changing the matrix abx bb the matrix consists of fp

14
Changing the Matrix So far, only discussed changing the RHS, i.e. Ax = b ! Ab x = b b. The matrix consists of FP numbers, too—it, too, is approximate. I.e. Ax = b ! b Ab x = b. What can we say about the error due to an approximate matrix? 58

Upload: others

Post on 14-Nov-2021

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Changing the Matrix Abx bb The matrix consists of FP

Changing the MatrixSo far, only discussed changing the RHS, i.e. Ax = b ! Abx = bb.The matrix consists of FP numbers, too—it, too, is approximate. I.e.

Ax = b ! bAbx = b.

What can we say about the error due to an approximate matrix?

58

Page 2: Changing the Matrix Abx bb The matrix consists of FP

Changing Condition NumbersOnce we have a matrix A in a linear system Ax = b, are we stuck with itscondition number? Or could we improve it?

What is this called as a general concept?

59

Page 3: Changing the Matrix Abx bb The matrix consists of FP

In-Class Activity: Matrix Norms and Conditioning

In-class activity: Matrix Norms and Conditioning

60

Page 4: Changing the Matrix Abx bb The matrix consists of FP

Singular Value Decomposition (SVD)

What is the Singular Value Decomposition of an m ⇥ n matrix?

61

Page 5: Changing the Matrix Abx bb The matrix consists of FP

Computing the 2-Norm

Using the SVD of A, identify the 2-norm.

Express the matrix condition number cond2(A) in terms of the SVD:

62

Page 6: Changing the Matrix Abx bb The matrix consists of FP

Not a matrix norm: FrobeniusThe 2-norm is very costly to compute. Can we make something simpler?

What about its properties?

63

Page 7: Changing the Matrix Abx bb The matrix consists of FP

Frobenius Norm: Properties

Is the Frobenius norm induced by any vector norm?

How does it relate to the SVD?

64

* Errata: the square root was missing from theoriginal scribbles

Page 8: Changing the Matrix Abx bb The matrix consists of FP

Solving Systems: Simple casesSolve Dx = b if D is diagonal. (Computational cost?)

Solve Qx = b if Q is orthogonal. (Computational cost?)

Given SVD A = UΣV T , solve Ax = b. (Computational cost?)

65

Page 9: Changing the Matrix Abx bb The matrix consists of FP

Solving Systems: Triangular matricesSolve

a11 a12 a13 a14a22 a23 a24

a33 a34a44

xyzw

=

b1b2b3b4

.

Demo: Coding back-substitution [cleared]What about non-triangular matrices?

66

Page 10: Changing the Matrix Abx bb The matrix consists of FP

Gaussian Elimination

Demo: Vanilla Gaussian Elimination [cleared]What do we get by doing Gaussian Elimination?

How is that different from being upper triangular?

What if we do not just eliminate downward but also upward?

67

Page 11: Changing the Matrix Abx bb The matrix consists of FP

LU Factorization

What is the LU factorization?

68

Page 12: Changing the Matrix Abx bb The matrix consists of FP

Solving Ax = b

Does LU help solve Ax = b?

69

Page 13: Changing the Matrix Abx bb The matrix consists of FP

Determining an LU factorization

Demo: LU Factorization [cleared]70

Page 14: Changing the Matrix Abx bb The matrix consists of FP

Computational Cost

What is the computational cost of multiplying two n ⇥ n matrices?

I u11 = a11, uT12 = aT

12.I l 21 = a21/u11.I L22U22 = A22 − l 21uT

12.

What is the computational cost of carrying out LU factorization on ann ⇥ n matrix?

Demo: Complexity of Mat-Mat multiplication and LU [cleared]

71