lecture 6

Topics:
F = ma Euler-Lagrange equations
Hamiltons principle
Functions of functions
Calculus of variations
Functional derivatives
Finding functional derivatives
Back to Hamiltons principle
More degrees of freedom
The Lagrangian and the action
Quantum mechanics and the classical trajectory
Appendix: On the functional Taylor series
F = ma Euler-Lagrange equations
This week, we will introduce some difcult, but beautiful material - Lagrangian mechanics. Essen-
tially, having spent the last couple of weeks getting a better understanding of classical mechanics in
terms of second order differential equations, we are now going to introduce a completely different
formulation of mechanics based on the calculus of an innite number of variables! This probably
sounds really scary. But DONT PANIC! I am introducing this for several reasons.
1. I hope that you will nd it fascinating.
2. I hope that it will give you a different philosophical slant on classical mechanics that will
help you to understand why classical mechanics works the way it does.
3. While the mathematics will seem unfamiliar to almost all of you, in the end, we will not
actually be doing anything with it that you do not already know how to do.
4. Finally, and perhaps most importantly, it is really useful not just theoretically but also for
solving real problems.
The result of the painful analysis I am about to subject you is actually pretty simple. The
statement is this. To solve many classical problems (and we will discuss later why this works in
the most interesting cases) you construct a function (called the Lagrangian) which is the kinetic
energy T minus the potential energy V as a function of the coordinates and their time derivatives
(the velocities).
L
_
q, q
_
= T
_
q, q
_
V
_
q, q
_
(1)
(T and V may not depend on both q and q, but that will depend on the problem). Then for each
coordinate, you get an equation of motion as follows:
d
dt
_

