Sie sind auf Seite 1von 29

mat210

INTRO TO LINEAR EQUATIONS AND SOME APPLICATIONS Introduction to linear equations: An equation of a line in 2D space is comprised of two variables. ax + by = c a, b, and c are constants. This is a linear equation. An equation of a plane in 3D space is made up of 3 variables. ax + by + cz = d a, b, c, and d are constants. This is a linear equation. A linear equation in n variables: a1x1 + a2x2 + .. + anxn = b Linear equations have no products or roots of variables, and no variables involved in trigonometric, exponential, or logarithmic functions. Variables only appear to the 1st power for these equations. Each variable in the linear equation has a real number for its solution that satisfies the equation. The solution is a sequence of numbers that substitute for the variables. There happens to be more than one set of solutions in some cases; the set of all solutions is called the solution set. Parametric equations: Parametric equations are used to describe the entire solution set. To get an equation in parametric form you must solve for one variable and plug t into the independent variable. For example: x = 4 - 2y x = 4 - 2t y = t x is already solved, so use y as the independent variable and let it equal t the parameter x could have been the independent variable instead of y. This set up allows you to plug values into t, so you can discover the solution set. These independent variables can be called free variables. In a three variable equation you will have 2 free variables, so this requires two parameters. System of linear equations: A solution for a system of linear equations is a sequence of numbers that satisfies each of the equations in the system. A system of linear equations is two or more linear equations that may have some type of relationship such as intersecting at a point, being the same lines, or not intersecting at all. A group of equations are consistent if they intersect, and inconsistent if they do not. The lines are coincidental if they are the same line. Below is a system of linear equations: x - 2y + 3z = 9 -x + 3y = -4 2x - 5y + 5z = 17 Here is an equivalent system which is in row-echelon form: x - 2y + 3z = 9

file:///Users/admin/Desktop/linear%20algebra/MAT210-1.HTM[10/10/09 5:05:00 PM]

mat210

y + 3z = 5 z=2 Using back substitution you can easily solve the system. It is important to note that two systems of linear equations, like above, are equivalent if they have the same solution set. To get the system in row-echelon form you can use the three operations (elementary row operations): Interchange two equations Multiple an equation by a nonzero constant Add a multiple of an equation to another equation Using these operations is following the Gaussian elimination process. The key to finding an inconsistent system is reaching a false statement such as 0 = 7. A system is said to homogeneous if the constants are all zero. Such a system has at least one solution which is when each variable is zero. Matrices: A matrix is a m x n rectangular array. Where m is the number of rows, and n is the number of columns. If a matrix is just composed of real numbers then it is a real matrix. If m = n, then the matrix is called square of order n. The main diagonal entries of a square of order n are a11, a22, a33, ... A matrix consisting of coefficients and constants from a system of linear equations is called an augmented matrix, and a coefficient matrix just has the coefficients from the system. Two matrices are said to be row-equivalent if a finite number of elementary row operations can be used to make one like the other.

The above picture shows a matrix representation of a system of linear equations and the steps to get it into rowechelon form. A matrix is in reduced row-echelon form if every leading coefficient has zeroes above and below. Matrices in row echelon form that have rows with all zeroes in them must be at the bottom. Gauss-Jordan elimination is the process of taking a system of linear equations and putting it into reduced rowechelon form. This process will give you the solution directly instead of having to use back substitution. Polynomial curve fitting:

file:///Users/admin/Desktop/linear%20algebra/MAT210-1.HTM[10/10/09 5:05:00 PM]

mat210

If you have a bunch of points, then it is possible to find the polynomial for that collection of data. You should have one point for each coefficient plus one in order to find the polynomial. A point makes one linear equation, so you may have 3 points and this means 3 linear equations. For example: a0 + a1x + a2x^2 Find the coefficients (1,4),(2,0),(3,12) Your ordered pairs p(1) = a0 + a1(1) + a2(1^2) = 4 p(2) = a0 + a1(2) + a2(2^2) = 0 p(3) = a0 + a1(3) + a2(3^3) = 12 p(x) = 24 - 28x + 8x^2 The equations above show the system you must solve in order to reach the destination polynomial. For large values of x you may want to substitute something for x to make the computation easier. This technique loses its accuracy the farther you get away from the points you used to define the polynomial. Network analysis: The total flow entering a junction equals the flow exiting it.

The above picture represents a network with 5 junctions. The arrows represent the direction of the flow. The equations that describe it: x1 + x2 = 20 x3 + 20 = x4 x2 + x3 = 20 x1 + 10 = x5 x5 + 10 = x4 You can solve the system using matrices. It turns out this system of linear equations has infinite solutions. This
file:///Users/admin/Desktop/linear%20algebra/MAT210-1.HTM[10/10/09 5:05:00 PM]

mat210

strategy can assist with solving an electrical network. More specifically, this can be done with Kirchoff's laws. These are: 1. All the current flowing into a junction must flow out of it 2. The sum of the products IR (I is the current and R is the resistance) around a closed path is equal to the total voltage in the path.

The equations for this circuit are: I1 + I3 = I2 (look at junction 2 to easily see that current 2 is the sum of one and three) I1R1 + I2R2 = 7 or I1(3) + I2(2) = 7 I3R3 + I2R2 = 8 or I3(4) + I2(2) = 8 The solution is I1 =1, I2 = 2 and I3 = 1.

file:///Users/admin/Desktop/linear%20algebra/MAT210-1.HTM[10/10/09 5:05:00 PM]

MATRICES

MATRICES

Matrix operations: A matrix can be represented by an uppercase letter, or a representative element in enclosed in brackets like [aij]. Two matrices can only be equal if they have the same dimensions and same elements. A matrix with one column is called a column vector or column matrix, and a matrix with one row is called a row vector or row matrix. If you were to add two matrices, they must have the same size and this goes for subtraction too. Adding or subtracting matrices of different dimensions amounts to undefined matrices. To complete an add or subtraction operation you simply add or subtract the corresponding elements and put them into the result. For instance: [1 2] + [1 3] = [2 5] [1 2] - [1 3] = [0 -1] Scalar multiplication allows you to multiply each element of the matrix with a real number. cA = [caij] Scalar multiplication A + B = [aij + bij] Matrix Addition 6[1 2] = [6 12] Multiplying matrices is a more complicated matter. Let's look at an example:

So there are three small matrices above and the two on top are being multiplied to get the bottom one. In order multiply two matrices they must have certain dimensions. Matrix one is 1 x 2, and matrix two is 2 x 1. You can multiply these two, because the number of columns in the first matrix is equal to the number of rows in the second matrix. So the resulting matrix is 1 x 1, since the first matrix has one row and the second matrix has one column. To actually come up with the value for an element you follow these steps: a11*b11 + a12*b21 + ... + a1n*bn1 = c11 a11*b12 + a12*b22 + ... + a1n*bn2 = c12 continue for each column in matrix B or the number of columns in C -- making the necessary modifications to the subscripts go to the next row: a21*b11 + a22*b21 + ... + a2n*bn1 = c21 repeat the steps from the top while paying attention to the subscripts again until you reach the last column in B. Continue this general procedure until you have done all of A's rows.

file:///Users/admin/Desktop/linear%20algebra/MAT210-2.HTM[10/10/09 5:05:01 PM]

MATRICES

The dimensions for Matrix A are 2 x 3, and Matrix B is 3 x 3, so the resulting matrix is 2 x 3. Element C11 is 1*-2 + 0*1 + 3*-1, which is -5, and C23 is 2*2 + -1*0 + -2*-1 or 6. With Matrix multiplication you can partition the matrix into a few matrices. Partitioned matrices follow the form Ax = b; A is the matrix of coefficients, x is the matrix of variables, and b is the constants, as you can see below.

