51

Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear Algebra and MATLAB

Pre-Semester Course Mathematics

Christian [email protected]

Page 2: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Organisation

Lectures:

I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF1.28/Shanghai)

I Tuesday 25/09: 10.00 - 12.00 and 13.00 - 15.00 (HoFE.20/DZ Bank)

I Wednesday 26/09: 10.00 - 12.00 and 13.00 - 15.00 (HoFE.01/Deutsche Bank)

I Thursday 27/09: 10.00 - 12.00 and 13.00 - 15.00 (HoFE.01/Deutsche Bank)

I Friday 28/09: 10.00 - 12.00 and 13.00 - 15.00 (HoFE.01/Deutsche Bank)

Tutorials:

I Daily problem sets

I Need not be handed in

Page 3: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Outline

Linear Algebra Overview

MATLAB Introduction

MATLAB Vectors and Matrices

MATLAB Plotting

MATLAB Root �nding and Optimization,Integration and Interpolation,Random numbers,Ordinary di�erential equations

Page 4: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear Algebra

Page 5: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear equations

Matrix notationSystem of linear equations:

a11x1 + ...+ a1nxn = b1

... = ...

am1x1 + ...+ amnxn = bm

Can be represented by: a11 . a1n. . .

am1 . amn

x1.xn

=

b1.bm

Page 6: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Fundamentals

FieldA Field is a set of not fewer than two numbers, which is closed withrespect to the four rational operations of addition, subtraction,multiplication, and division by any non-zero number.

→ e.g. real numbers, complex numbers, rational numbers

VectorLet F be a �eld. A vector x of order n > 0 over F is an ordered set(x1, x2, ..., xn) of n numbers which belong to F .

Page 7: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Multiplication by a scalar and addition

Multiplication of a vector by a scalar

Let x = (x1, x2, ..., xn) be de�ned on �eld F . Then for α ∈ F wehave

αx = (αx1, αx2, ..., αxn)

Vector additionVector addition is de�ned for vectors of the same order by thefollowing formula

(x1, x2, ..., xn) + (y1, y2, ..., yn) = (x1 + y1, x2 + y2, ..., xn + yn)

Page 8: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Relations

Commutative and Associative relationFor x , y and z being vectors the following relations hold:

x + y = y + x

x + (y + z) = (x + y) + z

α(x + y) = αx + αy

(α+ β)x = αx + βx

Page 9: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear independence

Linear independence

The vectors x1, x2, ..., xk are linearly independent if the linearcombination

α1x1 + α2x2 + ...+ αkxk = 0

implies thatα1 = α2 = ... = αk = 0.

Examples

1. Are (1, 2) and (2, 4) linearly independent?

2. Are (2, 1) and (2, 4) linearly independent?

Page 10: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear independence cont.

Example 1

α11+ α22 = 0

α12+ α24 = 0

⇒ α1 = −2α2

⇒ Linear dependence of (1, 2) and (2, 4)

Example 2

α12+ α21 = 0

α12+ α24 = 0

⇒ α1 = α2 = 0⇒ Linear independence of (2, 1) and (2, 4)

Page 11: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Dot product

Dot product

The dot product of two vectors with the same length (x1, x2, ..., xn)and (y1, y2, ..., yn) is de�ned as:

x · y =n∑

i=1

xiyi = x1y1 + x2y2 + ...+ xnyn

Example

Dot product of x = (1, 2,−5) and y = (2,−1,−2):

x · y = x1y1 + x2y2 + x3y3 = 1 · 2− 2 · 1+ 5 · 2 = 10

Page 12: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vector space and Inner product space

Vector space

A vector space of order n over F is a set B of vectors of order nover F with the property that whenever x , y ∈ B , α ∈ F , we have

I x + y ∈ B

I αx ∈ B

I 0 = (0, 0, ..., 0) ∈ B

I −x = (−x1,−x2, ...,−xn) ∈ B

Page 13: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vectorspace and Inner product space cont.

Inner product space

An inner product space is a vector space B over F together with amap 〈·, ·〉 : B × B → F which satis�es the following for x , y , z ∈ Band α ∈ F :

I 〈x , y〉 = 〈y , x〉I 〈αx , y〉 = α 〈x , y〉I 〈x + y , z〉 = 〈x , z〉+ 〈y , z〉I 〈x , x〉 ≥ 0

→ In Euclidean spaces (Rn) inner product is equal to dot product

Page 14: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vector space and Inner product space cont.

