Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
ii
CONTENTS
iii
Contents
1 2 Introduction Geometry in two dimensions 2.1 Vector operations . . . . 2.2 Lines . . . . . . . . . . . 2.3 Triangles . . . . . . . . . 2.4 Distances . . . . . . . . 2.5 Dot product . . . . . . . 2.6 Equation of a line . . . . 2.7 Perpendicular distance . 2.8 Two famous inequalities Linear inequalities Conics 4.1 Transforming conics . 4.1.1 Translation . 4.1.2 Rotation . . . 4.2 The Theory . . . . . 4.3 Geometrical denition 1 3 3 4 6 7 7 8 9 10 13 15 18 18 19 20 22
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
3 4
. . . . . . . . . . . . . . . . . . . . . . . . of conics
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Matrices and linear equations 25 5.1 Method Gaussian elimination . . . . . . . . . . . . . . . . . . . . . . 27 Some applications of matrix algebra 6.1 Linear equations . . . . . . . . . . . 6.2 Population Distribution . . . . . . . 6.3 Matrices and Geometry . . . . . . . 6.3.1 Rotation . . . . . . . . . . . 6.3.2 Reection . . . . . . . . . . 6.3.3 Combining Transformations 37 37 38 41 41 41 41 43 43 44 45
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
Inverses 7.1 Relation to linear equations . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Finding inverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Finding inverses in general . . . . . . . . . . . . . . . . . . . . . . . .
iv 8
CONTENTS Determinants 47 8.1 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 8.2 Eects of row operations on determinant . . . . . . . . . . . . . . . . 49 8.3 n n determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Eigenvalues and eigenvectors 9.1 Eigenvectors . . . . . . . . . 9.2 How to nd eigenvectors and 9.2.1 Back to Fibonacci . 9.3 Diagonalization . . . . . . . 9.3.1 Repeated eigenvalues . . . . . . . eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 53 54 55 56 60 63 65 67 67 70 70
10 Conics(again) and Quadrics 10.1 Reduction of conics . . . . . . . 10.1.1 Orthogonal matrices . . 10.2 Quadric surfaces . . . . . . . . . 10.3 Linear Geometry in 3 dimensions 10.4 Geometry of planes . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Chapter 1
Introduction
Some Greek highlights
Chapter 2
2.1
Vector operations
Denition 2.2. 1) Addition dene the sum of two vectors by (x1 , x2 ) + (y1 , y2 ) = (x1 + y1 , x2 + y2 ) 2) Scalar multiplication A scalar is a real number. If x = (x1 , x2 ) is a vector and R is a scalar x = (x1 , x2 ) = (x1 , x2 ) Subtraction x y = x + (y ) = (x1 y1 , x2 y2 )
4 Note. If , R and x , y R2 i) (x + y ) = x + y ii) ( + )x = x + x Proof. (i) (x + y ) = (x1 + y1 , x2 + y2 ) = ((x1 + y1 ), (x2 + y2 )) = (x1 + y1 , x2 + y2 ) = (x1 , y1 ) + (y1 , y2 ) = x + y (ii) excercise
2.2. LINES
2.2
Lines
In Euclid, points and lines are undened. But we need to dene points and lines. We have already dened points. Denition 2.3. Let a, u be vectors in R2 with u = O. The set of vectors {a + u | R} is called a line. Denition 2.4. Two lines {a + u | R} and {b + v | R} are parallel i v = u for some R. Next we will prove the three famous propositions from Euclid. Proposition 2.1. If L, L are parallel lines which have a point in common, then L = L . Proposition 2.2. Through any two points, there is exactly one line. Proposition 2.3. Two non parallel lines meet in exactly one point. Proof. Proposition 2.1. Let L = {a + u | R}
2.2. LINES Given that L and L have x in common, we have x = a + 1 u = b + 2 u = a + (1 2 )u b = a + 1 u 2 u b = a + 3 u Finally, we get L = {b + u | R} = {a + 3 u + u | R} = {a + u | R} = L 1 , 2 R
(3 = (1 2 ) 3 R)
= {a + (3 + )u | R}
Proof. Proposition 2.2. Let u, v R2 to be our two points. Let us take the line Suppose now that L is some line through u and v Since points u and v line on L , we get v L = {u + (v u ) | R} L = {a + w | R} u = a + 1 w = a + 2 w v u = (2 1 )w
Therefore, lines L and L are parallel. As they also have a common point, they are (by proposition 1) identical. Example 2.1. Find common point of L1 = {(0, 1) + (1, 1) | R} L2 = {(4, 0) + (2, 1) | R} x x = (0, 1) + 1 (1, 1) = (4, 0) + 2 (2, 1)
1 = 6
2.3. TRIANGLES
2.3
Triangles
Denition 2.5. A triangle is a set of 3 non-colinear points {a, b, c | a, b, c R2 }. Edges of a triangle are the line segments ab, ac , bc , where ab = {a + (b a) | 0 1} Midpoint of ab is the point 1 1 a + (b 1) = (a + b) 2 2 A median of a triangle is a line joining one of a, b, c with the midpoint of the oposite side (line segment). Proposition 2.4. The 3 medians of a triangle meet in a common point. Proof. Let the three medians be Ma , Mb and Mc 1 Ma = {a + ( (b + c ) a) | R} 2 1 Mb = {b + ( (a + c ) b) | R} 2 1 Mc = {c + ( (a + b) c ) | R} 2 We just show, that for = 2 3 , medians meet at the same point 1 Ma = a + b + 3 1 Mp = b + a + 3 1 Ma = c + a + 3 1 c 3 1 c 3 1 b 3 2 a= 3 2 b= 3 2 c= 3 1 (a + b + c ) 3 1 (a + b + c ) 3 1 (a + b + c ) 3
Therefore, all the medians contain the point (sometimes called centroid ) 1 3 (a + b + c ). Note. Other interesting properties of triangles are 3 altitudes meet at a point (sometimes called orthocentre ) 3 perpendicular bisectors meet at a point (sometimes called circumcentre ) the 3 centres (orthocentre, circumcentre, centroid) are colinear, (they lie on Euler line )
2.4. DISTANCES
2.4
Distances
The distance between x and y (x , y R2 ) is the length of (x y ) called dist(x , y ) dist(x , y ) = (x1 y1 )2 + (x2 y2 )2
2.5
Dot product
Denition 2.7. For x , y R2 , dot product (scalar product) of x and y is a real number x .y = x1 y1 + x2 y2 Example 2.2. x = (1, 2), y = (1, 0) x x .y Proposition 2.5. (i) x
2
12 + 22 =
dist(x , y ) =
5 22 + 22 = 2 2
= 1 + 0 = 1 x y
2
+ y
= 2x .y y cos
(iii) The position vectors of x and y are at right angles i x .y = 0 Proof. (i)
2 2 2 2 LHS = x1 + x2 + y1 + y2 (y1 x1 )2 (y2 x2 )2
= (2x1 y1 ) (2x2 y2 )
+ y 2 y x 2 x y
= cos
2x .y x .y
= 2 x = x
y cos y cos
2.6
Equation of a line
Consider a line L = {a + u | R}, u = (u1 , u2 ), dene n = (u2 , u1 ), so that n.u = 0. For any x L, x = a + u x .n = (a + u ).n = a.n + u.n = a.n x .n = a.n Proposition 2.6. Let L = {a + y | R}, n = (u2 , u1 ). Then (i) Every x in L satises x .n = a.n (ii) Every solution of x .n = a.n lies on L Proof. (ii) Suppose x .n = a.n (we need to show that x = a + u , i.e. x a = u ). Then (x a).n = 0. Let y = x a, and say y = (y1 , y2 ). As y .n = 0 y1 u2 + y2 u1 = 0
y1 (u1 , u2 ) if u1 = 0, or y = So, y = u 1 Therefore y2 u2 (u1 , u2 )
y2 u1 = y1 u2 if u2 = 0.