This partitioning can go even deeper, so that it is easier to see the solution. In a situation like this each unknown is a scalar for a column vector which contains each coefficient for that unknown. All these column vectors are added up to equal another column vector which holds all the constants. Properties of matrix operations: A, B, and C are matrices. c and d are scalars. Commutative property of addition: A + B = B + A Associative property of addition: A + (B + C) = (A + B) + C; c(dA) = (cd)A Distributive property: c(A + B) = cA + cB, (c + d)A = cA +dA

file:///Users/admin/Desktop/linear%20algebra/MAT210-2.HTM[10/10/09 5:05:01 PM]

MATRICES

A zero matrix consisting of an m x n matrix of zeroes added to any matrix of m x n dimensions is that matrix. This is called an additive matrix. Properties of zero matrices: A + 0mn = A A + (-A) = 0mn if cA = 0mn, then c = 0, or A = 0mn -A is an additive inverse. Matrix algebra works very similar to regular algebra. If X + A = B, then X = B -A.This was achieved by subtracting matrix A from both sides. If 3X = B - A, then X = 1/3(B - A).This is possible since 1/3 is a scalar. Properties of matrix multiplication: Associative property of multiplication: A(BC) = (AB)C Distributive property: A(B + C) = AB + AC, (A+ B)C = AC + BC, c(AB) = (cA)B = A(cB) Matrix multiplication is noncommutative. Also, just because AC = BC, does not mean A = B. There is a special type of matrix called the identity matrix; this is a square matrix with 1s in its main diagonal, and zeroes elsewhere. Identity matrices are denoted like In where n is the columns and rows. Properties of the identity matrix: A is an m x n matrix AIn = A ImA = A Exponential notation can be applied to matrices too, so you can simply express A*A*A like A^3. A^0 is In. Matrices can be transposed, so that their rows are switched to their columns. If the matrix was m x n, then it will be n x m. A transposed matrix is denoted by A^T. If you transpose a matrix and find that it equals the original one, then it is symmetric.A*A^T is symmetric. Properties of transposes: (A^T)^T = A (A + B)^T = A^T + B^T Transpose of a sum (cA)^T = c(A^T) Transpose of a scalar multiple (AB)^T = (A^T)(B^T) Transpose of a product Inverse of a matrix: An n x n matrix A is invertible (nonsingular), if it is a square matrix and there exists an n x n matrix B such that AB = BA = In. B is the inverse of A; a matrix that has no inverse is called noninvertible or singular.Nonsquare matrices do not have inverses. If a matrix is invertible, then its inverse is unique. An inverse can be denoted as A^-1. A useful trick is adjoining matrices to find the inverse matrix. If you have a matrix A, and an identity matrix In, you can put them into one matrix. Append the identity matrix to the right side, and use Gauss-Jordan elimination to reduce the A side to an identity matrix, and the other side to the inverse matrix. If you cannot reduce the matrix A to In, then it is singular. For 2 x 2 matrix you may want to use the formula A^-1 = {1/(ad - bc)}*[d -b, -c a] (comma denotes start of second row). A = [a b, c d] (comma denotes start of second row). Note that (ad- bc) cannot be zero, otherwise there is no inverse. This small expression is called the determinant of A. Properties of inverses: c is a scalar, and A is an invertible matrix. k is a positive integer. A^-1, A^k, cA, and A^T are also invertible matrices.
file:///Users/admin/Desktop/linear%20algebra/MAT210-2.HTM[10/10/09 5:05:01 PM]

MATRICES

(A^-1)^-1 = A (A^k)^-1 = A^-1 * A^-1 ... A^-1 (k factors) = (A^-1)^k (cA)^-1 = (1/c)(A^-1) c is not zero (A^T)^-1 = (A^-1)^T A^-2 is (A^-1)^2 or (A^2)^-1. If A and B are invertible matrices of size n, then AB is invertible and (AB)^-1 = (B^1)*(A^-1). If C is an invertible matrix, then the following properties hold: If AC = BC, then A = B Right cancellation property If CA = CB, then A = B Left cancellation property If A is an invertible matrix, then the system of linear equations Ax = b has a unique solution given by x = (A^-1)b. This allows you to determine whether a system of linear of equations has a unique solution. Elementary matrices: An n x n matrix is called an elementary matrix if it can be obtained from the identity matrix by a single elementary row operation. An identity matrix is also an elementary matrix since you can multiply one of its rows by 1 to obtain itself. Using elementary matrices you can achieve the same results as performing elementary row operations. For instance, E is an elementary matrix acquired by performing some elementary row operation -- that same operation can be performed on another matrix A (of the same dimensions) by doing E*A. You can actually solve a system of linear equations using these elementary matrices. The matrix that results from doing several multiplications like E3*E2*E1*A is row-equivalent to the matrix A, and so are the intermediate steps. All elementary matrices are invertible.To find the inverse matrix you simply reverse the operation used to get the elementary matrix and do it to an identity matrix. All invertible matrices can be written as a product of elementary matrices. If A is an n x n matrix then the following statements are equivalent: A is invertible Ax = b has a unique solution for every n x 1 column matrix b Ax = 0 has only the trivial solution A is row-equivalent to In A can be written as the product of elementary matrices An efficient way of solving a system of linear equations is LU-Factorization. For this method the square matrix A is expressed as product, A = LU. L is the lower triangle and all entries above the main diagonal are zero, and U is the upper triangle where below its main diagonal are all zeroes. Start by writing Ax = LUx, and letting Ux = y; solve for y by using b = Ly, and then x with Ux = y.

file:///Users/admin/Desktop/linear%20algebra/MAT210-2.HTM[10/10/09 5:05:01 PM]

The determinant of a matrix:

DETERMINANTS AND THEIR APPLICATIONS The determinant of a matrix: Every square matrix has a real number called a determinant. The general solution for a system: a11x1 + a12x2 = b1 a21x1 + a22x2 = b2 x1 = (b1a22 - b2a12)/(a11a22 - a21a12) x2 = (b2a11 - b1a21)/(a11a22 - a21a12) The denominator for each of these fractions is called the determinant of the coefficient matrix. The determinant is denoted as |A| or det(A); A is the matrix. The determinant of a matrix can be positive, negative or zero. The determinant of a matrix of order 1 is simply that one element. Finding a determinant for a matrix of order larger than 2 is more difficult. You need to find its minors and cofactors. A minor of the term aij is found by removing the ith row and the jth column. You do this for each element in the first row. The cofactor is defined by Cij = (-1)^(i+j)Mij. In order to plug Mij into the formula you must find the determinant of that matrix. Finally, the determinant is (j=1 to n)(a1j*C1j). The determinant can be found by doing this to any row of the matrix. Usually, you expand by cofactors in the first row, but this can be done with the other rows and the columns too. You do not need to evaluate the cofactors of zero entries, so it is best to find the row or column with the most zeroes. For matrices larger than 3x3 it can get complicated easily. Pick a row or column with the most zeroes. Find the minor of each element in that row or column. The minor is equal to the determinant of the matrix it represents, so if your minor is 3x3, you need to find the determinant of that and then plug into the cofactor formula. There is a trick to help find the 3x3 matrix determinant. This involves copying the 2nd and 3rd column to the 4th and 5th columns, and adding or subtracting the products of the six diagonals. |A| = a11a22a33 + a12a23a31 + a13a21a32 a31a22a13 - a32a23a11 - a33a21a12. A matrix that is an upper and lower triangle is called a diagonal. To find the determinant of an upper, lower, or diagonal matrix just find the product of the main diagonal entries. Evaluation of a determinant using elementary operations: Different elementary row operations have different results for the determinant. Switching the rows of a 2x2 matrix changes the determinant's sign. Adding a multiple of a row to another row does not change the value of the determinant. Multiplying a row by a nonzero constant changes the determinant to that constant times the original determinant. These rules apply to square matrices. Note that when you multiply a row by a nonzero constant, you can have a factor that is the reciprocal of that constant, so your determinant equals the original determinant. For instance, suppose you have a square matrix A and its determinant is |A|; you multiply one of the rows by k, so then your new matrix is B, and its determinant is |B|. Furthermore, |A| = (k^-1)|B|. In other words, if you have a row you want to divide by 2, you can factor out that 2, so |A| = (2)|B|. You can also have a -1 factor when you perform a row swap. These rules also apply for elementary column operations. Two matrices are said to be column-equivalent if one can be obtained from the other by elementary column operations. If you have a column or row of zeros, then the determinant is zero. Conditions that yield a zero determinant (A is a square matrix): An entire row or column consists of zeros Two rows or columns are equal
file:///Users/admin/Desktop/linear%20algebra/MAT210-3.HTM[10/10/09 5:05:02 PM]

