sphere relevant to spe-
cic heat? To carry thermal energy, an electron has to be able to
absorb and deposit energy; the typical package size is kT. Con-
sider an electron that wants to absorb a thermal-energy package.
When it does so, it will change its energyit will move outward
in the Fermi sphere. But can it? If it is in most of the sphere, it
cannot, because the sphere is packed solidall interior levels are
lled. Only if the electron is near the surface of the sphere can
it absorb the package. How near the surface must it be? It must
have energy within kT of the Fermi energy E
F
(the Fermi energy
is m
e
v
2
F
). The fraction of electrons within this energy range
4. Materials II 83
Substance K
_
cal
cms K
_
Al 0.5
Cu 1.0
Fe 0.2
Hg 0.02
W 0.5
Table 4.7. Thermal conductivities for
common metals at room temperature.
Metals should have thermal conductiv-
ities of roughly 1 erg cm
1
s
1
K
1
, so
we measure the actual values in that
unit. We are being gentle with our neu-
ral hardware, giving it the kind of num-
bers that it handles with the least di-
culty: numbers close to 1. Copper has a
large thermal conductivity, because the
electron mean free path is long (which is
the same reason that copper has a high
electrical conductivity). Mercury has a
low thermal conductivity, because it is a
liquid at room temperature, so the elec-
tron mean free path is short (only one or
two lattice spacings).
is f kT/E
F
. Typically, kT 0.025 eV and E
F
few eV, so
f 10
2
. This fraction is also the specic-heat ratio. To see why,
consider the case where f = 1 (every electron counts). Each atom
contributes 3k to the specic heat, and each electron contributes
one-half of that amount3k/2because there is no spring poten-
tial for the electrons (they contribute only translational degrees of
freedom). The number of free electrons is typically twice the num-
ber of atoms, so the total electron and phonon contributions to
specic heat are roughly equal. When f 10
2
, the contributions
have ratio 10
2
.
Now we have all the pieces. The conductivity ratio is
K
metal
K
dielectric
10
4
10
2
= 10
2
. (4.63)
We use (4.60) to estimate K
dielectric
, and nd that
K
metal
1
cal
cms K
. (4.64)
Note that this estimate is for a good conductor, such as copper,
with a large electron mean free path. Table 4.7 contains data on
the thermal conductivities of common metals. Our estimate in
(4.64) is quite accurate for coppermore accurate than it ought
to be given the number of approximations that we made to derive
it.
The method of successive approximation: How to solve compli-
cated equations one step at a time. We saw two examples of
this method: in estimating the thermal-expansion coecient,
and in solving for the boiling point as a function of cohesive
energy.
How to handle logarithms: Every (natural) log is 10. This rule
is often helpful for beginning a successive-approximation solu-
tion.
The microscopic basis of thermal diusivity and viscosity: Par-
ticles (or phonons) move in steps whose size is equal to the
mean free path, . The particles velocity v determines the
time to take one step, and therefore the diusion constant,
which is v.
85
5 Waves
Ocean covers most of the earth, and waves roam most of the ocean.
What makes waves move, and what determines their speed? We
now apply and extend our techniques of approximation to answer
such questions, as we study water waves. The theme of this chapter
is: Consider limiting cases.
5.1 Dispersion relations
We study waves using dispersion relations. A dispersion rela-
tion states what values of frequency and wavelength a wave can
have; see for example, Crawfords book [5] for more information
on dispersion relations. In their usual form, dispersion relations
connect frequency , and wavenumber k. Wavenumber is dened
as 2/, where is the wavelength. As an example, for electro-
magnetic waves in a vacuum, the frequency and wavenumber are
related by the dispersion relation = ck, which states that waves
travel at velocity /k = c, independent of frequency. Dispersion
relations contain a vast amount of information about waves. They
tell us, for example, how fast crests and troughs travel: the phase
velocity. They tell us how fast wave packets travel: the group
velocity. They tell us how these velocities depend on frequency:
the dispersion. And they tell us the rate of energy loss: the at-
tenuation.
You usually nd a dispersion relation by solving a wave equa-
tion, which is an unpleasant partial dierential equation. For wa-
ter waves, you obtain a wave equation by linearizing the Euler
equation of hydrodynamics. This procedure is mathematically in-
volved, particularly in handling the boundary conditions. Being
impatient and lazy, we instead derive dispersion relations with di-
mensional analysis, then complete and complement the derivation
with physical arguments. We cannot evaluate the dimensionless
constants with our sloppy methods, but the beauty of studying
waves is that these constants are often unity.
How can we connect frequency and wavenumber? The two
quantities have dimensions [T]
1
and [L]
1
, respectively. As glue,
we need to include more variables, with various dimensions. So, as
usual, we consider what physical properties of the system deter-
mine wave behavior, and thereby form a set of relevant variables.
Waves on the open ocean are dierent from waves in your bath-
tub, presumably because of the dierence in the depth of water,
which we call h. The width of the tub or ocean could matter, but
we neglect such eects, and consider waves that move in only one
5. Waves 86
dimension, perpendicular to the width.
To determine what other variables are important, we note use
the principle that waves are like springs, because every physi-
cal process contains a spring. This blanket statement cannot be
strictly correct. However, as a broad generalization, it is use-
ful. We can get a more precise idea of when this assumption is
useful by considering the characteristics of spring motion. First,
springs have an equilibrium position. If your system has an undis-
turbed, resting state, consider looking for a spring. For example,
for waves on the ocean, the undisturbed state is a calm ocean.
For electromagnetic wavessprings are not conned to mechan-
ical systemsthe resting state is an empty vacuum, with no ra-
diation. Second, springs oscillate. In mechanical systems, oscilla-
tion depends on inertia carrying the mass beyond the equilibrium
position. Equivalently, it depends on kinetic energy turning into
potential energy, and vice versa. Water waves store potential en-
ergy in the disturbance of the surface, and kinetic energy in the
motion of the water. For electromagnetic waves, energy sloshes
(oscillates) between the electric eld and the magnetic eld. A
magnetic eld can be generated by moving charges, so we call
the magnetic eld the reservoir of kinetic (motion) energy. An
electric eld can be generated by static charges, so we call the
electric eld the reservoir of potential energy. With these iden-
tications, the electromagnetic eld is a set of springs, one for
each frequency. These examples are positive examples. A negative
examplea marble oozing its way through corn syrup, which we
studied in Section 2.3.2illustrates when springs are not present.
The marble moves so slowly that the kinetic energy of the corn
syrup, and therefore its inertia, is miniscule and irrelevant. In such
a system, there is no reservoir of kinetic energy, so a spring is not
present.
We now return to the problem of waves, which have the neces-
sary reservoirs. In its lowest-energy state, the surface of the water
is level; deviations from atnesswavesare opposed by a restor-
ing force. In a real spring, the restoring force comes from inter-
atomic bonds. Once we know the origin of a spring constant, we
can calculate it; from the spring constant, we can calculate the
oscillation frequency. Our model that waves are springs suggests
that we study the restoring force in a wave. Distorting the sur-
face is like stretching a rubber sheet; the surface tension of water
opposes the distortion. Distorting the surface also requires rais-
ing the average water level, a change that gravity opposes. The
restoring force is therefore a combination of gravity and surface
tension. In our list of variables, then, we include surface tension,
, and gravity, g. We consider how gravity (or surface tension) re-
stores the water level when we introduce a simple physical model
in Section 5.2.1.
In a wave, like in a spring, the restoring force ghts inertia, rep-
resented here by the uid density . Gravity does not care about
5. Waves 87
density, because gravitys stronger pull on denser substances is ex-
actly balanced by their greater inertia. This exact cancellation is
a restatement of the equivalence principle, on which Einstein
based his theory of general relativity. In pendulum motion (Sec-
tion 2.2), the mass of the bob drops out of the nal solution for the
same reason. However, surface tension does not grow stronger in
proportion to density. So we expect to determine the properties
of waves in which surface tension provides the restoring force. We
therefore include in the list of variables. For concreteness, we
often refer in this chapter to water waves; the results are general,
and work for any uid. We pretend that waves are lossless, and
exclude viscosity from the set of variables. For simplicity, we also
exclude c
s
, the speed of sound in the medium (usually water); with
this approximation, we ignore sound waves. We also exclude the
wave amplitude from our dimensional analysis by assuming that
is small compared to the other length scales (wavelength and
depth): We study only small-amplitude waves. Table 5.1 shows
the list of variables that we have collected.
These six variables consist of three fundamental dimensions.
We therefore hunt for 6 3 = 3 dimensionless groups. One group
is easy: k is an inverse length, and h is a length, so we can form
1
kh. (5.1)
This group is the dimensionless depth of the water:
1
1 means
shallow and
1
1 means deep water. A second dimensionless
group comes from gravity. Gravity, represented by g, has the same
units as
2
, except for a factor of length. The wavenumber xes
this decit:
2
=

