Inner Products and Norms (Part III)
Total Page:16
File Type:pdf, Size:1020Kb
Inner Products and Norms (part III) Prof. Dan A. Simovici UMB 1 / 74 Outline 1 Approximating Subspaces 2 Gram Matrices 3 The Gram-Schmidt Orthogonalization Algorithm 4 QR Decomposition of Matrices 5 Gram-Schmidt Algorithm in R 2 / 74 Approximating Subspaces Definition A subspace T of a inner product linear space is an approximating subspace if for every x 2 L there is a unique element in T that is closest to x. Theorem Let T be a subspace in the inner product space L. If x 2 L and t 2 T , then x − t 2 T ? if and only if t is the unique element of T closest to x. 3 / 74 Approximating Subspaces Proof Suppose that x − t 2 T ?. Then, for any u 2 T we have k x − u k2=k (x − t) + (t − u) k2=k x − t k2 + k t − u k2; by observing that x − t 2 T ? and t − u 2 T and applying Pythagora's 2 2 Theorem to x − t and t − u. Therefore, we have k x − u k >k x − t k , so t is the unique element of T closest to x. 4 / 74 Approximating Subspaces Proof (cont'd) Conversely, suppose that t is the unique element of T closest to x and x − t 62 T ?, that is, there exists u 2 T such that (x − t; u) 6= 0. This implies, of course, that u 6= 0L. We have k x − (t + au) k2=k x − t − au k2=k x − t k2 −2(x − t; au) + jaj2 k u k2 : 2 2 Since k x − (t + au) k >k x − t k (by the definition of t), we have 2 2 1 −2(x − t; au) + jaj k u k > 0 for every a 2 F. For a = kuk2 (x − t; u) we have: 1 1 −2(x − t; (x − t; u)u) + j (x − t; u)j2 k u k2 k u k2 k u k2 1 1 = −2(x − t; (x − t; u)u) + j (x − t; u)j2 k u k2 k u k2 k u k2 j(x − t; u)j2 j(x − t; u)j2 = −2 + k u k2 k u k2 j(x − t; u)j2 = − 0; k u k2 > which is a contradiction. 5 / 74 Approximating Subspaces Theorem A subspace T of an inner product linear space L is an approximating ? subspace L if and only if L = T T . 6 / 74 Approximating Subspaces Proof Let T be an approximating subspace of L and let x 2 L. We have x − t 2 T ?, where t is the element of T that best approximates x. If y = x − t, we can write x uniquely as x = t + y, where t 2 T and ? ? y 2 T , so L = T T . ? Conversely, suppose that L = T T , where T is a subspace of L. Every x 2 L can be uniquely written as x = t + y, where t 2 T and y 2 T ?, so x − t 2 T ?. Thus t is the element in T that is closest to x, so T is an approximating subspace of L. 7 / 74 Approximating Subspaces Theorem Any subspace T of a finite-dimensional inner product linear space L is an approximating subspace of L. 8 / 74 Approximating Subspaces ? Let T be a subspace of L. It suffices to show that L = T T . ? If T = f0Lg, then T = L and the statement is immediate. Therefore, we can assume that T 6= f0Lg. We need to verify only that every x 2 L can be uniquely written as a sum x = t + v, where t 2 T and v 2 T ?. Let t1;:::; tm be an orthonormal basis of T , that is, a basis such that ( 1 if i = j; (ti ; tj ) = 0 otherwise, for 1 6 i; j > m. Define t = (x; t1)t1 + ··· + (x; tm)tm and v = x − t. 9 / 74 Approximating Subspaces Proof (cont'd) The vector v is orthogonal to every vector ti because (v; ti ) = (x − t; ti ) = (x; ti ) − (t; ti ) = 0: Therefore v 2 T ? and x has the necessary decomposition. To prove that the decomposition is unique suppose that x = s + w, where s 2 T and ? w 2 T?. Since s + w = t + v we have s − t = v − w 2 T \ T = f0Lg, which implies s = t and w = v. 10 / 74 Approximating Subspaces Theorem Let T be a subspace of an inner product space L. We have (T ?)? = T . 11 / 74 Approximating Subspaces Observe that T ⊆ (T ?)?. Indeed, if t 2 T , then (t; z) = 0 for every z 2 T ?, so t 2 (T ?)?. To prove the reverse inclusion, let x 2 (T ?)?. We can write x = u + v, where u 2 T and v 2 T ?, so x − u = v 2 T ?. Since T ⊆ (T ?)?, we have u 2 (T ?)?, so x − u 2 (T ?)