The determinant of a matrix:

One row or column is a multiple of another row or column Elementary row/column operations is a preferred way to finding the determinant as opposed to using cofactor expansion, because the number of operations required for cofactor expansion grows like n!. A good way of finding the determinant is to use elementary row/column operations to reduce a row or column to one element, and then continue with cofactor expansion. Properties of determinants: The determinant of A and B is a and b respectively; the determinant of the product AB is a*b. This can include the product of any finite number of matrices like |A1A2A3 ... Ak| = |A1|*|A2|*|A3|*...*|Ak|. If A is an n x n matrix and c is a scalar, then the determinant of cA is given by det(cA) = (c^n)*det(A). It is important to note that the sum of the determinants of two matrices is not usually equal to the determinant of the sum of two matrices. If the determinant of a matrix is equal to zero then it is singular. In other words, a square matrix A is invertible if and only if det(A) != 0. If A is invertible then det(A^-1) = 1/det(A). Equivalent conditions for a nonsingular matrix (A is an n x n matrix, then the follow statements are equivalent): 1. 2. 3. 4. 5. 6. A is invertible Ax = b has a unique solution for every n x 1 matrix b Ax = 0 has only the trivial solution A is row-equivalent to In A can be written as the product of elementary matrices det(A) != 0

This list can help you determine whether or not a system has a unique solution (determinant is not zero). If A is a square matrix, then det(A) = det(A^T). Introduction to eigenvalues: If you have some n x n matrix A, then there may exist some number l such that A times some n x 1 matrix x is equal to l times x. l is called the eigenvalue and x is the eigenvector. Any nonzero multiple of x can be used with l. To find the eigenvalues and eigenvectors use the equivalent equation (lI - A)x = 0. This homogeneous system of equations has nonzero solutions if and only if the coefficient matrix (lI - A) is singular. Once you have the eigenvalues you can use Gaussian elimination to get the eigenvectors. Example:

file:///Users/admin/Desktop/linear%20algebra/MAT210-3.HTM[10/10/09 5:05:02 PM]

The determinant of a matrix:

|lI - A| = 0 is called the characteristic equation. The adjoint of a matrix: The transpose of a matrix of cofactors is called the adjoint of a matrix. For each element aij of the matrix A there is a cofactor cij; the related cofactor goes in the same part of the cofactor matrix. The utility of adjoints is that they can be used to find the inverse. The inverse of the matrix A is equal to the adjoint of A over the determinant of A. This method is not the fastest way to get the inverse, but it's another way. Cramer's rule: Cramer's rule is named after Gabriel Cramer (1704-1752) and it is a formula to solve a system of linear equations. This method is perfect for calculators, since you can quickly solve a system. First you find the determinant of the coefficient matrix. Then you find a special determinant for each variable you want to find. It is that same coefficient matrix, but you replace the column represented by the variable you are solving for with the column vector of constants. For example:

file:///Users/admin/Desktop/linear%20algebra/MAT210-3.HTM[10/10/09 5:05:02 PM]

The determinant of a matrix:

Some applications of determinants: A triangle has three points, because it has two legs and a base; the legs meet at one point and they also touch the opposite ends of the base. These three points can be used to find the area, if you plug them into a matrix. Each x will go in the first column and each y will go in the next column. The last column just has ones. Find the determinant of this and multiply it by 1/2 to get the area (it may be -1/2 depending on the sign). If this same formula is used for any three points and the determinant happens to be zero, then these points are on the same line. Suppose you have two points, and you want to find the line for them, you can use the last formula too. You have to add a row in there to make your matrix 3 x 3. This row contains x in a11, y in a12, and 1 in a13. Setting the determinant of this matrix to zero, and doing cofactor expansion will give you the equation. The same kind of formula has application to 3D space too. If you have 4 points and each one represents a vertex of a tetrahedron (a 3D triangle), then you can find the volume of the tetrahedron. Like the triangle, each x is in the first column, each y is in the 2nd column, each z is in the 3rd column and the last column has 1s. This will give you a 4 x 4 matrix. The determinant of this times 1/6 is the volume (multiply by -1/6 if negative). If these 4 points are on the same plane then the determinant of the matrix is zero (just like for the collinear points). You can get the equation of a plane if you have three points. The first row has x for a11, y for a12, z for a13, and 1 for a14. The second row has the first coordinates, the third row has the second coordinates, and forth row has the last coordinates. Remember that last column (column 4) is all 1s. Set the determinant of this to zero, and use expansion of cofactors to find the linear equation.

file:///Users/admin/Desktop/linear%20algebra/MAT210-3.HTM[10/10/09 5:05:02 PM]

VECTOR SPACES

VECTOR SPACES Vectors in R^n: A vector in the plane is represented by an initial point and terminal point. The initial point is at the origin and these two points are connected by a directed line segment. The vector is represented by the ordered pair of the terminal point, if its initial point is at the origin. The x and y coordinates of the vector are called the components.Two vectors in a plane are equal if and only if their components are equal. Vector addition is accomplished by adding the corresponding components. To do this graphically, draw the two vectors on a coordinate plane, and make a parallelogram from them. The diagonal of that parallelogram is the sum of those vectors.A zero vector has a terminal point at (0,0). If you add this vector to another vector, you get that other vector, so it doesn't change it. If you add two vectors and their corresponding components are opposite of each other, then you get a zero vector. Scalar multiplication is accomplished by multiplying a constant by each component. You would be interested in doing this if you are multiplying a vector by a constant. A scalar is a real number, and if it is negative then the vector will point in the opposite direction. Vector subtraction is achieved with vector addition, the one you are subtracting with is multiplied by the scalar -1. Properties of vector addition and scalar multiplication (u, v, and w are vectors in a plane, and c and d are scalars): 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. u + v is a vector in a plane (closure under addition) u + v = v + u commutative property of addition (u + v) + w = u + (v + w) associative property of addition u+0=u u + (-u) = 0 cu is a vector in a plane (closure under scalar multiplication) c(u + v) = cu + cv distributive property (c + d)u = cu + du distributive property c(du) = (cd)u 1(u) = u

Vectors can exist in different spaces, like 1 space, 2 space, 3 space, 4 space, and so on. What space the vector exists in is represented by R^n, where n is the space it is in. Vector addition and scalar multiplication are called standard operations in R^n. A vector in n-space is represented by an ordered n-tuple. For instance, a vector in 4-space can be shown with (x1,x2,x3,x4). A zero vector in R^n is (0,0,0, ... , 0). The properties listed above, also apply to vectors in R^n, just replace the word plane with R^n. 0 (zero vector) is called the additive identity, and -v is the additive inverse of v in R^n. Properties of additive identity and additive inverse (v is a vector in R^n, and c is a scalar): 1. 2. 3. 4. 5. 6. The additive identity is unique. That is, if v + u = v, then u = 0 The additive inverse of v is unique. That is, if v + u = 0, then u = -v 0v = 0 c0 = 0 if cv = 0, then c = 0, or v = 0 -(-v) = v

Writing a vector as a sum of scalar multiples of other vectors is called a linear combination. For instance, if you have four vectors x, u, v, and w, and x = au + bv + cw, then solve for a, b and c using a system of equations. au1 + bv1 + cw1 = x1, au2 + bv2 + cw2= x2, and au3 + bv3 + cw3 = x3. Since you have 3 equations you can solve for a, b, and c, and represent x as the linear combination au + bv + cw.

file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

VECTOR SPACES

You can represent vectors in R^n, as column or row matrices, so you can perform vector addition, and scalar multiplication. Vector spaces: The ten properties above can apply to other quantities such as functions, polynomials, and matrices. Any set of vectors that satisfies these properties is called a vector space. Let V be a set on which two operations (vector addition and scalar multiplication) are defined. If the following axioms are satisfied for every u, v, and w in V and every scalar (real number) c and d, then V is called a vector space. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. u + v is in V (closure under addition) u + v = v + u (commutative property) u + (v + w) = (u + v) + w (associative property) V has a zero vector 0 such that for every u in V, u + 0 = u (additive identity) For every u in V, there is a vector in V denoted by -u such that u + (-u) = 0 (additive inverse) cu is in V (closure under scalar multiplication) c(u + v) = cu + cv (distributive property) (c + d)u = cu + du (distributive property) c(du) = (cd)u (associative property) 1(u) = u (scalar identity)