y x
= u = x a = a + u L
Denition 2.8. For a line L = {a + u | R}, the vector n = (u2 , u1 ) (or any scalar multiple of it) is called a normal to L. The equation x .n = a.n is called the equation of L. Proposition 2.7. The linear equation px1 + qx2 + r = 0 (1) is an equation of a line with normal (p, q ) and with direction (q, p ). Proof. Suppose q = 0. Then the solution of (1) r p x1 + x2 + = 0 q q Let x1 = , x2 =
r q
p q ,
so , p r q q p r + 1, 0, q q
p 1, q
(x1 , x2 ) = =
1 q (q, p ).
Normal is clearly
2.7. PERPENDICULAR DISTANCE When q = 0. Equation (1) is px1 + r = 0, with solution (x1 , x2 ) = r , 0 + (0, 1) p
Denition 2.9. Lines L1 = {a + u | R} and L2 = {b + v | R} are perpendicular i u.v = 0. Proposition 2.8. Let lines L1 , L2 have equations L1 : p1 x1 + q1 x2 + r1 = 0 L2 : p2 x1 + q2 x2 + r2 = 0
Then (i) L1 , L2 are parallel i (p1 , q1 ) = (p2 , q2 ) for some scalar (ii) L1 , L2 are perpendicular i (p1 , q1 ) (p2 , q2 ) = 0 Proof. (i) By Proposition 2.7 L1 = {a + (q1 , p1 ) | R}
L2 = {b + (q2 , p2 ) | R}
These are parallel i R such that (q2 , p2 ) = (q1 , p1 ), i.e. (p1 , q1 ) = (p2 , q2 ). (ii) excercise
2.7
Perpendicular distance
Denition 2.10. For a line L and point p , p / L, dist(p, L) is the length of the perpendicular from p to L. Proposition 2.9. If a is a point on the line and n is a normal to the line, then dist(p, L) = (p a).n n
w w
Note. A unit vector is a vector of length 1. If w is any vector, then w = vector. Therefore we can write dist(p, l ) = |(p a).n |
is a unit
10 Proof.
2.8
Proposition 2.10. (Cauchy - Schwartz inequality) Let x , y R2 . (i) |x .y | x (ii) If |x .y | = x Proof. (i) |x .y | (x1 y1 + x2 y2 )
2 2 x1 y1 2
2 2 2 2 ) + y2 )(y1 + x2 (x1 2 2 2 2 2 2 2 2 x1 y1 + x1 y2 + x2 y1 + x2 y2 2 2 2 2 x1 y2 + x2 y 1 2x 1 y 1 x 2 y 2
+ 2x 1 y 1 x 2 y 2 +
2 2 x2 y2
0 (x1 y2 x2 y1 ) (ii) If |x .y | = x
2
Therefore one of x and y is a scalar multiple of the other. 2 + x 2) = x 2. Note that x .x = (x1 2
11
= (x + y )(x + y ) = x .x + y .y + 2x .y = x x
2 2
+ y + y
2 2
+ 2x .y +2 x y by Cauchy - Schwartz
( x + y ) Therefore x + y x + y .
12
13
Chapter 3
Linear inequalities
14
15
Chapter 4
Conics
In R2 , a linear equation px1 + qx2 + r = 0 denes a line. Now we dene a curve in R2 by quadratic equation. Example 4.1. Circle of center c , radius r . It contains all the points x such that x c = r
2
2 2 2 2 x1 + x2 2c2 x2 + c1 + c2 r2 = 0
(x1 c1 )2 + (x2 c2 )2 = r 2
x c
= r2
Denition 4.1. A conic section in R2 is the set of poitns x R2 , x = (x1 , x2 ), satisfying a quadratic equation
2 2 ax1 + bx2 + cx1 x2 + dx1 + ex2 + f = 0
(4.1)
where not all of a, b and c are 0. Here are some basic examples (1)
2 2 x1 x2 + =1 a2 b2
is an ellipse b a
(2)
2 2 x1 x2 =1 a2 b2
is a hyperbola
16
(3)
2 x2 = ax1 +b
(a = 0) is a parabola.
(4)
2 2 x2 x1 =0 a2 b2
is a pair of lines.
(5)
2 x1 =1 a2
is a pair of lines.
17
(6)
2 x1 =0 a2
is a point (8)
2 2 x2 x1 + = 1 a2 b2
is the empty set (empty conic) We call the conics (6), (7) and (8) degenerate conics. Well see how to reduce an arbitrary conic to one of these basic examples.
4.1
Transforming conics
Aim start with an arbitrary conic in the form (4.1), do some geometrical chages to coordinates so that the equation becomes one of our standard equations (1) (8). Two kinds of changes are allowed.
4.1.1
Translation
To change coordinates from (x1 , x2 ) to (y1 , y2 ) with new origin (, ) y1 = x1 y2 = x2 This is called a translation. Example 4.2. What type is the following conic?
2 2 x1 + 2x 2 2x1 + 4x2 10 = 0
We complete the square (x1 1)2 + 2(x2 + 1)2 13 = 0 So the new coordinates are y1 = x1 1
y2 = x2 + 1 Now, we can easily see that the our conic is an ellipse. Remark If (4.1) has no x1 x2 term (c = 0), then we can nd a translation which reduces the equation to one of the standard ones (examples (1) (6)).
19
4.1.2
Rotation
Rotate axes anticlockwise through . What happens to the coordinates of a general point P ? Say P has old coordinates (x1 , x2 ) and new coordinates (y1 , y2 ).
x2 y2 r
y1
x1
Now y1 = r cos y2 = r sin and x1 = r cos( + ) x2 = r sin( + ) Hence x1 = r cos cos r sin sin = y1 cos y2 sin and x2 = r sin cos + cos sin = y2 cos + y1 sin Summarizing change of coordinates when we do a rotation through angle is x1 = y1 cos y2 sin
x2 = y1 sin + y2 cos
x1 = x2 =
1 (y1 y2 ) 2 1 (y1 + y2 ) 2
Example 4.4. What is the following conic (nd the rotation and translation to change coordinates and get standard equation)
2 2 x1 + x2 + 4x 1 x 2 = 1
4.