2
gk
. (5.2)
Notice that this group does not contain density. If we exclude
surface tension, then
1
and
2
are the only dimensionless groups
that we can form; without surface tension, the waves propagate
because of gravity alone. The equivalence principle tells us that
the way in which gravity aects motion is independent of density.
Therefore, density cannotand does notappear in either group.
Now we consider what happens when we allow surface tension
back into the playpen of dimensionless variables. It must belong in
the third group. Knowing only that belongs to
3
, we still have
great freedom in choosing the form of
3
. We choose it so that it
measures the relative importance of gravity and surface tension in
determining the wave motion. Surface tension has units of force
per length, whereas g has units of force per volume, or force per
length cubed. So the ratio /g has units of length squared. We
can neutralize the two excess lengths with two factors of k:
3

k
2
g
. (5.3)
We could also have built this group without using gravity, choos-
ing
3
k
3
/
2
. There are many possibilities for dimension-
less groups. The choices that we made, which emphasize gravity
5. Waves 88
Var. Dimensions Description
[T]
1
angular frequency
k [L]
1
wavenumber
g [L][T]
2
acceleration due to gravity
h [L] depth of water
[M][L]
3
density
[M][T]
2
surface tension
behavior of water waves.
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.3. Deep-water portion of the
world of waves.
over surface tension, are convenient for studying waves restored
by gravity; most ocean waves are of that ilk.
We want to solve for frequency, , as a function of wavenumber,
k. We could also solve for k versus , but the relations for phase
and group velocity are simpler with as a function of k. Only the
group
2
contains , so we write the most general dimensionless
relation as
2
= f(
1
,
3
), (5.4)
or
2
gk
= f
_
kh,
k
2
g
_
. (5.5)
Solving for
2
, we nd that
2
= gk f(kh,
k
2
g
). (5.6)
This relation is valid for waves in shallow or deep water; for waves
propagated by surface tension or by gravity; and for waves in
between. Figure 5.1 shows how the two groups
1
and
3
di-
vide the world of waves into four regions. Figure 5.2 shows how
wavelength and depth partition the world, and gives examples of
dierent types of waves.
We do not as yet know the form of the magic function f in
any region, let alone for all regions. To determine its form, and to
understand its consequences, we make a simple physical model, by
considering limiting cases. Like a jigsaw-puzzle-solver, we work
out the corners of the world rst, because the physics is simplest
there. Then, we connect the solutions along the edges, where the
physics is the almost as simple. Finally, we crawl our way inward
to assemble the complicated, more complete solution. This chapter
is a large example of the divide-and-conquer approach to solving
problems, where limiting cases are the pieces.
5.2 Deep water
We rst study deep water, where kh 1, as shaded in the map in
Figure 5.3. Deep water is dened as water suciently deep that
waves cannot feel the bottom of the ocean. How deep do waves
feelers extend? The only length scale in the waves is the wave-
length, = 2/k. We expect, therefore, that the feelers extend
to a depth d 1/k (as always, we neglect constants, such as 2).
We can justify this guess by using Laplaces equation, which is the
spatial part of the wave equation. Suppose that the waves are pe-
riodic in the x direction, and z measures depth below the surface,
5. Waves 89
10
4
10
2
10
2
10
4
1
hk
10
4
10
2
10
2
10
4
3

k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.1. Map of the world of waves.
The dimensionless groups
1
and
3
partition the world into four regions. We
study them in turn, and combine our
analyses to understand the whole world
(of waves). The group
1
measures the
depth of the water: Are we in a puddle
or an ocean? The group
3
measures
the relative contribution of gravity and
surface tension: Are the waves ripples or
gravity waves?
The division into deep and shal-
low water (left and right) follows from
the interpretation of
1
= kh as di-
mensionless depth. The division into
surface-tension- and gravity-dominated
waves (top and bottom) is more sub-
tle. You can understand the division
by looking at the form of
3
, which is
k
2
/g. Large g, or small k, result in
the same consequence: small
3
. There-
fore, the physical consequence of longer
wavelength is similar to that of stronger
gravity: Longer-wavelength waves are
gravity waves. The large-
3
portion of
the world (top) is therefore labeled with
surface tension.
as shown in Figure 5.4. Then, Laplaces equation becomes
x
2
+

2
z
2
= 0, (5.7)
where is the velocity potential. Its not important what exactly
is. You can nd out more about it in a uid mechanics textbook.
Our favorite is Fabers Fluid Dynamics for Physicists [7]; Lambs
Hydrodynamics [12] is a classic, but more dicult. For this argu-
ment, all that matters is that measures the eect of the wave;
that where = 0, theres no wave; and that satises Laplaces
equation. The wave is periodic in the x direction, with a form such
as sin kx. We take
Z(z) sin kx. (5.8)
The function Z(z) measures how the wave decays with depth.
The second derivative in x brings out two factors of k, and a
minus sign:
x
2
= k
2
. (5.9)
So that satises Laplaces equation, the z-derivative term must
contribute +k
2
. Therefore,
z
2
= k
2
, (5.10)
so Z(z) e
kz
. The physically possible solutionthe one that
does not blow up exponentially at the bottom of the oceanis
Z(z) e
kz
. Therefore, relative to the eect of the wave at the
surface, the eect of the wave at the bottom of the ocean is e
kh
.
When kh 1, the bottom might as well be on the moon. The
5. Waves 90
1
= 1
3
= 1
10
4
10
2
10
2
10
4
depth(m)
10
4
10
2
10
2
10
4
10
6
wavelength (m)
Tide
Storm wave at sea
Storm wave at shore
Ripple on pond
Ripple on thin puddle
Deep water
Gravity waves
Shallow water
Gravity waves
Deep water
Ripples
Shallow-water
Ripples
Figure 5.2. Water waves on the earth.
Here, we partition the world using the
familiar variables of depth and wave-
length for the axes. The thick dividing
lines are based on the dimensionless
groups
1
= hk and
3
= k
2
/g. Each
region contains one or two examples of
its kind of waves. With g = 1000cms
1
and 1 g cm
3
, the border wavelength
between ripples and gravity waves is just
over 1 cm (the horizontal,
3
= 1
dividing line).
dimensionless factor khit has to be dimensionless to sit alone in
an exponent as it doescompares water depth with feeler depth
d 1/k:
water depth
feeler depth

