Merced College; Don Power

 

LINEAR ALGEBRA - CHAPTER 6, LECTURE

 

6.1  Inner Products           Hwk:  (1-4)b, 5a, (6-12)b, 13a, 16b, 17a, 20, 22, 24, 27b, 28abc, 31, 32, 34

 

This generalizes the dot product, now called the "Euclidean inner product"

Any operation which satisfies these properties of the "dot product" is defined as an "inner product":

      1. <u,v> = <v,u> "Symmetry"  (commutative)

      2.  <u+v,w> = <<u,w> + <v,w>  "Additivity"   (distributive)

      3.  <ku,v> = k<u,v>   "Homogeneity"  (scalar mult is associative)

      4.  <v,v> ³ 0, and <v,v> = 0 iff v = 0  "Positivity"

 

The examples in the text are specific inner products you should be familiar with:

      Ex 1.  Euclidean inner product on Rn

      Ex 2.  Weighted Euclidean inner product           Application:  mean of a frequency dist

 

Ex 3:  Norm and distance:  (these yield the ordinary norm and dist in Rn with the Euclidean inner product)

norm = <u,u>^(1/2)

dist(u,v) = norm(u-v)

 

Ex 4:  shows how these change when using a weighted Euclidean inner product

 

Unit circle or unit sphere:

            set of points in V that satisfy norm(u) = 1

 

            Ex 5 shows a "unit circle" with a weighted inner product:  result is an ellipse

For different inner products, distortions occur, but many results of ordinary geometry still apply (such as triangle inequality)

 

More specific inner product spaces:

 

Before Ex 6:  inner product generated by a matrix A:  <u,v> = Au · Av 

      Alternate forms that use only matrix multiplication, not the inner product notation:

            <u,v> = (Av)TAu

                        =  vTATAu

 

      Ex 6:  The identity matrix generates the ordinary Euclidean dot product

 

After Ex 6:  The weighted Euclidean inner product is generated by a diagonal matrix whose entries are the square roots of the weights.

 

      Ex 7:  For any 2´2 matrices, the sum of the product of corresponding elements is an inner product:

            <U,V> = tr(UTV) = tr(VTU) = u1v1 + u2v2 + u3v3 + u4v4

 

      Ex 8:  On P2, <p,q> = sum of products of coefficients of like terms

            Text shows what the norm and the unit circle look like

 

      Ex 9 (calculus)  For f,g in C[a,b], define <f,g> = int(f(x)·g(x),x=a..b)

            Text proves axioms hold

            Ex 10 describes the norm and unit sphere for this space

                  Norm:  int(f2(x),x=a..b)

                  Unit sphere:  the set of all functions for which norm = 1, i.e.  int(f2(x),x=a..b) = 1

            Caution:  the norm in this inner product space is not the same as the arc length

 

Thm 1:  Basic algebraic properties of inner products

      These are extensions of the axioms, and match basic properties of the dot product

 

Ex 11:  Shows how FOIL works for the inner product in any inner product space 

 

6.2  Angle and Orthogonality in Inner Product Spaces        Hwk:  1b, 2, 3, 4, 5b, 6b, 7, 8b, 10b, 13b, 15b, 17, 18b, 19, 21, 23, 28, 30, 34

 

Thm 1:  Cauchy-Schwarz Inequality

      See Proof

 

Thm 2 and 3:  Properties of length and distance.  Notice positivity, homogeneity (for length), symmetry (for distance), and triangle inequality (for both)

 

      Use Cauchy-Schwarz to prove  triangle inequality

 

Thm 4:  Generalized Pythagorean Thm

     

      Proof follows from <u,v> = 0

 

      Ex:  show x and x3 are orthogonal, and demonstrate

            Cauchy-Schwarz

            Triangle Ineq

            Pythagorean Thm

     

Def:  Orthogonal complement:  Let W be subspace of vector space V

      1.  A vector u in V is orthogonal to W if it is orthogonal to every vector in W

      2.  The set of all such vectors in V is the orthogonal complement of W

      3.  We write W┴

 

Thm 5:  Properties of W┴

      1.  It's a subspace

      2.  The only vector common to W and W┴ is 0

      3.  (W┴ )┴ = W

 

Thm 6:  If A is mxn matrix

      1.  Nullspace(A)    rowspace(A) in Rn

      2.  Nullspace(AT)    colspace(A) in Rm

 

Ex 6:  To find a basis for the orthogonal complement of a set of vectors

      1.  Let A be the matrix with the given row vectors

      2.   Find the basis for the nullspace (solve Ax=0)

      3.   Express the result as column vectors.

 

Thm 7 adds Thm 6 to the "big thm"

 

 

6.3  Orthonormal Bases; Gram-Schmidt Process; QR-Decomposition            Hwk:  1ab, 2ab, 4ab, 5a, 6b, 7b, 10b, 11b, 12, 15b, 17a, 18, 20, 21, 29 ,30     [omit QR decomposition]

 

Contrast thm 1 with thm 5a.  Let W be a subset of vectors in a vector space V. 

 

In thm 1, u is in span(W), so we get coefficients that give u itself

 

In thm 5, u is in general not in span W, so we get the closest approx to u that is possible in span(V), i.e. projwu

 

So, to find vector orthog to u, take v = u – projwu.  Adding v / norm(v) to the set W, we get a new set V that includes the new vector in its span and is still orthonormal.

 

Good example:  X19.  Calculate q1 and q2 by Gram Schmidt.  Then, when you calculate projw2u3, the result is u3 again, showing that u3 is in span(W2).  Hence the orthonormal basis is {q1,q2}

 

6.4  Best Approximation;  Least Squares           Hwk:  1a, 2b, 3b, 4b, 5b, 6, 8b, 10, 13,    (see 14)

      Also do lesson 9.3 #2, 3, 4 (check with calculator LinReg, P2Reg, P3Reg)

 

6.5  Change of Basis        Hwk:  1b, 2b, 3b, 4, 6, 8, 10

 

Change of basis problem:  If we change the basis of a vector space V from an old basis B to a new basis B', how is the old coordinate vector [v]B of a vector v related to the new coordinate vector [v]B'

 

Solution:  Find the coordinate vectors of the new basis vectors in terms of the old basis vectors.  Form a "transition matrix" P whose columns are these coordinate vectors.  Then [v]B = P [v]B'

 

It should be clear that the transition matrix from B' to B will be P–1

 

 

6.6  Orthogonal Matrices Hwk:  1ab, 2ab, 3abcd, 5, 6, 19

 

Be able to prove thms 1 and 2

 

 

Return to:  Merced College; Don Power               Updated 04/25/07 by Don Power