NOTES
ON LINEAR ALGEBRA
CONTENTS:
[10] BASIS
VECTORS - PART II
[11] LINEAR
TRANSFORMATIONS
[10] BASIS VECTORS - PART II
We’ll
now give a procedure to determine when two vectors W1 and W2 are a basis for
the plane. Not only will our method say if they’re a basis, but it will also
tell us how to find a and b.
The
Equation we’re trying to solve is:
Let (R) (U)
W1 = (S) W2 = (V)
Find
a, b so that
(x) (R) (U)
(y) = a (S) + b (V)
Then
(x) (aR) (bU)
(y) = (aS) + (bV)
(x) (Ra) (Ub)
(y) = (Sa) + (Vb)
(x) (Ra
+ Ub)
(y) = (Sa + Vb)
(x) (R
U) (a)
(y) = (S
V) (b)
But
this equation is just begging us to use Gaussian Elimination.
We need to find a number m such that, if we multiply the first row by m and add
it to the second, we get the new row will be (0 something).
So Rm + S = 0
hence m = -S / R
So,
we carry out the Gaussian Elimination. It can be shown that if the two vectors
(R,S) and (U,V) do not lie on the same line, then
Gaussian Elimination will never yield the last row all zero.
Let’s
do some examples:
FIRST EXAMPLE
(2) (3)
W1 = (4) W2 = (7)
Then
we must solve
(x) (2) (3)
(y) = a(4) + b (7)
Then
(x) (2a) (3b)
(y) = (4a) + (7b)
Hence
(x) (2
3) (a)
(y) = (4 7) (b)
NOW
WE DO GAUSSIAN ELIMINATION:
So,
what should we multiply the first row by? We need 2m + 4 = 0, so m = -4/2 = -2.
Hence
we get
(x) (2
3) (a)
(y-2x) = (0 1) (b)
Or 2a + 3b = x; 0a + 1b = y
Therefore
b = y.
2a + 3b = x à a = (x-3b)/2 = (x-3y)/2
So,
given a vector (x,y) we can
find a,b such that (x,y) =
aW1 + bW2. So W1, W2 is a basis!
SECOND EXAMPLE
(2) (4)
W1 = (4) W2 = (8)
Then
we must solve
(x) (2) (4)
(y) = a(4) + b (8)
Then
(x) (2a) (4b)
(y) = (4a) + (8b)
Hence
(x) (2
4) (a)
(y) = (4 8) (b)
NOW
WE DO GAUSSIAN ELIMINATION:
So,
what should we multiply the first row by? We need 2m + 4 = 0, so m = -4/2 = -2.
Hence
we get
(x) (2
4) (a)
(y-2x) = (0 0) (b)
So,
the two equations we must solve are:
2a + 4b = x and 0a
+ 0b = y - 2x
Now,
regardless of what a and b are, 0a + 0b is always zero. If y
- 2x is not zero, it will be impossible to
solve the second equation. So, can we find x and y such that y - 2x ¹ 0? Sure. Take x
= 0, y non-zero. Or take x nonzero, y = 0. Or take y = 22, x = 12. Almost any
choice works.
So we see that W1, W2 are not a basis. We ended up with a row of
zeros. Let’s look at our two vectors again:
(2) (4)
W1 = (4) W2 = (8)
Notice that
(2*2) (2)
W2
= (2*4) = 2(4)
= 2W1
Not
only are W1 and W2
not a basis, but they lie on the same line!
THIRD EXAMPLE
(1) (2) (0)
W1 = (2) W2 = (2) W3 = (1)
(1) (2) (1)
Are
W1, W2, W3 a basis?
(x) (1) (2) (0)
(y) =
a (2) + b (2) + c(1)
(z) (1) (2) (1)
(x) (1a) (2b) (0c)
(y) = (2a) + (2b) + (1c)
(z) (1a) (2b) (1c)
(x) (1a
+ 2b + 0c)
(y) = (2a + 2b + 1c)
(z) (1a
+ 2b + 1c)
(x) (1
2 0) (a)
(y) = (2 2 1) (b)
(z) (1
2 1) (c)
Now
we do Gaussian Elimination! We multiply the first row by -2 and add it to the
second row. (1m + 2 = 0, m = -2).
(x) (1
2 0) (a)
(y-2x) = (0 -2 1) (b)
(z) (1
2 1) (c)
Now
we multiply the first row by -1 and add it to the third row (1m + 1 = 0, m =
-1).
(x) (1 2 0) (a)
(y-2x) = (0 -2 1) (b)
(z-x) (0 0 1) (c)
We
don’t have to do any more work, as this matrix is UPPER
TRIANGULAR. This means the matrix is all zeros below the main diagonal.
We can now solve the three equations, one at a time.
0a + 0b + 1c = z
- x à c = z - x
0a - 2b + 1c = y - 2x à b = (y-2x - z + x) / -2
1a + 2b + 0c =
x à a = y - 2x - z
+ x
So these three vectors are a basis.
In general, to determine if something is a basis for 3space:
(L) (P) (S)
W1 = (M) W2 = (Q) W3 = (T)
(N) (R) (U)
Are
W1, W2, W3 a basis?