h
1/k
= hk, (5.11)
which is the dimensionless group
1
.
In deep water, where the bottom is hidden from the waves,
the water depth h does not aect the waves propagation, so h
disappears from the list of relevant variables. When h goes, so
does
1
= kh. There is one caveat. If
1
is the only variable that
contains k, then we cannot blithely discard it, just because we no
longer care about h. Fortunately, k appears in
3
= k
2
/g as
well. So its safe to toss
1
for deep water. We have just argued
that h is irrelevant based on a physical argument. This argument
has a mathematical equivalent that we can express in the lan-
guage of dimensionless groups and functions. The statement that
h is large is meaningless, because h has units. The question that
remains is, large compared to what length? When we choose 1/k
as the standard of comparisonbased on the Laplaces-equation
argumentwe can rephrase the meaningless h is large statement
to
1
= kh is large. Using the Laplaces-equation argument, we
drop
1
because h is irrelevant. Mathematically, we are assuming
that the function f(kh, k
2
/g) from (5.6) has a limit as kh .
Without
1
, the general dispersion relation (5.6) simplies to
2
= gk f
deep
(
k
2
g
). (5.12)
This relation describes the deep-water edge of the world of waves.
The edge has two corners, labeled by whether gravity or surface
5. Waves 91
y
x
z
w
2
h
Figure 5.4. Piece of an innite wave.
The wave, with a sinusoidal prole, has
wavelength . The water has depth h,
and we study a width w in the y direc-
tion.
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.5. Deep-watergravity-wave
portion of the world of waves.
1. When the reporters Woodward and
Bernstein [2] were investigating crim-
inal coverups in the Nixon administra-
tion, they received help from the mys-
terious Deep Throat, whose valuable
advice was to follow the money.
tension provides the restoring force. Although we do not know
the form of f
deep
, it is a function simpler than the original f. To
determine the form of f
deep
, we partition deep-water waves into
its two limiting cases: gravity waves and ripples.
5.2.1 Gravity waves on deep water
Now we specialize to regions of the wave world where water is
deep and gravity is strong (Figure 5.5). This category includes
most waves generated by wind, and wakes generated by ships.
With gravity much stronger than surface tension, the dimension-
less group k
2
/g limits to 0 (pretend that we cranked up g, or,
by dumping soap on the water, that we turned down ). So the
general deep-water dispersion relation (5.12) simplies to
2
= f
deep
(0)gk = C
1
gk, (5.13)
where f
deep
(0) is an as-yet-unknown constant, C
1
. [We do not
know that f
deep
(x) has a limit as x 0. The slab argument, which
follows shortly, shows that it does.] The constant remains unknown
to our lazy methods, because the methods trade evaluation of
dimensionless constants for comprehension of physics. We usually
assume that such constants are unity. In this case, we get lucky:
An honest calculation does produce C
1
= 1. Therefore,
2
= gk. (5.14)
Such results from dimensional analysis seem like rabbits jump-
ing from a hat. The relation (5.14) is correct, but our gut is not
happy with this magical derivation; it asks, Why is the result
true? We can understand the origin of the dispersion relation
by making a simple physical model of the forces or energies that
drive the waves. The rst step is to understand the mechanism:
How does gravity make the water level rise and fall? Taking a hint
from the Watergate investigators,
1
we follow the water. The water
in the crest does not move into the trough. Rather, the water in
the crest, being higher, creates a pressure underneath it higher
than that of the water in the trough, as shown in Figure 5.6. The
higher pressure forces water underneath the crest to ow toward
5. Waves 92
p
>
p
0
+g(z +) p
<
p
0
+g(z )
z +
z
Flow
===
p
0
=
1
a
t
m
Figure 5.6. Higher pressure under the
crest. The crest rises above the trough
by roughly , the wave amplitude. This
extra water creates a pressure under-
neath the crest larger than that un-
der the trough, by roughly p g.
This pressure dierence makes water
ow from under the crest to under the
trough: The wave advances.
the trough, making the water level there rise. Like a swing sliding
past equilibrium, the surface overshoots the equilibrium level, to
produce a new crest, and the cycle repeats.
Our next step is to make this model quantitative, by estimating
sizes, forces, speeds, and energies. In Section 1.1.4, we replaced a
messy mortality curve with a more tractable shape: a rectangle.
We use the same trick repeatedly for this model. Water underneath
the surface moves quickly because of the pressure gradient. Farther
down, it moves more slowly. Eventually, it does not move at all.
We replace this fallo with a step function, and pretend that water
down to a certain depth moves as a block, and deeper water stays
still (Figure 5.7). How deep should this slab of water extend?
By the Laplace-equation argument, the pressure variation falls o
exponentially with depth, with length scale 1/k. So we assume
that the slab has depth 1/k. (What choice do we have? On an
innitely deep ocean, there is no length scale other than 1/k.) How
long should the slab be? Its length should be roughly the peak-to-
trough distance, because the surface height changes signicantly
in that distance. This distance is 1/k. Actually, its /k, but we
ignore the constants, such as . All such constants combine into a
constant at the end, which we cannot determine with dimensional
analysis anyway, so why not discard it now? The width of the slab,
w, is arbitrary. It drops out in all the analyses.
So the slab of water has depth 1/k, length 1/k, and width w.
What forces act on it? We can determine the forces by estimating
the pressure gradients. Across the width of the slab (the y direc-
tion), the water surface does not vary, so there are no pressure
variations in that direction. Across the depth (the z direction),
the pressure varies because of gravity, but that variation is just
sucient to prevent the slab from sinking. We care about only the
pressure dierence across the length. This dierence depends on
the height of the crest, , and is p g (see Figure 5.6). This
pressure dierence acts on a cross-section with area A w/k, to
produce a force
F w/k
..
area
g
..
p
= gw/k. (5.15)
5. Waves 93
w
1/k
1/k
A
top
w/k
Lower
pressure
Higher
pressure
Figure 5.7. Slab of water. The pres-
sure is higher under the crest (shaded
side on the left) than toward the trough
(shaded side on the right). The shaded
sides have area A w/k. From Fig-
ure 5.6, this extra pressure is p g,
and it acts on an area A w/k to pro-
duce a force F Ap wg/k. The
slab has mass m w/k
2
. In the text,
we use this mass to compute the acceler-
ation and kinetic energy of the slab.
The slab (Figure 5.7) has mass
m = w/k
2
. .
volume
, (5.16)
so this force produces an acceleration
a
slab