q
L
_
q, q
_
_
=

q
L
_
q, q
_
(2)
This gives you a set of straightforward rules that will allow you to write down the second order
differential equations for the time evolution of a physical system in a single step, without identing
1
all the separate forces acting on the system. So I hope that even if you never fully appreciate all
the deep philosophical implications of the Lagrangian approach, you will love it as a labor-saving
device.
I could of course just write down (1) and skip all the difcult math that leads up to it, because
the hard stuff wont really be important in the way we use these results most of the time. I think
it is good to go through it once, but really what I am really most interested in is the philosophical
implications of all this. The deeper question that we are heading to is Why are energy and
momentum conserved? Next week, I will try to convince you that these conservation laws follow
directly from fundamental symmetries of the world plus the assumption that the laws of mechanics
arise from a Lagrangian by the simple set of rules you will learn today. This will move us quite a
way along in our quest to understand the fundamental meaning of classical mechanics.
Hamiltons principle
We will start by showing in what sense Newtons laws can be formulated in a way that appears very
different, not as a differential equation, but as a condition that picks out the classical trajectory of
a particle by comparing with all other motions the particle might have had! This probably sounds
pretty wierd. Later we will argue that this new formulation is actually much deeper and more
fundamental. It is also a gorgeous piece of mathematics. So hold on to your hats, and remember
that it is not going to be as complicated as it seems at rst.
Consider the one dimensional motion of a particle with coordinate x in a potential V (x). Call
the particles kinetic energy
T( x) =
m
2
x
2
(3)
Now we would like to nd a particular trajectory x(t) are such that the particle moves from x
1
at
time t
1
to x
2
at time t
2
or
x(t
1
) = x
1
x(t
2
) = x
2
(4)
Hamiltons principle is the statement that if you compute the quantity S[x] (called the action
Im not really sure why)
S[x]
_
t
2
t
1
_
T
_
x(t)
_
V
_
x(t)
_
_
dt (5)
depending on a function x(t) satisfying
1
then the variation of S[x] with respect to the function
x(t) vanishes for the actual trajectory. The hard part of Hamiltons principle will not be proving
this statement, but guring out exactly what it means to talk about the variation of something with
respect to a function.
The rst thing to notice is that T V is an ordinary function of two variables, x and x,
2
but
that S[x] is actually a function of a function it depends on x(t) for all values of t from t
1
to t
2
1
We will see later that the restriction to functions satisfying (4) is important and has to do with the initial conditions
that we have spent so much time talking about in the rst couple of weeks.
2
In the simple example we are working out now, T is a function of one of them and V of the other, but in general,
both T and V could depend on both x and x.
2
where x(t) can be any function satisfying (4) something like
x
1
x
2
t
1
t
2
..................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
x(t)
Thus S[x] is a function whose argument is itself a function (such an object is sometimes called a
functional and I may sometimes use that term). I have put the x in square brackets to remind you
that in this case x is a function rather than a number. The value of S[x], on the other hand, is just
a number. It doesnt depend on t. The variable t in (5) is a dummy variable. Now Hamiltons
principle is a statement about the variation of S[x] as we let x(t) vary over all possible functions
satisfying (4).
3
We will nd that when x(t) is very near to a solution to Newtons second law, then
S[x] changes slowly as a function of whatever parameters you use to specify the function x(t).
When x(t) is a solution, S[x] is not varying at all.
Functions of functions
I imagine that the idea of a functional that is a function of a function is pretty unfamiliar to
most of you, but you shouldnt get too worried about it. Most of you have learned or are beginning
to learn about the calculus of functions of several variables, and dealing with functionals is not
really much different. The main differences are in notation. In fact, I would argue that the big jump
is going from one or two variables to more than that. Once you have made it to three variables, it
really doesnt get any harder to think about more not even an innite number more.
Here is what I mean by the peculiar statement that going from two variables to three variables
is what is hard. This has to do with visualization. As you have probably noticed, it gets harder to
make a mental picture of what a function means as the number of variables increase. A function
of a single variable is easy. We naturally associate a function f(x) with a graph of y = f(x).
This allows us to do things that bring the function alive to us, such as relating the derivative of the
function to the slope of the line in the graph. We can sort of do something similar with a function
of two variables f(x, y) by imagining a surface in three dimensional space with z = f(x, y). This
3
Really? All possible functions? One has to be reasonable here it might be better to say all possible functions
for which the integral in (5) makes sense differentiable functions in this example. We will ignore such issues and
leave them for the mathematicians.
3
can be quite helpful, because it allows you to have a visual representation of features that only
appear with more than one variable, such as the gradiant vector, which for two variables x and y
looks like the two dimensional vector

=
_

x
,

y
_
(6)
The vector

f(x, y) points in a direction that corresponds to going upwards on the surface z =
f(x, y).
But what do you do about visualizing functions of three variables, or more? There is really no
ideal way of doing this. So already by the time you get to functions of three variables, you have to
stop relying on visual crutches and just develop an analytic sense of what the function means.
A functional like S[x] is just a function of an innite number of variables where the variables
are the values of x(t) at all possible values of t. The really peculiar new thing about a functional is
that the variables are labeled by a continuous parameter, rather than having different names like x,
y and z, or different discrete indices, like a
1
, a
2
, etc. This is why functionals look so different and
why we have to invent some new notation to deal with them.
A mathematical example of a functional is the length of a path described by a curve, y = f(x).
The path length from x
1
to x
2
is
P[f] =
_
x
2
x
1
_
dx
2
+ dy
2
=
_
x
2
x
1

_
1 +
_
dy
dx
_
2
dx =
_
x
2
x
1
_
1 + f

(x)
2
dx (7)
The path length depends on the function f that denes the shape of the curve. This example is
particularly interesting because it depends on the endpoints in same way that (5) does.
A specic example of path length may be useful. Consider a graph y = f(x) for the function
f(x) =

