-
We begin our study of the basic principals
of optimization by looking at a very
-
general form of optimization problem which
is given by this. Maximize over a variable
-
x. An objective function t of x. To
provide some motivation think of the
-
problem as a one face by a firm that once
maximize total profits given by t of x. So
-
these are total profits. That's a function
of an action that in can take x, which x
-
is any number. A little bit of terminology
in a general optimization problem the
-
variable that we can control is called the
control variable. And this is a variable
-
that which is over and the function that
we are trying to maximize in this case T
-
of X is called the objective function.
Using this terminology the goal of the
-
optimum maximization problem is to select
the level of the control variable that
-
generates the maximum valuable of the
objective function. To provide some
-
intuition consider a graphical
presentation of the problem. We have some
-
sort of function, let's say something that
looks like this, is the function t of x.
-
And what we are trying to do is find the
level of x in this axis that provides the
-
maximum level of a function. In this
graphical representation this would be
-
this level. Fortunately there is a recipe
that you learn in calculus that allow us
-
to take a function t apply some
derivatives and find the general body of x
-
star that maximizes the function. And the
recipe is given by what are known as the
-
first order. Or necessary conditions and
second order or sufficient conditions. The
-
first order conditions state that at any
point that is a local maximum must satisfy
-
the condition that each derivative is
equal to zero. However as you probably
-
remember for calculus necessary conditions
are not sufficient some points may satisfy
-
them and it's still not the maximum and we
will see an example of this in a second.
-
However, if a point also satisfies what I
call second order of sufficient conditions
-
which in this case is given by second
derivative is less than zero so then such
-
a point is a local max imum.
To warm up a practice one which we have
-
just learned lets do a simple example.
Consider the function ten minus x minus
-
five squared. Think of this as a profit
function that we are trying to maximize.
-
In order to find the maximum we need to
first apply the first order conditions
-
which would say that the derivative of the
objective function which in this case is
-
minus two times x minus five. Has to be
equal to zero and doing a little bit of
-
algebra this implies that x and the
optimum. Has to be equal to five. Now we
-
know that these conditions are necessary
but not sufficient so lets check the
-
second order conditions which are the
sufficient ones that requires taking the
-
second derivative of the function or the
derivative of this which in this case
-
means. The second derivative of this
function is just minus two which is less
-
than zero which implies that the second
order conditions are satisfied and that
-
the unique local optimum or local maximum
therefore the global maximum is at x star
-
equals five. Let me provide you with three
different types of intuition for why the
-
first and second order conditions look
like they do. Let's just start with a
-
graphical intuition of the first order
conditions. Suppose that you have a
-
function that looks more or less like
this. That is the objective function that
-
we're trying to maximize. Remember, that
the first order condition states that at
-
any local maximum, the derivative at the
optimum, at the at local maximum has to be
-
equal to zero. Now, to see why this has to
be the case, consider a point for example
-
like at x, at which the derivative.
Clearly, it's not equal to zero. In this
-
particular case it's positive. It is easy
to see that the value of the function of
-
that point can not be a maximum. In other
words x can not be maximizing the function
-
because given that this lobby is positive
if you move a little bit to the right,
-
let's say to a point like x plus delta x
you can increase the value of the function
-
approximately by the value this low times
delta x. Now, it's also easy to see, that
-
if you're at a point, like X-hat, at
which, in this case, the value of the
-
slope is negative, you can also be, novy
maximizing, because if you move to a point
-
like X-hat, minus D-X, minus delta X, you
can also increase the value of the
-
function for similar resource. Now, it's
very easy to see, that the only point at
-
which you can not play this trick, of
improving the value of the objective
-
function, by changing the value X, is a
point, like, X-star, at which, the slope,
-
is zero, because if you move locally,
either to the left, or to the right, the
-
value of the objective function would not
change. This provides an intuition, for
-
why, the first order conditions, have to
satisfy, this property at a global, at a
-
local maximum. Now consider a graphical
intuition for why the second order
-
conditions look like they do. To remind
you, the second order condition is given
-
by, the second derivative and the local
optimum has to be less than zero. That
-
provides together with the first other
conditions, sufficient conditions for
-
optimization. Now, to see why that has to
be the case, compare the following two
-
functions. The one on the left looks like
this, and it does satisfy the, Second
-
order conditions, are the point X star. In
fact, it is easy to see that in this case,
-
the second derivative, at X star is
greater than zero. This low is becoming
-
progressive more positive as you move to
the right. Just like the one before and at
-
which an x star our candidate for the
local optimum the second derivative
-
satisfies the second order condition let's
see why it has to be the case that x star
-
is not a maximum here but is a maximum
here and the key. Nature of the difference
-
is captured by the second order condition.