gw
k
. .
force
_
w
k
2
..
mass
= gk. (5.17)
The factor of g says that the gravity produces the acceleration. Full
gravitational acceleration is reduced by the dimensionless factor
k, which is roughly the slope of the waves.
From the acceleration of the slab, we can calculate the acceler-
ation of the surface. If the slab moves a distance x, it sweeps out a
volume of water V xA. This water moves under the trough, and
forces the surface upward a distance V/A
top
. Because A
top
A
(both are w/k), the surface moves the same distance, x, that
the slab moves. Therefore, the acceleration of the slab, a
slab
, given
in (5.17), is equal to the acceleration a of the surface:
a a
slab
gk. (5.18)
As we discover in Section 5.3, this equivalence of slab and surface
acceleration does not hold in shallow water, where the bottom at
depth h cuts o the slab before 1/k.
With this slab argument, we hope to obtain the deep-water dis-
persion relation that we derived by dimensional analysis in (5.14).
That equation contains frequency; the acceleration relation (5.18),
however, does not. So we massage it until appears. The accel-
eration relation contains a and , whereas the dispersion relation
does not. We therefore manufacture an alternative expression for
the acceleration. With luck, the expression will contain
2
and
also , so that we can cancel and insert
2
. In simple harmonic
motion (springs!), acceleration is a
2
, where is the ampli-
tude. In waves, which behave like springs, we can use the same
5. Waves 94
w
1/k
Figure 5.8. Energy required to make a
wave. The shaded block of water has to
be lifted up (and over, but sliding over
takes no energy) a height . As ex-
plained in the text, water does not actu-
ally move from trough to crest, but as-
suming that it does makes a convenient
picture, and does not aect the energy
calculation.
expression for a. We can also derive it for waves: In time 1/,
the surface moves a distance d , so a/
2
, or a
2
. Then
(5.18) becomes
..
a
gk, (5.19)
or
2
gk, (5.20)
which is the dispersion relation (5.14).
This equation lacks the dimensionless constant, which as usual
we assume is unity. Fortunately, an exact calculation conrms this
lucky guess. Our luck suggests that our procedures for choosing
how to measure the lengths were reasonable. We made two ap-
proximations:
We replaced an exponentially falling variation in velocity po-
tential by a step function with size equal to the length scale of
the exponential decay.
By taking the length of the slab to be 1/k instead of /k, we
used only 1 radian of the cycle as the characteristic length,
instead of using a half cycle.
Both approximations are usually accurate in order-of-magnitude
calculations. Rarely, however, you get killed by a factor of (2)
6
,
and wish that you had used a full cycle instead of only 1 radian.
The derivation that resulted in (5.20) analyzed the motion of
the slab using forces. We can also derive the dispersion relation
using energy, by balancing kinetic and potential energy as we did
for the Bohr atom in Section 3.1.2. To make a wavy surface, we
require energy, as shown in Figure 5.8. The crest rises a character-
istic height above the zero of potential, which is the level surface.
The volume of water moved upward is w/k. So the potential en-
ergy is
PE
gravity
w/k
. .
m
g gw
2
/k. (5.21)
The kinetic energy is contained in the sideways motion of the slab
and in the upward motion of the water pushed by the slab. Be-
cause slab and surface move at approximately the same speed,
5. Waves 95
the sideways and upward motions contribute similar energies. We
evaluate the energy for just the sideways motion. Because we are
ignoring constants, such as 2, we do not need to compute the en-
ergy contributed by the upward motion. The surface, and therefore
the slab, moves a distance in a time 1/, so its velocity is .
The kinetic energy is
KE
deep
w/k
2
. .
m
slab
2
..
v
2

2
2
w/k
2
. (5.22)
We balance this energy against the potential energy (5.21):
2
w/k
2
. .
KE
gw
2
/k
. .
PE
. (5.23)
Canceling the common factor of w
2
, we nd, once again, that
2
gk. (5.24)
The energy method agrees with the force method, as it should,
because energy can be derived from force (by integration). The
energy derivation tell us what the dimensionless group
2
means:
2

kinetic energy in slab
gravitational potential energy

2
gk
. (5.25)
The gravity-wave dispersion relation,
2
= gk, is equivalent to
2
1, or to the assertion that kinetic and gravitational poten-
tial energy are comparable in wave motion. This rough equality is
not a surprise for waves, because waves are like springs; in spring
motion, kinetic and potential energies have equal averages (which
is a consequence of the virial theorem, which we discussed in Sec-
tion 3.2.1).
We have now derived the same dispersion relation in three
ways: with dimensional analysis, and with two physical arguments
applied to a simple model. Using multiple methods increases our
condence not only in our result, but also in our methods. We gain
condence in our methods of dimensional analysis, and in our slab
model for waves. Were we to study nonlinear waves, for example,
where we cannot assume that the wave height is innitesimal,
we would be able to use our techniques and model with more
condence.
Now that we believe the dispersion relation, lets study its
consequences. We rst determine the phase and group velocities.
The crests move at the phase velocity: v
ph
= w/k. For deep-water
gravity waves, this velocity becomes
v
ph
=
_
g
k
, (5.26)
or, using the dispersion relation to replace k by ,
v
ph
=
g
. (5.27)
5. Waves 96
In an innite, single-frequency wave train, the crests and troughs
move at this speed. However, a nite wave train contains a mix-
ture of frequencies, and the various frequencies move at dierent
speeds, as given by (5.27): Deep water is dispersive. Dispersion
makes a nite wave train travel with the group velocity, given by
v
g
= w/k, as explained in Fluid Dynamics for Physicists [7].
Within a wave train, the crests move at the phase velocity, which
can be dierent from the group velocity, shrinking and growing to
t under the envelope of the wave train. The group velocity is
v
g
=

k
_
gk =
1
2
_
g
k
=
1
2
v
ph
. (5.28)
The group velocity is one-half of the phase velocity. The everyday
consequence is that ship wakes trail a ship. A ship, moving with
velocity v, creates gravity waves with v
ph
= v. (Wind also gener-
ates waves, and the wind speed corresponds roughly to the phase
velocity, as we discover shortly.) The waves combine to produce
wave trains that propagate forward with the group velocity, which
is only one-half of the velocity of the ship. From the ships point
of view, these gravity waves travel backward. In fact, they form a
wedge, and the opening angle of the wedge depends on the factor
of 1/2.
5.2.1.1 Surfing. Lets apply the dispersion relation to surng. Following
one winter storm reported in the Los Angeles Timesthe kind of
storm that brings cries of Surfs upwaves arrived at Los Ange-
les beaches roughly every 18 s. How fast were the storm winds that
generated the waves? Wind pushes crests, as long as they move
more slowly than the wind. After a long-enough push, the crests
move with nearly the wind speed. Therefore, the phase velocity of
the waves is an accurate approximation to the wind speed. The
phase velocity, from (5.27), is g/. In terms of the wave period T,
the velocity is v
ph
= gT/2, so
v
wind
v
ph

g
..
10 ms
2
T
..
18 s
2 3
30 ms
1
. (5.29)
In units more familiar to Americans, this wind speed is roughly
60 mpha hefty storm. The wavelength is given by
= v
ph
T 30 ms
1
18 s 500 m. (5.30)
The crests are separated by half a kilometer. They bunch up near
shore, because they feel the bottom; this bunching is a conse-
quence of the shallow water dispersion relation, which we discuss
in Section 5.3.1.
In this same storm, the following days waves arrived more
frequently, at 17 s intervals. Lets rst check that this decrease in
period is reasonable. This precaution is a simple sanity check.
5. Waves 97
United States
Los Angeles
storm
distance 20, 000 km
travel time 17 days
Figure 5.9. Map of the actual world
(roughly drawn). A storm, almost at the
south pole, generates waves that prop-
agate outward (like ripples in a pond).
Some waves make it to Los Angeles, to
gladden the hearts of surfers and text-
book writers.
If our theory is wrong about a physical eect as fundamental as
this signwhether the period should decrease or increasethen
we have certainly neglected important physics. Fortunately, our
theory passes the test. Group velocity varies as 1/, which is pro-
portional to the period. The storm winds generate waves of dier-
ent wavelengths and periods, and the dierent wavelengths sort
themselves in the trip from the far ocean to Los Angeles. Longer-
period waves move faster, so the 18 s waves should arrive before
the 17 s waves, and they did. We can extract a surprising piece of
information from the decline in the interval; we can determine the
distance to the storm. In their long journey, the 18 s waves raced
ahead 1 day. The ratio of speeds is
velocity(18 s waves)
velocity(17 s waves)
=
18
17
= 1 +
1
17
. (5.31)
so the race must have lasted roughly t 17 days 1.5 10
6
s. The
wave train moves at the group velocity, v
g
= v
ph
/2 15 ms
1
, so
the storm distance was d tv
g
2 10
4
km, or roughly halfway
around the world (Figure 5.9)!
5.2.1.2 Speedboating. Our next application of the dispersion relation
is to speedboating: How fast can a boat travel? We exclude hy-
droplaning boats from our analysis (even though some speedboats
can hydroplane). Longer boats generally move faster than shorter
boats, so it is likely that the length of the boat, l, determines
the top speed. The density of water might matter. From only v
(the speed), , and l, however, we cannot form any dimensionless
groups. So we need to add at least one more variable. Viscosity
5. Waves 98
2. Catamarans and hydrofoils skim the
water, so this kind of drag does not limit
their speed. The hydrofoil makes a much
quicker trip across the English channel
than the ferry makes, even though the
hydrofoil is much shorter.
is irrelevant because the Reynolds number (Section 2.3) for boat
travel is gigantic. Even for a small boat of length 5 m, creeping
along at 2 ms
1
,
Re
500 cm200 cms
1
10
2
cm
2
s
1
10
7
. (5.32)
At such a huge Reynolds number, the ow is turbulent, and in-
dependent of viscosity. Surface tension is also irrelevant, because
boats are much longer than a ripple wavelength (roughly 1 cm).
Our search for new variables is not meeting with success. Perhaps
gravity is relevant. From v, , g, and l, we can form one dimen-
sionless group (four variables, three dimensions), which is v
2
/gl,
also called the Froude number:
Fr
v
2
gl
. (5.33)
The critical Froude number, which determines the maximum boat
speed, is a dimensionless constant. As usual, we assume that the
constant is unity:
v
_
gl. (5.34)
A rabbit has just jumped out of our hat. What mechanism
justies this formula? Consider what happens when a boat plows
through water, and follow the waves. The moving boat generates
waves (the wake), and it rides on one of those waves. Lets analyze
this bow wave: It is a gravity wave with v
ph
v
boat
. Because
v
2
ph
=
2
/k
2
, the dispersion relation tells us that
v
2
boat