Example

u = (u1, u2, ..., un), v = (v1, v2, ..., vn) ∈ Rn and w1,w2, ...,wn arepositive real numbers.Then the weighted Euclidean inner product is given by

〈u, v〉 = w1u1v1 + w2u2v2 + ...+ wnunvn

Check that ful�lls axioms of inner product space:

1.axiom: 〈u, v〉 = 〈v , u〉

〈u, v〉 = w1u1v1 + w2u2v2 + ...+ wnunvn

= w1v1u1 + w2v2u2 + ...+ wnvnun = 〈v , u〉

Page 15: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vector space and Inner product space cont.

Example cont.

2. axiom: 〈cu, v〉 = c〈u, v〉 with c being a scalar.

〈cu, v〉 = w1cu1v1 + w2cu2v2 + ...+ wncunvn

= c(w1u1v1 + w2u2v2 + ...+ wnunvn) = c〈u, v〉

3.axiom: 〈u + v , a〉 = 〈u, a〉+ 〈v , a〉 with a = (a1, a2, ..., an) ∈ Rn

〈u + v , a〉 = w1(u1 + v1)a1 + w2(u2 + v2)a2 + ...+ wn(un + vn)an

= (w1u1a1 + w2u2a2 + ...+ wnunan)

+ (w1v1a1 + w2v2a2 + ...+ wnvnan)

= 〈u, a〉+ 〈v , a〉

4.axiom: 〈u, u〉 ≥ 0

〈u, u〉 = w1u21 + w2u

22 + ...+ wnu

2n ≥ 0

Page 16: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vector space and Inner product space cont.

NormSuppose that V is an inner product space. The norm of a vector uin V is given by

||u|| = 〈u, u〉1/2.

DistanceSuppose that V is an inner product space and that u and v are twovectors in V . The distance between u and v ,denoted by d(u, v) isgiven by

d(u, v) = ||u − v ||.

Page 17: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Vector space and Inner product space cont.

Cauchy-Schwarz Inequality

Suppose u and v are two vectors in an inner product space, then

|〈u, v〉| ≤ ||u|| · ||v ||.

Triangle Inequality

Suppose u and v are two vectors in an inner product space , then

||u + v || ≤ ||u||+ ||v ||.

Page 18: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Orthogonality

Orthogonality

Two vectors of the same length x = (x1, x2, ..., xn) andy = (y1, y2, ..., yn) in an inner product space B are orthogonal iftheir dot product x · y = 0.

Example

Are x = (1, 5,−2) and y = (3,−1,−1) orthogonal?

x · y = x1y1 + x2y2 + x3y3 = 1 · 3− 5 · 1+ 2 · 1 = 0

⇒ x and y are orthogonal

Page 19: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrices

De�nitionA m × n matrix A over a �eld F is a rectangular array of numbersin F consisting of m rows and n columns.

A =

a11 a12 . a1na21 a22 . a2n. . . .

am1 am2 . amn

Page 20: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Special matrices

Square matrix

A square matrix of order n is a n× n matrix. The elements a11, a22,... , ann are called the diagonal elements.

Zero matrixA zero matrix is a m × n matrix where all elements are 0.

Identity matrix

A identity matrix is a n × n matrix where the diagonal elements are1 and the rest 0 (denoted by In).

Page 21: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Special matrices cont.

Diagonal matrix

A diagonal matrix is a n× n matrix where all non-diagonal elementsare 0.

Triangular matrix

I An upper triangular matrix is a n × n matrix where allelements below the diagonal are 0.

I A lower triangular matrix is a n × n matrix where all elementsabove the diagonal are 0.

Idempotent matrix

The matrix A is idempotent if A = A2.

Page 22: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations

SummationIf both A and B are m × n matrices, then (A+ B)ij = Aij + Bij .

Example

A =

[1 34 1

], B =

[3 −1−2 1

][1 34 1

]+

[3 −1−2 1

]=

[1+ 3 3− 14− 2 1+ 1

]=

[4 22 2

]

Page 23: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Multiplication

Let A be a k ×m and B a m × n matrix. Then (AB)ij =m∑l=1

AilBlj .

