P. 1
1540-notes2

|Views: 0|Likes:

See more
See less

10/19/2013

pdf

text

original

Math 1540

Spring 2011
Notes #2
Chapter 6, Di¤erentiable Mappings
This chapter is about the derivative of a function , : 1
n
÷ 1
m:
First, though,
we review some linear algebra.
1 Linear transformations and their matrix repre-
sentations
De…nition 1 A linear transformation from 1
n
to 1
m
is a function 1 : 1
n
÷ 1
m
such that for any r and ¸ in 1
n
. and any scalars c and ,.
1(cr + ,¸) = c1(r) + ,1(¸) .
We can associate 1 with a matrix once we have chosen bases for 1
n
and 1
m
.
Recall that a basis for 1
n
space is a set ¦·
1
. .... ·
n
¦ of vectors in 1
n
which is lin-
early independent and spans the space. Most commonly we use the standard bases,
_
_
_
_
1
0

0
_
_
_
_
.
_
_
_
_
0
1

0
_
_
_
_
.
_
_
_
_
0
0

1
_
_
_
_
for each space. Of course the : basis vectors for
1
n
have : components and the : basis vectors for 1
m
have : components.
As an example, consider the linear transformation from 1
3
to 1
2
de…ned by
1
_
_
r
¸
.
_
_
=
_
r + 2¸ + 3.
4r + 5¸ + 6.
_
.
The standard basis for 1
3
is
_
_
_
_
_
1
0
0
_
_
.
_
_
0
1
0
_
_
.
_
_
0
0
1
_
_
_
_
_
and the standard basis
for 1
2
is
__
1
0
_
.
_
0
1
__
and the matrix for 1 with respect to these bases is the
“obvious” matrix
¹ =
_
1 2 3
4 5 6
_
.
1
But we should keep in mind that 1 is not the same thing as ¹. In order to emphasize
this we will compute that matrix of 1 with respect to another pair of bases. For 1
3
we choose the basis
_
_
_
_
_
1
0
0
_
_
.
_
_
1
1
0
_
_
.
_
_
1
1
1
_
_
_
_
_
.
For 1
2
we choose the basis
__
1
1
_
.
_
1
0
__
.
Suppose that r ¸ 1
3
and r = c
1
·
1
+c
2
·
2
+c
3
·
3
. where ¦·
1
. ·
2
. ·
3
¦ is the basis chosen
for 1
3
. And suppose that 1r = d
1
n
1
+d
2
n
2
. where ¦n
1
. n
2
¦ is the basis chosen for
1
2
. Then the matrix for 1 with respect to these two bases is the matrix 1 such that
_
d
1
d
2
_
= 1
_
_
c
1
c
2
c
3
_
_
.
In other words, the matrix …nds the coe¢cients of 1r with respect to the chosen
basis for 1
2
in terms of the coe¢cients of r with respect to the chosen basis for 1
3
.
In your linear algebra text you can probably …nd the formula:
1 =
_
1 1
1 0
_
1
¹
_
_
1 1 1
0 1 1
0 0 1
_
_
=
_
1 1
1 0
_
1
_
1 2 3
4 5 6
_
_
_
1 1 1
0 1 1
0 0 1
_
_
=
_
4 9 15
÷3 ÷6 ÷9
_
.
We can check this on an example: Let r = 2
_
_
1
0
0
_
_
÷
_
_
1
1
0
_
_
+3
_
_
1
1
1
_
_
=
_
_
4
2
3
_
_
.
Then
1r =
_
17
44
_
= 44
_
1
1
_
÷27
_
1
0
_
.
And
_
4 9 15
÷3 ÷6 ÷9
_
_
_
2
÷1
3
_
_
=
_
44
÷27
_
We will encounter linear transformations that are not de…ned originally by a
matrix; the matrix comes later after the basis is speci…ed.
2
2 The de…nition of Derivative
Derivatives have not appeared previously in this book, but from earlier classes you
should be familiar with the formula
,
0
(r) = lim
h!0
, (r + /) ÷, (r)
/
. (1)
though perhaps you used di¤erent symbols, such as “r” for “/”. This is for a
function , : 1 ÷ 1. Another version of this is probably also familiar:
,
0
(r
0
) = lim
x!x
0
, (r) ÷, (r
0
)
r ÷r
0
.
But either of these formulas is rather misleading when it comes to considering , :
1
n
÷ 1
m
. Whichever formula we use, the message that comes across from the
: = : = 1 case is that the derivative of , at a point r
0
is a number. True, ,
0
is
a function, but for each r
0
. ,
0
(r
0
) is a number. As an example, if , (r) = r
2
then
,
0
(2) = 4.
Now, however, consider the following function from 1
2
to 1
1
:
, (r
1
. r
2
) = r
2
1
+ 2r
2
2
. (2)
Both of the formulas above are di¢cult to interpret, because now r ¸ 1
2
, and so to
make sense of the terms in these formula, such as r +/, both / and r
0
must be also
be in 1
2
. And we don’t know how to divide by an element of 1
2
.
To get around this di¢culty, we recall that one important interpretation of the
derivative in 1
1
is as the slope of a tangent line. If , (r) = r
2
. then ,
0
(2) is the
slope of the line which is tangent to the graph of , at the point (2. 4). We can easily
write down the equation of this tangent line as ¸ = 4r ÷4. The ÷4 insures that this
line passes through the point (2. 4) .