2
k
2
=
g
k
= g, (5.35)
where 1/k = /2. So the wavelength of the waves is
v
2
boat
/g. The other length in this problem is the boat length; so
we can interpret the Froude number (5.34) as
Fr =
v
2
boat
/g
l

wavelength of bow wave
length of boat
. (5.36)
Why is Fr 1 the critical number? Interestingand often
dicultphysics occurs when a dimensionless number is near
unity. In this case, the physics is as follows. The wave height
changes signicantly in a distance ; if the boats length l is com-
parable to , then the boat rides on its own wave and tilts upward.
It then presents a large cross-section to the water, and the drag
becomes huge.
2
So the top speed is given by
v
boat

_
gl. (5.37)
For a small motorboat, with length l 5 m, this speed is roughly
7 ms
1
, or 15 mph. Boats can go faster than the nominal top speed
given in (5.37), but it takes plenty of power to ght the drag, which
is why police speedboats have huge engines.
5. Waves 99
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.10. Deep-waterripple portion
of the world of waves.
We meet the Froude number in surprising places. It deter-
mines, for example, the speed at which an animals gait changes
from a walk to a trot or, for animals that do not trot, to a run.
Here, it determines maximum boating speed. The Froude number
is, in a sense, the ratio of potential energy to kinetic energy, as we
can see by massaging v
2
/gl:
Fr =
v
2
gl
=
mv
2
mgl

kinetic energy
potential energy
. (5.38)
In this case, the length l is a horizontal length, so gl is not a
potential energy. In the Froude number for walking speed, l is leg
length, and gl is a potential energy. Then, for a human with leg
length l 1 m, the condition Fr 1 implies that v 3 ms
1
or 6 mph; this speed is a rough estimate for the top speed for a
race walker (the world record for mens race walking is held by
Bernado Segura of Mexico, who walked 20 km in 1:17:25.6, for a
speed of 4.31 ms
1
).
5.2.2 Ripples on deep water
For small wavelengths (large k), surface tension provides the re-
storing force: We are now studying the shaded region in Fig-
ure 5.10. If surface tension, instead of gravity, provides the restor-
ing force, then g drops out of the nal dispersion relation. We
could argue, as we did in Section 5.2.1, that g shows up in the
dimensionless group
3
k
2
/g, so
3
is irrelevant. In that ar-
gument, however, lies infanticide. It throws out the variable that
determines the restoring force: surface tension. To retrieve the
baby from the bathwater, we do not throw out k
2
/g directly.
We instead choose the form of f
deep
so that gravity, represented
by g, vanishes from the dispersion relation.
The deep-water dispersion relation (5.12) contains one power
of g in front. The argument of f
deep
also contains one power of g,
in the denominator. To cancel g, we choose f
deep
to have the form
f
deep
(x) x. Then
2
k
3
. By luck, the dimensionless constant

is unity, which we would have assumed anyway, so
2
=
k
3
. (5.39)
Lets try to derive this relation (up to a dimensionless constant)
using the slab argument.
In the slab picture, we replace gravitational by surface-tension
energy, and again balance potential and kinetic energies. The sur-
face of the water is like a rubber sheet. A wave stretches the sheet
and creates area. This stretching requires energy. To estimate the
energy, we rst estimate the extra area that a wave of amplitude
and wavenumber k creates. The extra area depends on the extra
length in a sine wave compared to a at line. The typical slope in
the sine wave sin kx is k. Instead of integrating to nd the arc
length, we approximate the curve as a straight line with slope k
5. Waves 100
l
0
1/k
l0
(
1
+
(
k
)
2
)
slope k
Figure 5.11. Approximating a sine
wave by a straight line. We want to
compute the arc length of the sine wave
so that we can compute the extra surface
area required to make a wave. Were
too lazy to integrate to nd arc length.
(We here means the authors
our readers may be more diligent; the
purpose of this text is to correct that
habit.) So we replace the curve by a
straight line with the same character-
istic slope. The characteristic slope of
the sine wave is k (which happens to be
the exact slope at the origin): In a dis-
tance of roughly 1 radwhich is a length
1/kit rises a height . The length of
the hypotenuse is l l
0
/ cos , which is
roughly l
0
(1 +(k)
2
) for small slopes k.
(Figure 5.11). Relative to the level line, the tilted line is longer by
a factor 1 + (k)
2
, for small slopes. As before, we study a piece
of a wave, with characteristic length 1/k in the x direction, and
width w in the y direction. The extra area is
A w/k
..
level area
(k)
2
. .
fractional increase
w
2
k. (5.40)
The potential energy stored in this extra surface is
PE
ripple
A w
2
k. (5.41)
The kinetic energy in the slab is the same as it is for gravity waves,
which is given in (5.22):
KE
2
2
w/k
2
. (5.42)
We balance the energies,
2
w/k
2
. .
KE
w
2
k
. .
PE
, (5.43)
and nd that
2
k
3
/. (5.44)
This dispersion relation agrees with the one that we found in (5.39)
using dimensional analysis. For deep-water gravity waves, we used
energy and force arguments to rederive the dispersion relation.
For ripples, we have just worked out the force argument; you are
invited to work out the corresponding force argument.
We have already interpreted the rst two dimensionless groups:
1
is the dimensionless depth, and
2
is ratio of kinetic energy to
gravitational potential energy. We described
3
as a group that
compares the eects of surface tension and gravity. Having com-
puted potential energy for gravity waves in (5.21), and for ripples
in (5.41), we can make the comparison more precise:
3

potential energy in a ripple
potential energy in a gravity wave
w
2
k
gw
2
/k
k
2
g
.
(5.45)
5. Waves 101
Alternatively,
3
compares k
2
/ with g:
3