⇒ A · B 6= B · AFor a square matrix A (n × n) we have A0 = In and Ar = Ar−1A.

Example

A =

[1 34 1

], B =

[3 −1−2 1

][1 34 1

]·[

3 −1−2 1

]=

[1 · 3+ 3 · (−2) 1 · (−1) + 3 · 14 · 3+ 1 · (−2) 4 · (−1) + 1 · 1

]=

[−3 210 −3

]

Page 24: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Transpose

The transpose A′ of a matrix A is a matrix obtained from A byinterchanging rows and columns.

Example

A =

[1 3 24 −1 2

]⇒

A′ =

1 43 −12 2

Page 25: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Epsilon sign

For x = (x1, x2, ..., xn) we have ε(x) = sgn∏

1≤r<s≤n(xs − xr ).

Determinant of a matrixThe determinant of a n × n matrix A = [aij ] is the number∑

(λ1,λ2,...,λn)

ε(λ1, λ2, ..., λn)a1λ1a2λ2 ...anλn

where the summation is over all the n! arrangements(λ1, λ2, ..., λn) of (1, 2, ..., n). This determinant is denoted by

|A| =

∣∣∣∣∣∣∣∣a11 a12 . a1na21 a22 . a2n. . . .

an1 an2 . ann

∣∣∣∣∣∣∣∣or det(A) or |aij |n.

Page 26: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example

A =

2 4 51 2 32 1 3

, x = (1, 2, 3)

Compute ε(x) for all arrangements:

x = (1, 2, 3) : ε(x) = sgn((2− 1)(3− 1)(3− 2)) = 1

x = (1, 3, 2) : ε(x) = sgn((3− 1)(2− 1)(2− 3)) = −1x = (2, 1, 3) : ε(x) = sgn((1− 2)(3− 1)(3− 2)) = −1x = (2, 3, 1) : ε(x) = sgn((3− 2)(1− 3)(1− 2)) = 1

x = (3, 1, 2) : ε(x) = sgn((2− 1)(1− 3)(2− 3)) = 1

x = (3, 2, 1) : ε(x) = sgn((2− 3)(1− 3)(1− 2)) = −1

Page 27: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example cont.

det(A) = 1 · 2 · 2 · 3+ (−1) · 2 · 3 · 1+ (−1) · 4 · 1 · 3+ 1 · 4 · 3 · 2+ 1 · 5 · 1 · 1+ (−1) · 5 · 2 · 2 = 3

Page 28: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Determinant of a matrix cont.Further, if (µ1, µ2, ..., µn) is any �xed arrangement of (1, 2, ..., n),then the determinant can be expressed as∑

(λ1,λ2,...,λn)

ε

(λ1, λ2, ..., λnµ1, µ2, ..., µn

)aλ1µ1aλ2µ2 ...aλnµn

with ε

(λ1, λ2, ..., λnµ1, µ2, ..., µn

)= ε(λ1, λ2, ..., λn) · ε(µ1, µ2, ..., µn).

Page 29: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Determinant - Special cases

2× 2 matrix - Example

|A| =∣∣∣∣ 4 2−3 1

∣∣∣∣ = 4 · 1− 2 · (−3) = 10

3× 3 matrix - Example

|A| =

∣∣∣∣∣∣4 2 −12 3 15 −2 −1

∣∣∣∣∣∣= 4 · 3 · (−1) + 2 · 1 · 5+ 2 · (−2) · (−1)− 5 · 3 · (−1)− 2 · 2 · (−1)− 4 · (−2) · 1= 29

Page 30: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Properties of the determinant

I det(In) = 1

I det(AT ) = det(A)

I det(A−1) = 1

det(A)

I det(AB) = det(A) det(B)

I det(cA) = cn det(A) for n × n matrix A

I If A is a n × n triangular matrix then

det(A) = a1,1 · a2,2 · ... · an,n =n∏

i=1

ai ,i

Page 31: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

MinorFor a n × n matrix A, the minor of it's entry aij is denoted by Mij

and is de�ned to be the determinant of the submatrix obtained byremoving from A it's i th row and j th column.