R
2
x
2
(8)
f

(x) = x/

R
2
x
2
(9)
Then the path length from x
1
to x
2
is
_
x
2
x
1
_
1 + f

(x)
2
dx =
_
x
2
x
1
R/

R
2
x
2
dx (10)
which with the substitution x = Rsin becomes
R
_
arcsinx
2
/R
arcsinx
1
/R
d = R
_
arcsin
x
2
R
arcsin
x
1
R
_
(11)
4
And this is right because this is just the arc length along a circle, as show in the diagram below.
............................................................................................................................................................................................................................................................................................................................................................................................................................. . . . . . . . . . . . . . .
. . . . . . . . . . . . . .
............................................................................................................................................................................................................................................................................................................................................................................................................................. ..............
. . . . . . . . . . . . . .
arcsin
x
1
R
arcsin
x
2
R
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..............
..............
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...............
..............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
...........
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
.............
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

y
x x
1
x
2
................................
......................................................................
.. ... .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Here is a more physical example. Suppose that the function y = f(x) in (7) describes the
height of the track of of roller coaster. If a roller car is moving along the track with mass m and
energy
E =
1
2
mv
2
+ mgy =
1
2
mv
2
+ mg f(x) (12)
the time the car takes to get from x to x + dx is the innitesimal path length,

dx
2
+ dy
2
=
dx
_
1 + f

(x)
2
divided by the speed v which from (12) is
v =
_
2
_
E mg f(x)
_
/m (13)
Thus the time the car takes to get from x
1
to x
2
is
[f] =
_
x
2
x
1

dx
2
+ dy
2
v
=
_
x
2
x
1
_
1 + f

(x)
2
_
2
_
E mg f(x)
_
/m
dx (14)
The time (14) is a function of E, m and g, but it is a functional of the function f(x) that describes
the height of the track. This is an important example to understand and we will try to come back
to it several times over the next few weeks.
Calculus of variations
Now lets return to Hamiltons principle. The proof and indeed the more precise mathematical
statement of the principle is an exercise in what is called the calculus of variations, which is sort of
calculus with an innite number of variables. It probably wont surprise you that what we need to
do is to generalize the notion of a derivative to functionals and to explore the nature of the Taylor
series. After all, these are our main tools in understanding how things vary.
The actual example we are interested in is complicated because the integrand on the right hand
side of (5) depends on both x(t) and x(t), so let us rst discuss a couple of simpler examples.
5
Consider the quantity
W[x] =
_
t
2
t
1
_
x(t) t
_
2
dt (15)
and nd the value of x(t) for which the variation of W[x] with respect to x(t) vanishes. We can do
this by the following trick. Suppose that we have already found the soltuion x(t). Now consider
small variations about this particular function:
x(t) + x(t) (16)
Ive used the symbol rather than just to remind us that this is a small function rather than a
small number. Now look at (15) as function of x + x,
W[x + x] =
_
t
2
t
1
_
x(t) + x(t) t
_
2
dt
=
_
t
2
t
1
_
x(t) t
_
2
dt +
_
t
2
t
1
x(t) 2
_
x(t) t
_
dt +
_
t
2
t
1
_
x(t)
_
2
dt
= W[x] +
_
t
2
t
1
x(t) 2
_
x(t) t
_
+O(x
2
) dt
(17)
Now we look at the coefcient of the linear term in x(t), which is 2
_
x(t) t
_
. This vanishes for
all possible small functions x(t) if
x(t) = t (18)
and this is the answer we are looking for. Around the function x(t) = t, there is no linear term in
the expansion of W[x]. Note that while t in (17) is a dummy variable, we can say something about
the t dependence because we require that the variation vanishes for any possible x(t). We will
come back to this below when discuss functional derivatives.
This result is quite reasonable and maybe obvious in this particular example because W[x] is
actually minimized for x(t) = t. The variation of a functional at a minimum vanishes for the
same reason that the derivative of a smooth function vanishes at a minimum. If there were a
linear variation, then on one side or the other, the result would be larger, which is impossible at a
minimum value. But what we want to extract from this calculation is not the result, but the general
technique.
Functional derivatives
What we did to solve the variational problem for (15) was to set to zero the coefcient of the linear
term in x(t) in W[x + x]. It is useful to give this coefcient a name, so we call it a functional
derivative, and denote it by
W
x(t)
[x] or

