Math 103, Summer 2006 Eigenvectors and Eigenvalues, Geometrically August 1, 2006 EIGENVECTORS AND EIGENVALUES, GEOMETRICALLY 1. Announcements There are things about the test people might be interested in. The first is that the material covered on this coming midterm will focus mainly on material that we’ve covered since the first midterm (including the lecture or two before the first midterm). Math exams are almost always cumulative in some way or another, but while you will be required to know the topics we discussed in the first few weeks of class the test will ask questions centered around the more recent material. As for the first midterm, we’ll be holding a review session for this midterm. It will be held in room 380-C (basement of the math building) and will start at 6pm. As last time I’ll bring pizza and drinks, and you should bring lots of questions to ask. 2. Recap Last class period we talked about how one goes about finding eigenvalues for a matrix A, and at the very end of class we talked about a strategy for computing the corresponding eigenspaces. As a refresher, I’ll just state the following Definition 2.1. The eigenspace corresponding to an eigenvalue λ is the set → → → {− v : A− v = λ− v }. It is denoted Eλ . We said in class yesterday that this is actually a subspace, and we asked the question: how does one compute the eigenspace? More precisely, how can I find a basis for the eigenspace Eλ when I have an eigenvalue λ? → − − → By the definition of Eλ we know that any − v ∈ Eλ satisfies A→ v = λ→ v , which means that − v ∈ ker(A − λIn ). Hence to find Eλ , one just computes ker(A − λIn ). Example. In class yesterday we said that the eigenvalues of the matrix 1 9 3 11 1 4 5 A= 3 2001 17 are {1, 3, 17}. Find a basis for E1 . Solution. To find a basis for E1 we need to find a basis for ker(A − In ). But notice 0 9 3 11 1 0 0 0 4 5 1 0 = rref(A − In ) = rref 2 2001 1 16 aschultz@stanford.edu that , http://math.stanford.edu/~aschultz/summer06/math103 Page 1 of 3 Math 103, Summer 2006 Eigenvectors and Eigenvalues, Geometrically August 1, 2006 1 1 0 0 and hence ker(A − In ) = Span( 0 ). This means that E1 has a basis given by the vector 0 0 0 . ¤ Notice in this example that dim(E1 ) = 1, but that the algebraic multiplicity of λ = 1 was 2. The dimension of an eigenspace captures important geometric information about the action of the matrix A, and hence has a special name. Definition 2.2. The geometric multiplicity of an eigenvalue λ is dim(Eλ ). With this new language, our observation is that the geometric multiplicity of 1 was less than the algebraic multiplicty of 1. In fact this is an example of a more general relationship. Theorem 2.1. For an eigenvalue λ, the geometric multiplicty of λ is at most the algebraic multiplicty of λ. Let’s do another example, but this time we’ll do the whole kit and kaboodle. Example. Let µ A= 7 −3 −3 7 ¶ . → Find all eigenvalues of A, a basis for all eigenspaces, and a formula for − vt if µ ¶ 1 − → . v0 = 2 Solution. To find the eigenvalues of A we must compute the roots of the characteristic polynomial ¶ µ 7−λ −3 = (7 − λ)2 + 9 = ((7 − λ) − 3)((7 − λ) + 3) = (4 − λ)(10 − λ). det(A − λI2 ) = det −3 7−λ Since we’ve factored the characteristic polynomial we can ‘see’ the eigenvalues are {4, 10}. Now to find the corresponding eigenspaces, we find a basis for the kernels of appropriate matrices. For instance, to find E4 I compute a basis for ¶ ¶ µ µ ¶ µ 3 −3 1 −1 7 − 4 −3 = ker = ker . ker 0 0 −3 3 −3 7 − 4 But I can read off a basis of the kernel on the µ right ¶ since my matrix is in reduced row echelon form, and I see 1 that E4 has a basis given by the the vector . 1 Similarly, to find E10 I compute a basis for µ ¶ µ 7 − 10 −3 −3 ker = ker −3 7 − 10 −3 −3 −3 ¶ µ = ker 1 0 1 0 ¶ . I can read µ off a basis ¶ of the kernel on the right just like last time, and I see that E10 has a basis given by the 1 the vector . −1 aschultz@stanford.edu http://math.stanford.edu/~aschultz/summer06/math103 Page 2 of 3 Math 103, Summer 2006 Eigenvectors and Eigenvalues, Geometrically August 1, 2006 ¶ 1 → − . Now I know that − v t = At → v0 , but I also know this isn’t 2 − → much of an answer: computationally, it’s quite hard to calculate vt in this way when t gets large. However, since I can write µ ¶ µ ¶ µ ¶ 3 1 1 1 1 = − 2 −1 1 2 2 I have ¶ µ ¶¶ µ µ 1 3 1 1 − → → t− t − vt = A v0 = A −1 1 2 2 → − Finally, I am to find a formula for − vt if → v0 = = 3 t A 2 µ 1 −1 ¶ µ 1 − At 2 µ 1 1 ¶ = 3 t 4 2 µ 1 −1 ¶ 1 − 10t 2 µ 1 1 ¶ . ¤ 3. Sketching Trajectories We talked in class today about how one can sketch solutions to linear dynamical systems by computing eigenvectors and eigenvalues. The motivating example was the stretch of iron-rusted road that I ran down in Columbus, Ohio earlier this summer. You can find pictures of the road up on the web at http://math.stanford.edu/~aschultz/summer06/math103/coursenotes_and_handouts/Rust_Spots The point of the example was this: by looking how the rust seeped across the road through time, you can reconstruct the lay of the road. In particular since the rust will move downhill you can tell which way is uphill on the road. You can also see that the road bows slightly in the middle since rust spots trail to the right on the right side of the road and to the left on the left side of the road. The point of looking at these pictures was this: if you see trajectories, your brain knows how to take that information and reconstruct the lay of the road. If I showed you these pictures and didn’t explain anything to you about them, you’d be able to tell me what is uphill and what is downhill. What we’re going to start doing is something like the reverse: the eigenvectors and eigenvalues will tell you how the corresponding matrix acts on your space, and you will use your physical intuition to sketch trajectories. Examples of the intuition we described are shown in today’s Mathematica example, so you should check it out on the web. You can find it in its normal spot, but since it was so central to today’s class I’ll also link to it from this document http://math.stanford.edu/~aschultz/summer06/math103/couresnotes_and_handouts/mathematica_060801 The conclusion we came to was a way to sketch trajectories for the action of a matrix A by plotting the lines spanned by eigenvectors and determining whether the corresponding eigenvalues where greater than or less than 1. See the sketch on page 302 as your guide. aschultz@stanford.edu http://math.stanford.edu/~aschultz/summer06/math103 Page 3 of 3