Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Matrix Calculations: Inner Products & Orthogonality A. Kissinger Institute for Computing and Information Sciences Radboud University Nijmegen Version: spring 2017 A. Kissinger Version: spring 2017 Matrix Calculations 1 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Outline Inner products and orthogonality Orthogonalisation Application: computational linguistics Wrapping up A. Kissinger Version: spring 2017 Matrix Calculations 2 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Length of a vector • Each vector v = ( x 1 , . . . , x n ) ∈ R n has a length (aka. norm), written as � v � • This � v � is a non-negative real number: � v � ∈ R , � v � ≥ 0 • Some special cases: • n = 1: so v ∈ R , with � v � = | v | • n = 2: so v = ( x 1 , x 2 ) ∈ R 2 and with Pythagoras: � � v � 2 = x 2 1 + x 2 x 2 1 + x 2 and thus � v � = 2 2 • n = 3: so v = ( x 1 , x 2 , x 3 ) ∈ R 3 and also with Pythagoras: � � v � 2 = x 2 1 + x 2 2 + x 2 x 2 1 + x 2 2 + x 2 and thus � v � = 3 3 • In general, for v = ( x 1 , . . . , x n ) ∈ R n , � x 2 1 + x 2 2 + · · · + x 2 � v � = n A. Kissinger Version: spring 2017 Matrix Calculations 4 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Distance between points • Assume now we have two vectors v , w ∈ R n , written as: v = ( x 1 , . . . , x n ) w = ( y 1 , . . . , y n ) • What is the distance between the endpoints? • commonly written as d ( v , w ) • again, d ( v , w ) is a non-negative real • For n = 2, � ( x 1 − y 1 ) 2 + ( x 2 − y 2 ) 2 = � v − w � = � w − v � d ( v , w ) = • This will be used also for other n , so: d ( v , w ) = � v − w � A. Kissinger Version: spring 2017 Matrix Calculations 5 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Length is fundamental • Distance can be obtained from length of vectors • Interestingly, also angles can be obtained from length! • Both length of vectors and angles between vectors can de derived from the notion of inner product A. Kissinger Version: spring 2017 Matrix Calculations 6 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Inner product definition Definition For vectors v = ( x 1 , . . . , x n ) , w = ( y 1 , . . . , y n ) ∈ R n define their inner product as the real number: � v , w � = x 1 y 1 + · · · + x n y n � = x i y i 1 ≤ i ≤ n Note : Length � v � can be expressed via inner product: � � v � 2 = x 2 1 + · · · + x 2 n = � v , v � , so � v � = � v , v � . A. Kissinger Version: spring 2017 Matrix Calculations 7 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Inner products via matrix transpose Matrix transposition For an m × n matrix A , the transpose A T is the n × m matrix A obtained by mirroring in the diagonal: T a 11 · · · a 11 · · · a m 1 a 1 n . . . . = . . a m 1 · · · a mn a 1 n · · · a mn In other words, the rows of A become the columns of A T . The inner product of v = ( x 1 , . . . , x n ) , w = ( y 1 , . . . , y n ) ∈ R n is then a matrix product: y 1 . = v T · w . . � v , w � = x 1 y 1 + · · · + x n y n = ( x 1 · · · x n ) · . y n A. Kissinger Version: spring 2017 Matrix Calculations 8 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Properties of the inner product 1 The inner product is symmetric in v and w : � v , w � = � w , v � 2 It is linear in v : � v + v ′ , w � = � v , w � + � v ′ , w � � a v , w � = a � v , w � ...and hence also in w (by symmetry): � v , w + w ′ � = � v , w � + � v , w ′ � � v , a w � = a � v , w � 3 And it is positive definite: v � = 0 = ⇒ � v , v � > 0 A. Kissinger Version: spring 2017 Matrix Calculations 9 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Inner products and angles, part I For v = w = (1 , 0), � v , w � = 1. As we start to rotate w , � v , w � goes down until 0: � v , w � = 4 � v , w � = 3 � v , w � = 1 � v , w � = 0 5 5 ...and then goes to − 1: � v , w � = − 3 � v , w � = − 4 � v , w � = 0 � v , w � = − 1 5 5 ...then down to 0 again, then to 1, then repeats... A. Kissinger Version: spring 2017 Matrix Calculations 10 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Cosine Plotting these numbers vs. the angle between the vectors, we get: It looks like � v , w � depends on the cosine of the angle between v and w . Let’s prove it! A. Kissinger Version: spring 2017 Matrix Calculations 11 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Recall: definition of cosine a y γ x cos( γ ) = x = ⇒ x = a cos( γ ) a A. Kissinger Version: spring 2017 Matrix Calculations 12 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up The cosine rule a b y γ x c cos( γ ) = a 2 + b 2 − c 2 Claim: 2 ab Proof: We have three equations to play with: x 2 + y 2 = a 2 ( c − x ) 2 + y 2 = b 2 x = a cos( γ ) ...lets do the math. � A. Kissinger Version: spring 2017 Matrix Calculations 13 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Inner products and angles, part II Translating this to something about vectors: � v � d ( v , w ) := � v − w � γ � w � gives: cos( γ ) = � v � 2 + � w � 2 − � v − w � 2 2 � v � � w � Let’s clean this up... A. Kissinger Version: spring 2017 Matrix Calculations 14 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Inner products and angles, part II Starting from the cosine rule: cos( γ ) = � v � 2 + � w � 2 − � v − w � 2 2 � v � � w � n − ( x 1 − y 1 ) 2 − · · · − ( x n − y n ) 2 = x 2 1 + · · · + x 2 n + y 2 1 + · · · + y 2 2 � v � � w � = 2 x 1 y 1 + · · · + 2 x n y n 2 � v � � w � = x 1 y 1 + · · · + x n y n � v � � w � � v , w � � v , w � = remember this: cos( γ ) = � v � � w � � v � � w � Thus, angles between vectors are expressible via the inner product � (since � v � = � v , v � ). A. Kissinger Version: spring 2017 Matrix Calculations 15 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Linear algebra in gaming, part I • Linear algebra plays an important role in game visualisation • Here: simple illustration, borrowed from blog.wolfire.com ( More precisely: http://blog.wolfire.com/2009/07/ linear-algebra-for-game-developers-part-2 ) • Recall: cosine cos function is positive on angles between -90 and +90 degrees. A. Kissinger Version: spring 2017 Matrix Calculations 16 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Linear algebra in gaming, part II • Consider a guard G and hiding ninja H in: � 1 � • The guard is at position (1 , 1), facing in direction D = , 1 with a 180 degree field of view • The ninja is at (3 , 0). Is he in sight? A. Kissinger Version: spring 2017 Matrix Calculations 17 / 48
Inner products and orthogonality Orthogonalisation Radboud University Nijmegen Application: computational linguistics Wrapping up Linear algebra in gaming, part III � 2 � 3 � � 1 � � • The vector from G to H is: V = − = 0 1 − 1 • The angle γ between D and V must be between -90 and +90 � D , V � • Hence we must have: cos( γ ) = � D �·� V � ≥ 0 • Since � D � ≥ 0 and � V � ≥ 0, it suffices to have: � D , V � ≥ 0 • Well, � D , V � = 1 · 2 + 1 · − 1 = 1. Hence H is within sight! A. Kissinger Version: spring 2017 Matrix Calculations 18 / 48
Recommend
More recommend