x(t)
W[x] (19)
so in this case, we can write
W
x(t)
[x] = 2
_
x(t) t
_
(20)
Notice that the t dependence of the RHS of (20) comes from the t dependence of the differential
in the denominator of the LHS again W[x] for some particular x is just a number with no t
dependence.
6
This denition of functional derivative is generally useful, so lets discuss it in general:

x(t)
W[x] is the coefcient of the linear term in x(t) in W[x + x]. (21)
This is a reasonable denition because it makes the functional Taylor series work in the same way
that the ordinary Taylor series does. In terms of the functional derivative, the functional Taylor
series starts like this:
W[x + x] = W[x] +
_
t
2
t
1
x(t)

x(t)
W[x] dt + (22)
In words, this says that the total change in the functional is obtained by adding up (actually in-
tegrating because the variable is continuous) the small changes x(t) in x(t) at each point times
the rate of change of the functional at that point. This is just what we always say, except that the
number of variables has become continuously innite. (22) is precisely analogous to the Taylor
series for functions of several variables that we will discuss in the appendix and in future lectures:
F(x
1
+ x
1
, x
2
+ x
2
, ) = F(x
1
, x
2
, ) +

j
x
j

x
j
F(x
1
, x
2
, ) + (23)
The only difference is that instead of the sum over all the different variables that we have in (23),
in (22) we need an integral over t because in effect, the value of x(t) at each value of t is a separate
variable.
Finding functional derivatives
In general, functional derivatives can be difcult to nd, but it is easy to nd them for functionals
like W[x] that have the form of an integral of an ordinary function
W[x] =
_
t
2
t
1
F
_
x(t)
_
dt (24)
In this case, the functional derivative of W[x] is related to the ordinary derivative of F
W
x(t)
[x] = F

_
x(t)
_
(25)
The reason this is so simple is that we can calculate the functional Taylor expansion of W using
the ordinary Taylor expansion of F, and looking for the coefcient of x(t). Lets show how this
works.
W[x + x] =
_
t
2
t
1
F
_
x(t) + x(t)
_
dt (26)
But using the ordinary Taylor expansion for the ordinary function F(x), the right hand side be-
comes
=
_
t
2
t
1
_
F
_
x(t)
_
+ x(t) F

_
x(t)
_
+
_
dt (27)
7
Picking out the coefcient of x(t) on the right hand side gives (25). Notice the way that the
variable t gets promoted from being a dummy variable in (24) and (26) to being a real variable in
(27). We single out a particular value of t when we nd the coefcient of x(t) for that t.
The function F in (24) may depend on other functions of t, in which case we can generalize
(24) as follows:

x(t)
_
t
2
t
1
F
_
x(t

), y(t

),
_
dt

=

x
F(x, y(t), )|
x=x(t)
(28)
where . . . denotes and other functions that F depends on. Note that in (28), t

is a dummy index,
integrated over so that it could be called anything except t. The dependence on t is real (not
dummy) because the process of functional differentiation picks out a value of t. Here are some
examples:
F[x] =
_
t
2
t
1
x(t

