18.03 Final Exam Review

Transcription

1 8.3 Final Exam Review Chris Kottke Contents Definitions & Terminology. Single Equations Systems of Equations Initial Value Problems First Order Equations 4. Separable Equations Linear Equations Graphical Methods Isoclines Phase Portraits Numerical Methods Higher Order Linear Equations 3. Homogeneous Solutions Particular Solutions Exponential Response Undetermined Coefficients Fourier Series for Periodic Inputs Laplace Transform Methods Weight Function, Transfer Function, & Convolution Systems of Equations 4. Relationship to Higher Order Equations Homogeneous Solutions Fundamental Matrices & Initial Value Problems Inhomogeneous Equations & Variation of Parameters Phase Diagrams for Linear Systems Nonlinear Systems A Complex Variables 34 A. Multiplication and Division A. Powers and roots A.3 Trigonometric Identities A.4 Sinusoidal Identity B Fourier Series 38 B. Periodic Extensions B. Operations on Series

2 C Laplace Transform 4 C. Convolution and the Laplace Transform C. Piecewise Functions & Delta Functions D Matrices 45 D. Eigenvalues & Eigenvectors D. Matrix Exponentials Definitions & Terminology. Single Equations A general differential equation looks like ) F t, xt), x t),..., x n) t) The order of the equation is the highest order derivative of xt) which occurs in the equation. In the above, the order is n. A linear equation is one which looks like a t)x n) t) + a t)x n ) t) + + a t)xt) ft) The a i t) are the called the coefficients. We say the equation is constant coefficient if a i t) a i constant for all i. If ft), then the equation is homogeneous. For constant coefficient linear equations, we sometimes use the operator notation where pd) is the operator. For a linear equation pd)xt) a D n + + a )xt) a x n) t) + + a xt) ft) pd)xt) ft) we often refer to ft) as the input, and the to solution xt) as the response. An important property of linear equations is that solutions to different inputs add. That is, } pd)x f t) pd)ax pd)x f t) t) + bx t)) af t) + bf t) The solution to an nth order equation will have a general solution with n parameters. If the equation is linear, the general solution looks like xt) x h t) + x p t) c x t) + + c n x n t) +x p t) }{{} x h t) where the c i are the parameters. The factors x t),... x n t) are independent solutions to the associated homogeneous equation which is when we set ft). We call x h t) c x t) + + c n x n t) the homogeneous solution and x p t) a particular solution Note that is is not quite correct to say the particular solution since particular solutions are not unique. x pt) can be any solution to the equation. Any two particular solutions will differ by an element of the homogeneous solution. However, most of the techniques we use for finding an x pt) do give us a preferred one, so there s not too much harm in thinking of the x pt) we get as the particular solution.

3 . Systems of Equations A system of differential equations looks like which can be expressed in vector notation A linear system is one which can be written x f t, x,..., x n ) x f t, x,..., x n ) x n f n t, x,..., x n ) x ft, x), x x x. x n x a t)x t) + a t)x t) + + a n t)x n t) + f t) x a t)x t) + a t)x t) + + a n t)x n t) + f t) x n a n t)x t) + a n t)x t) + + a nn t)x n t) + f n t) which can be expressed using matrix notation a a t) a n t) x a t) a t) a n t) At)xt) + ft), At)...., f a n t) a n t) a nn t) f t) f t). f n t) If ft), the system is homogeneous. As in the case of single linear equations, the general solution to a linear system has the form xt) x h t) + x p t) c x t) + + c n x n t) +x p t) }{{} x h t) where x h t) is the associted homogeneous solution where we set f ), and x p t) is any particular solution. The general homogeneous solution x h consists of an arbitrary linear combination of n independent solutions x t),..., x n t)..3 Initial Value Problems As we know, an nth order equation or first order system of n equations) has a general solution depending on n parameters. To fix these parameters uniquely, we often require the solution to satisfy an initial condition. A differential equation together with an initial condition is known as an initial value problem. For an nth order linear) equation, a typical initial value problem looks like a x n) t) + a x n ) t) + + a xt) ft), xt ) v, x t ) v,..., x n ) t ) v n where the v i are some specified constants. Note that for an nth order equation, we need to specify n conditions on x and its first n derivatives) at t t to get a unique solution. In many cases, t. For a first order system of n equations, a typical initial value problem looks like where x is a specified constant vector. x Ax + f, xt ) x 3