4.2
The Theory
Aim (i) nd the rotation which gets rid of x1 x2 (ii) complete the square to nd the translation which changes the equation to one of our six standard equations Part(i). If c = 0, we dont need to rotate. So assume that c = 0. When we do general rotation through we change coordinates to y1 , y2 x1 = y1 cos y2 sin x2 = y1 sin + y2 cos We aim to nd so that the new equation has no y1 y2 term.
2 bx2 = b(y1 sin + y2 cos )2 2 ax1 = a(y1 cos y2 sin )2
cx1 x2 = c (y1 sin + y2 cos )(y1 cos y2 sin ) So the y1 y2 term when we change coordinates will be 2a sin cos + 2b sin cos + c (cos2 sin2 ) = (b a) sin 2 + c cos 2
4.2. THE THEORY So we want to choose to make this expression zero. (a b) sin 2 = c cos 2 If a = b then tan 2 = c ab
21
Step 1 Rotation. If c = 0 in (4.1), we rotate through , where = 4 when a = b or c tan 2 = ab when a = b. Step 2 Translation. After Step 1, equation becomes
2 2 a y1 + b y2 + d y1 + e y2 + f = 0
Now we complete the square to nd a translation which changes equation to one of the standard ones. Weve proved: Theorem 4.1. Every conic in the form (4.1) can be changed by rotation and translation of the axes to one of the standard equations (1) (8). Thus every conic is either an ellipse, hyperbola, parabola, or one of the degenerate conics. Example 4.5. Reduce conic 2 2x 1 + 2 3x1 x2 + 3 3 1 x1 + 3 + 3x2 = 0 to standartd form by rotation and translation. Step 1 Rotation Here a = 2, b = 0, c = 2 3. So rotate through , where tan 2 = 3 Therefore 2 = 3 , so =
6.
x1 = y1 cos y2 sin 1 3 = y1 y2 2 2 x2 = y1 sin + y2 cos 1 3 y1 + y2 = 2 2 So, the equation becomes 2 1 3y 1 y 2 + 2 3 1 3y1 y2 (y1 + 3y2 ) 24 4 1 1 +(3 3 1) 2 ( 3y1 y2 ) + (3 + 3) 2 (y1 + 3y2 ) = 0 Which is
2 2 3y 1 y2 + 6y 1 + 2y 2 = 0
22
4.3. GEOMETRICAL DEFINITION OF CONICS Step 2 Translation complete the square 3(y1 + 1)2 (y2 y1 )2 = 2 So put z1 = y1 + 1 , z2 = y2 1, and equation is now standard equation
2 2 3 z1 z2 =2
This is a hyperbola. Sketch new origin has y1 , y2 coordinates (1, 1), so has 1 1 x1 x2 coordinates 2 ( 3 + 1), 2 (1 + 3) Note. Usually tan 2 = use formula
c a b
is not so convinient (cant write what 2 is). In general, tan 2 = 2 tan 1 tan2
2 + b . What about parabola x = ax 2 + b ? Note. Standard parabola is x2 = ax1 1 2 We can rotate this through 2 to the standard equation.
4.3
Ingredients of this denition a line L a point p not on L a real number e > 0 Denition 4.2. Curve C is a set of all points x R2 , such that x p = e dist(x , L) i.e. C = x R2 | x p = e dist(p, L) Example 4.6. Let e = 1. 1 Let e = 2 Theorem 4.2. Curve C (from previous denition) is a conic. It is a parabola if e = 1 an ellipse if e < 1 a hyperbola if e > 1 Proof. Do a rotation and translation to make p the origin and L vertical line x1 = s . Then the equation deninig C is x p = e dist(x , L)
23
= e 2 (x1 s )2
(1 e 2 ) x1 + So put y1 = x1 +
se 2 1 e 2
se 2 1 e2
2 2 + x2 s 2e 2
s 2e 4 =0 1 e2
i.e
e2 1 e2 (4.2)
Denition 4.3. The conic has focus p , directric L, excentricity e from the previous proof. Example 4.7. Find e , p and L for the ellipse
2 x1 2 + x2 =1 2 This is the standard equation. We compare it with (4.2) 2 y1 + 2 s 2e 2 y2 = 1 e2 (1 e 2 )2 se 2 1 e 2 , 0
(4.3) , directrix is x1 = s +
se 2 1 e 2 .
2 y2 s 2e 2 = 1 e2 (1 e 2 )2 2 2 x 1 + 2x 2 = 2
So e =
1 , 2
1 , 2
24
25
Chapter 5
26 1) 3x 1 + 4x 2 = 5 8x 1 x 2 = 2 is a system of 2 linear equations in x1 , x2 . 2) x1 + x2 + x3 + x4 = 0 2x 1 x 2 + 5x 4 = 2 x1 + x2 x4 = 3 is a system of 3 equations in x1 , x2 , x3 , x4 . Denition 5.6. General system a21 x1 + a22 x2 + + a2n xn am1 x1 + am2 x2 + + amn xn a11 x1 + a12 x2 + + a1n xn = = = b1 b2 bm
where x1 , . . . , xn are unknowns and aij , bbi are constants. A solution to this system is a vector (k1 , . . . , kn ) which satises all the equations. Aim is to nd the method to nd all solutions of any system of the form from previous denition. Example 5.3. System x1 + x1 = 1 2x 1 + 3x 2 = 5 Eliminate x1 take twice rst, add to second 5x 2 = 7 One solution (x1 , x2 ) = Example 5.4. 2x 1 + x 2 = 2 6x 1 3x 2 = 1 These are just two parallel (nonidentical) lines the system has no solution. Example 5.5. System x1 + x2 = 0 x 1 x 2 + 2x 3 = 0 Notice (cleverly), that third equation is equal to rst minus two times second. So any solution of system of rst and second equation will automatically satisfy third. So general solution is (x1 , x2 , x3 ) = (a, a, a) x2 x3 = 0
2 7 ( 5 , 5 ).
for any a R. This system has innitely many solutions. Soon we will see that every system has either no solution, one solution, or unlimited number of solution.