)
3
dt

x(t)
F[x] = 3 x(t)
2
F[x] =
_
t
2
t
1
sin
_
x(t

)
_
dt

x(t)
F[x] = cos
_
x(t)
_
F[x, y] =
_
t
2
t
1
_
x(t

) y(t

)
_
3
dt

x(t)
F[x, y] = 3 x(t)
2
y(t)
3
F[x, y] =
_
t
2
t
1
sin
_
x(t

) y(t

)
_
dt

x(t)
F[x, y] = y(t) cos
_
x(t) y(t)
_
(29)
We can summarize these ideas by saying that when the functional is the integral of an ordi-
nary function, what the functional derivative does is to eliminate the integral and differentiate the
function. Think about it in the context of the examples in (29).
Back to Hamiltons principle
With this new tool of the calculus of variations, we can go back and consider Hamiltons principle.
Remember that what we want to do is to show that the functional S[x] has zero variation around
the same trajectory that we get by using F = ma.
According to our discussion, what we want to do to impose vanishing variation on S[x] is to
set the functional derivative of S[x],
S
x(t)
(30)
to zero. As in the simpler example, we calculate the functional derivative by performing a func-
tional Taylor series and picking out the coefcient of x(t). For pedagogical purposes, we will
break this up into two pieces:
S = S
T
S
V
(31)
where
S
T
[x]
_
t
2
t
1
_
T
_
x(t)
_
_
dt =
_
t
2
t
1
m
2
_
x(t)
_
2
dt (32)
8
and
S
V
[x]
_
t
2
t
1
_
V
_
x(t)
_
_
dt =
_
t
2
t
1
V
_
x(t)
_
dt (33)
For the S
V
term, because V does not depend on x, this is just like the W[x] example,
S
V
x(t)
= V

_
x(t)
_
(34)
This is a good sign because the derivative of the potential is related to the force.
For the S
V
, term, we will need an extra step, so lets write out the functional Taylor series in
detail.
S
T
[x + x] =
_
t
2
t
1
m
2
_
x(t) + x(t)
_
2
dt (35)
=
_
t
2
t
1
m
2
_
x(t)
_
2
dt +
_
t
2
t
1
x(t) m x(t) dt +O(x
2
) (36)
= S
T
[x] +
_
t
2
t
1
x(t) m x(t) dt +O(x
2
) (37)
But the x(t) in (37) is the change in x(t) when we make a change x(t) in x(t). Thus since
d
dt
_
x(t) + x(t)
_
= x(t) +
d
dt
x(t) (38)
x(t) is given by
x(t) =
d
dt
x(t) . (39)
The trouble with (37) is that it depends on x(t), while we have dened the functional derivative
to be the coefcient of x(t). To take care of that, we can integrate by parts:
_
t
2
t
1
x(t) m x(t) dt =
_
t
2
t
1
_
d
dt
x(t)
_
m x(t) dt
=
_
t
2
t
1
d
dt
_
x(t) m x(t)
_
dt
_
t
2
t
1
x(t) m x(t) dt
=
_
t
2
t
1
x(t) m x(t) dt + x(t
2
) m x(t
2
) x(t
1
) m x(t
1
)
(40)
The last line in (40) has two terms from the endpoints of the integration. These dont look like
everything else, and we want to get rid of them. This is where (4) comes in. We are supposed to
be restricting our attention to functions satisfying (4). But if x(t) and x(t) +x(t) both satisfy (4),
then
x(t
1
) = x(t
2
) = 0 . (41)
In words, we are only interested in variations that vanish at the endpoints. These functions might
9
look something like this
x
1
x
2
t
1
t
2
..................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
....................................................................................................................................................................................................................................................................................................................................................................................................................................
x(t)
x(t) + x(t)
x(t)
So in fact, the funny looking terms vanish because of (4). Then (40) becomes
_
t
2
t
1
x(t) m x(t) dt =
_
t
2
t
1
x(t) m x(t) dt (42)
Now we can pick out the coefcient of x(t) and see that
S
T
x(t)
= m x(t) (43)
Now, nally, we can verify that Hamiltons principle is correct. Putting (34) and (43) together,
we have that the condition for vanishing variation of S[x] is
S
x(t)
= m x(t) V

_
x(t)
_
= 0 (44)
or
m x(t) = V

_
x(t)
_
(45)
which, since F = V

, is just Newtons second law!!!
More degrees of freedom
It is easy to see that Hamiltons principle works just as well for more particles, or in more dimen-
sions. Suppose, for example, that there are n particles, so that
T =
n

j=1
m
j
2
x
2
j
and V = V (x
1
, , x
n
) (46)
Then it is easy to see by the same sort of arguments that
S
x
j
(t)
= m
j
x
j
(t) V
j
_
x
1
(t), , x
n
(t)
_
(47)
10
where
V
j
(x
1
, , x
n
) =