So now we look at the graph of the function , in equation (2). This is a surface,
r
3
= r
2
1
+2r
2
2
. One point on this surface is (1. 2. 9) . We can ask: what is the equation
of the plane which is tangent to the surface at this point?
You probably saw this in Calc 3. To answer it we use partial derivatives. We have
@x
3
@x
1
= 2r
1
and
@x
3
@x
2
= 4r
2
. and substituting r
1
= 1. r
2
= 2 we get the “gradient”:
_
@x
3
@x
1
.
@x
3
@x
2
_
= (2. 8) . The equation of the tangent plane is then
r
3
= 2 (r
1
÷1) + 8 (r
2
÷2) + 9.
which is the equation of a plane which passes through the point (1. 2. 9) .
3
So we ask: what is the derivative of the function , (r
1
. r
2
) = r
2
1
+2r
2
2
at the point
(1. 2) . And the answer is: The derivative is the linear function of two new variables
n and · de…ned by the formula
1, (1. 2) (n. ·) = 2n + 8·.
You can think of n and · as representing (r
1
÷1) and (r
2
9 to get the point (1. 2. 9), the derivative is the function de…ning the tangent plane
at this point.
And now we come to the formula for derivative (De…nition 6.1.1 in the text). But
we need a little more motivation. We will rewrite the one dimensional formula above
to get
lim
x!x
0
_
, (r) ÷, (r
0
)
r ÷r
0
÷,
0
(r
0
)
_
= 0.
and rewriting again, as
lim
x!x
0
, (r) ÷, (r
0
) ÷,
0
(r
0
) (r ÷r
0
)
r ÷r
0
= 0.
Still, this wouldn’t extend easily to higher dimensions, because we are dividing by
r ÷r
0
. and we don’t know how to do this when r and r
0
are vectors. But the limit
above is zero if and only if
lim
x!x
0
[, (r) ÷, (r
0
) ÷,
0
(r
0
) (r ÷r
0
)[
[r ÷r
0
[
= 0.
Now we need only a small change in order to interpret this in higher dimensions:
De…nition 2 The derivative of a function , : 1
n
÷ 1
m
at a point r
0
is a linear
transformation 1, (r
0
) : 1
n
÷ 1
m
such that
lim
x!x
0
[[, (r) ÷, (r
0
) ÷1, (r
0
) (r ÷r
0
)[[
[[r ÷r
0
[[
= 0. (3)
To make sense of this, notice that 1, (r
0
) is a function de…ned on 1
n
. so by
1, (r
0
) (r ÷r
0
) we mean this function evaluated at the vector r ÷r
0
. It is a linear
function, or a linear transformation, from 1
n
to 1
m
. so after agreeing on a basis for
each space, usually the standard bases, we can think of it as a matrix. You can
probably guess the matrix with respect to the standard basis. If r = (r
1
. .... r
n
) and
, = (,
1
. .... ,
m
) . then the matrix is
1, (r
0
) =
_
J,
i
Jr
j
(r
0
)
_
.
4
which we call the “Jacobian of , at r
0
.” It is an : : matrix, as be…ts a linear
transformation from 1
n
to 1
m
.
We should test this out on the example above, namely , (r
1
. r
2
) = r
2
1
+ 2r
2
2
.
Then : = 2. : = 1. It is convenient here to let r
0
= (r
1
. r
2
) and to use (n. ·) as
the argument to which we apply the function 1, (r
1
. r
2
) . We switch to vector and
matrix notation in the 2nd expression below:
1, (r
1
. r
2
) (n. ·) = (2r
1
. 4r
2
)
_
n
·
_
= 2r
1
n + 4r
2
·.
This shows that when : = 1. the matrix for 1, (r
0
) is the gradient of , at r
0
. With
r
0
= (1. 2) . we get
1, (r
0
) = (2. 8) .
which is a 1 2 matrix. Now we want to check the equation (3).
We are lucky here, because the norm [[ [[ in the numerator is fairly simple. That
is because : = 1. so the term inside [[ [[ on the top is a scalar. To make sense of
the second term we have to write “r ÷r
0
” as
_
r
1
÷1
r
2
÷2
_
. We get
r
2
1
+ 2r
2
2
÷1 ÷8 ÷2 (r
1
÷1) ÷8 (r
2
÷2) (4)
which simpli…es to (r
1
÷1)
2
+2 (r
2
÷2)
2
. If we let n = r
1
÷1. · = r
2
÷2. then the
ratio in (3) becomes
n
2
+ 2·
2
_
n
2
+ ·
2
_ 2
_
n
2
+ ·
2
which obviously tends to zero as (n. ·) ÷ (0. 0) . Thus we have veri…ed that 1, (1. 2) =
(2. 8) .
3 6.2; Matrix representation
As discussed earlier, each linear transformation from 1
n
to 1
m
has a matrix repre-
sentation relative to a choice of bases for the two spaces. By far the most frequent
case is when the standard basis is used for both 1
n
and 1
m
. In this case the matrix
representation is easy. Theorem 6.2.2 tells us that if 1, (r) exists and is continuous
in an open set. then the matrix representation of 1, (r) in this set is the Jacobian
matrix
_
_
_
_
_
@f
1
@x
1
(r)
@f
1
@x
2
(r)
@f
1
@xn
(r)
@f
2
@x
1
(r)
@f
2
@x
2
(r)
@f
2
@xn
(r)