Now, if you start x star in the field on
-
the left and move a tiny bit to the right
by delta X, since the slope is zero you
-
don't really change the value of the
function, so that doesn't help you,
-
however and this is the key intuition that
you have to see, because the second
-
derivative is positive the value of t he
log changes when you move from Xstar to
-
Xstar + delta X. And becomes positive here
at the new point. That means that if you
-
carry out a further move, let's say by
another delta x since this slope was
-
positive at x star plus delta x, the
function now has increased, and that means
-
that by moving two delta x to the right,
you can improve the value of the function
-
which means that it cannot be an optimum.
In contrast see what happens if you tried
-
to play the same game at x star in the
field on the right. As before the first
-
order conditions are satisfied so eh this
lobby zero at x star but if you tried to
-
move to the right because lets say by
delta x because the second derivative is
-
less then zero this lobe actually goes
down. Unit becomes negative therefore if
-
you try to move a farther the x the value
of the function goes down. It should be
-
easy to see and you should convince
yourself that it doesn't matter for this
-
argument whether the movement is to the
right or to the left. In both cases they
-
go through in an analogous way. This
demonstrates the intuition for why the
-
second order condition t double prime of x
star less than zero provide sufficient
-
conditions for the point to be a local
maximum. Let's look at intuition for why
-
the optimization conditions look like they
do from a different perspective in this
-
case a mathematical one. By the way I'm
trying different intuitions because some
-
of you may find some of them more natural
then the others. Now you will recall from
-
basic calculus that the value of the
objective function in fact of any function
-
at the point x plus dx can be very well
approximated by the value of the function
-
at x. Class the derivative of the
functional x times delta x class the
-
second derivative of the function at x
times the x square plus a bunch of high
-
order terms that are approximate zero for
a small changes in the value of x. Now,
-
given this, the change in the fun in the
objective function produced by the trans
-
delta x can be written as DT. The change
in the function is approximately equal to
-
the derivative at x times delta x, plus
the second derivative at x times the x
-
squared. Now for a function to be
maximized at X, it has to be the case that
-
we cannot find a small change delta X,
either positive or negative, that
-
increased the value of the function, in
other words that makes the change in the
-
function positive. But notice that if the
point X satisfies the first order
-
conditions, the necessary conditions, this
term is zero, sees the first derivative is
-
equal to zero. Furthermore, if it
satisfies the second order conditions,
-
this term is negative, and therefore, the
doesn't change the function has to be less
-
than zero. In other words, when the first
order conditions and second order
-
condition are satisfied, it is not
possible to find a change, a small lock
-
and chance, delta x, that increases the
value of the function. Finally, consider
-
the economic intuition behind the
maximization problem. Recall the
-
reputation of the representative function,
T of X, as some sort of benefit of taking
-
action x measuring dollars. For example if
you would affirm t of x made the note the
-
profit of taking action x. Now under this
interpretation of the problem the
-
derivative of the function x is called the
marginal benefit, which is a concept that
-
we will use repeatedly throughout the
course and that I will abbreviate with the
-
acronym MB. Now, the marginal benefit of x
is equal to either the increase in benefit
-
produced by increasing X by one unit, or
alternative, they decrease in benefit
-
produced by decreasing X by one unit. Now
an important property of an optimum has to
-
be that at the optimum or maximum, the
marginal benefit has to be equal to zero.
-
Now, why is this the case? Well, suppose
that is not the case? For example, suppose
-
that the marginal benefit is greater than
zero then clearly profits or benefits were
-
not maximized at x star since you could
increase them by increasing x by a little
-
bit. So if you increase x, you increase
profits, or the benefit function t.
-
Similarly, if the marginal benefit is l
ess then zero, you can not be maximizing
-
the benefit either, because in this case
by decreasing x you can also increase
-
benefits. It follows that at an optimum
marginal benefit has to be equal to zero.
-
Now the last principle that we have seen
in the economic intuition is so important
-
that is water everything. An economic
actor that is performing the optimal
-
action would choose its level of activity
to the point where marginal benefit is
-
equal to zero. Marginal benefit equal to
zero ingrain that in your heads because
-
otherwise it cannot be optimizing if
marginal benefit was not equal to zero it
-
could increase its benefit or its profits
by either increasing its level of activity
-
or decreasing its level of activity.