27
5.1
Example 5.6. System x 1 + 2x 2 + 3x 3 = 9 4x1 + 5x2 + 6x3 = 24 3x 1 + x 2 2x 3 = 4 Step 1 Eliminate x1 from second and third, using rst. We get equivalent system x 1 + 2x 2 + 3x 3 = 9 5x2 11x2 = 23 3x2 6x3 = 12 (3) 3(1) (2) 4(1)
This system has the same solutions as the original one as new equations are combinations of orignal ones, and vice versa. Step 2 Eliminate x2 from the third equation using only second equation. x 1 + 2x 2 + 3x 3 = 9 3x2 6x3 = 12 x 3 = 3 5 (3) (2) 3
Step 3 Solve! By third, x3 = 3. By the second, 3x2 = 12 18, so x2 = 2. By the rst, x1 = 4. So the system has one solution (4, 2, 3). For bigger systems, we need better notation. This is provided by matrices. Denition 5.7. A matrix is a rectangular array of numbers. Eg 1 2 3 4 5 6 This is 2 3 matrix. 1 5
This is 3 1 matrix. Call matrix m n if it has m rows, n collumns. We use matrices to encapsulate systems of linear equations. System from previous example has the coecients matrix 1 2 3 4 5 6 3 1 2 and augmented matrix 1 2 3 9 4 5 6 24 3 1 2 4
coe matrix =
a11 am 1 a11 am 1
a 1n amn
b1
a 1n
augmented matrix =
amn bm
A row of augmented matrix corresponds to an equation in the system. The number in the i -th row and j -th collumn of a matrix is called the ij -entry. Operation on the equations in a system are operations matrix. 1 2 3 9 1 2 4 5 6 24 0 3 3 1 2 4 0 5 1 2 0 3 0 0 on the rows of the augmented 3 9 6 12 11 23 3 9 6 12 1 3
Denition 5.8. Elementary row operations are the following operations on the rows of an (augmented) matrix 1) Add a scalar multiple of one row to another r1 ri + rj , R 2) Swap two rows ri rj 3) Multiply any row by a nonzero scalar ri ri , = 0 Doing these elementary row operations to an augmented matrix does not change the solution of the system. Idea of Gaussian elimination We start with a hard system. Then we do a couple of row operations and get an easy system. What makes it easy are zeros under the main diagonal. Denition 5.9. An m n matrix is in echelon form if (i) The rst non-zero number in each row occurs to the right of the rst non-zero number in any higher row (ii) All rows (0, 0, . . . , 0) appear at the bottom.
5.1. METHOD GAUSSIAN ELIMINATION Example 5.7. The following matrices 1 2 3 9 0 3 6 12 0 0 3 9 0 0 1 0 0 0 are in echelon form. This one is not 0 1 3 1 0 0 0 0 0
29
The point If a system has its augmented matrix in echelon form, the system is easy to solve. Example 5.8. Solve the system of equations with augmentated matrix 2 1 3 0 0 1 1 1 0 0 0 0 This system is 2x 1 x 2 + 3x 3 = 0 x2 + x3 = 1
Solve from bottom up. Equation 3 tells us nothing. Let x3 = a (any a R. Then eq 2 implies that x2 is 1 a. 4a Equation 1 implies 2x1 = x2 3x2 , therefore x1 = 1 2 . 1 Therefore the solutions are (x1 , x2 , x3 ) = ( 2 (1 4a), 1 a, a). E.g when a = 0, we 1 , 1, 0 . get solution 2 Example 5.9. Solve the system with augmentated matrix 2 1 3 0 0 1 1 0 0 0 0 2 System thus is 2x 1 x 2 + 3x 3 = 0 x2 + x3 = 1 0 = 2 Third equation instantly implies no solution at all.
30 Example 5.10.
The system is
1 1 1 3 2 4 0 0 0 0 2 2 0 3 0 0 0 0 1 1 0 x 1 + x 2 + x 3 + 3x 4 + 2x 5 + 4x 6 = 0 2x 4 x 5 = 3 x5 + x6 = 0
1 (x5 + 3) = Equation 3 sets x6 = a (any a). Then x5 = a. From 2 we get x4 = 2 1 2 2 (3 a ). From 1 we get x3 = b and x2 = c . Then x1 = c b 3 (3 a ) + 2a 4a 1 9 = 2 2 a b c . The general solution is
(x1 , x2 , x3 , x4 , x5 , x6 ) = ( 9 2 for any a, b, c R. In general, if augmentated matrix is in echelon form, solve the system by solving the last equation, then the next last, and so on. This method is called back substitution. The variables we can put free equal to a, d, c etc are free variables. E.g. in previous example, the free variables are x6 , x3 , x2 . Theorem 5.1. (Gausian elimination theorem) Any matrix can be reduced by elementary row operations to a matrix which is in echelon form. Method 5.1. (Gausian elimination method) System of linear equations (augmented matrix). We put the augmented to echelon form using elementary row operations. Solve the new system by back substitution. Example 5.11. Solve the system with augmentation matrix 1 0 3 2 5 1 1 0 3 5 8 1 A= 1 1 4 5 7 0 1 1 2 1 3 3 Answer.
1 2a
b c, c, b, 1 2 (3 a ), a, a )
Step 1 clear rst collumn using top left hand entry (i.e. equations 2, 3, 4 take away equation 1) 1 0 3 2 5 1 0 0 0 0 3 3 A 0 1 1 3 2 1 0 1 1 3 2 1
5.1. METHOD GAUSSIAN ELIMINATION Step 2 swap rows 2 and 4 1 0 3 2 5 1 0 1 1 3 2 1 0 1 1 3 2 1 0 0 0 3 3 0 Step 3 clear second collumn using the row 2 1 0 3 2 5 0 1 1 3 2 0 0 0 0 0 0 0 0 3 3 Step 4 1 0 3 2 5 0 1 1 3 2 0 0 0 3 3 0 0 0 0 0
31
1 1 0 0 1 1 0 0
Matrix Algebra
Denition 5.10. Matrix multiplication Dot product in Rn y1 x1 . . x =. . . , y =. yn xn x .y = x1 y1 + + xn y n It is convenient given a row vector z = (za , . . . , zn ) to also dene z .y = z1 y1 + +zn yn . Then one can dene matrix multiplication by a1 A = . . . , B = b1 . . . bn an c11 . . . c1p ... AB = . . . cm1 . . . cmp
where cij = ai bj
32
3) Matrix multiplication, like for on R, is associative. Theorem 5.2. Given any matrices A m n, B n p , C p q (AB )C = A(BC ) Example 5.13. A= Then AB = BC = 2 5 4 11 3 0 5 2 (AB )C = , A(BC ) = 13 4 29 8 13 4 29 8 1 2 , B= 3 4 0 1 , C= 1 2 1 2 3 0
Proof. First, we do special case, when C is p 1 matrix, i.e. column vector. Let A is m n, B is n p , x is p 1. We show that (AB )x = A(Bx ) Let y = Bx be a vector and z = Ay . So z is the right hand side. Claim is z = (AB )x . y = B x (n 1) (n p ) (p 1) x1 y1 b11 . . . b1p . . . . . y =. . .= . . . . xp yn bn1 . . . bnp
33
bmn
For i i n zi = ai 1 y1 + + ain yn , yj = bji x1 + + bjp xp . Then zi = ai 1 (b1i x1 + + b1p ) + ai 2 (b2i x1 + + b2p )+ ... +ain (bni x1 + + bnp ) = (ai 1 b11 + ai 2 b21 + + ain bn1 )x1 + (ai 2 b12 + ai 2 b22 + + ain bn2 )x2 + + (ai 2 b1p + ai 2 b2p + + ain bnp )xp Thus coecient of xi in zi is (ai 1 bij + ai 2 b2j + + ain bnj ) which is ai .bj , where ai 1 b1 j . . ai = . . , bj = . . ain bnj
which by denition is the (i, j ) entry of AB ! Claim (AB )C = A(BC ) . . . . . . cq Write C = c1 Then . . . . . . BC = Bc1 . . . Therefore A(BC ) = A(Bc1 ) . . .