A vector space consists of four entities: a set of vectors, a set of scalars, and two operations. R^2 with the standard operations is a vector space, and so is R^n with standard operations, because you are referring to the set of ordered ntuples of real numbers. m x n matrices are also in the vector space. Polynomials can be shown to exist in the vector space. If you have a vector space of polynomials of degree n, then it also includes polynomials of degree less than n, and the zero polynomial (which is 0x^n + 0x^n-1 + ... 0). The set of continuous functions from - infinity to positive infinity is in the vector space, because the sum of two continuous functions is continuous, and a scalar times a continuous function is continuous. Here is a list of common vector spaces: R = set of all real numbers R^2 = set of all ordered pairs R^3 = set of all ordered triples R^n = set of all ordered n-tuples C(-,) = set of all continuous functions on the real line C[a,b] = set of all continuous functions in the closed interval [a,b] P = set of all polynomials Pn = set of all polynomials with degree less than or equal to n Mmn = set of all m x n matrices Mnn = set of all n x n matrices Properties of scalar multiplication: 1. 2. 3. 4. 0v = 0 if cv = 0 then c = 0 or v = 0 c0 = 0 (-1)v = -v

To show that a set is not vector space you only need to find one axiom that is false. The set of integers is not a vector space, because a scalar could be a rational number, and scalar multiplication yields a rational number which is not an integer. The set of all polynomials with degree of exactly two is not a vector space either, because it isn't closed under addition. x^2 + -x^2 + x is not a second degree polynomial (v1 is x^2 and v2 is -x^2 + x). Vector subspaces:

file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

VECTOR SPACES

A subset of a vector space is a subspace if it's a vector space itself. A nonempty subset W of a vector space V is called a subspace if W is a vector space under the operations of addition and subtraction defined in V. If W is a subspace then it must be closed under the operations inherited from V. Specifically, if W is a subspace of V, then it must be a nonempty subset of V, and this implies that W is a subspace of V if and only if: 1. If u and v are in W, then u + v is in W 2. If u is in W and c is any scalar, then cu is in W The above 2 tests can save you some work in determining if W is a subspace, but this assumes that W is already a subset of V. If you did not know that W was a subset, then you must prove all 10 axioms, so you know that W is a vector space. If W is a subspace of V, then they both have the same zero vector. A vector space with just the zero vector is called the zero subspace. V is a subspace of itself, so all vector spaces have at least two subspaces (the other is zero subspace). Proper subspaces are the other subspaces besides the two previously mentioned. The intersection of two subspaces is also a subspace. Spanning sets and linear independence: Sometimes vectors in a set can be written as a linear combination of other vectors in that set. If every vector can be written as a linear combination of other vectors in some subset S from the vector space V, then S is called a spanning set. A spanning set could be {(1,0,0),(0,1,0),(0,0,1)} for the vector space R^3. When you are testing for spanning sets, set up the linear combination, and make it equal to any vector from V (use variables, so it is clear that it could be any vector). If this coefficient matrix has a determinant then the set S spans V (is a spanning set), otherwise the determinant is zero and S is not a spanning set. The span of a set is the set of all linear combinations of the vectors in S. The span of a set S is denoted by span(S) or span{v1,v2, ..., vk}. The span of any finite nonempty subset of a vector space V is a subspace of V. Span(S) is the smallest subspace that contains S -- any other subspace that contains S, must contain Span(S). A set of vectors {v1,v2,... vk} (S) in a vector space V is called linearly independent if the vector equation cv1 + c2v2 + ... ckvk = 0 only has the trivial solution, otherwise it also has a nontrivial solution and it is called linearly dependent. A set S is linearly dependent if and only if at least one of the vectors in that set can be written as a linear combination of the other vectors in S. Two vectors in a vector space are linearly dependent if and only if one is a scalar multiple of the other. The zero vector is always a scalar multiple of another vector from the vector space. Basis and dimension: A set of vectors S in a vector space V is called a basis for V, if S spans V, and S is linearly independent. If a vector spaceV has a basis consisting of a finite number of vectors, then V is finite dimensional. If the basis is an infinite number of vectors then V is infinite dimensional. A vector space V consisting of all polynomials is infinite dimensional, and the vector space consisting of only the zero vector is finite dimensional. {(1,0,0),(0,1,0),(0,0,1)}, this set is the standard basis for R^3. That means it spans the entire vector space R^3, and it is linearly independent. There is a standard basis for each R^n -- just follow the pattern implied by the standard basis for R^3. The standard basis for a P3 vector space (polynomials of degree 3 and less) is (1,x,x^2,x^3). These 4 vectors span the entire vector space P3. There is only one term in each vector for that spanning set. If you were to write a system of equations, so you can prove if it actually spans P3, you would see that:

file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

VECTOR SPACES

c1 + c2x + c3x^2 + c4x^3 = a0 + a1x + a2x^2 + a3x^3 The right side represents every possible vector from that vector space. Because you have this equality, c1 = a0, c2 = a1, c3=a2, and c4 = a3. This means you can rewrite the linear combination in the general form for a third degree polynomial. (1,x,x^2,...,x^n) happens to be the standard basis for polynomials of degree n and less. The standard basis for as 2 x 2 matrix is:

You can probably guess what the standard basis is for other square matrices and m x n matrices (you have n*m matrices in your spanning set, and each matrix has only a 1, and all zeros). If S is the basis for a vector space V, then each vector in V can only be made from one linear combination from S. If S is the basis for a vector space V, then if you define a linear combination using more vectors than what is in S, then that linear combination is linearly dependent. It is useful to note that every vector space that has a basis with n vectors must only have a basis with n vectors. This number n is called the dimension of V, denoted as dim(V). If the vector space only has the zero vector then dim(V) is zero. The dimension of R^n is n, Mmn (vector space of matrices) is m*n, and Pn (vector space of polynomials of power n and less) is n+1. A subspace is finite dimensional, and has dimension less than or equal to the dimension of the vector space that contains it. You can find the dimension by creating a spanning set of linearly independent vectors for that subspace. Rank of a Matrix and system of linear equations: The row and column vectors of an n x m matrix can span a subspace of R^n or R^m. The subspace spanned by the row vectors of a m x n matrix is a subspace of R^m and is called a row space. Furthermore, a column space is the subspace of R^n spanned by the column vectors of an m x n matrix. If two matrices are row-equivalent, then their row spaces are equal. This does not apply to column spaces, because elementary row operations will change the column space. If a matrix A is row-equivalent to a matrix B in reduced row-echelon form, then the nonzero row vectors of B form a basis for the row space of A. The nonzero row vectors of B form a linearly independent set, if B is in reduced rowechelon form. In order to find the basis for a row space you should transform that matrix of row vectors into reduced row-echelon form. The row vectors of this matrix in row-echelon is the basis for the row space. To find the basis for a column space transpose the matrix, and put it in reduced row-echelon form. You can do this, because the column space of A is equal to the row space of A^T. An m x n matrix A has the same dimension for its row and column space. This dimension of the row or column vector is called the rank of the matrix A, denoted rank(A). The set of all solutions to the homogenous system Ax = 0 is a subspace of R^n called the nullspace (solution space) of A (m x n matrix) and is denoted N(A). The dimension of the nullspace is called the nullity of A. When you are trying to find the solution to this homogenous system, you will probably find that the matrix reduces to a set of equations less than the number of variables. So you will have to assign one or more of the variables as free variables to make it work. All solutions for Ax = 0 are linear combinations of the vectors from that basis. These vectors are linearly independent if the matrix was in reduced row-echelon form. The number of columns of a matrix is equal to the rank plus the
file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