k
2
/
g
. (5.46)
This form of
3
may seem like a trivial revision of k
2
/g. How-
ever, it suggests an interpretation of surface tension: that surface
tension acts like an eective gravitational eld with strength
g
surface tension
= k
2
/, (5.47)
In a balloon, the surface tension of the rubber implies a higher
pressure inside than outside. Similarly, in wave, the water skin
implies a higher pressure underneath the crest, which is curved
like a balloon; and a lower pressure under the trough, which is
curved opposite to a balloon. This pressure dierence is what a
gravitational eld with strength g
surface tension
. We have met this
idea of eective gravity already, when we studied marbles falling
in corn syrup (Section 2.3.2); in that problem, we replaced g by
an eective g that included the eect of buoyancy.
If we replace g in the gravity-wave potential energy (5.21) with
this eective g, we get the ripple potential energy (5.41):
gw
2
/k
. .
PE(gravity wave)
gk
2
/
====== w
2
k
. .
PE(ripple)
. (5.48)
The left side becomes the right side after we make the substitu-
tion above the arrow. If we make the same replacement in the
gravity-wave dispersion relation (5.14), we get the ripple disper-
sion relation (5.39):
2
= gk
gk
2
/
======
2
=
k
3
. (5.49)
The interpretation of surface tension as eective gravity is useful
when we combine our solutions for gravity waves and for ripples, in
Sections 5.2.3 and 5.3.3. We could have reversed the analysis and
interpreted gravity as eective surface tension; however, gravity is
the more familiar force, so we chose to use eective gravity.
As we did for gravity waves, we now use the dispersion relation
to calculate phase and group velocities. The phase velocity is
v
ph

k
=
, (5.50)
and the group velocity is
v
g

k
=
3
2
v
ph
. (5.51)
The factor of 3/2 is a consequence of the form of the dispersion
relation: k
3/2
; for gravity waves, k
1/2
, and the corre-
sponding factor is 1/2. In contrast to deep-water waves, a train
5. Waves 102
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.12. Deep-water portion of the
world of waves (same as Figure 5.3).
of ripples moves faster than the phase velocity. So, ripples steam
ahead of a boat, whereas gravity waves trail behind.
Lets work out speeds for typical ripples, such as the ripples
that we create by dropping a pebble into a pond. These ripples
have wavelength 1 cm, and therefore have wavenumber k =
2/ 6 cm
1
. The surface tension of water, which we estimated
in Section 3.3.3, is 72 erg cm
2
. So the phase velocity, given
by (5.50), is
v
ph
=
_
_
_
_
_
..
72 erg cm
2
k
..
6 cm
1
1 g cm
3
. .
_
_
_
_
_
1/2
21 cms
1
. (5.52)
The group velocity is v
g
30 cms
1
. This wavelength of 1 cm is
roughly the longest wavelength that still qualies as a ripple, as
shown in Figure 5.2; the third dimensionless group, which distin-
guishes ripples from gravity waves, has value
3

k
2
g

..
72 erg cm
2
k
2
..
36 cm
2
1 g cm
3
. .
1000 cms
2
. .
g
2.6. (5.53)
With a slightly smaller k, the value of
3
would slide into the
gray zone where
3
1. If k were yet smaller, the waves would
be gravity waves. Other ripples, with a larger k, have a shorter
wavelength, and therefore move faster: 21 cms
1
is roughly the
minimum phase velocity for ripples. This minimum speed explains
why we see mostly 1 cm ripples when we drop a pebble in
a pond. The pebble excites ripples of various wavelengths; the
shorter ones propagate faster and the 1 cm ones straggle, so we
see the stragglers clearly, without admixture of other ripples.
5.2.3 Combining ripples and gravity waves on deep water
Having studied two corners of the puzzlegravity waves and rip-
ples in deep waterwe connect the corners, and study the deep-
water edge, as shown in Figure 5.12. The dispersion relations
(5.14) and (5.39), for convenience restated here, are
2
=
_
gk, gravity waves;
k
3
/, ripples.
(5.54)
We would like to combine the relations in these two extreme
regimes, to produce a dispersion relation valid for gravity waves,
for ripples, and for waves in between.
Both functional forms came from the same physical argument
of balancing kinetic and potential energies. The dierence was the
source of the potential energy: gravity or surface tension. Gravity
and surface tension are active in both regimes. On the top por-
tion of the world of waves (Figure 5.19), surface tension dominates
5. Waves 103
gravity; on the bottom portion (Figure 5.18), gravity dominates
surface tension. You might therefore guess that, in the intermedi-
ate region, the two contributions to the potential energy simply
add, and that the combination dispersion relation is just the sum
of the two extremes:
2
= gk +k
3
/. (5.55)
Your guess would be correct (which is why we used an equality);
when in doubt, try the simplest solution.
We can restate our reasoning by using the eective gravi-
tational eld produced by surface tension, which we derived in
(5.47). The two sources of gravityreal and eectivesimply add,
to make
g
total
= g +g
surface tension
= g +
k
2
. (5.56)
When we replace g in
2
= gk with g
total
, we rederive the deep-
water dispersion relation (5.55):
2
=
_
g +
k
2
_
k = gk +k
3
/. (5.57)
With this dispersion relation, we can calculate wave speeds for
all wavelengths or wavenumbers. The phase velocity is
v
ph

k
=
+
g
k
. (5.58)
As a function of wavenumber, the two terms in the square root
compete to increase the speed. The surface-tension term wins at
high wavenumber; the gravity term wins at low wavenumber. So
there is an intermediate, minimum-speed wavenumber, k
0
, which
we can estimate by balancing the surface tension and gravity con-
tributions:
k
0

g
k
0
. (5.59)
This computation is another example of order-of-magnitude min-
imization, which we introduced in Section 3.1.2 to calculate the
Bohr radius. This minimum-speed wavenumber is
k
0

_
g
. (5.60)
At this wavenumber,
3
= 1: These waves lie just on the border
between ripples and gravity waves. Their phase speed is
v
0

_
2g/k
0

_
4g
_
1/4
. (5.61)
In water, the critical wavenumber is k
0
4 cm
1
, so the critical
wavelength is
0
1.5 cm; the speed is
v
0
23 cms
1
. (5.62)
5. Waves 104
We derived the speed (5.61) dishonestly. Instead of using the
maximumminimum methods of calculus, we balanced the two
contributions. A calculus calculation would agree with our mini-
mum phase velocity. A tedious calculus calculation shows that the
minimum group velocity is
v
g
17.7 cms
1
. (5.63)
Lets do the minimizations honestly. The calculation is not
too messy if its done properly, and the proper method is useful in
many physical maximumminimum problems. We rst nd the
minimum of the phase velocity (5.58). That equation contains
constants, , and gwhich we have to carry through all the
dierentiations. To simplify the manipulations, we instead choose
a convenient set of units, in which
= = g = 1. (5.64)
Our treatment of waves uses three basic dimensions: mass, length,
and time. Choosing three constants equal to unity is the most
freedom that we have; it is equivalent to choosing canonical mass,
length, and time, and thereby making all quantities dimensionless.
We restore the constants at the end of the minimization. In
addition to constants, the phase velocity also contains a square
root. So we minimize v
2
ph
, which in our new unit system is
v
2
ph
= k +
1
k
. (5.65)
For this minimization, we do not need calculus. The two terms are
both positive, so we may apply the arithmetic-meangeometric-
mean inequality (aectionately known as AMGM) to k and 1/k.
The inequality states that, for positive a and b,
(a +b)/2
. .
AM
ab
..
GM
, (5.66)
with equality when a = b. Figure 5.13 is a geometric proof of this
inequality. You are invited to convince yourself, just for fun, that
the gure is a proof. Using a = k and b = 1/k, we nd that the
geometric mean is unity, so the arithmetic mean is 1. Therefore,
k +
1
k
2, (5.67)
with equality when k = 1/k, or when k = 1. At this wavenum-
ber, the phase velocity is
2. In this unit system, the dispersion