?. Consequently, x − u 2 T ? \ (T ?)? = f0g, so x = u 2 T . Thus, (T ?)? ⊆ T , which concludes the argument. 12 / 74 Approximating Subspaces Corollary n ? ? Let Z be a subset of R . We have (Z ) = hZi. 13 / 74 Approximating Subspaces Proof n ? ? Let Z be a subset of R . Since Z ⊆ hZi it follows that hZi ⊆ Z . Let ? now y 2 Z and let z = a1z1 + ··· + apzp 2 hZi, where z1;:::; zp 2 Z. Since (y; z) = a1(y; z1) + ··· + ap(y; zp) = 0; it follows that y 2 hZi?. Thus, we have Z ? = hZi?. ? ? ? ? n This allows us to write (Z ) = (hZi ) . Since hZi is a subspace of R , we have (hZi?)? = hZi, so (Z ?)? = hZi. 14 / 74 Approximating Subspaces Let W = fw1;:::; wng be a basis in the real n-dimensional inner product space L. If x = x1w1 + ··· + xnwn and y = y1w1 + ··· + ynwn, then n n X X (x; y) = xi yj (wi ; wj ); i=1 j=1 due to the bilinearity of the inner product. n×n Let A = (aij ) 2 R be the matrix defined by aij = (wi ; wj ) for 1 6 i; j 6 n. The symmetry of the inner product implies that the matrix A itself is symmetric. Now, the inner product can be expressed as 0 1 y1 B . C (x; y) = (x1;:::; xn)A @ . A : yn We refer to A as the matrix associated with W . 15 / 74 Approximating Subspaces Theorem n Let S be a subspace of R such that dim(S) = k. There exists a matrix n×k A 2 R having orthonormal columns such that S = Ran(A). 16 / 74 Approximating Subspaces Proof Let v1;:::; vk be an orthonormal basis of S. Define the matrix A as A = (v1;:::; vk ). We have x 2 S, if and only if x = a1v1 + ··· + ak vk , which is equivalent to x = Aa. This amounts to x 2 Ran(A), so S = Ran(A). 17 / 74 Approximating Subspaces For an orthonormal basis in an n-dimensional space, the associated matrix is the diagonal matrix In. In this case, we have (x; y) = x1y1 + x2y2 + ··· + xnyn for x; y 2 L. Observe that if W = fw1;:::; wng is an orthonormal set and x 2 hWi, which means that x = a1w1 + ··· + anwn, then ai = (x; wi ) for 1 6 i 6 n. 18 / 74 Approximating Subspaces Let W = fw1;:::; wng be an orthonormal set and let x 2 hWi. The equality x = (x; w1)w1 + ··· + (x; wn)wn is the Fourier expansion of x with respect to the orthonormal set W . Furthermore, we have Parseval's equality: n 2 X 2 k x k = (x; x) = (x; wi ) : i=1 Thus, if 1 6 q 6 n we have q X 2 2 (x; wi ) 6k x k : i=1 19 / 74 Gram Matrices Gram Matrices Definition Let V = (v1;:::; vm) be a sequence of vectors in an inner product space. m×m The Gram matrix of this sequence is the matrix GV = (gij ) 2 R defined by gij = (vi ; vj ) for 1 6 i; j 6 m. Note that GV is a symmetric matrix. 20 / 74 Gram Matrices Theorem Let V = (v1;:::; vm) be a sequence of m vectors in an inner product linear space (L; (·; ·)). If fv1;:::; vmg is linearly independent, then the Gram matrix GV is positive definite. 21 / 74 Gram Matrices Proof m Suppose that V is linearly independent and let x 2 R . We have m m 0 X X x GVx = xi (vi ; vj )xj i=1 j=1 0 m m 1 X X = @ xi vi ; xj vj A i=1 j=1 m 2 X = xi vi > 0: i=1 0 Therefore, if x GVx = 0, we have x1v1 + ··· + xnvn = 0. Since fv1;:::; vmg is linearly independent it follows that x1 = ··· = xm = 0, so x = 0n. Thus, GV is indeed, positive definite. 22 / 74 Gram Matrices Example S Let S = fx1;:::; xng be a finite set, L a C-linear space and L be the linear space that consists of function defined on S with values in L. We defined the linear basis of this space as fe1;:::; eng consisting of the functions ( 1 if x = xi ; ei (x) = 0 otherwise; for 1 6 i 6 n. If E = (e1;:::; en), the Gram matrix of E is positive definite Pn Pn and the inner product of two functions f = i=1 ai ei and g = j=1 bj ej is n n ! X X (f ; g) = ai ei ; bj ej i=1 i=1 n n X X = ai (ei ; ej )bj : i=1 i=1 23 / 74 Gram Matrices The Gram matrix of an arbitrary sequence is positive semidefinite, as the reader can easily verify.