VECTOR SPACES

nullity. The solution set for Ax = b is not a subspace because it does not contain the zero vector. But x = xh + xp, so this contains the solutions for x; xh is the x from Ax = 0, and xp is a particular solution for Ax = b. The system of linear equations Ax = b is consistent if and only if b is in the column space of A. Coordinates and change of basis: If B is a basis for a vector space V, then every vector x from V can be expressed by only one linear combination from B. The coefficients in the linear combination are called coordinates of x relative to B.When dealing with coordinates the order of the basis is important. The ordered basis B has a set of coordinates that make up a coordinate matrix (coordinate vector or column matrix) for each vector x in V. There is a way you can switch between two bases using an equation. Suppose B and B' are bases that span some vector space V. The x coordinate relative to B is [x}B and the x coordinate relative to B' is [x]B'. Let P be the transition matrix; [x]B = P[x]B', and P^-1[x]B = [x]B'. If P is the transition matrix from a basis B' to a basis B in R^n, then P is invertible and the transition matrix from B to B' is given by P^-1. If B is {v1, v2, v3) and B' is (u1, u2, u3), and they are bases for a vector space V, and v1 = c11u1 + c21u2 + c31u3, and v2 = c12u1 + c22u2 + c32u3, and v3 = c13u1 + c23u2 + c33u3, then the transition matrix from B to B' is equal to Q. Q is the set of row vectors that are the coefficients of the previously mentioned equations. So the first row vector is (c11,c12,c13), and the second one is (c21,c22,c23), and the third one is (c31,c32,c33). This pattern can be expanded for any dimension. Another way to find the transition matrix from B to B' is to adjoin B' and B. So you start with the n x 2n matrix [B' | B] and use Gauss-Jordan elimination to get [In | P^-1]. P^-1 is the transition matrix from B to B'. If B is the standard basis, and B' is a nonstandard basis, then [B' | B] can be written as [B' | In], and the answer is [In | P^-1]. If B' is the standard basis and B is the nonstandard basis, then [B' | B] can be written as [In | B], and the answer is right there, because B is P^-1. To find the transition matrix from B' to B you just switch it around to [B | B']. These coordinate matrices can be used to represent vectors, so instead of referring to some vector the normal way, you can represent it this way, as long as the relevant basis is known. Applications of vector spaces: There may be some point where you need to find the solution for some homogenous differential equation that follows the form y^(n) + g(n-1)(x)y^(n-1) + ... + g1(x)y' + g0(x)y = 0. This equation has a number of functions that can be used to solve this. The linear combination of these functions is equal to y, the general solution. The n denotes what order differential equation it is. The order of these equations determines how many solutions (functions) you need. If the linear combination of these solutions is linearly independent, then every solution to this equation is of the form y = c1y1 + c2y2 + ... + cnyn. If you find the Wronskian for these functions, it will tell you whether they are linearly independent or not. The Wronskian is the determinant of the matrix which has y1 in c11, y2 in c12, y1' in c21, and y2' in c22. If this determinant is not zero then the functions are linearly independent. You can see that the first row of this matrix contains each of the functions, and the next row contains their derivatives, and the rows after that contain the derivatives of the functions above them. The number of rows in this matrix is equal to n. If the determinant is zero, then the Wronskian is said to be identically equal to zero. Every conic in the xy-plane has an equation of the form ax^2 + bxy + cy^2 + dx + ey + f = 0. Drawing this graph is easy as long as b is zero, because the conic axes are parallel to the coordinate axes. For circles, ellipses and hyperbolas

file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

VECTOR SPACES

the center is (h,k). (h,k) is the vertex for parabolas. Standard forms of equations of conics: Circle: (r = radius): (x - h)^2 + (y - k)^2 = r^2 Ellipse: (2a = major axis length, 2b = minor axis length) [(x - h)^2]/(a^2) + [(y - k)^2]/(b^2) = 1 (fat ellipse) [(x - h)^2]/(b^2) + [(y - k)^2]/(a^2) = 1 (thin ellipse) Hyperbola: (2a = transverse axis length, 2b = minor axis length) [(x - h)^2]/(a^2) - [(y - k)^2]/(b^2) = 1 (horizontal hyperbola) [(y - k)^2]/(a^2) - [(x - h)^2]/(b^2) = 1 (vertical hyperbola) Parabola: (p = directed distance from vertex to focus) (x - h)^2 = 4p(y - k) focus (h, k+p) (vertical parabola) (y - k)^2 = 4p(x - h) focus (h + p, k) (horizontal parabola) You can rewrite the conic equation by rotating the coordinate axes counterclockwise through angle theta, which is cot2q = (a - c)/b.So the new function is a'(x')^2 + c'(y')^2 + d'x' + e'y' + f' = 0. x = x'cosq - y'sinq, and y = x'sinq + y'cosq.

file:///Users/admin/Desktop/linear%20algebra/MAT210-4.HTM[10/10/09 5:05:03 PM]

INNER PRODUCT SPACES

INNER PRODUCT SPACES Length and Dot Prodct in R^n: The length of a vector v = (v1, v2), denoted ||v||, is defined as [v1^2 + v2^2]^(1/2). This is found by using the Pythagorean theorem with the length as the hypotenuse, and the legs are the components of the vector. The length of vector is also called the magnitude or norm. The norm is zero if and only if the vector is the zero vector. The norm cannot be negative.The length of a scalar multiple ||c * v|| is |c| * ||v||. If the norm of a vector is 1, then that vector is a unit vector. Each vector in the standard basis has a length of 1 and is called a standard unit vector. {i,j} = {(1.0),(0,1)}. Two nonzero vectors are parallel if one is a scalar multiple of the other. To find the unit vector that is going in the same direction as some nonzero vector v divide it by its length. To find the distance between two vectors u and v, denoted d(u,v), is [(u1 - v1)^2 + (u2 - v2)^2]^(1/2). This distance can also be viewed as the length of the vector u - v. Using the law of cosines you can find the angle between two vectors. Cosq = (u1v1 + u2v2)/(||u||*||v||).The numerator of that fraction is called the dot product of u and v. A dot product is a scalar quantity. Properties of the dot product (u, v and w are vectors in R^n and c is a scalar): u dot v = v dot u u dot (v + w) = u dot v + u dot w c(u dot v) = (cu) dot v = u dot (cv) v dot v = ||v||^2 v dot v 0 and v dot v = 0 if and only if v = 0 If u and v are vectors in R^n, then |u dot v| ||u|| * ||v||. This is called the Cauchy-Schwarz Inequality. This inequality must work otherwise you cannot get an angle from the above equation. The angle is between or equal to 0 and p. Two vectors meet at a right angle if and only if their dot product is zero. These vectors are orthogonal (or perpendicular). The length of one side of a triangle cannot be larger than the sum of the other two sides. The Cauchy-Schwarz can be used to prove this traingle inequality ||u + v|| ||u|| + ||v||. If u and v are vectors in R^n, then u and v are orthogonal if and only if ||u + v||^2 = ||u||^2 + ||v||^2. They can be orthogonal if u dot v is zero. It is interesting to note that the dot product of two vectors can be represented by matrix multiplication if you transpose the first matrix (both started as column vectors). So u dot v is u^Tv. Inner product spaces: The dot product is a type of inner product for vectors. It is also called the Euclidean inner product. There are inner products for vector spaces in general. The general inner product is denoted <u,v>. In order for a function to qualify as an inner product it must follow a set of axioms (u, v and w are vectors in vector space V, and c is any scalar): 1. 2. 3. 4. <u,v> = <v,u> <u,v + w> = <u,v> + <u,w> c<u,v> = <cu,v> <v,v> 0, and <v,v> = 0 if and only if v = 0

file:///Users/admin/Desktop/linear%20algebra/MAT210-5.HTM[10/10/09 5:05:04 PM]

INNER PRODUCT SPACES

A vector space that has an inner product is called an inner product space.Properties of inner products (u, v, and w are vectors in an inner product space V, and let c be any real number): 1. <0,v> = <v,0> = 0 2. <u + v, w> = <u,w> + <v,w> 3. <u,cv> = c<u,v> If u and v are vectors in an inner product space V, then the norm of u is [<u,u>]^(1/2). Also, the distance between u and v is ||u - v||. The angle between the nonzero vectors u and v is given by cosq = <u,v>/(||u||*||v||).u and v are orthogonal if <u,v> = 0. If ||v|| = 1 then v is a unit vector. The Cauchy-Schwarz Inequality, triangle inequality, and pythagorean theorem all apply to vectors in an inner product space. If u and v are vectors in an inner product space V such that v is not zero, then the orthogonal projection of u onto v is projuv = [<u,v>/<v,v>]*v. The orthogonal projection of u onto v is a vector that is a scalar multiple of v that is orthogonal with u. The distance of this orthogonal projection is less than the d(u,cv) (c is not equal to <u,v>/<v,v>) Orthonormal bases: Gram-Schmidt process: The standard basis for R^3 has vectors that are orthogonal with each other and they are unit vectors. The GramSchmidt orthonormalization process is used to create bases that have mutually orthogonal unit vectors. Sets of vectors that have these 2 characteristics(orthogonal & unit vectors) are called orthonormal.