relation is
=
_
k
3
+k, (5.68)
and the group velocity is
v
g
=

k
_
k
3
+k, (5.69)
5. Waves 105
ab
a +b
2
a
b
Figure 5.13. Proof without words: A
geometric proof of the arithmetic-mean
geometric-mean inequality. We leave to
you the fun of translating the picture
into a traditional proof. (Hint: Draw a
triangle.)
which is
v
g
=
1
2
3k
2
+ 1
k
3
+k
. (5.70)
At k = 1, the group velocity is also
2: These borderline waves

have equal phase and group velocity.
To convert k = 1 back to our usual units, we multiply it by
1, where we form the 1 using a product of , , and g. How can
we make a length from , , and g? The form of the result (5.60)
tells us that
_
g/ has units of [L]
1
. So k = 1 really means
k = 1
_
g/, which is just what we found in (5.60), but now
we know that the missing constant is 1.
The minimum group velocity is more complicated. We mini-
mize v
2
g
, to get rid of the square root. The derivative is
k
9k
4
+ 6k
2
+ 1
k
3
+k
. .
v
2
g
=
(3k
2
+ 1)(3k
4
+ 6k
2
1)
(k
3
+k)
2
. (5.71)
Equating this derivative to zero, we solve 3k
4
+6k
2
1 = 0, which
is a quadratic in k
2
, and has positive solution
k
1
=
_
1 +
_
4/3 0.393. (5.72)
At this k, the group velocity is
v
g
(k
1
) 1.086. (5.73)
In more usual units, this minimum velocity is
v
g
1.086
_
g
_
1/4
. (5.74)
When we put in the density and surface tension of water, we get
a minimum group velocity of 17.7 cms
1
, as we claimed in (5.63).
After dropping a pebble in a pond, you see a still circle sur-
rounding the drop point. Then, the circle expands at the minimum
group velocity given in (5.63). If you do not have a pond handy,
5. Waves 106
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.14. Shallow-water portion of
the world of waves.
you can try the experiment in your kitchen sink: Fill it with water,
and drop in a coin. The existence of a minimum phase velocity,
given in (5.62), is useful for bugs that walk on water. If they move
slower than 23 cms
1
, they generate no waves, and thereby reduce
the energy cost of walking.
5.3 Shallow water
In shallow water, the height h, absent in the deep-water calcula-
tions, returns to complicate the set of relevant variables. We are
now in the shaded region of Figure 5.14. With this extra length
scale, we have too much freedom; using only dimensional analysis,
we cannot deduce the shallow-water form of the magic function f
in the dispersion relation (5.6). However, by modifying the slab
argument, we can.
In deep water, the slab has depth 1/k. In shallow water, how-
ever, where h 1/k, the bottom of the ocean arrives before that
depth. The shallow-water slab has depth h. Its length is still 1/k,
and its width is still w. However, the continuity argument changes.
In deep water, where the slab has depth equal to length, the slab
and surface move the same distance. In shallow water, with a slab
thinner by hk, the surface moves more slowly than the slab, by
the factor hk. With wave height and frequency , the surface
moves with velocity , so the slab moves (sideways) with velocity
v
slab
/hk. The kinetic energy in the water is contained mostly
in the slab, because the upward motion is much slower than the
slab motion. This energy is
KE
shallow
wh/k
. .
mass
(/hk)
2
. .
v
2
w
2
2
hk
3
. (5.75)
We balance this energy against the potential energy, which we
compute for the two limiting cases: ripples and gravity waves.
5.3.1 Gravity waves on shallow water
We rst specialize to gravity wavesthe shaded region in Fig-
ure 5.15, where water is shallow and wavelengths are long. These
conditions include tidal waves and waves generated by undersea
earthquakes. For gravity waves, we already computed the poten-
tial energy, in (5.21), and found that
PE gw
2
/k. (5.76)
This energy came from the distortion of the surface, and it is the
same in shallow water (as long as the wave amplitude is small
compared with the depth and wavelength). [The dominant force
(gravity or surface tension) determines the potential energy. As
we see when we study shallow-water ripples, in Section 5.3.2, the
water depth determines the kinetic energy.]
We balance this energy against the kinetic energy (5.75):
w
2
2
hk
3
. .
KE
gw
2
/k
. .
PE
. (5.77)
5. Waves 107
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.15. Shallow-watergravity-
wave portion of the world of waves.
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.16. Shallow-waveripple por-
tion of the world of waves.
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.17. Shallow-water portion of
the world of waves (same gure as Fig-
ure 5.14).
So
2
= ghk
2
. (5.78)
The equation has an equality, because, once again, the dimension-
less constant is unity. So, for gravity waves on shallow water, the
function f has the form
f
shallow
(kh,
k
2
g
) = kh. (5.79)
The group and phase velocities are equal to each other, and
are independent of frequency:
v
ph
=

k
=
_
gh,
v
g
=

k
=
_
gh.
(5.80)
Shallow water is nondispersive: All frequencies move at the
same velocity, so pulses composed of various frequencies propa-
gate without smearing. Undersea earthquakes illustrate the dan-
ger in such unity. If there is an earthquake o the coast of Chile,
and the ocean bottom drops, it generates a shallow-water wave,
which travels without distortion to Japan. The wave speed is v
3000 m10 ms
2
170 ms
1
: The wave can cross a 10
4
km
ocean in 1 day. As it approaches shore, where the depth decreases,
the wave slows, grows in amplitude, and becomes a large, destruc-
tive wave when it reaches land.
5.3.2 Ripples on shallow water
Ripples on shallow waterthe shaded region in Figure 5.16are
rare. They occur when raindrops land in a shallow rain puddle,
one whose depth is less than 1 mm. Even then, only the longest-
wavelength ripples, where 1 cm, can feel the bottom of the
puddle. What is the potential energy of the surface? We have al-
ready computed the potential energy, in (5.41). Although we de-
rived that formula for deep water, the water depth does not aect
the potential energy; the dominant forcehere, surface tension
determines the potential energy. We equate the potential energy
from (5.41) with the kinetic energy (5.75):
w
2
2
hk
3
. .
KE
w
k
(k)
2
. .
PE
. (5.81)
We then nd that
hk
4
. (5.82)
The phase velocity is
v
ph
=
hk
2
, (5.83)
and the group velocity is v
g
= 2v
ph
(the form of the dispersion
relation is k
2
). For h 1 mm, this speed is
v
_
72 erg cm
2
0.1 cm36 cm
2
1 g cm
3
_
1/2
16 cms
1
. (5.84)
5. Waves 108
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.18. Gravity-wave portion of
the world of waves.
10
4
10
2
10
2
10
4
1 hk
10
4
10
2
10
2
10
4
3
k
2
g
Deep water
Surface tension
Deep water
Gravity
Shallow water
Surface tension
Shallow water
Gravity
Figure 5.19. Ripple portion of the world
of waves.
5.3.3 Combining ripples and gravity waves on shallow water
We have studied the two corners of the world of shallow-water
waves: ripples and gravity waves. Now we connect the corners to
make an edge: We study general shallow-water waves. This region
of the world of waves is shaded in Figure 5.17. We can combine the
dispersion relations for ripples with that for gravity waves using
two equivalent methods. We either can add the two extreme-case
dispersion relations, (5.78) and (5.82); or, can use the eective
gravitational eld given in (5.56) in the gravity-wave dispersion
relation (5.78). Either method produces
2
k
2
_
gh +
hk
2
_
. (5.85)
5.4 Combining deep- and shallow-water gravity waves
We now examine the gravity-wave edge of the world, as shown in
Figure 5.18. The deep- and shallow-water dispersion relations are,
from (5.14) and (5.78):
2
= gk
_
1, deep water;
hk, shallow water.
(5.86)
To interpolate between the two regimes, we want a function f(hk)
that asymptotes to 1 as hk , and to hk as hk 0. One such
simple function is tanh hk, so we guess that the One True Gravity
Wave Dispersion Relation is:
2
= gk tanh hk. (5.87)
Our guess is plausible, because tanh hk falls o exponentially as
h , in agreement with the argument based on Laplaces equa-
tion. In fact, our guess is correct.
5.5 Combining deep- and shallow-water ripples
We now examine the nal edge: ripples, in both shallow and deep
water, as shown in Figure 5.19. In Section 5.4, we found that
tanh kh did the yeoman work of interpolating between hk and
1, as hk went from 0 to (as the water went from shallow to
deep). We guess that the same trick holds for ripples, because the
Laplace-equation argument, which justied the tanh kh, does not
depend on the restoring force. The relevant dispersion relations
are (5.82), for shallow water, and (5.39), for deep water:
2
=
_
k
3
/, if kh 1;
hk
4
/, if kh 1.
(5.88)
If we factor out k
3
/, the necessary transformation becomes
clear:
2
=
k
3