4 First Order Equations. Separable Equations A differential equation is separable if x ft, x) ft, x) gt)hx) We can solve it by collecting everything involving x on the left and t on the right, to get x hx) dx hx) dt gt) Now we can integrate both sides dx hx) dt dt dx hx) gt) dt and solve the result for x. The constant of integration serves as the free parameter in the general solution for x. Example. Solve x t x ) for xt) i.e. find the general solution). Solution. Write x x t and integrate. Use partial fraction decomposition to write x A x + B A x) + B + x) + x x and so We get A + B B A } B A dx + x + dx x t dt ) + x ln + x) ln x) ln t3 x 3 + c ) + x ke t3 /3 x x ket3 /3 ke t3 /3 +. Linear Equations A first order linear equation has the general form x + pt)x qt) 4

5 We can solve any linear equation even with variable coefficients) by the method of integrating factors. We multiply the equation on both sides by a function ut) and we want to write the left hand side as ut)x + ut)pt)x ut)qt) ut)x + ut)pt)x ut)x) ut)x + u t)x which will be true if u ut)pt). Thus we take ut) exp pt) dt ). Then the equation ut)x) e R pt) x) dt R e pt) dt qt) is separable and can be integrated. solution. Example. Find the solution to x + x t, where x). Solution. The integrating factor ut) above must satisfy Thus the equation can be written The constant of integration serves as the parameter for the general u ut) et ut) e t x ) te t We integrate, using integration by parts d e t x ) dt te t dt dt t et e t dt ) te t et + c So e t xt) is the general solution. To satisfy the initial condition, set ) te t et + c xt) t 4 + ce t so the solution to the initial value problem is x) 4 + c c 4 xt) t 4 + e t 4 When dealing with sines and cosines, it is usually easiest to complexify the problem as illustrated in the following example. Example. What is the response of x + kx qt) to the input qt) k sin ωt? Assume k >, so the system is stable; we re only interested in the steady state, so assume a convenient value for x).) Express your answer in the form xt) A sin ωt φ) A is is the amplitude response, and φ is the phase shift. 5

6 Solution. It is convenient to consider this as half of the complex problem see appendix A for complex variables techniques) x + k x ke iωt k cos ωt + ik sin ωt where we are interested in the imaginary part of the solution x Im x). We solve according to the methods described above, taking the arbitrary constant to be for simplicity as noted, we are interested in the form of the steady state response, so the value of x) is unimportant). e kt x ) ke kt e iωt ke k+iω)t e kt x kek+iω)t k + iω) + c x keiωt k + iω eiωt + iω/k There are a few methods to find the imaginary part of the right hand side. However, because we want to write our solution in the specified form, it is best to do the following. Write + iω/k in polar form, Then + iω/k + ω /k e i tan ω/k) e iωt + iω/k + ω /k ) / e iωt φ) where φ tan ω/k. Then the imaginary part and therefore the solution we seek) is y Imỹ) + ω /k ) / sin ωt φ), φ tan ω/k We obtain A + ω /k ) / for the amplitude response, and φ tan ω/k for the phase shift. We note in passing that repeating the same calculation for qt) cos ωt would be identical, but for taking the real part at the end instead of the imaginary part, which gives a similar answer with sin replaced by cos. Often a nonlinear first order equation can be transformed into one which is either linear or separable by an appropriate change of variables. The appropriate choice of variables depends on the equation and will be specified in a given problem. Note: When applying a change of variables, you must be careful to transform derivatives correctly. Use the chain rule to get a formula for derivatives of the old variable in terms of the new. Example. Find the general solution to xy y x 3 +y 3, by using the change of variables yx) zx) y/x. Solution. First we divide both sides of the equation by xy to get y x3 + y 3 xy Then we notice that if we divide the top and bottom of the right hand side by y 3 we get y x 3 y + 3 x x y y + y x Taking z y/x as suggested, we see that the right hand side can be written as F z) z + z. To write the left hand side in terms of the new variable, we have to use the chain rule: y xzx)) xz + z 6

7 Plugging in, we get an equation which is separable in z and x: Separate as xz z + z z z z z x dx z dz x z 3 lnx) + c lnx) + lnc ) lnc x) 3 z ln [ c x) 3) /3 Now remember to substitute back y xz, so that our general) solution is y x ln [ c x) 3) /3.3 Graphical Methods.3. Isoclines We can obtain a graphical picture of the solutions to a first order equation x fx, t) by drawing isoclines. A solution to the above equation will be a curve xt) in the x-t plane, and note that the equation says that the slope of such a curve at a point t, x ) must be equal to fx, t ). So the method of isoclines works as follows. For various choices of c, plot the level curves defined by fx, t) c these are the isoclines. By the above discussion, any solution curve touching the curve fx, t) c must have slope equal to c there. So we draw little hashes with slope equal to c along the corresponding isocline.. Sketch solution curves in the plane by demanding that their slope at each isocline match the slope of the hashes on that isocline. Note that solutions cannot cross one another. 3. In particular, if an isocline has hashes which are tangent to the isocline itself, then that isocline is also a solution curve, and no other solution may cross it. Example. Draw isoclines with slope,, -, and -, and sketch some solution curves for the equation x xt Solution. The plot appears in figure. The isocline corresponds to xt whose solutions are the both the x and t axis. In particular the t axis itself is a solution x. The other curves are hyperbolas. 7