(x) (L) (P) (S)
(y) =
a (M) +
b (Q) +
c(T)
(z) (N) (R) (U)
(x) (L a) (P b) (S c)
(y) = (M
a) + (Q b) + (T c)
(z) (N a) (R b) (U c)
(x) (L a + P b + S c)
(y) = (M a
+Q b + Tc)
(z) (N a + R b + Uc)
(x) (L
P
S) (a)
(y) = (M Q T) (b)
(z) (N
R U)
(c)
The
reason for all the colour is to (hopefully) show how things are
going. To determine if W1, W2, W3 are a basis, we are led to solving a matrix
equation. The first column of our matrix is W1, the second column is W2, the third column is W3. Call this matrix W. We then have
(x)
(y) = a W1 +
b W2 + c W3
(z)
(x) ( ) (a)
(y) = ( W1 W2 W3)(b)
(z) ( ) (c)
[11] LINEAR TRANSFORMATIONS
Linear
Transformations are very useful in mathematics. The reason is they allow us to
understand functions at complicated values by understanding them at simpler
values. First, the definition for functions, then we’ll generalize to matrices:
We
say a function is a linear function if two conditions hold:
(1)
f(x + y) = f(x) + f(y) for
all x,y
(2)
f(ax) = af(x)
Now,
it is very unusual for a function to be linear. Take f(w)
= Sin[w].
Then
f(x) = Sin[ax], which usually is not a Sin[x]. For
example, if x = 180, then a Sin[x] is always zero. But if a = ½, Sin[a x] = Sin[90] = 1.
Let’s
try f(w) = w2. Does f(ax)
= af(x)?
Well,
f(ax) = (ax)2 = a2 x2
= a2 f(x) ¹ a f(x) unless a = 1 or 0.
Also,
f(x+y) = (x+y)2 = x2 + 2xy + y2 = f(x) +
2xy + f(y) ¹ f(x) + f(y)
unless x or y = 0.
How about f(w) = 3w + 1?
Well, f(ax) = 3(ax) + 1 = a(3x)
+ 1
= a(3x + 1 - 1)
+ 1
= a(f(x) - 1) +
1
=
a f(x) - a + 1
¹ a f(x) unless
a = 1
Just in case you’re wondering if any function is
linear, here’s one that is:
f(w) = 3w
Then f(ax) =
3(ax) = a(3x)
= a f(x)
f(x+y) = 3(x+y) = 3x + 3y = f(x) + f(y)
[NOTE: one can prove that the only linear functions are f(x) = cx, where c is any real or complex number].
We now generalize this to higher dimensions. Why do we care about
higher dimensions? Well, matrices act on vectors (you’ve seen this in your
force / stress diagrams) and it turns out that matrices will be linear
transformations.
Let V and W be any two vectors with the same
number of components, and let e be a real number. Then any matrix (that is the
correct size to act on V and W) is a linear transformation, namely,
(1) A (V
+ W)
= A V +
A W
(2) A(c V) =
c A V
I’ll sketch the proof for the 2x2 case:
(v1) (w1)
Let V = (v2) W = (w2)
(a b)
Let the matrix
A = (c d)
Then
(a b) ( (v1)
(w1) )
A (V + W) = (c
d) ( (v2) +
(w2) )
(a b) ( v1 + w1)
= (c
d) ( v2 + w2)
( a(v1 +
w1) +
b(v2 + w2) )
= ( c(v1 + w1) + d(v2 + w2) )
( av1 +
bv2 + aw1 + bw2 )
= ( cv1 + dv2 + cw1 + dw2 )
( av1 +
bv2 ) ( aw1 + bw2 )
= ( cv1 + dv2 ) + ( cw1 + dw2 )
(a b) (v1)
(a b) (w1)
= (c d) (v2) + (c d) (w2)
= A
V + A W
The
other condition is even easier to check:
(a b) ( (v1) )
A (eV) = (c d) ( e (v2) )
(a b) (e v1)
= (c
d) (e v2)
(ae v1 + be v2)
= (ce v1 + de v2)
(a v1 + b v2)
= e (c
v1 + d v2)
(a b) (v1)
= e (c
d) (v2)
= e A V
A
similar proof works for any size matrix, concluding the proof.
COMING
ATTRACTIONS:
WHY
DO WE CARE ABOUT BASES? WHY DO WE CARE ABOUT LINEAR TRANSFORMATIONS? WHAT’S THE
CONNECTION BETWEEN THE TWO?
Eventually, we’ll see that certain matrices have natural ‘bases’
attached to them. They (and powers of them) may look very ugly as given. But if
we changes bases, using something else other than the x-axis and the y-axis, we
can often make the matrices look good.
For symmetric matrices, this will be the case. In fact, the Principle Axis Theorem says we will be able to find a
basis where, if we write our matrix relative to that basis, it will be diagonal!
Also, let’s say W1 and W2 are a basis. Then we can write any
vector V = (x,y) in terms of
the two, or
V
= a W1 + b W2
Then if A is a matrix, we have
A
V = A ( aW1 + bW2 )
A
V = A (aW1) + A (bW2)
A
V = a (A W1) + b (A W2)
Or, more generally, AN
V = a (AN W1) + b (AN W2)
Real
Symmetric Matrices have what is called a ‘basis of eigenvectors’. That means
there are real numbers c1 and c2 such that
A W1 = c1 W1 A W2 = c2 W2
Applying
A multiple times yields
AN W1 = c1N
W1 AN W2 = c2N
W2
Hence
AN V = a (AN W1) + b (AN W2)
= a c1N
W1 + b c2N W2 (Eq 11.1)
So
here’s the advantage: Let’s say N is real big, say a billion. If we were to
calculate AN V we would have to multiply A by itself one billion
times, and then have that act on V. That’s a lot of calculation to do.
But,
if our matrix is symmetric, we’ll be able to find W1 and W2 (two calculations),
c1 and c2 (two more calculations) and numbers a and b (two more calculations), and then we just take N =
one billion in (Eq 11.1), and we’re done!
See how much we saved!