Bcq
A(Bcq )
4) Again, as for + and on R, distributivity holds Proposition 5.3. Given matrices A m n, B , C , both n p A(B + C ) = AB + AC
34
Powers of matrices
Denition 5.11. A square matrix is one which is n n for some n. If A is n n, dene A2 A
3 4
= = = ... =
A A Example 5.14. 1 2 0 1
2
1 2 0 2
1 2 0 2
1 4 0 1
n n In
0 ... 1 ... . 0 .. 0 0
Proposition 5.4. If A is m n and B is n p then AIn = A In B = B Link between addition and multiplication Proposition 5.5. (i) If A is m n and B , C are n p , then
A(B + C ) = AB + AC (ii) If D , E are m n and F is n p , then (D + E )F = DF + EF Proof. (i) Let A = (aij ), B = (bij ), C = (cij ). The ij -entry of AB is bij . (aj 1 . . . ain ) . . = ai 1 bij + + ain bnj bnj
5.1. METHOD GAUSSIAN ELIMINATION So ij -entry of A(B + C ) is ai 1 (b1j + c1j ) + + ain (bnj + cnj ) = ai 1 b1j + ai 1 c1j + + ain bnj + ain cnj = (ai 1 b1j + + ain bnj ) + (ai 1 c1j + + ain cnj )
35
36
37
Chapter 6
where A = aij
Proof. Suppose the number of solutions is not 0 or 1, i.e. there are at least two solutions. We prove that this implies that there are innitely many solutions. Let p and q be two dierent solutions (p, q Rn ), so Ap = b Aq = b and p = q . Then A(p q ) = Ap Aq = 0 For any scalar R A(p q ) = A(p q ) = 0 So A(p + (p q )) = Ap + A(p q ) = Ap + 0 = b So p + (p q ) is a solution of the original system. Since p = q , p q = 0 and so each dierent scalar gives a dierent solution. So there are solutions.
38
Structure of solutions
Proposition 6.2. (i) System Ax = 0 either has one solution (which must be x = O) or it has an innite number of solutions. (ii) Suppose p is a solution of a system Ax = b (i.e. p Rn , Ap = b). Then all solutions of Ax = b take the form p+h where h is a solution of the system Ax = 0. x 1 0 3 1 x2 = 0 1 1 x3 2 . The general solution is 1
x = (2 3a, 1 a, a) Particular solution p = (2, 1, 0). So general solution is (2 3a, 1 a, a) = (2, 1, 0) + (3a, a, a) = p + h Where h = (3a, a, a) is the general solution of Proof. (i) clear from previous proposition (ii) Let q be a solution of A. So Ap = b, Aq = b. So A(q p ) = 0. Put q p = h, a solution to Ax = 0. Then q = p + h. 1 0 3 x= 0 1 1 0 . 0
6.2
Population Distribution
Example 6.2. Population classied in 3 income states (1) Poor (P) (2) Middle incom (M) (3) Rich(R) Over one generation (20 year period) P 20%M, 5%R rest stay P
6.2. POPULATION DISTRIBUTION Summarize this information in a matrix. Its 3 3 matrix T , T = (tij ), where tij = proportion moving from state j state i So 0.75 0.25 0.5 T = 0.20 0.65 0.30 0.05 0.10 0.65
39
This is called the transition matrix. All entries are 0, and column sums are 1. Say we start with proportions p1 , p2 , p3 in states 1, 2, 3. p1 p (0) = p2 p3 is the initial population vector. After 1 generation
Continuiing, see that after n generations, population vector is p (n) = T n p (0) This is an example of a Markov chain a population is divided into states 1, . . . , n, and were given the proportion tij moving from state j to state i over a generation. The transition matrix T = (tij ) is n n with properties 1) tij 0 2) all column sums are 1 Markov chain is regular if some power of T has no zero entries. Example 6.3. The above example is regular. T = is not regular. 0 1 1 0
40
Basic Fact m (0) In a regular Markov chain, as n grows, the vector T p gets closer and closer to a s1 s2 steady state vector s = . where . . sn 1) s1 + + sn = 1 2) T s = s This is true, whatever the initial population vector p (0) . Proof is not hard. Example 6.4. On a desert island, a veg-prone community dies according to (1) no-one eats meat 2 days in a row (2) if someone doesnt eat meat one day, they toss a coin: heads eat meat next day, tails dont What proportion can be expected to eat meat on a given day? Answer Markov chain. State 1: meat State 2: no meat generation: 1 day Transition matrix T =
1 2 1 2 1 4 3 4
0 1
1 2 1 2
Notice T 2 = vector s = So
s1 , where s1 + s2 = 1 and T s = s . s2 1 s2 = s1 2 1 s1 + s2 = s2 2 s1 + s2 = 1
So in long run,
1 3
41
6.3
6.3.1
Consider a rotation about the origin through angle . Then y1 = r cos( + ) = x1 cos x2 sin = r (cos cos sin sin )
y2 = r sin( + )
= r (sin cos + cos sin ) = x1 sin + x2 cos So y1 y2 = cos sin sin cos x1 x2
Call R =
6.3.2
Reection
x1 x2 x1 x2 1 0 0 1 x1 x2
6.3.3
Combining Transformations
Reection+Rotation
Now consider sr , sending x s (r (x )). This sends x S (R x ) = = = (SR )x 1 0 0 1
42
43
Chapter 7
Inverses
Denition 7.1. Let A be a square matrix (n n). Say another n n matrix B is an inverse of A i AB = BA = In . Inverse of A is denoted A1 . If A has an inverse, A is invertible. Example 7.1. A = 1 1 0 1 . Then 1 1 0 1 1 1 0 1 1 0 0 1 1 0 0 1
= =
a b c d
a + 2c b + 2d 0 0
Proposition 7.1. If A is invertible then its inverse is unique. Proof. Suppose B , C are inverses of A. Then AB = BA = I AC = CA = I B = BI = B (AC ) = (BA)C = IC = C
7.1
Proposition 7.2. Suppose A is invertible. Then any system Ax = b has a unique solution x = A1 b
44 Proof. Ax A
1
= = = =
b A1 b A1 b A1 b
(Ax ) x
(A1 A)x
By previous proposition x= =
7.2
Finding inverses
a b . Observe c d a b c d d b c a = ad bc 0 0 ad bc = (ad bc )I
2 2 matrices
Let A =
Using this, we can prove the following Proposition 7.3. 1) If ad bc = 0 then A is invertible and A1 = 1 ad bc d c b a
A1 (AB ) = A1 0 = 0 A1 (AB ) = (A1 A)B = IB Therefore IB = 0 and thus A = B = 0. But zero matrix is not invertible.
45
7.3
Let A = (aij ) is n n matrix. We want n n matrix X = xij such that AX = I . To solve, we write one giant matrix a 1n 1 . . . 0 . . .. . . . . . . . . = (A|I ) ann 0 . . . 1
a11 . . . . . . an 1 . . .
