Finding Eigenvalues and Eigenvectors
Date: 06/27/2007 at 17:20:32 From: Paul Subject: linear algebra eigenvalues/eigenvectors (5,-2,2) (5-x,-2,2) (4,-3,4) = (4,-3-x,4) = -x^3+9x^2-23x+15 = 0 (4,-6,7) (4,-6,7-x) In this 3x3 square matrix, how would you find the eigenvalues and eigenvectors? We believe the eigenvalues are 5,1,3 but we cannot seem to come up with eigenvectors. Thanks, Paul
Date: 06/28/2007 at 00:16:44 From: Doctor Ricky Subject: Re: linear algebra eigenvalues/eigenvectors Hi Paul, Thanks for writing Dr. Math! Eigenvalues and eigenvectors are actually somewhat interesting in that they provide bases for the null set. Remember, the null set is the set of vectors that satisfy the homogeneous system (where all equations are equal to zero). So we will define our matrix A as your matrix: [5 -2 2] A = [4 -3 4] [4 -6 7] Now, remember that an eigenvalue is a scalar that, when multiplied by a column vector X, will give us the same as our matrix A multiplied by the same column vector. i.e. AX = LX where L is the scalar (written as lambda, typically) This means that: AX - LX = 0 where 0 is the zero matrix which is the same as: (A-LI)(X) = 0 where I is the identity matrix since X is multiplied on the right for both. Also, remember that matrix multiplication is not generally commutative, so (A-L)(X) is not necessarily the same as (X)(A-L), so the order here is important. If we take the determinant of both sides, we get: det[(A - LI)(X)] = 0 where 0 is now the scalar 0 because the determinant of the zero matrix is 0 A little review on theory would tell you that if you have two square matrices M and N, then det(M*N) = det(M)*det(N) We know that our original matrix A was square, and since the matrix LI is just a square matrix with L on the diagonal and zeros everywhere else, LI is a square matrix also. Writing this using our notation above, this means that: det[(A-LI)(X)] = det(A-LI)*det(X) = 0 Dividing both sides by det(X) [since it is irrelevant], we get: det(A-LI) = 0 Remember, we are trying to find what values of L make this statement true, so obviously we wouldn't divide by the determinant of the matrix that involves L. This may have seemed like a lot of explanation to get to this step, but a nice review of the topic and why we do each step is important to truly grasping the process. So now let's look at A - LI: [5 -2 2] [L 0 0] [5-L -2 2] A-LI = [4 -3 4] - [0 L 0] = [4 -3-L 4] [4 -6 7] [0 0 L] [4 -6 7-L] Using Lagrange's theorem (which tells us that we can find determinants of large matrices using minors and cofactors to break down the matrix into smaller submatrices) with the top row of the matrix, we get: det(A-LI) = (5-L)[(-3-L)(7-L)-(4)(-6)] - (-2)[(4)(7-L)-(4)(4)] +(2)[(4)(-6)-(-3-L)(4)] = 0 [because we said the det(A-LI)=0] which simplifies to what you found, 15 - 23L + 9L^2 - L^3 = 0 At this point, we can use the rational root theorem (if we are without a graphing utility that can help us find roots) to find the solutions of this equation. The rational root theorem says that the only possible rational roots of this equation are all the factors of the constant term divided by all the factors of the coefficient of the largest exponentiated variable. In this case, the constant and the coefficient respectively are 15 and -1, so the possible rational roots are: +-1, +-3, +-5, +-15 Using synthetic division or by simply plugging these possible values into lambda, we find that the solutions are 1, 3, and 5. The matrix associated with L = 1 is (A-I), which is: [4 -2 2] A-I = [4 -4 4] [4 -6 6] The eigenvector is the column vector X that we can right-multiply (because of our initial setup) by our matrix (A-LI) such that it will always be equal to our zero matrix. In other words, [4 -2 2] [X1] (A-I)(X) = [4 -4 4]*[X2] = 0 [4 -6 6] [X3] We see on inspection that this will occur when X1=0, X2=1, and X3=1, which give us the eigenvector associated with L = 1 to be the matrix X = [0 1 1]^T, where ^T means transposed to be a column vector. Keep in mind that the eigenvector is just ONE basis that makes this true. We could have just as easily used the column vector X = [0 -1 -1]^T. However, notice that this vector is not linearly independent to our previous eigenvector, so each will form a basis but not necessarily a linearly independent basis. Now, to find the eigenvector associated with L = 3, we have the matrix: [2 -2 2] [X1] (A-3I)(X) = [4 -6 4]*[X2] = 0 [4 -6 4] [X3] Now, since the last two rows are the same, that means that we can eliminate the last row so we really only have: [2 -2 2] [X1] (A-3I)(X) = [4 -6 4]*[X2] = 0 [X3] On inspection, we see that the vector [1 0 -1]^T satisfies this relationship, so it is an eigenvector for L = 3 and forms a basis for the null set. Now, for our final eigenvalue L = 5. Substituting this in for L in our matrix, we get: [0 -2 2] [X1] (A-5I)(X) = [4 -8 4]*[X2] = 0 [4 -6 2] [X3] Although this matrix might seem a little more difficult to find an eigenvector on inspection, the first row tells us that X2 = X3 since they must cancel out. Using that in one of our other rows, we see that an eigenvector for L = 5 could be [1 1 1]^T. Note that the eigenvectors form a basis of the null space of A-LI, otherwise known as the eigenspace. This means that the three column vectors we found all constitute a (linearly independent) basis of the eigenspace. Hopefully this helped illustrate the process of finding eigenvalues and eigenvectors. If you are having any more difficulty, the Dr. Math Archive does have some articles on both topics. Otherwise, if you have any more questions for me, feel free to write back! - Doctor Ricky, The Math Forum http://mathforum.org/dr.math/
Search the Dr. Math Library:
Ask Dr. MathTM
© 1994- The Math Forum at NCTM. All rights reserved.