_
1, if kh 1;
hk, if kh 1.
(5.89)
5. Waves 109
Shallow water
Ripples
2
=
hk
4
Deep water
Ripples
2
=
k
3
Shallow water
Gravity waves
2
= ghk
2
Deep water
Gravity waves
2
= gk
Ripples
2
=
k
3
tanhkh
shallow water
2
= k
2
_
gh +
hk
2
_
deep water
2
= gk +
k
3
Gravity waves
2
= gk tanhkh
2
=
_
gk +
k
3
_
tanhkh
1
2
3
4
5
6
7
8
9
Figure 5.20. What we did. Each box
represents a limit. The numbers next to
the boxes mark the order in which we
studied that limit. In the nal step (9),
we combined all the analyses into the
superbox in the center, which contains
the dispersion relation for all waves:
gravity waves or ripples, shallow wa-
ter or deep water. The arrows show
how we combined smaller, more spe-
cialized corner boxes into the more gen-
eral edge boxes (double ruled), and the
edge regions into the universal center
box (triple ruled).
This ripple result looks similar to the gravity-wave result (5.86),
so we make the same replacement:
_
1, if kh 1,
hk, if kh 1,
becomes tanh kh. (5.90)
Then we get the general ripple dispersion relation:
2
=
k
3
tanh kh. (5.91)

This dispersion relation does not have much practical interest,
because, at the cost of greater complexity than the deep-water
ripple dispersion relation (5.39), it adds coverage of only a rare
case: ripples on ponds. We include it for completeness, so that we
have treated all four edges of the world, in preparation for the
grand combination coming up next.
5.6 Combining all the analyses
Now we can replace g with g
total
from (5.56), to nd the One True
Dispersion Relation:
2
= (gk +k
3
/) tanh kh. (5.92)
5.7 What we did
We studied water waves by investigating dispersion relations. We
mapped the world of waves, explored the corners, then the edges,
and then assembled the results to form an understanding of the
complex, complete solution. We can now draw the whole map,
shown in Figure 5.20. Considering limiting cases, as we did, and
then stitching them together, makes the analysis tractable, and,
more important, comprehensible.
110
Bibliography
[1] Howard C. Berg. Random Walks in Biology. Princeton University Press, Princeton, NJ, 1993.
[2] Carl Bernstein and Bob Woodward. All the Presidents Men. Simon and Schuster, New York,
1974.
[3] S. Chandrasekhar. An Introduction to the Study of Stellar Structure. University of Chicago
Press, Chicago, Ill., 1939.
[4] S. Chandrasekhar. Hydrodynamic and Hydromagnetic Stability. Oxford University Press, Oxford,
1961.
[5] Frank S. Crawford. Waves, volume 3 of Berkeley Physics Course. McGraw-Hill, New York,
1968.
[6] Lyn Dupre. BUGS in Writing: A Guide to Debugging Your Prose. AddisonWesley, Reading,
Mass., revised edition, 1998.
[7] T. E. Faber. Fluid Dynamics for Physicists. Cambridge University Press, Cambridge, England,
1995.
[8] Richard Feynman, Robert B. Leighton, and Matthew Sands. The Feynman Lectures on Physics,
volume I. Addison-Wesley, Reading, MA, 1963.
[9] William Elmer Forsythe. Smithsonian Physical Tables, volume 120 of Smithsonian Miscellaneous
Collections. Smithsonian Institution, Washington, DC, ninth revised edition, 1964.
[10] H. W. Fowler and F. G. Fowler. The Kings English. Oxford University Press, Oxford, 3rd
edition, 1931. Reprinted 1962.
[11] Stephen Gasiorowicz. Quantum Physics. Wiley, New York, second edition, 1996.
[12] Horace Lamb. Hydrodynamics. Dover, New York, sixth edition, 1945.
[13] David R. Lide, editor. CRC Handbook of Chemistry and Physics: A Ready-Reference Book of
Chemical and Physical Data. CRC Press, Boca Raton, FL, 74th edition, 1993.
[14] R. C. Pankhurst. Dimensional Analysis and Scale Factors. Chapman and Hall, London, 1964.
[15] William L. Riordan. Plunkitt of Tammany Hall. E. P. Dutton, New York, 1963.
[16] Bertrand Russell. A History of Western Philosophy, and Its Connection with Political and
Social Circumstances from the Earliest Times to the Present Day. Simon and Schuster, New
York, 1945.
[17] Knut Schmid-Nielsen. Scaling: Why Animal Size is So Important. Cambridge University Press,
Cambridge, England, 1984.
[18] Steven Vogel. Life in Moving Fluids: The Physical Biology of Flow. Princeton University Press,
Princeton, NJ, 1994.

Estimation

Uploaded by

Copyright:

Available Formats

Estimation

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Estimation

Uploaded by

Copyright:

Available Formats

What is the main topic of the book?

What is the main topic of the book?

What method is emphasized for problem solving?

What method is emphasized for problem solving?

Order-of-Magnitude Physics: Understanding the World with Dimensional

Analysis, Educated Guesswork, and White Lies

1 1000 30. Interpolating

h, and h is roughly constant. So

, because of a vague hunch about

a. Although dimensionally correct by construction, this

. Then, there are two dimensionless

a. The most general relation using these

gl; we use this

inside the product

; equations that contain

. How can we make an energy from

= 0, then gravity retains its full

. Why must the variation be linear? Gravity makes pres-

. Therefore, we make the

: The terminal velocity is smaller in a denser uid (try run-

should be in the denominator.

. We put these numbers

is [E][L], we make an energy from e

E Figure 3.3. Order-of-magnitude calcu-

(r) = E/r. We assume that E r

(r) E(r)/r and F E/r. What is the dipole moment?

1.1. Here is another example: Minimize the function f(r) = e

A, for n 3; for n > 3, the diameter is still 3

A for a typical atomic diameter.

A (Figure 3.4), and has mass Am

A. For covalent and ionic bondstheir bind-

, then lead pencils would be diamond

slopes. The shaded top block, which has

6000 km, so h/R

/30 (roughly 200 km), the fractional height is roughly 1.

r. The drop moves at constant

. For most atoms, =

, R is roughly 0.3 earth radii. A more exact cal-

, the mountain is 20 km tall. It would be a many-

T (in dimensionless units). A typical T

T; the plus sign repre-

T and use the approximation that

the latent heat of vaporizationcomes

A (triple bonds are shorter

A. After a scattering, the phonon

N; this square-root dependence

C) cooks in a few minutes; we also know

C, the meat will never cook (50

C, the turkey will cook in a few minutes after

A; the number density is n

The specic heat per volume is C

. By luck, the dimensionless constant

2. In this unit system, the dispersion

2: These borderline waves

tanh kh. (5.91)

You might also like