x
j
V
j
(x
1
, , x
n
) (48)
is the force on particle j. For S[x] to have vanishing variation, we must have (47) vanish for each
j, which just gives F = ma for each particle.
The Lagrangian and the action
As we will see, Hamiltons principle really captures more of what is going on in the world than
F = ma. It is worth going over the consequences of Hamiltons principle again in a more general
language. In general, the combination T V is called the Lagrangian, L. S[x], as I said above,
is called the action,
S[x] =
_
t
2
t
1
L
_
x(t), x(t)
_
dt (49)
Here, x might actually have indices that allows it to represent more than one particle or dimension
or both. We wont write them explicitly in this formal derivation. Now we want the functional
derivative of S[x] with respect to x(t) to vanish. In general, the functional derivative is
S
x(t)
[x] =

x(t)
L
_
x(t), x(t)
_

d
dt

x(t)
L
_
x(t), x(t)
_
(50)
The rst term arises from the Taylor expansion of the x(t) dependence. The second term arises in
the same way as (43), fromthe Taylor expansion of the x(t) dependence, followed by an integration
by parts, which gives the minus sign. Thus Hamiltons principle implies that the solution for the
motion satises

x(t)
L
_
x(t), x(t)
_

d
dt

x(t)
L
_
x(t), x(t)
_
= 0 (51)
If x has several components, (51) must be true for each component separately.
The equation (or set of equations) (51) is called the Euler-Lagrange equation. We have seen
that in simple situations this equation is just a trivial rewriting of F = ma. However, there are
several important advantages to thinking about mechanics in terms of Hamiltons principle, and
deriving the Euler-Lagrange equation, rather than using Newtons second law directly.
1. The mathematics is kind of cool! In this course, we will just give you a small taste of
the beautiful mathematical structure of mechanics that was worked out in the 19th century.
If you like this sort of thing, you can look forward to a feast in Physics 151, where it is
described in more depth.
2. It is often easier to write down the kinetic and potential energies than to understand all of the
forces. Thus Hamiltons principle can be a labor saving device.
3. In general, there may be many different ways of representing the conguration of a physical
system, and thus many ways of choosing the variables that describe the system. Hamiltons
principle makes it obvious that the particular choice of variables doesnt matter, because
S[x], whose vanishing variation determines the trajectories, is just a single number that
doesnt depend on how we choose to describe the system. This will be especially useful
when we work in spherical or cylindrical coordinate, which are more appropriate to some
problems.
11
4. More generally, Hamiltons principle makes it very easy to understand the consequences of
symmetry. We will use it to see that there is a deep connection between symmetries and
conservation laws like conservation of momentum or angular momentum. I happen to love
symmetry (as will become apparent as the course goes on if it is not already obvious), so this
is one of my favorites.
5. Sometimes, we dont know or care about all the forces in a mechanical system. This of-
ten happens when there is a constraint. There are many examples, like a bead sliding on a
frictionless wire, or a particle moving on the surface of the earth, where the system is con-
strained by some forces that we dont understand in detail. We can use Hamiltons principle
to solve such problems without even computing the constraining forces by simply choosing
coordinates that automatically incorporate the constraints well do some examples, and
there are more in David Morins notes.
The most compelling argument for Hamiltons principle is none of the above, but a deeper, philo-
sophical one.
Quantum mechanics and the classical trajectory
In my view, the most important reason that Hamiltons principle and the Lagrangian are so im-
portant has to do not with classical mechanics alone, but with quantum mechanics and the way in
which classical physics emerges as an approximation to the quantum world.
From the classical point of view, Hamiltons principle is actually a little peculiar. Why should
it matter to a classical particle what the value of the action is for paths that the particle does not
actually take? OK - so Hamiltons principle works to give the classical equation of motion, but
it is hard to gure out what it means physically. But in quantum mechanics, it has a very denite
meaning, because in quantum mechanics, the particle really takes all paths! This is absolutely
nutty, but this is really the way the world works. Roughly, the way it works is this. When a
quantum mechanical particle moves from point x
1
at time t
1
to point x
2
at time t
2
, it takes all
trajectories from the starting point to the end simultaneously. But the different trajectories can add
together like the different ripples in a wave on a pond. Associated with each trajectory there is a
complex number A whose phase is the action divided by h, Plancks constant over 2.
e
iS[x]/h
(52)
The most likely trajectories are those that are near the classical trajectory, because the action is
changing very slowly for these trajectories, they have approximately the action of the classical
trajectory, and therefore all the As have the same phase and all the nearby trajectories add up
coherently. Trajectories very far from the classical trajectory are unlikely because the phase is
changing rapidly and nearby trajectories have different phases and add nearly to zero.
There is another way of putting this that is perhaps more interesting. Euler-Lagrange equations
are differential equation, but not every differential equation is an Euler-Lagrange equation. The
fact that the world is quantum mechanical explains why the classical physics that we see can
be described by the solutions not just to any old differential equation, but specically to Euler-
Lagrange equations. We will see why this distinction is important as we go along.
12
Appendix: On the functional Taylor series
It might be useful for some of you if I expand on the statement of the functional Taylor series
because it a pretty piece of math. Some of you, on the other hand, may nd this rather terrifying.
Please DONT PANIC. This section is called an appendix because it is completely optional. I
wont get to it at all in lecture. But for some of you it may be fun.
The picture of the functional Taylor series that I will give you is related to the following state-
ment that you may nd useful. The t in a function x(t) can be thought of as a kind of index,
labeling components of an innite dimensional vector. Or to put this the other way around, a vec-
tor r can be thought of as a function of the index that labels the component r
j
is a number for
each j just like x(t) is a number for each t. In this way of thinking, a functional, W[x] is like a
function of several variables, f(r
1
, , r
n
) (depending on an n-dimensional vector), except that
W[x] depends on an innite number of variables, the values of x(t) for some range of t. Now what
does the Taylor expansion look like for a function of several variables, like f(r
1
, , r
n
)? We can
build up the Taylor series by looking at one variable at a time:
f(r
1
+ a
1
, , r
n
) =
_
1 + a
1