To create a nonstandard orthonormal basis from a standard basis use this formula: B = {(cosq,sinq,0),(-sinq,cosq,0),(0,0,1)} The new basis is called B. This formula just rotates the standard basis about the z-axis. Two vectors u and v are orthogonal if the inner product <u,v> is 0. A vector is a unit vector if its length is 1. So if a vector u is a unit vector then (<u,u>)^(1/2) is 1. If a set of nonzero orthogonal vectors is from an inner product space, then this set is linearly independent. If V is an inner product space of dimension n, then any orthogonal set of n nonzero vectors is a basis for V. If B is an orthonormal basis for the inner product space V, then the coordinate representation of a vector w is <w,v1>v1 + <w,v2>v2 + ... + <w,vn>vn. The coordinates of w relative to B are called the Fourier coefficients of w relative to B. {w}B = [<w,v1>,<w,v2>,...,<w,vn>]^T. One of the advantages of orthonormal bases is that allow easy coordinate representations. The process of finding an orthonormal basis involves 3 steps: 1. Begin with a basis for the inner product space. 2. Convert the basis to an orthogonal basis 3. Normalize each vector in the orthogonal basis to form an orthonormal basis. Specifically, it goes like this: 1. Let B = {v1,v2, ..., vn} be a basis for an inner product space V 2. Let B' = {w1,w2, ..., wn}, where wi is given by w1 = v1 w2 = v2 - [<v2,w1>/<w1,w1>]w1 . . .

file:///Users/admin/Desktop/linear%20algebra/MAT210-5.HTM[10/10/09 5:05:04 PM]

INNER PRODUCT SPACES

wn = vn - [<vn,w1>/<w1,w1>]w1 - [<vn,w2>/<w2,w2>]w2 - ... - [<vn,wn-1>/<wn-1,wn-1>]wn-1 B' is an orthogonal basis for V. 3. Let ui = wi/||wi||. Then B'' = {u1,u2,...,un}. B'' is an orthonormal basis for V, and span(B) is equal to Span(B'').

There is another way to apply this orthonormalization process. You can get the unit vector for w1 right away, and then you plug that into the equation to get w2. If you do it this way, w2 is v2 - <v2,u1>u1. u2 is then equal to w2/||w2||. If you have a inner product space that is the definite integral from -1 to 1 of p(x) times q(x), and you are trying to convert the standard basis for P3 to a orthonormal basis, the first three polynomials of the orthonormal basis are called normalized Legendre polynomials. Mathematical models and least squares analysis: Trying to find the best line for a set of points involves an inconsistent set of equations. This means that the points are not collinear. It is impossible to find a x that fits such a group of equations, but you can find a x that minimizes the norm of the error ||Ax- b||. The solution to this problem is the least squares regression line. y = c1x + c0. Finding the minimum length from ||Ax - b|| is the same as ||Ax - b||^2. Two subspaces of R^n are orthogonal to each other if every vector from the first one is orthogonal with every vector from the second one. If two subspaces of R^n are orthogonal, then their only intersection is the zero vector. The set of vectors that are orthogonal to every vector in subspace S from R^n is called the orthogonal complement of S. The orthogonal complement of R^n is the trivial subspace {0}, and the orthogonal complement of the zero vector is R^n. In general, the orthogonal subspace of a subspace of R^n is a subspace of R^n. To find the orthogonal complement of a subspace of R^n you must find the nullspace of a matrix. If a vector from a complement of subspace S is dotted with the vectors that span S, it is zero. The nullspace of this equation is the subspace that is the complement of S. If R^n has two subspaces S and Sc (complement of S), then every vector from R^n be written as a sum of vectors from S and Sc (s1 + s2 is x). So R^n is a direct sum of subspaces S and Sc SSc = R^n. Properties of orthogonal subspaces(S is a subspace of R^n and S^ is the orthogonal complement of S): 1. Dim(S) + Dim(S^) = n 2. S S^= R^n 3. (S^)^ = S

Every vector from R^n can be acquired from the sum of a vector from S and its complement, so v is in R^n, and v = v1 + v2. v1 S, and v2 S^. It is possible to find the projection of a vector onto a vector space. Suppose you have a vector v that represents every possible vector from R^n. v = v1 + v2, v1 is from subspace S1, and v2 is from subspace S2. S1 and S2 are subspaces of R^n. The projection of v onto S1 is the closest vector from S1 to v. Since v and v1 are sides of a right triangle, v2 is perpendicular to v1, and it happens to be from a subspace that is orthogonal to S1 anyway. Formula to a projection onto a subspace: projvs = (v dot u1)u1 + (v dot u2)u2 + ... + (v dot ut)ut (projection of v onto s) {u1,u2, ... , ut} is a orthonormal basis for the subspace S The distance between v and the orthogonal projection of v onto S is smaller than the distance between v and any vector from S. In other words, ||v - projvs|| < ||v - u||.

file:///Users/admin/Desktop/linear%20algebra/MAT210-5.HTM[10/10/09 5:05:04 PM]

INNER PRODUCT SPACES

From an m x n matrix A you derive 4 important subspaces. They are: N(A) nullspace of A N(A^T) = nullspace of A^T R(A) = column space of A R(A^T) = column space of A^T

More on fundamental subspaces (A is an m x n matrix): 1. 2. 3. 4. R(A) and N(A^T) are orthogonal subspaces of R^m R(A^T) and N(A) are orthogonal subspaces of R^n R(A)N(A^T) = R^m R(A^T)N(A) = R^n

Finally, we are up to the least squares problem. Suppose you want to find the best fit for a set of points (1,0), (2,1), and (3,3). In order to find this line we must set up a matrix. The first column of the matrix is just three rows of 1s. The 2nd column is the x coordinates. This makes up the matrix A. Matrix b is a column vector of the y coordinates. Since this is not a consistent system of equations Ax is not equal to b. However, Ax draws from a vector space of possibilities, and the closest thing to b is the projection of b onto Ax's vector space. Instead of writing Ax, we will write projbs (let the column space of A be S). If we let Ax' = projbs, then we can say b = Ax' + b - projbs or Ax' - b = projbs - b. projbs - b is orthogonal to Ax' (draw a triangle), so it is in the nullspace of A^T. Therefore, A^T(Ax' b) = 0. This of course can be simplified to A^TAx' = A^Tb. Solve it and x1 is the constant, and x2 is the coefficient for x. Applications of inner product spaces: The cross product is a function that produces a vector (in R^3) that is orthogonal to the two vectors that it uses as parameters. Suppose you have two vectors u and v from R^3. u cross v is (u2v3 - u3v2)i - (u1v3 - u3v1)j + (u1v2 u2v1)k. This may be hard to remember, so there is another way to go about this. u cross v is also equal to:

Algebraic properties of the cross product (u, v, and w are vectors in R^3, and c is a scalar): 1. 2. 3. 4. u cross v = -(v cross u) u cross (v + w) = (u cross v) + (u cross w) c(u cross v) = cu cross v = u cross cv u cross 0 = 0 cross u = 0

file:///Users/admin/Desktop/linear%20algebra/MAT210-5.HTM[10/10/09 5:05:04 PM]

