Lecture 11 Fundamental Theorems of Linear Algebra Orthogonalily and Projection Shang-Hua Teng The Whole Picture • Rank(A) = m = n Ax=b has unique solution R I • Rank(A) = m < n Ax=b has n-m dimensional solution R I F I R 0 I F • Rank(A) < n, Rank(A) < m R 0 0 Ax=b has 0 or n-rank(A) dimensions • Rank(A) = n < m Ax=b has 0 or 1 solution Basis and Dimension of a Vector Space • A basis for a vector space is a sequence of vectors that – The vectors are linearly independent – The vectors span the space: every vector in the vector can be expressed as a linear combination of these vectors Basis for 2D and n-D • (1,0), (0,1) • (1 1), (-1 –2) • The vectors v1,v2,…vn are basis for Rn if and only if they are columns of an n by n invertible matrix Column and Row Subspace • C(A): the space spanned by columns of A – Subspace in m dimensions – The pivot columns of A are a basis for its column space • Row space: the space spanned by rows of A – – – – Subspace in n dimensions The row space of A is the same as the column space of AT, C(AT) The pivot rows of A are a basis for its row space The pivot rows of its Echolon matrix R are a basis for its row space Important Property I: Uniqueness of Combination • The vectors v1,v2,…vn are basis for a vector space V, then for every vector v in V, there is a unique way to write v as a combination of v1,v2,…vn . v = a1 v1+ a2 v2+…+ an vn v = b1 v1+ b2 v2+…+ bn vn • So: 0=(a1 - b1) v1 + (a2 -b2 )v2+…+ (an -bn )vn Important Property II: Dimension and Size of Basis • If a vector space V has two set of bases – v1,v2,…vm . V = [v1,v2,…vm ] – w1,w2,…wn . W= [w1,w2,…wn ]. • then m = n – Proof: assume n > m, write W = VA – A is m by n, so Ax = 0 has a non-zero solution – So VAx = 0 and Wx = 0 • The dimension of a vector space is the number of vectors in every basis – Dimension of a vector space is well defined Dimensions of the Four Subspaces Fundamental Theorem of Linear Algebra, Part I • • • • • Row space: C(AT) – dimension = rank(A) Column space: C(A)– dimension = rank(A) Nullspace: N(A) – dimension = n-rank(A) Left Nullspace: N(AT) – dimension = m –rank(A) Orthogonality and Orthogonal Subspaces • Two vectors v and w are orthogonal if vw v w w v 0 T T • Two vector subspaces V and W are orthogonal if for all v V and w W , v w 0 T Example: Orthogonal Subspace in 5 Dimensions 1 1 0 0 0 0 1 1 0 0 C 0, 0, 1 C 0, 0 0 0 0 0 1 0 0 0 1 1 The union of these two subspaces is R5 Orthogonal Complement • Suppose V is a vector subspace a vector space W • The orthogonal complement of V is V {w W such that w V } • Orthogonal complement is itself a vector subspace dim( V ) dim( V ) dim( W ) Dimensions of the Four Subspaces Fundamental Theorem of Linear Algebra, Part I • • • • • Row space: C(AT) – dimension = rank(A) Column space: C(A)– dimension = rank(A) Nullspace: N(A) – dimension = n-rank(A) Left Nullspace: N(AT) – dimension = m –rank(A) Orthogonality of the Four Subspaces Fundamental Theorem of Linear Algebra, Part II • The nullspace is the orthogonal complement of the row space in Rn • The left Nullspace is the orthogonal complement of the column space in Rm • N ( A) C ( A ) T N ( AT ) (C ( A)) Proof • The nullspace is the orthogonal complement of the row space in Rn N ( A) C ( A ) N ( A) x : Ax 0 C ( AT ) AT y : y R m implying T x A y A y x y T Ax y T Ax 0 T T T T The Whole Picture dim r C(AT) xr A xr = b C(A) b A x= b dim r x xr xn Rn dim n- r Rm xn N(A) A xn = 0 N(AT) dim m- r Uniqueness of The Typical Solution • Every vector in the column space comes from one and only one vector xr from the row space • Proof: suppose there are two xr , yr from the row space such that Axr =A yr =b, then Axr -A yr = A(xr -yr ) = 0 (xr -yr ) is in row space and nullspace hence must be 0 • The matching of dim in row and column spaces Deep Secret of Linear Algebra Pseudo-inverse • Throw away the two null spaces, there is an r by r invertible matrix hiding insider A. • In some sense, from the row space to the column space, A is invertible • It maps an r-space in n space to an r-space in m-space Invertible Matrices • Any n linearly independent vector in Rn must span Rn . They are basis. • So Ax = b is always uniquely solvable • A is invertible Projection • Projection onto an axis (a,b) x axis is a vector subspace Projection onto an Arbitrary Line Passing through 0 Projection on to a Plane Projection onto a Subspace • Input: 1. Given a vector subspace V in Rm 2. A vector b in Rm… • Desirable Output: – – – A vector in x in V that is closest to b The projection x of b in V A vector x in V such that (b-x) is orthogonal to V How to Describe a Vector Subspace V in Rm • If dim(V) = n, then V has n basis vectors – a1, a2, …, an – They are independent • V = C(A) where A = [a1, a2, …, an] Projection onto a Subspace • Input: 1. Given n independent vectors a1, a2, …, an in Rm 2. A vector b in Rm… • Desirable Output: – – – A vector in x in C([a1, a2, …, an]) that is closest to b The projection x of b in C([a1, a2, …, an]) A vector x in V such that (b-x) is orthogonal to C([a1, a2, …, an]) Think about this Picture dim r C(AT) xr A xr = b C(A) b A x= b dim r x xr xn Rn dim n- r Rm xn N(A) A xn = 0 N(AT) dim m- r