Now use Gaussian elimination to reduce to Gaussian form. There are two possibilities
1) There is a row with zeros in the left side and non zero elements in the right side. There is no inverse in this case.
2) We dont get into the rst situation. Therefore we can put the left matrix into Echelon form. Then, we can get rid of non-zero elements above the main diagonal of the left matrix and get a matrix
(I |E ) So AE = I . Start with (E |I ) and reverse everythung, we end up with (I |A). So solving EX = I gives X = A, completing proof that E = A1 .
Proposition 7.4. 1) If we can reduce (A|I ) to (I |E ) using elementary row operations, then E = A1 . 2) If we can reduce (A|I ) to matrix with zeros in the left side of the last line and a non-zero element in the right side of the last line, then A is not invertible.
46 Augmented matrix is
A result linking inverses, linear equations and echelon forms Proposition 7.5. Let A be a square matrix n n. The following four statements are equivalent (1) A is invertible (2) Any system Ax = b has a unique solution (3) The system Ax = 0 has the unique solution x = 0 (4) A can be reduced to the identity In using ERO. Proof. We prove (1) (2), (2) (3), (3) (4), (4) (1). (1) (2) Is done, x = A1 b (2) (3) Is obvious (3) (4) Needs to be proved. (4) (1) Is proved earlier.
47
Chapter 8
Determinants
Recall that matrix a b c d is invertible i ad bc = 0.
a b c d
or |A|. Example 8.1. 1 2 = 2 3 4 |R | = Recall Proposition 8.1. For a 2 2 matrix A |A| = 0 A is invertible Ax = b has unique solution. Aim is to dene det(A) for 3 3 and larger matrices in such a way that this result is still true. Denition 8.2. For a 3 3 matrix cos sin =1 sin cos
a a a a a22 a23 a12 21 23 + a13 21 22 a31 a32 a31 a33 a32 a33 = a11 (a22 a33 a23 a32 ) a12 (a21 a33 a23 a31 ) + a13 (a21 a32 a22 a31 )
48 Example 8.2. 1 2 3 4 5 6 1 1 2 = 1 4 5 4 6 5 6 +3 2 1 1 1 2 1 2
8.1. PROPERTIES
8.1
Properties
Denition 8.3. If A is a 3 3 matrix, the ij -minor of A is the 2 2 matrix Aij obtained by deleting i th row and j th collumn of A. Then |A| = a11 |A11 | a12 |A12 | + a13 |A13 | This is called the expansion of |A| by the rst row. Example 8.4. 1 2 3 A = 4 5 6 , A11 = 1 1 2 5 6 1 2
Proposition 8.2. Expansion by the second row |A| = a21 |A21 | + a22 |A22 | a23 |A23 | third row |A| = a31 |A31 | a32 |A32 | + a33 |A33 | Proof. The second row. RHS = a21 (a12 a33 a13 a32 ) + a22 (a11 a33 a13 a31 ) a23 (a11 a32 a12 a31 ) = |A| Check third row. Proposition 8.3. If A (3 3) has two equal rows, then |A| = 0. Proof. WLOG1 say that a1 a2 a3 A = a1 a2 a3 c1 c2 c3
49
8.2
2) Swapping two rows changes |A| to |A|. 3) Row operation ri ri changes |A| to |A| Proof. 1) Say i = 2, j = 1, so the row. op. sends a1 a2 a3 a1 a2 a3 A = b1 b2 b3 A = b1 + a1 b2 + a2 b3 + a3 c1 c2 c3 c1 c2 c3 Expand by the second row |A | = (b1 + a1 ) a a a a a2 a3 + (b2 + a2 ) 1 3 (b3 + a3 ) 1 2 c1 c2 c1 c3 c2 c3
= a1
= = 9
50 Example 8.6.
1 x 1 y 1 z
x2 y2 z2
1 x x2 = (y x )(z x ) 0 1 y + x 0 1 z +x 1 x x2 = (y x )(z x ) 0 1 y + x 0 0 z y = (y x )(z x )(z y ) This is the 3 3 Vandermande determinant. Proposition 8.5. Let A be 3 3 matrix, and let A be obtained from A by el. row. ops. Then |A| = 0 |A | = 0 Proof. Doing a row op. changes |A| to |A|, -|A| or |A| (where = 0). Main result Theorem 8.6. Let A be 3 3 matrix. Then |A| = 0 A is invertible (or the system Ax = 0 has unique solution, or A I3 by row ops.) Proof. Suppose |A| = 0. Reduce A to echelon form A by row operations. By 8.5, 1 |A | = 0. If A has a zero row, |A| = 0. Hence A = 0 1 so can be 0 0 1 reduced to I . So A is invertible. Suppose A is invertible. A can be reduced to I by row ops. Since |I | = 1, so |A| = 0. Example 8.7. For which values of a is A invertible? 1 2 5 A = 1 3 7 1 4 a |A| = So by 8.6, A is invertible i a = 9. 1 2 5 0 1 2 =a9 0 2 a5
1 x 0 y x 0 z x
x2 y2 x2 z2 x2
51
Corollary 8.7. Let A be 3 3 matrix. Suppose the system Ax = 0 has a non-zero solution x = O. Then |A| = 0. 1 2 5 Example 8.8. Let A = 1 3 7. System Ax = 0 has augmented matrix 1 4 a 1 2 5 0 1 2 5 1 3 7 0 0 1 2 1 4 a 0 0 2 a5 1 2 5 0 1 2 0 0 a9 0 0 0 0 0 0
8.3
n n determinants
It is possible to dene det(A) for any n n matrix A, and to prove that all the results of this section are true for n n matrices. Wait for proofs until next year. Denition 8.4. 4 4 determinant. If A = (aij ) is 4 4, |A| = a11 |A11 | a12 |A12 | + a13 |A13 | a14 |A14 | Similarly, dene n n determinant in terms of (n 1) (n 1) determinants (recursive denition). Example 8.9. 1 2 1 0 2 1 0 1 3 1 0 1 4 5 2 3 0 1 5 1 0 2 1 1 3 1 1 5 2 0 2 0 1 3
+ 3 + ...
52
8.3. N N DETERMINANTS
53
Chapter 9
and so on. = An F1 F0
9.1
Eigenvectors
v1 . Denition 9.1. Let A be a n n matrix. Vector v = . . R is an eigenvector of vn A if (1) v = 0 (2) Av = v ( R) The scalar is an eigenvalue of A. Example 9.1. 3 2 2 0 So 1 2 1 2 = 1 2 = 1 1 2
is an eigenvector of A =
3 2 2 0
54 Example 9.2.
3 2 2 0 So 1 1 is not an eigenvector of
1 1
5 2
1 1
3 2 . 2 0
9.2
Let A be a n n matrix. Vector x is a non-zero solution of the system Ax = x Example 9.3. 3 2 2 0 We can get (3 )x1 + 2x2 = 0 So the equation is 3 2 =0 2 2 3 4 = 0 Eigenvalues are 1 and 4. For = 1, eigenvectors are non-zero solution of A + Ix 4 2 x 2 1 So eigenvectors are For = 4 a 2a (a R, a = 0) 1 2 2 4 So eigenvectors are 2b (b R,b = 0). b x =0 = 0 = 0 x1 x2 = x1 x2
2x1 x2 = 0
55
(1) If A is a 3 3 or 2 2, then the eigenvalues of A are the solutions of |A I | = 0. (2) If is an eigenvalue, then the corresponding eigenvectors are the non-zero solutions of (A I )x = 0 Denition 9.2. The equation |A I | = 0 is the characteristic equation of A, and |A I | is the characteristic polynomial of A.