8 Figure : Isoclines Note that solutions appear to be gaussians, which can be verified by solving this separable!) equation exactly: dx t dt x ln x t + c x c e t.3. Phase Portraits If we have an autonomous equation, i.e. x fx) where the right hand side doesn t depend on t, we use a slightly different method, known as a phase portrait. The isocline picture for an autonomous equation is somewhat redundant, since all of the isoclines are horizontal constant in t)! So we can collapse the picture to be a one dimensional picture consisting of the x-axis only not the t-axis!). We proceed as follows. Plot the critical points These correspond to constant solutions xt) x. x such that fx ). In between the critical points, we draw arrows going to the left or right according to whether fx) > xt) increasing arrow right fx) < xt) decreasing arrow left It is useful to plot fx) on a vertical axis over the x-axis to make it obvious where the critical points are and where f is positive/negative. 8

9 Figure : Phase diagram & stability Note y should be x) This characterizes the behavior of the solutions. If they start in between critical points, they will procede left or right according to the arrows, and asymptotically approach either the next critical point, or ± as t. We classify critical points as stable if solutions approach the critical point from either side unstable if solutions depart from the critical point on either side semistable if solutions approach from one direction and depart on another Example. For what values of a does x ax x have a stable, positive solution? Solution. Complete the square on the right hand side to get ax x x a ) a + 4 If a is large enough, there are two critical points on the x-axis, corresponding to the points where the graph of fx) crosses it since by definition, this is where fx) ). From the plot ), it is clear that unless a /4 >, the parabola will lie completely below the x axis and there will be no fixed points at all. The lower root is an unstable fixed point, and the upper one is stable. We therefore need a condition on a for which the larger root exists and is positive. The solutions to x a ) + a 4 are x a a ± 4 and we are interested in the larger root x a + a 4. Thus the requirements on a that this exist i.e. not be complex) and be positive are a a + 4 > a 4 > 9

10 .4 Numerical Methods The basic numerical method for differential equations is Euler s method. Although much more sophisticated methods are used in practice, this is the simplest and the one we require students to know for this class. The idea is as follows. Suppose we have the initial value problem x fx, t), xt ) x and we want to approximate the value of the solution xt) at a point t n.. Divide the interval [t, t n into n even subintervals of width h called the step size, to get t < t < < t n, t i t i h. Start with the initial point x xt ). We approximate the equation by x x )/h fx, t ) so we iteratively solve for x i x i + fx i, t i )h 3. Stop when you get to x n Euler s method will tend to underestimate solutions when they are upward sloping, and overestimate when the solutions are downward sloping. Example. Estimate x), where xt) is the solution to using a step size of h.5. Solution. It is best to make a table as follows x xt, x) n t n x n fx n, t n ) hfx n, t n ) / + / /4 5/4 + /4 5/4 5/8 3 3/ 5/8 5/4 + 5/8 45/6 45/3 4 5/3 5/8 + 45/3 so we compute x) 5/ This is likely to be an underestimate since the solution is increasing exponentially. Indeed, the exact solution is xt) e t / x) e Higher Order Linear Equations This section will review the methods for solving higher order constant coefficient linear equations. See section for definitions and terminology. To solve an nth order constant coefficient linear initial value problem pd)x x n) + a x n ) + + a x ft), x) v, x ) v,..., x n ) ) v n the steps are as follows.

