Sie sind auf Seite 1von 3

Interesting Questions About the Definition of a Linear Transformation

Definition 1. T : Rn → Rm is called a linear transformation if both of the following hold:


(1) T (sx) = sT (x) for all scalars s and all vectors x in Rn .
(2) T (x + y) = T (x) + T (y) for all vectors x, y in Rn .

Question. Does (1) imply (2)?


Answer. No.
 
2 x1 p
Example. Consider T : R → R defined by T = 3 x31 + x32 . Then (1) holds because
x2
p p p
T (sx) = 3 (sx1 )3 + (sx2 )3 = 3 s3 (x31 + x32 ) = s 3 x31 + x32 = sT (x).
However, (2) does not hold because

     
1 0 1 3
T + =T = 2
0 1 1
and
√ √
   
1 0 3 3
T +T = 1 + 1 = 2.
0 1
Therefore (1) does not imply (2).
Question. Does (2) imply (1)?
Answer. No. But it’s a very interesting story that leads quickly to the cutting edge of mathematical
knowledge.
Let’s assume (2) and try to prove (1).
T (sx) = T (x + (s − 1)x)
= T (x) + T ((s − 1)x) By (2)
?
= T (x) + (s − 1)T (x) How?!?!
= sT (x)

It seems the above argument doesn’t work without already having (1). However, by building step-by-
step, we can show that (2) almost implies (1).
Theorem 2. If T : Rn → Rm satisfies
(2) T (x + y) = T (x) + T (y) for all vectors x, y in Rn ,
then
(10 ) T (sx) = sT (x) for all rational numbers s and all vectors x in Rn .

Proof. Let x be an arbitrary vector in Rn . We have


T (2x) = T (x + x) = T (x) + T (x) = 2T (x).
1
2

Moreover, if k is a postive integer, then


T (kx) = T (x + · · · + x) = T (x) + · · · + T (x) = kT (x)
k times k times
Since
T (0) = T (0 + 0) = T (0) + T (0),
we must have
T (0) = 0.
Therefore
T (0x) = T (0) = 0 = 0T (x)
and
T (−x) = T (−x) + T (x) − T (x) = T (−x + x) − T (x) = T (0) − T (x) = −T (x).
So if k is a postive integer we have
T (−kx) = −T (kx) = −kT (x).
So far we’ve shown that, for all integers s,
T (sx) = sT (x).
If a, b are non-zero integers, then
a       
1 ab 1 a 1 a
T x = aT x = T x = T b x = T (x).
b b b b b b b
This proves
T (sx) = sT (x)
for all rational numbers s. 

If we assume that T is continuous at even one point, then (2) does imply (1).
Theorem 3. If T : Rn → Rm is continuous at one or more points and satisfies
(2) T (x + y) = T (x) + T (y) for all vectors x, y in Rn ,
then
(1) T (sx) = sT (x) for all scalars s and all vectors x in Rn .

Proof. Assume first that T is continuous everywhere and (2) holds. We know that (2) implies
T (sx) = sT (x) for all rational numbers s and all vectors x in Rn .
Let s be any scalar (i.e., real number). Let sk be s up to k decimal places. For example, if s = π =
3.14159265 . . . and k = 4, then sk = 3.1415. Clearly lim sk = s and each sk is rational. Therefore
n→∞

T (sx) = T ( lim sk x) = lim T (sk x) = lim sk T (x) = sT (x).


k→∞ k→∞ n→∞

Now assume only that T is continuous at some point x0 and (2) holds. If we can show that T must
be continuous everywhere, then we’ll be done because of the argument above. Let x be any vector
in Rn . Then
lim T (y) = lim T (x + z − x0 ) = lim (T (x) + T (z) − T (x0 )) = T (x).
y→x z→x0 z→x0

This proves T is continuous everywhere. 


3

As a consequence of the previous theorem, any transformation T that satisfies (2) but not (1) must
be discontinuous everywhere. In fact, such transformations are necessarily so badly behaved that you
will almost certainly never meet them outside of certain advanced mathematics courses where you
specifically go looking for monsters.
In the next example we will construct a transformation T : Rn → Rm that satisfies (2) but not
(1), hence showing (2) does not imply (1). The example uses some advanced ideas that are beyond
the scope of the current course, so I won’t try to explain them here. However, you are welcome
to look them up or ask me about them. Unfortunately, a simpler example is currently unknown to
mathematics.
Example. Choose a Hamel basis for Rn over Q, then choose a countably infinite subset {v1 , v2 , . . .}
of the basis. Let T : Rn → Rm be the Q-linear transformation defined by setting T (vk ) = kkvk ke1
for k = 1, 2, . . . and T (x) = 0 for all other bases vectors x. Here e1 is the first standard basis vector
for Rm . Since kT (vk )k/kvk k = k → ∞, T is unbounded. Since R-linear transformations between
finite dimensional vector spaces must be bounded, T is not R-linear. Therefore (1) or (2) must fail.
The Q-linearity of T implies (2) holds. Thus (1) must fail.
Complex Vectors. Let C be the set of complex numbers. These are the numbers of the form a + ib,
where a and b are real and i has the property i2 = −1. Later in the course, we will work with
transformations T : Cn → Cm and the scalars will be in C.
Definition 4. T : Cn → Cm is called a linear transformation if both of the following hold:
(1.C) T (sx) = sT (x) for all scalars s and all vectors x in Cn .
(2.C) T (x + y) = T (x) + T (y) for all vectors x, y in Cn .

In this setting, it is easy to come up with an example that shows (2.C) does not imply (1.C).
Example. Consider the transformation T = Re : C → C defined by Re(a + ib) = a. For any
x = a + ib and y = c + id in C, we have
Re(x+y) = Re(a+ib+c+id) = Re(a+c+i(b+d)) = a+c = Re(a+ib)+Re(c+id) = Re(x)+Re(y).
Therefore (2.C) holds. However, if x = 2 + i3 and s = i, then
Re(sx) = Re(i(2 + 3i)) = Re(−3 + i2) = −3
and
sRe(x) = iRe(2 + 3i) = i2.
So (1.C) does not hold. Thus (2.C) does not imply (1.C).

Das könnte Ihnen auch gefallen