9.2.1
We had
Back to Fibonacci
Fn+1 Fn = An 1 0 where A = 1 1 . 1 0
Strategy (1) Find eigenvalues 1 , 2 and eigenvectors v1 and v2 of A. Observe Av1 = 1 v1 Av2 = 2 v2 Then A2 v1 = A(Av1 ) = A(1 v1 ) = 1 Av1 = 2 1 v1 Similarly An v1 = n 1 v1 An v2 = n 2 v2 (2) Express 1 0 as a combination of v1 and v2 . 1 0 Then An 1 0 1 0 = An (v1 + v2 ) = An v1 + An v2 An
n = n 1 v1 + 2 v2
= v1 + v2 (, R)
(9.1)
9.3. DIAGONALIZATION
Similarily v2 =
(2) We now nd and such that 1 0 = = = Putting all this into (9.1) Fn+1 Fn = An 1 0 n = 1 v1 + n 2 v2 1 1 = 1 5 5 2 1 2 + 1 1 1 1 2 1 2 1
1 1
1 5
1 1 5 2
2 1
To get formula 1 Fn = 5 1+ 5 2
n
1 5 2
9.3
Diagonalization
1
9.3. DIAGONALIZATION Denition 9.3. An n n matrix D is diagonal matrix if 1 O .. D = . O n Example 9.4. 1 0 0 2 Its easy to nd powers of diagonal matrices. Proposition 9.2. Let 1 .. O 1 O . O
57
D = E =
Then
n O .. . n O .. .
and
1 1 DE = O Dk = k 1 O
n n O .. . k n
Proof. DE given by denition of matrix multiplication. Take E = D to get D 2 and repeat to get D k . Aim - To relate an arbitrary square matrix A to a diagonal matrix, and exploit this to nd An , etc. 22 Let A be 2 2 and suppose A has eigenvalues 1 , 2 with eigenvectors v1 , v2 . Assume 1 = 2 . We get Av1 = 1 v1 Av2 = 2 v2 Cleverly dene 2 2 matrix P (v1 , v2 are collumn vectors) P = v1 v2
9.3. DIAGONALIZATION
So if we write D =
We claim that P is invertible. For if not, then |P | = 0, which means that v1 = v2 , which is false as Av1 = 1 v1 = 1 v2 Av2 = 2 v2 and these are not equal as 1 = 2 . So from (9.2) P 1 AP = P 1 P D = D Summary Proposition 9.3. Let A be 2 2 with distinct eigenvalues 1 , 2 , eigenvectors v1 , v2 . Let P = v1 v2 Then P is invertible and P 1 AP = D = Note. Also true for 3 3, . . . matrices. Example 9.5. Let A = 0 1 . 2 3 1 0 0 2
1) Find P such that P 1 AP is diagonal. 2) Find the formula An . 3) Find a 5th root of A, i. e. nd B such that B5 = A Answer.
59
( 1)( 2) = 0 So eigenvectors are =1 a =2 b Let P = 1 , (a = 0). 1 1 , (b = 0) 2 1 1 . Then by Proposition 9.3, 1 2 P 1 AP = D = Note that many other P s work, e. g. Q1 AQ = 1 0 0 2
2 1 2 2 2 0 0 1
. Or if Q =
1 1 , then 2 1
1 0 0 2n
= P 1 AP P 1 AP . . . P 1 AP P 1 AP
n
= P 1 An P So P 1 An P = Dn P P 1 An P P 1 = P D n P 1 An = P D n P 1 So An = 1 1 1 2 1 0 0 2n 2 1 1 1 = 1 2n 1 2n+1 2 1 1 1
9.3. DIAGONALIZATION
(P CP 1 )5 = P CP 1 . . . P CP 1 = P C 5 P 1 = P DP 1 = A So take B = P CP 1 = = 2 25 6 2 25
1
1 1 1 2 25 1 6 25 1
1
1 0 1 0 25
2 1 1 1
Note. Usually a matrix has many square roots, fth roots, etc. Eg, I has innitely many. Note. Similarly can calculate polynomial functions p (A) = an An + an1 An1 + + a1 A + a0 I
Summary
If a square matrix A has distinct eigenvalues, then it can be diagonalized, i. e. there exists an invertible P such that P 1 AP is diagonal.
9.3.1
Repeated eigenvalues
If the characteristic polynomial of A has a repeated root , we call a repeated eigenvalue of A. Some As with the repeated eigenvalue can be diagonalised and some cant. Example 9.6. Let A = 1 1 . Then the characteristic polynomial is 0 1 1 1 = (1 )2 0 1 so 1 is a repeated eigenvalue. Claim A cannot be diagonalized (i.e. no invertible P exists such that P 1 AP is diagonal). Proof. Assume there exists an invertible P such that P 1 AP = D= 1 0 0 2
9.3. DIAGONALIZATION Then AP Writing P A v1 v2 = = = Hence Av1 = 1 v1 , Av2 = 2 v2 . So v1 , v2 are eigenvectors of A. 0 1 0 0 0 0 eigenvectors are a 1 , a = 0. Hence 0 P = But this is not invertible. Contradiction. a b 0 0 v1 v2 v1 v2 D 1 v 1 2 v 2 = PD
61
Point P must have evectors of A as collumns, but A does not have enough independent evectors to make invertible P . 1 0 0 2 0 . Then the characteristic polynomial is Example 9.7. Let A = 1 1 1 1 1 0 0 1 2 0 1 1 1 So 1 is a repeated evalue. Can A be diagonalized? Lets try. Evectors for = 2
= (1 )2 (2 )
0 eigenvectors a 1 , a = 0. 1 For = 1
1 0 0 0 1 0 0 0 1 1 1 0
0 0 0 0 1 1 0 0 1 1 0 0
9.3. DIAGONALIZATION
Summary If A has enough independent vectors for the repeated eigenvalue, A can be diagonalized. If not, it cant.
63
Chapter 10
x1 x2
ax1 + b 2 x2
b 2 x1
+ cx2
x1 x2
e x +f =0
b 2
Digression Transposes
Denition 10.1. If A = (aij ) is m n, the transpose of A is the m n matrix AT = (aji ). 1 2 1 3 5 Example 10.1. A = 3 4, AT = 2 4 6 5 6 x x = 1 , x T = x1 x2 x2 Note. (AT )T x .y = A = x1 x2 y1 y2 = xT y
is symmetric.