Example

A =

4 2 −12 3 15 −2 −1

M23 =

∣∣∣∣ 4 25 −2

∣∣∣∣ = 4 · (−2)− 2 · 5 = −18

Page 32: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

CofactorFor a n × n matrix A with Mij being the minor of it's entry aij , Cij

is called the cofactor of aij and is de�ned as

Cij = (−1)i+jMij

Example

A =

4 2 −12 3 15 −2 −1

C23 = (−1)(2+3)M23 = (−1)5

∣∣∣∣ 4 25 −2

∣∣∣∣ = (−1)(4·(−2)−2·5) = 18

Page 33: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Cofactors and DeterminantFor a n× n matrix A with Cij being the cofactor of it's entry aij thefollowing holds:

|A| =n∑

i=1

aijCij for any j

=n∑

j=1

aijCij for any i

Page 34: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Cofactors and Determinant

Example

A =

4 2 −12 3 15 −2 −1

C13 = (−1)(1+3)M13 = (−1)4∣∣∣∣ 2 35 −2

∣∣∣∣ = 2 · (−2)− 3 · 5 = −19

C23 = (−1)(2+3)M23 = (−1)5∣∣∣∣ 4 25 −2

∣∣∣∣ = (−1)(4 · (−2)− 2 · 5) = 18

C33 = (−1)(3+3)M33 = (−1)6∣∣∣∣ 4 22 3

∣∣∣∣ = (4 · 3− 2 · 2) = 8

|A| =3∑

i=1

ai3Ci3 = (−1) · (−19) + 1 · 18+ (−1) · 8 = 29

Page 35: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Singularity

A n × n matrix A is said to be singular if |A| = 0.→ Matrix is singular if linearly dependent vectors.

InverseThe inverse of a n × n matrix A is a matrix A−1 such that

A · A−1 = I ,

where I is the identity matrix.

Adjugate

The adjugate matrix A∗ of the n × n matrix A is the transpose ofthe matrix of cofactors of the elements of A.

Page 36: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example adjugate

A =

4 2 −12 3 15 −2 −1

C11 = M11 = −1,C12 = (−1)M12 = 7,C13 = M13 = −19C21 = (−1)M21 = 4,C22 = M22 = 1,C23 = (−1)M23 = 18

C31 = M31 = 5,C32 = (−1)M32 = −6,C33 = M33 = 8

A∗ =

−1 7 −194 1 185 −6 8

′ = −1 4 5

7 1 −6−19 18 8

Page 37: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Methods to compute Inverse

Search inverse of

A =

4 2 −12 3 15 −2 −1

Gauss-Jordan-Algorithm 4 2 −1 1 0 0

2 3 1 0 1 05 −2 −1 0 0 1

1 0 0 − 1

29

4

29

5

29

0 1 0 7

29

1

29− 6

29

0 0 1 −19

29

18

29

8

29

Page 38: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Methods to compute Inverse cont.

With the help of the Adjugate

A−1 =1

|A|A∗

A−1 =1

29

−1 4 57 1 −6−19 18 8

=

− 1

29

4

29

5

297

29

1

29− 6

29

−19

29

18

29

8

29

Special case 2× 2 matrix

A =

[4 2−3 1

]A−1 =

1

|A|A∗ =

1

10

[1 −23 4

]=

[1

10− 2

103

10

4

10

]

Page 39: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Vec operator

Let A be a m × n matrix. Then B = vec(A) is a mn × 1 matrixobtained from stacking the columns of A.

Tracetr(A) is the trace of the n × n matrix A and is the sum of thediagonal elements of A.

Example

A =

4 2 −12 3 15 −2 −1

tr(A) = 4+ 3− 1 = 6

Page 40: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Minor of order kLet A be a m × n matrix. If k ≤ m and l ≤ n, then any k rows andl columns of A determine a k × l submatrix of A. The determinantof a k × k submatrix of A is called a k-rowed minor of A, or aminor of order k .

RankThe rank R(A) or rank(A) of a non-zero matrix A is the maximumvalue of r for which there exists a non-vanishing r -rowed minor ofA. Therefore the rank of A is smaller or equal to the minimum ofthe numbers of rows and columns.

→ Rank (A) is the maximum number of linearly independent rowor column vectors.

