12.07.2015 Views

Stat 5101 Lecture Notes - School of Statistics

Stat 5101 Lecture Notes - School of Statistics

Stat 5101 Lecture Notes - School of Statistics

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

234 <strong>Stat</strong> <strong>5101</strong> (Geyer) Course <strong>Notes</strong>Pro<strong>of</strong>. This is just linearity <strong>of</strong> matrix multiplication.The second property means that all the eigenvectors corresponding to oneeigenvalue constitute a subspace. If the dimension <strong>of</strong> that subspace is k, thenit is possible to choose an orthonormal basis <strong>of</strong> k vectors that span the subspace.Since the first property <strong>of</strong> eigenvalues and eigenvectors says that (E.1)is also satisfied by eigenvectors corresponding to different eigenvalues, all <strong>of</strong> theeigenvectors chosen this way form an orthonormal set.Thus our orthonormal set <strong>of</strong> eigenvectors spans a subspace <strong>of</strong> dimension mwhich contains all eigenvectors <strong>of</strong> the matrix in question. The question thenarises whether this set is complete, that is, whether it is a basis for the wholespace, or in symbols whether m = n, where n is the dimension <strong>of</strong> the wholespace (A is an n × n matrix and the x i are vectors <strong>of</strong> dimension n). It turnsout that the set is always complete, and this is the third important fact abouteigenvalues and eigenvectors.Lemma E.3. Every real symmetric matrix has an orthonormal set <strong>of</strong> eigenvectorsthat form a basis for the space.In contrast to the first two facts, this is deep, and we shall not say anythingabout its pro<strong>of</strong>, other than that about half <strong>of</strong> the typical linear algebra book isgiven over to building up to the pro<strong>of</strong> <strong>of</strong> this one fact.The “third important fact” says that any vector can be written as a linearcombination <strong>of</strong> eigenvectorsn∑y = c i x ii=1and this allows a very simple description <strong>of</strong> the action <strong>of</strong> the linear operatordescribed by the matrixn∑n∑Ay = c i Ax i = c i λ i x ii=1i=1(E.4)So this says that when we use an orthonormal eigenvector basis, ifyhas therepresentation (c 1 ,...,c n ), then Ay has the representation (c 1 λ 1 ,...,c n λ n ).Let D be the representation in the orthonormal eigenvector basis <strong>of</strong> the linearoperator represented by A in the standard basis. Then our analysis above saysthe i-the element <strong>of</strong> Dc is c i λ i , that is,n∑d ij c j = λ i c i .j=1In order for this to hold for all real numbers c i , it must be that D is diagonald ii = λ id ij =0,i ≠ j

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!