6.3 - 6.3 Orthogonal Projections Review: {u1 , · · ·, un...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 6.3 Orthogonal Projections Review: {u1 , · · ·, un } orthogonal basis for Rn , y = c1 u1 + · · · + cn un , ci = y · ui y ∈ Rn ⇒ ∴ {u1 , · · ·, un } orthonormal basis ⇒ ui · ui y = (y · u1 )u1 + (y · u2 )u2 + · · · + (y · un )un . Theorem: m × n matrix U has orthonormal columns ⇔ UTU = I uT 1 . Why? Because if U = [ u1 | · · · | un ] , U T = . . T un 1 0 ... U T U = uT uj = i 0 1 ⇒ for any x, y ∈ Rn , a. b. c. Ux = x (U x) · (U y ) = x · y Ux · Uy = 0 ⇔ x · y = 0 Theorem: m × n matrix U with orthonormal columns ∴ a. Preserves lengths b. Preserves angles cosθ = 1 Ux Ux · Uy Uy = x x·y y c. Preserves orthogonality Definition: if its columns are orthonormal. n × n matrix U is an orthogonal matrix Question: Which basic matrices in R2 are orthogonal? Sc = 10 0k 10 00 Ro = s c −s c Sh = 2cs 2s2 − 1 11 01 P= Re = 2c2 − 1 2cs EX: Note U −1 = U T ← important √ √ √ 1/ 6 1/ 3 1/ 2 √ √ U = 1/ 3 0 −2/ 6 √ √ √ 1/ 3 −1/ 2 1/ 6 W subspace of Rn with orthogonal basis Definition: {u1 , · · ·, up } as For any y ∈ Rn , define orthogonal projection of y onto W y = ProjW y = c1 u1 + c2 u2 + · · · + cp up ˆ , or ci = y · ui if it is orthonormal basis ci = ui · ui y · ui 2 Orthogonal Decomposition Theorem: y = y + z (or z = y − y ), then z ∈ W ⊥ . ˆ ˆ W sub- space of Rn with orthogonal basis {u1 , · · ·, up } and Proof: z · ui = (y − y ) · ui = y · ui − y · ui ˆ ˆ = y · ui − (c1 u1 + · · · + cp up ) · ui = y · ui − ci ui · ui = 0 for all i 1 1 3/2 3/2 −1/2 1/2 EX: In R2 , W =Span{u1 }, u1 = y= ˆ y · u1 u1 = 3 2 u1 = , y= 1 2 u1 · u1 z =y−y = ˆ 1 2 − 3/2 3/2 y z = y ˆ u1 3 EX: 1 u1 = 1 0 y= ˆ y · u2 y · u1 u1 + u2 u1 · u1 u2 · u2 3 2 = u1 + u2 2 3 3/2 + 2/3 13/6 = 3/2 − 2/3 = 5/6 0 + 2/3 2/3 1 u2 = −1 1 1 y = 2 3 1 13/6 −7/6 z = y − y = 2 − 5/6 = 7/6 ˆ 3 2/3 7/3 So z · u1 = 0 z · u2 = 0 y u2 y2 ˆ y= ˆ y1 ˆ y · u1 u1 + y · u2 u2 = y 1 + y 2 u1 · u1 u2 · u2 u1 Figure 1: The orthogonal projection of y is the sum of its projections onto one-dimensional subspaces that are mutually orthogonal. 4 EX: Above y1 = ˆ 3 2 u1 , y2 = ˆ 2 3 u2 Properties of orthogonal projections 1. Best approximation theorem: If W subspace of Rn , is closest to y in the sense that y−y < y−v ˆ y ∈ Rn , y = ProjW (y ) then y is the vector in W which ˆ ˆ for all other v ∈ W (v = y ). ˆ Question: Why? Answer: y y−y ˆ 0 y ˆ y−v y−v ˆ W v Figure 2: The orthogonal projection of y onto W is the closet point in W to y Take v = y , so y − v ∈ W is orthogonal to y − y . ˆ ˆ ˆ ∴ y−v 2 = y−y ˆ 2 + y−v ˆ 2 > y−y ˆ 2 (Pythagorean Theorem) 2. y ∈ W then y = ProjW y = y itself. ˆ 5 ...
View Full Document

This note was uploaded on 04/09/2011 for the course MATH 232 taught by Professor Russel during the Spring '10 term at Simon Fraser.

Ask a homework question - tutors are online