r
1
+
1
2
a
2
1

2
r
2
1
+
_
f(r
1
, , r
n
)
f(r
1
+ a
1
, r
2
+ a
2
, , r
n
) =
_
1 + a
2

r
2
+
1
2
a
2
2

2
r
2
2
+
_
f(r
1
+ a
1
, r
2
, , r
n
)
and so on for all n variables
(53)
This looks complicated, but it can be simplied easily using a very beautiful form for the Taylor
expansion of one variable:
g(x + a) = exp
_
a

x
_
g(x) =

k=0
a
k
k!
d
k
dx
k
g(x) (54)
The expansion of the exponential precisely reproduces the terms in the Taylor expansion. Using
(54), equation (53) becomes
f(r
1
+ a
1
, , r
n
+ a
n
) =
_
_
n

j=1
exp
_
a
j

x
j
_
_
_
f(r
1
, , r
n
) (55)
But because the product of exponentials is the exponential of the sum of the exponents, this can be
written as
f(r
1
+ a
1
, , r
n
+ a
n
) = exp
_
_
n

j=1
a
j

x
j
_
_
f(r
1
, , r
n
) (56)
This is the Taylor series for more than one variable. It is rather neat that it looks just like the Taylor
series for a single variable, (54), except that we have to include derivatives with respect to all the
variables in the exponent. Now for an innite number of variables labeled by a continuous variable
t, in some range t
1
t t
2
, the argument goes the same way, but instead of summing over the
indices, as we do in (56), we must integrate over the continuous variable t, so that the functional
Taylor series looks like
W[x + x] = exp
_
_
t
2
t
1
dt x(t)

x(t)
dt
_
W[x] (57)
13
Note that if we expand the exponential, we get a rst term which is just (22), as expected from
our earlier discussion.
14