Page 41: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Quadratic formA quadratic form on Rn is a real valued functionQ(x1, x2, ..., xn) =

∑i≤j

aijxixj which can be represented using

matrices as Q(x) = x ′Ax with

A =

a11

1

2a12 . 1

2a1n

1

2a12 a22 . 1

2a2n

. . . .1

2a1n

1

2a2n . ann

→ Example for matrix A: Variance-Covariance-Matrix

Page 42: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Positive / negative (semi-)de�nite

For any nonzero (n × 1) vector x a n × n matrix A is said to be

I positive de�nite if x ′Ax > 0.

I positive semi-de�nite if x ′Ax ≥ 0.

I negative de�nite if x ′Ax < 0.

I negative semi-de�nite if x ′Ax ≤ 0.

Page 43: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example

A =

4 2 −12 3 15 −2 −1

, x =

121

x ′Ax = [ 1 2 1 ]

4 2 −12 3 15 −2 −1

121

= [ 13 5 0 ]

121

= 23

⇒ A is positive (semi-)de�nite.

Page 44: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Eigenvalue and Eigenvector

For the n × n matrix A a non-zero n × 1 vector v is an eigenvectorof A if there exists a scalar α such that

Av = αv .

Computation

The eigenvalues of A are precisely the solutions α to the equation

det(A− αI ) = 0.

Page 45: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example

Compute eigenvalues and eigenvectors for A =

[4 2−3 1

].

Eigenvalues:

det(A− αI ) =∣∣∣∣ 4− α 2

3 1− α

∣∣∣∣ = 4− 5α+ α2 − 6 = 0

α1/2 =5

2±√

25

4+ 2

α1 = 5.3723, α2 = −0.3723

Page 46: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Matrix operations cont.

Example cont.

Eigenvectors:

α = 5.3723:[4 23 1

] [v11

v12

]= 5.3723

[v11

v12

]⇒ v1 =

[0.82460.5658

]α = −0.3723:[

4 23 1

] [v21

v22

]= −0.3723

[v21

v22

]⇒ v2 =

[−0.41600.9094

]

Page 47: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear equations

Matrix of Coe�cients and augmented matrix

In the system of linear equations

a11x1 + ...+ a1nxn = b1

... = ...

am1x1 + ...+ amnxn = bm

the matrix A =

a11 . a1n. . .

am1 . amn

is called the matrix of coe�cients

and B =

a11 . a1n b1. . . .

am1 . amn bm

is called the augmented matrix.

Page 48: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear equations cont.

Theorem 1: Consistency

A necessary and su�cient condition for a system of linear equationsto be consistent is that the matrix of coe�cients has the same rankas the augmented matrix.

Theorem 2: SolutionsLet A be the matrix of coe�cients and B the augmented matrix.Then the associated system of linear equations possesses an in�nitenumber of solutions if and only if rank(A) = rank(B) < n. Itpossesses a unique solution if and only if rank(A) = rank(B) = n.It possesses no solution if and only if rank(A) < rank(B).

Page 49: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Linear equations cont.

Theorem 3: Nontrivial solution for homogeneous system

The homogeneous system Ax = 0 in n unknowns possesses anontrivial solution if rank(A) < n.

Theorem 4The solution to the homogeneous system Ax = 0 in n unknownsconstitutes a vector space of dimensionality n − rank(A).

Page 50: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Functions

Vector derivativesThe derivative of a m × 1 vector y with respect to a n × 1 vector xis the n ×m matrix

∂y

∂x=

∂y1∂x1

∂y2∂x1

. ∂ym∂x1

∂y1∂x2

∂y2∂x2

. ∂ym∂x2

. . . .∂y1∂xn

∂y2∂xn

. ∂ym∂xn

Page 51: Linear Algebra and MATLABbadarinza.net/download/christian/slides1.pdf · Organisation Lectures: I Monday 24/09: 10.00 - 12.00 and 13.00 - 15.00 (HoF 1.28/Shanghai) I uesdaTy 25/09:

Functions

Gradient and Hessian matrixLet f : Rn → R then the gradient (denoted by ∇ and is also called'del' or 'nabla') of f at x = (x1, ..., xn)

∇f (x) = ∂f (x)

∂x=

∂f (x)∂x1.

∂f (x)∂xn

and its Hessian is

H(x) =∂2f (x)

∂x ′∂x=

(∂

∂x1∇f (x), ..., ∂

∂xn∇f (x)

)→ The Hessian is a n × n matrix and symmetric.