11 . Solve the associated homogeneous problem pd)x. There will be n linearly independent solutions x t),... x n t), and the homogeneous solution is given by x h t) c x t)+c x t)+ +c n x n t). The solutions are determined by the characteristic roots of the operator as described below in section 3... Find a particular solution x p t) to pd)x ft), by using one of the methods in section 3., such as undetermined coefficients or Fourier series. 3. The general solution is given by xt) x h t) + x p t) c x t) + + c n x n t) + x p t) where the two parameters c,..., c n are yet unspecified. 4. Use the initial conditions x) v,..., x n ) ) v n to determine c,..., c n and find the unique solution to the intial value problem. Remark. If the equation is homogeneous to begin with, omit step. Alternatively, you can solve the initial value problem using the Laplace transform section 3.3, or by converting the equation to a first order system section 4). 3. Homogeneous Solutions Given a differential operator pd), the solutions to the homogeneous equation pd)x a D n + a D n + a n )x correspond to the roots of the associated characteristic polynomial ps) a s n + + a n It will have n roots r,... r n, some of which may be complex. Assuming the roots are distinct, we have corresponding independent solutions x i t) e rit, i,... n We typically write complex exponentials in terms of sines and cosines. The case of repeated roots is discussed below. In the case of the second order equation we have roots from the quadratic formula) There are three cases to consider. x + bx + cx D + bd + c)x s b ± ) b 4c. r and r are both real, and r r. In this case the solutions are e rt and e rt and x h t) c e rt + c e rt. r r is a repeated root. Write r r r. Then the solutions are e rt and te rt. In general, for an nth order ODE, if a root is a repeated root with multiplicity k, the solutions will be given by t j e rt, j,..., k). So x h t) e rt c + c t)

12 3. r, r are complex, and therefore given by a ± ib. Then the corresponding real valued solutions are e at cos bt and e at sin bt, and x h t) e at c cos bt + c sin bt) If the real part a is equal to zero, the homogeneous solution is periodic. If the real part is nonzero a, the solutions are not periodic, but we still refer to b as the pseudofrequency, with corresponding pseudoperiod π/b. Example. Find the general solution to the homogeneous ODE x 6x + x. Solution. We first write the equation as D 3 6D + D)x so ps) s 3 6s + s ss 6s + ). The roots are s and s r ±, where So x h t) c x t) + c x t) + c 3 x 3 t) with so x t) and x 3 t) have pseudofrequency. r ± 6 ± 36 44) 6 ± i) 3 ± i x t) x t) e 3t cos t x 3 t) e 3t sin t Example. Find the general solution to the homogeneous ODE x + 6x + 9x. Solution. We first write the equation as D + 6D + 9)y and the chacteristic root equation is ps) s + 6s + 9 s + 3) and has the double root s 3. Hence the homogeneous solution will be x h t) c e 3t + c te 3t 3. Particular Solutions For certain types of input functions ft), we have direct methods for finding a particular solution. 3.. Exponential Response In the case that the input is exponential ft) e αt, α a complex number then the particular solution will also have the form of an exponential. differential operator. Suppose then, that we seek a particular solution to Its exact form depends on the pd)x e αt The exponential response formula gives the result as follows

13 ERF: pα). In this case, a particular solution will be given by x p t) eαt pα) ERF : pα) but p α), where p s) d dsps). In this case, a particular solution is given by x p t) teαt p α) ERF k : pα) p α) p k ) α), but p k) α). Then a particular solution is x p t) tk e αt p k) α) Example. Given the ODE x + x 3x ft), find a particular solution when ft) e t and when ft) e 3t. Solution. First, we write x + x 3x D + D 3)x. The roots of ps) s + 3)s ) are and 3. In the first case, ft) e t we have p), so ERF applies. x p t) e t + 3) et 5 In the second case, α 3 is a single root, and so p 3), but p 3) 3) +, and we get x p t) te 3t te 3t 3) + 4 When ft) sin bt or cos bt, we consider cos bt Re e ibt) and sin bt Im e ibt) and proceed as above. For a general product of exponentials with sines or cosines, we use e at cos bt Re e a+ib)t), and can apply the exponential response formula where α a + ib. Remember to take the real or imaginary part of the resulting x p t) to get the correct solution! Example. Find a particular solution to x 4x + 5x e t sin t. Solution. The characteristic polynomial associated to pd) is Using the quadratic formula, we have the roots ps) s 4s + 5 s 4 ± ± i Writing ft) Im e +i)t), we see that the coefficient in the exponent is a single root so we can use the ERF formula. Taking a derivative of p we get p s) s 4. Thus ) ) te +i)t te +i)t x p t) Im Im + i) 4 i ) i)te +i)t Im 4 ) te t Im i cos t + sin t) tet cos t 4 3