@fm
@x
1
(r)
@fm
@xn
(r)
_
_
_
_
_
.
5
As an example, let , (r
1
. r
2
. r
3
) = (2r
1
r
2
3
. r
2
÷r
1
). Thus, , : 1
3
÷ 1
2
. Then
1, (1. 1. 1) (n. ·. n) =
_
2r
2
3
0 4r
3
r
1
÷1 1 0
_
[
(1;1;1)
_
_
n
·
n
_
_
=
_
2 0 4
÷1 1 0
_
_
_
n
·
n
_
_
=
_
2n + 4n
÷n + ·
_
. (5)
4 6.3; Di¤erentiability implies continuity.
This is obvious in one dimension, since lim
x!x
0
f(x)f(x
0
)
xx
0
cannot approach a limit un-
less the numerator approaches zero. For the case of , : 1
n
÷ 1
m
it is just as obvious;
recall that , is di¤erentiable at r
0
with derivative 1, (r
0
) if lim
x!x
0
jjf(x)f(x
0
)Df(x
0
)(xx
0
)jj
jjxx
0
jj
=
0. Assuming this is true, the second term in the numerator tends to zero since 1, (r
0
)
is a linear function and by de…nition of linearity, 1, (r
0
) (0) = 0. Therefore, the limit
of the ratio cannot be zero unless lim
x!x
0
, (r) ÷, (r
0
) = 0. which is the de…nition
of continuity at r
0
.
The text then discusses the case c : 1 ÷ 1
n
. In this case, c is called a curve.
Often,
1
the image of c. which is a set of points in 1
n
is also called a curve, with
it being understood that there is some function c involved. (The text says that c
“represents” a curve, but what “represents” means is not made clear. I prefer to say
that c is the curve, though this has the drawback that di¤erent functions can have
the same image set, which sometimes leads to confusion.)
For the case of a curve c : 1 ÷ 1
m
. di¤erentiability is simpler to discuss. There
are : coordinate functions, so
c (t) =
_
_
_
_
c
1
(t)
c
2
(t)