INNER PRODUCT SPACES

5. u cross u = 0 6. u dot (v cross w) = (u cross v) dot w

Geometric properties of the cross product (u and v are nonzero vectors in R^3): 1. 2. 3. 4. u cross v is orthogonal to both vectors The angle q between vectors v and u is ||u cross v|| = ||u|| ||v|| sinq. u and v are parallel if and only if u cross v is zero The parallelogram having u and v as adjacent sides has an area of ||u cross v||

file:///Users/admin/Desktop/linear%20algebra/MAT210-5.HTM[10/10/09 5:05:04 PM]

LINEAR TRANSFORMATIONS

LINEAR TRANSFORMATIONS

Introduction to linear transformations: A function that maps vector space V to vector space W is denoted T: V -> W. V is called the domain of T, and W is called the codomain of T. The function can also be written like T(v) = w; v is from V, and w is from W. w is called the image, and v is called the preimage of w. The set of all images is called the range of T. A function such as T: V -> W is a linear transformation of V into W if these two properties are true for all u and v in V, and any scalar c. 1. T(u + v) = T(u) + T(v) 2. T(cu) = cT(u)

Linear transformations are operation preserving, because you get the same result whether you do vector addition before or after, or scalar multiplication before or after. There are two simple linear transformations; they are the zero transformation and the identity transformation. The zero transformation is equal to the zero vector for all elements from the domain. The identity transformation's image is equal to the preimage for all images. Properties of linear transformations (T is a linear transformation from V to W, and u and v are in V): 1. 2. 3. 4. T(0) = 0 T(-v) = -T(v) T(u - v) = T(u) - T(v) If v = c1v1 + c2v2 + ... + cnvn, then T(v) = c1T(v1) + c2T(v2) + ... + cnT(vn)

You can use a matrix to define a linear transformation. Let A be an m x n matrix. The function T is defined by T(v) = Av and is a linear transformation from R^n into R^m. The kernel and range of a linear transformation: The kernel of a linear transformation is simply the set of vectors that make the function T equal the zero vector. The kernel is denoted ker(T). The kernel is a subspace of V, which is the domain of the function T: V -> W. If T is T: R^n -> R^m or T(x) = Ax, then the ker(T) is equal to the solution space of Ax = 0. The range of T, denoted range(T), is a subspace of W for the the function T: V -> W. If T is T: R^n -> R^m or T(x) = Ax, then the range(T) is the column space of A. The dimension of the kernel is called the nullity of T. The dimension of T's range is called the rank of T. The rank(T) + nullity(T) = n. T is one-to-one if and only if ker(T) = {0}. T is onto if and only if the rank of T is equal to the dimension of W (T is T: V -> W). If V and W both of dimension n for a function T: V -> W, then T is one-to-one if and only if it is onto. A linear transformation is an isomorphism if it is one-to-one and onto, and so V and W are isomorphic to each other (T is T: V -> W). Two finite dimensional vector spaces V and W are isomorphic is and only if they are the same dimension.

file:///Users/admin/Desktop/linear%20algebra/MAT210-6.HTM[10/10/09 5:05:05 PM]

LINEAR TRANSFORMATIONS

Matrices for linear transformations: The standard basis for R^n can be used to find a matrix for a linear transformation. This matrix replaces the algebraic expressions that make up T, so you can use matrix multiplication to get the image. Each column vector of the linear transformation's matrix comes from the image of standard basis' column vectors. This matrix is called the standard matrix. B, the standard basis, is {e1,e2,... , en}. T(e1) is the first column vector of the standard matrix, and T(e2) is the 2nd column vector, and so on. T is the function T: R^n -> R^m, or T(v) = Av for every v in R^n. The standard matrix for the zero transformation from R^n into R^m is the m x n zero matrix and the standard matrix for the identity transformation from R^n into R^m is In. The composition T of T1:R^n -> R^m with T2:R^m -> R^p is defined by T(v) = T2(T1(v)) where v is a vector in R^n. This composition is denoted by T = T2 T1. If this composition is a linear transformation, and the standard matrix for T1 is A1, and the standard matrix for T2 is A2, then the standard matrix for T is A2 times A1. If T1:R^n -> R^n and T2:R^n -> R^n are linear transformations such that for every v in R^n T2(T1(v)) = v and T1(T2(v)) = v , then T2 is called the inverse of T1 and T1 is said to be invertible. Let T:R^n -> R^n be a linear transformation with standard matrix A, then the following conditions are equivalent: 1. T is invertible 2. T is an isomorphism 3. A is invertible And, if T is invertible with standard matrix A, then the standard matrix for T^-1 is A^-1. You can find a matrix for bases that are nonstandard. The procedure is similar to finding the standard matrix. Just plug in each vector from the nonstandard basis to get each column vector for the matrix A. Specifically, T is a function from V to W, and they are both vector spaces. The basis for V is B, and the basis for W is B', so T is a function that goes from basis B to B'. Therefore, [T(v)]B' = A[v]B for every v in V. When you use this function, you start with some vector, and then it has to converted to its coordinates relative to B. This yields a coordinate matrix relative to B' that needs to be converted, so it is the real T(v). Note that A is the matrix relative to B, because B is used to find this matrix. Transition matrices and similarity: The matrix of T relative to B is A, and the matrix of T relative to B' is A'. The transition matrix from B' to B is P, and the transition matrix from B to B' is p^-1. There are two ways to get from coordinate matrix [v]B' to coordinate matrix [T(v)]B'. One is A'[v]B' = [T(v)]B', and the other is P^-1AP[v]B' = [T(v)]B'. A' = P^-1AP; this equation can be helpful in finding A', so you have a matrix relative to B'. For two square matrices A and A', of order n, A' is said to be similar to A if there exists an invertible matrix P such that A' = P^-1AP. Properties of similar matrices (A, B, and C are square matrices of order n): 1. A is similar to A 2. If A is similar to B, then B is similar to A 3. If A is similar to B, and B is similar to C, then A is similar to C

file:///Users/admin/Desktop/linear%20algebra/MAT210-6.HTM[10/10/09 5:05:05 PM]

Eigenvalues and eigenvectors

EIGENVALUES AND EIGENVECTORS

Eigenvalues and eigenvectors: If A is an n x n matrix do there exist nonzero vectors x in R^n such that Ax is a scalar multiple of x. The scalar is denoted by the greek character lambda, and is called the eigenvalue of the matrix A, and the nonzero vector x is called an eigenvector of A corresponding to l. The equation looks like this Ax = lx. lx is parallel to x. The eigenvector cannot be zero, but the eigenvalue can be zero. Every scalar multiple of the eigenvector is an eigenvector of A. That means A(cx) = c(Ax) = c(lx) = l(cx). If two eigenvectors corresponding to an eigenvalue l, then the sum is also an eigenvector, because A(x1 + x2) = l(x1 + x2). The set of all eigenvectors of a given eigenvalue l, together with the zero vector, is a subspace of R^n. This subspace is called the eigenspace of l. (lI - A)x = 0 (I is the n x n identity matrix). This has nonzero solutions if and only if the determinant of (lI - A) is zero. This equation, |(lI - A)| = 0, is called the characteristic equation of A. It is equal to this polynomial l^n + c(n-1)l^n-1 + ... + c1l + c0. That polynomial is called the characteristic polynomial of A. A can only have n distinct eigenvalues (degree of characteristic polynomial). The homogenous systems you must solve to get the eigenvector always have at least one row of zeros, since they must have more than the trivial solution. Steps to finding the eigenvalues and eigenvectors (A is a n x n matrix): 1. Form the characteristic equation; it will be a polynomial equation of degree n in the variable l 2. Find the real roots of the characteristic equation. These are the eigenvalues of A 3. For each eigenvalue li, find the eigenvectors corresponding to li by solving the homogeneous system (liI - A)x = 0