64 Proof. The ij -th entry of (AB )T is the ji -th entry of AB , which is the j -th row of A multiplied by i -th column of B . And ij -th entry of B T AT is i -th row of B T multiplied by j -th collumn of AT , i.e. i -th column of B multiplied by j -th row of A. These are equal.
Back to conics
We formulatied the equation of a conic as x T Ax + d where x = Notice x1 ,A= x2 a
b 2 b 2
e x +f =0
(10.1)
c xT
1 0 0 2
2 2 = 1 x1 + 2 x2
So, aim is to nd a rotation matrix P , such that the change of coordinates to y = where x = P y changes the equation (10.1) to (P y )T A(P y ) + d e Py + f = 0 = 0 y T P T AP y + d e y + f and P T AP is diagonal. We can do this. Theorem 10.2. Let A = a
b 2 b 2
y1 , y2
1) A has two real eigenvalues 1 , 2 , with 1 = 2 . 2) If v1 , v2 are eigenvectors of A corresponding to 1 and 2 , then v1 .v2 = 0 3) We can choose unit eigenvectors v1 , v2 , such that P = v1 v2
has determinant 1 and P is a rotation matrix. Moreover, P 1 = P T , and P T AP is diagonal. Proof. 1) Characteristic polynomial of A is a
b 2 b 2
= 2 (a + c ) + ac
b2 4
65
Since b = 0, (a c )2 + b2 > 0, so the roots are real and distinct. Call them 1 and 2 . 2) Let Av1 = 1 v1 Av2 = 2 v2 Consider
T v1 Av2 T ( v ) = v T v . (a) This is v1 2 2 2 1 2
So
T T 2 v1 v2 = 1 v1 v2 T v = 0, i.e. v .v = 0. As 1 = 2 , this forces v1 2 1 2
3) Choose unit v1 , v2 (see picture). Then v1 = v2 = So P = = = R Finally, by 9.3 , P T AP = P 1 AP = 1 0 0 2 v1 v2 cos sin sin cos cos sin sin cos
10.1
Reduction of conics
Start with equation (10.1). 1) Find the evalues and evectors of A 2) Find unit eigenvectors v1 , v2 , such that P = v1 v2 has determinant 1, so is a rotation matrix.
66 y1 y2
3) Change coordinates to y =
y T (P T AP )y + d e y + f = 0 i.e. yT i.e.
2 2 1 y1 + 2 y2 + d y1 + e y2 + f = 0
y1 0 y + d e y + f = 0 0 y2
This is an ellipse if 1 2 > 0, hyperbola if 1 2 < 0, parabola if 1 2 = 0 (or possible degenerate cases). Example 10.3. Reduce conic
2 2 5x 1 + 4x 1 x 2 + 2x 2 =1
to standard form. This is xT 1) Characteristic polynomial 5 2 = 2 7 + 6 = ( 1)( + 6) 2 2 For = 1, eigenvectors are a 1 . 2 2 For = 6, eigenvectors are b . 1 1 1 , 5 1 2 . 1 1 2 2 1
1 , 5
5 2 x 2 2
= 1
Take
sin =
2 . 5
1 (y +2y2 ), 5 1
equation becomes
2 2 y1 + 6y 2 =1
which is an ellipse.
67
10.1.1
Orthogonal matrices
T = R R
Recall, rotation matrix satises So does a reection matrix cos sin sin cos Denition 10.3. A square matrix P is an orthogonal matrix if P T = P 1 , i. e. P P T = I. The key property: Proposition 10.3. Orthogonal matrices preserve lengths, i. e. Pv Proof. Pv
2
= (P v ).(P v ) = (P v )T (P v ) = vT PT Pv = v T Iv = v
2
10.2
Quadric surfaces
(10.2)
Standard examples
1) Ellipsoid
2 2 2 x1 x2 x3 + + =1 a2 b2 c 2
2) Hyperboloid of 1 sheet
2 2 2 x1 x2 x3 + =1 a2 b2 c 2
3) Hyperboloid of 2 sheets
2 2 2 x2 x3 x1 + = 1 a2 b2 c 2
4) Elliptic cone
2 2 x 2 x3 x1 + 2 2 =0 2 2 a b c
8) Parabolic cylinder
2 x1 ax2 = 0
Aim is to nd a rotation and translation, which reduces (10.2) to one of the standard examples. Here is the procedure. 1. Write (10.2) in matrix form x T Ax + g h j x + k where x1 x2 x = x3 a d 2 b A = 2 2
e 2 f 2
= 0
e 2 f 2
Notice that A is symmetric. 2. Find the eigenvalues 1 , 2 , 3 of A (theory these are real). Find corresponding unit eigenvectors v1 , v2 , v3 which are perpendicular to each other (theory this can be done). Choose a directions for the vi so that the matrix P = v1 v2 v3
69
Finally, complete the square to nd translation reducing to a standard equation. Note. All the assumed bits of theory will be covered in Algebra II next year. Example 10.4. Reduce the quadric 2x 1 x 2 + 2x 1 x 3 x 2 1 = 0 to standard form. Answer. 1. Equation is 0 1 1 x T 1 0 0 x 1 0 0 2. Characteristic polynomial is 1 1 1 0 = (2 2) 1 0 = x2 + 1
2, 2.
Let
1 2 1 2 1 2
1 2 1 2 1 2
70
10.3. LINEAR GEOMETRY IN 3 DIMENSIONS Then |P | = 1 and P is a rotation. Change of coordinates x = P y changes equation to
2 2y2
1 1 1 2 2y3 = y1 + y2 + y3 + 1 2 2 2
Finally, complete the square 1 2(y2 )2 2(y3 )2 = (y1 ) 2 This is a hyperbolic paraboloid.
10.3
Denition 10.5. For two vectors x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ) in R3 , dene x + y = (x1 + y1 , x2 + y2 , x3 + y3 ) x = (x1 , x2 , x3 ), R length of x , denoted x =
2 + x2 + x2 x1 2 3 2)
y cos
+ y
2 x
y cos = x .x + y .y (y x )(y x ) = 2x .y
2 x
y cos
10.4
Geometry of planes
Fact there is a unique plane through any 3 points in R3 provided thay arent colinear. We want to describe planes in terms of vectors and equations. A plane in R3 is specied by 1) a point A on the plane 2) a vector n normal to the plane Then x (x a).n = n x .n = a.n. Proposition 10.5. If there are points A, B, C R3 that are not collinear, there exists a unique plane through A, B, C .
10.4. GEOMETRY OF PLANES Proof. Because both A B and B C are both in the plane n.(c a) = 0 i.e. r1 n1 + r2 n2 + r3 n3 = 0 s1 n1 + s2 n2 + s3 n3 = 0 n.(a b) = 0
71
where (r1 , r2 , r3 ) = a b, (s1 , s2 , s3 ) = c a. We want to show that all solutions are {n | R}. Because A, B , C are not colinear, r and s are not parallel. How to nd dist(P, )? Find foot of perpendicular, Q and arbitrary A in the plane. dist(P, Q) = a p cos But (a p ).n = dist(P, Q) = a p n cos (a p ).n n
72 Notes