c
m
(t)
_
_
_
_
.
and it is easy to show that c is di¤erentiable to t
0
if and only if c
1
. .... c
m
are all
di¤erentiable at t
0
. If c is di¤erentiable at t
0
. then we can …nd the value in 1
m
of
1
Indeed, more often, but not in analysis except informally.
6
the linear function 1c (t
0
) . which is
1c (t
0
) (n) =
_
_
_
_
c
0
1
(t
0
) n
c
0
2
(t
0
) n

c
0
m
(t
0
) n
_
_
_
_
= n
_
_
_
_
c
0
1
(t
0
)
c
0
2
(t
0
)

c
0
m
(t
0
)
_
_
_
_
.
In the text it is stated that 1c (t) is “represented by” the column vector
_
_
_
_
c
0
1
(t)
c
0
2
(t)

c
0
m
(t)
_
_
_
_
.
Interpreting all this in the image space 1
m
. this vector is tangent to the image of c
at c (t) . In fact, we usually say, a bit sloppily, that the vector 1c (t) is tangent to c
at the point c (t) .
De…nition 3 A curve c is “smooth” at t
0
if it is di¤erentiable at t
0
and 1c (t
0
) is
not the zero vector.
It is useful to look at the image of a non-smooth curve, such as
c (t) =
_
t
2
. t
3
_
.
For this example, 1c (0) = (0. 0) (using a row vector to save space), and so c is not
smooth at t
0
= 0. And a plot shows that the image of the curve doesn’t look smooth
at c (0) = (0. 0):
25 20 15 10 5
100
50
0
-50
-100
x
y
x
y
On the other hand, c (t) = (t
3
. t
3
) is also not smooth at (0. 0) . and yet its image
looks perfectly smooth, being a straight line.
4.1 Functions which have partial derivatives at a point but
are not di¤erentiable, or even continuous.
The text gives a number of these. One of the simplest is:
, (r. ¸) =
_
_
_
r if ¸ = 0
¸ if r = 0
1 if r ,= 0 and ¸ ,= 0.
7
The partial derivatives exist because , (r. 0) = 0 and , (0. ¸) = 0. But this function
isn’t even continuous at (0. 0) .
Another example is related to the idea of a “directional derivative”. This is
de…ned for a function from 1
n
to 1.
De…nition 4 If , : ÷ 1 where is an open subset of 1
n
. and r
0
¸ . then the
directional derivative of , in the direction of a unit vector c is given by
lim
t!0
, (r
0
+ tc) ÷, (r
0
)
t
.
Here is a picture when : = 2. r
0
= (0. 0) . and c =
1
p
5
(2. 1) . The line is in the
direction of c. If q (t) = , (r
0
+ tc) . then q is the value of , as a function of distance
along the line, and the directional derivative in the direction of c is q
0
(0) .
1 0 . 7 5 0 . 5 0 . 2 5 0 -0 . 2 5 -0 . 5
1
0 . 7 5
0 . 5
0 . 2 5
0
-0 . 2 5
-0 . 5
x
y
x
y
It is shown in the text that if , is di¤erentiable at r
0
. then q
0
(t) = 1, (r
0
) c.
But all of the directional derivatives (i.e. in every direction) can exist even if , is
not di¤erentiable. Examples are given on page 343 and will be discussed in class.
5 Homework, due January 19 at the beginning of
class
1. pg. 330, # 3.
2. pg. 334, #4, But in the …rst part …nd 1, (1. 2) (n. ·) and use (r. ¸) = (1. 2)
in the second part as well. It is assumed that you are using the given basis for
both 1
n
and 1
m
. since in this example : = : = 2.
3. pg. 344 # 2. Also investigate the existence of the directional derivatives of ,
at (0. 0) in every direction.
8
4. pg. 384, #5 b,f .
5. pg. 389, #40 a.
6. pg. 388, #36.
9

scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->