The dimension of an eigenspace is equal to the number of eigenvectors corresponding to the eigenvalue. The multiplicity of an eigenvalue is the number of occurences of an eigenvalue such as (l - l1)^k being a factor of the characteristic polynomial. In that case l1 has multiplicity k. The multiplicity is always greater than or equal to the dimension. The eigenvalues for a triangle matrix n x n are the entries on the main diagonal. Linear transformations can be used to define eigenvectors and eigenvalues. l is called the eigenvalue of a transformation T: V -> V, if there is a nonzero vector x such that T(x) = lx. The vector x is called an eigenvector of T corresponding to l, and the set of all eigenvectors of l (with the zero vector) is called the eigenspace of l. Diagonalization: The diagonalization problem deals with determining whether, for a square matrix A, an invertible matrix P such that P^-1AP is diagonal. A and B are similar if there exists an invertible matrix P such that B = P^-1AP. Matrices that are similar to diagonal matrices are called diagonalizable. Every diagonal matrix D is diagonalizable, because the identity matrix can play the role of P, so D = I^-1DI. If A and B are similar n x n matrices, then they have the same eigenvalues. An n x n matrix A is diagonalizable if and only if it has n linearly independent eigenvectors. The eigenvectors show as column vectors in P.
file:///Users/admin/Desktop/linear%20algebra/MAT210-7.HTM[10/10/09 5:05:05 PM]

Eigenvalues and eigenvectors

Steps for diagonalizing an n x n matrix (Let A be an n x n matrix) 1. Find n linearly independent eigenvectors p1, p2, ... pn for A with corresponding eigenvalues l1, l2, ... , ln. If n linearly independent eigenvectors do not exist, then A is not diagonalizable 2. If A has n linearly independent eigenvectors, let P be the n x n matrix whose columns consist of these eigenvectors. That is, P = [p1|p2|...|pn]. 3. The diagonal matrix D=P^-1AP will have the eigenvalues l1, l2, ..., ln on its main diagonal (and zeros elsewhere). Note that the order of the eigenvectors used to form P will determine the order in which the eigenvalues appear on the main diagonal D.

When you find the eigenvalues and eigenvectors, there should be equal amounts of each, if A is to be diagonalizable. However, you can have multiples of an eigenvalue, and an eigenvalue can have more than one eigenvector as its basis. For a square matrix A of order n to be diagonalizable, the sum of the dimensions of the eigenspaces must be equal to n. If an n x n matrix A has n distinct eigenvalues, then the corresponding eigenvectors are linearly independent and A is diagonalizable. Linear transformations can be done in relation to eigenvalues and eigenvectors. A linear transformation T: V -> V is defined by T(x) = Ax = lx. The matrix A is the standard matrix; it is derived for the equations from the linear transformation, and it is relative to the standard basis. As you can see the domain of T are the eigenvectors, and the eigenvalues can be determined by diagonalizing A. If you find the characteristic polynomial for A, and then get the eigenvalues, you can solve for the bases for the eigenspaces. The bases form the column vectors of P, and A' can be found by P^-1AP. The main diagonal of this matrix A' has the eigenvalues. This will only be the case if you have the same amount of eigenvalues and eigenvectors, and the eigenvectors are linearly independent. Symmetric matrices and orthogonal diagonalization: A symmetric matrix is guaranteed to be diagonalizable. A square matrix A is symmetric if it is equal to its transpose. Some properties of nonsymmetric matrices: A nonsymmetric matrix may not be diagonalizable A nonsymmetric matrix can have eigenvalues that are not real A nonsymmetric matrix can have less linearly independent eigenvectors corresponding to an eigenvalue than the multiplicity of the eigenvalue

If a is an n x n symmetric matrix, then the following is true: A is diagonalizable All eigenvalues of A are real If l is an eigenvalue of A with multiplicity k, then l has k linearly independent eigenvectors. That is, the eigenspace of l has dimension k.

The last three properties are part of a theorem called the Real Spectral Theorem, and the set of eigenvalues of A is called the spectrum of A. A square matrix P is called orthogonal if it is invertible and P^-1 is equal to the transpose of P. An n x n P is orthogonal if and only if its column vectors form an orthonormal set. Let A be an n x n symmetric matrix. If l1 and l2

file:///Users/admin/Desktop/linear%20algebra/MAT210-7.HTM[10/10/09 5:05:05 PM]

Eigenvalues and eigenvectors

are distinct eigenvalues of A, then their corresponding eigenvectors x1 and x2 are orthogonal. The fundamental theorem of symmetric matrices states that if A is an n x n matrix, then A is orthogonally diagonalizable if and only if A is symmetric. A is orthogonally diagonalizable if there exists an orthogonal matrix P such that P^-1AP is equal to a diagonal matrix. The process of orthogonal diagonalization of a symmetric matrix (A is an n x n symmetric matrix): 1. Find all eigenvalues of A and determine the multiplicity of each 2. For each eigenvalue of multiplicity 1, choose a unit eigenvector (choose any eigenvector and then normalize it) 3. For each eigenvalue of multiplicity k 2 , find a set of k linearly independent eigenvectors. If this set is not orthonormal, apply the Gram-Schmidt orthonormalization process 4. The composite of steps 2 and 3 produces an orthonormal set of n eigenvectors. Use these eigenvectors to form the columns of P. The matrix P^-1AP = P^TAP = D will be diagonal (the main diagonal entries of D are the eigenvalues of A)

Applications of eigenvalues and eigenvectors: Eigenvalues and eigenvectors can be applied to population growth. A matrix represents the different age classes, the survival possibility, and average offspring for each age class. The age distribution vector has n entries, where n is equal to the number of age classes. Each entry of the age distribution vector says how many are in each age class. The first entry corresponds to the first age class, the 2nd entry corresponds to the 2nd age class and so on. The first row of the matrix represents how many offspring is produced at each age class. The first entry of the row corresponds to the first age class, the 2nd entry of the row corresponds to the 2nd age class, and so on. Here is the matrix:

b stands for the average offspring at each age class. p stands for the possibility of survival, so the member can go into the next age class. Multiplying the age distribution column vector by this matrix yields the next age distribution vector. Basically, it will tell you what will happen after one age class has elapsed. Each age class has equal length. The matrix is called the age transition matrix. Just starting with some arbitrary number of members can result in undesired results such as having different ratios of members at each age class interval. Using eigenvalues and eigenvectors you can maintain a ratio between age class intervals. To begin, you must have an age transition matrix like before. Solving for the eigenvalues will tell you how many members should be in the initial age distribution vector. This initial age distribution vector can be any of the eigenvectors. When you multiply the matrix and this vector, it will give you the next age distribution vector.

file:///Users/admin/Desktop/linear%20algebra/MAT210-7.HTM[10/10/09 5:05:05 PM]

Eigenvalues and eigenvectors

Eigenvalues and eigenvectors can be applied to a system of first-order linear differential equations of the form: y1' = a11y1 + a12y2 + ... + a1nyn y2' = a21y1 + a22y3 + ... + a2nyn . . . yn' = an1y1 + an2y2 + ... + annyn Each yi is a function of t and each yi' = dyi/dt. If you let y = [y1 y2 . . . yn]^T and y' = [y1' y2' . . . yn']^T, then y' = Ay. As you can guess the coefficients of yi go into the matrix A, and if it is diagonal then you can get the eigenvalues right away and solve the system, otherwise you must find P, and diagonalize A. To diagonalize A, find the eigenvalues and eigenvectors like you normally do with a matrix. The column vectors of the matrix P are made up of the eigenvectors. Like before, D (the diagonal matrix) can found with the formula P^-1AP. This will change the variables which is given by y = Pw, and y' = Pw'. Pw' = y' = Ay = APw, and from those you can get w' = P^-1APw. Eigenvectors and eigenvalues can assist you with solving the rotation of axes problem. For a conic whose equation ax^2 + bxy + cy^2 + dx + ey + f = 0 given by X = PX' (X is [x y]^T) eliminates the xy-term if P is an orthogonal matrix, with |P| = 1, that diagonalizes A. That is P^TAP is made up of the matrix of column vectors [l1 0]^T and [0 l2]^T. l1 and l2 are eigenvalues of A. The equation of rotated conic is given by l1(x')^2 + l2(y')^2 + [d e]PX' + f = 0 (X'= [x' y']^T)

file:///Users/admin/Desktop/linear%20algebra/MAT210-7.HTM[10/10/09 5:05:05 PM]

Das könnte Ihnen auch gefallen