14 3.. Undetermined Coefficients When the input has the form of a polynomial times a possibly complex) exponential, ft) qt)e αt, qt) a polynomial the particular solution will have the same form; however, we may have to shift the degree of the resulting polynomial depending on whether or not pα), p α), etc., vanish. In any case, we guess x p t) to have the corresponding form, with undetermined coefficients, which we solve for by plugging x p t) into the equation. The method goes by the name of undetermined coefficients. Suppose qt) is a degree k polynomial it s highest power term is t k ). If pα) : then we guess x p t) to be x p t) general degree k poly)e αt a t k + a t k + + a k t + a k )e αt where a,..., a k are constants to be determined by plugging x p t) into the equation. pα), but p α) : then we guess x p t) general degree k poly)te αt a t k+ + a t k + + a k t + a k t)e αt and solve for a,..., a k by plugging in. pα) p α) p l ) α), but p l) α) : then we guess x p t) general degree k poly)t l e αt a t k+l + a t k+l + + a k t l+ + a k t l )e αt and solve for a,..., a k by plugging in. Remark. Note that the cases are the same as the exponential response formula. In fact, ERF is a special case of undetermined coefficients where qt) is a polynomial of degree ; in this case ERF tells you what the resulting undetermined coefficient a is, namely a pα), or p α), etc. Note also that undetermined coeffients also includes the case where ft) qt) is just a polynomial. In this case we take α since indeed, qt) qt)e t ), so the particular solution to guess depends on to what degree is a root of ps). Example. Find a particular solution to x + 3x + Ax t + when A and when A. Solution. Start with A. We have ps) s + 3s +, and since the input has the form t + )e t, and p), we guess a solution of the form where a, b and c are arbitrary. computation as follows: We conclude x p +3x p +x p t + x p t) at + bt + c When evaluating the left hand side on x p t), consider organizing your at + bt + c ) +3 at + b ) + a ) a)t + b + 6a)t + c + 3b + a) at + b + 6a)t + c + 3b + a) t +, a, b 3, c 9 4 Thus, for A, x p t) 4 t 6t + 9 ). 4

15 When A, ps) s + 3s, and we have p) but p ). So we guess a solution of the form and we compute We get 3x p +x p t + x p t) at 3 + bt + ct 3 3at + bt + c ) + 6at + b ) 9a)t + 6b + 6a)t + 3c + b) a 9, b 9, c 7 So for A, x p t) 9 t 3 t + 3 t). Example. Find a particular solution to x + 7x + x 6te t Solution. We have ps) s + 7s + s + )s + 5) so p ) but p ). Since qt) t is a degree polynomial, we guess x p t) of the form x p t) at + b)te t at + bt)e t We plug this into the equation to solve for a and b. It is convenient to organize the computation as follows remember also to include the effect of differentiating the e t when computing the derivatives!) ) x p at + bt + e t ) + 7x p +7 at + a b)t + b e t ) + x p + 4at + 4a a b))t + a b) b) e t ) 6te t t + 6at + a + 3b) e t so we must have We conclude that is a particular solution. 6a 6, a + 3b) a, b /3 x p t) t /3t)e t 3..3 Fourier Series for Periodic Inputs When the input function to a linear equation is periodic, that is, ft+p ) ft) for some period P, we can use Fourier series to construct a particular solution. For general techniques of Fourier series, see appendix B; here we will focus on the application to particular solutions. The idea is to express ft) by its Fourier series, and then construct the particular solution one term at a time. Suppose we want to solve pd)x ft) a + a n cos n ) nπt L + n b n sin We let x an t) and x bn t) be the particular solutions to the following ODE pd)x a a ) nπt pd)x an a n cos L ) nπt pd)x bn b n sin L 5 ) nπt L

16 Then, by the additive properties of linear differential equations see section ), a particular solution to the original problem will be given by x p t) x a t) + x an t) + x bn t) n Remark. Depending on the operator pd), it may be the case that each particular solution x an t) and x bn t) has the form of sines, cosines and constant terms. If this happens, then the sum x p t) is itself a Fourier series and so the response will also be periodic. If however, any one of the particular solutions x an t) or x bn t) has a different form such as t cosnπt/l) from the ERF formula when inπ/l) is a root of ps)), then the resulting x p t), while still a particular solution, will not be periodic. Example. Find a particular solution to x + kx ft), where ft) is the odd periodic extension of t with period 4 a sawtooth wave). For what k will the solution be periodic? Solution. The Fourier series for ft) is computed in an example in section B. We obtain ft) ) nπt b n sin, b n 4 )n+ nπ n We proceed term by term, seeking a particular solution to the equation ) nπt x n + kx n b n sin For this we can use the exponential response formula 3..), thinking of x n Im x n ), where x n + k x n b n e inπ/)t Provided p inπ/)) k nπ/), we can use ERF, to obtain ) b n b n nπt x n t) k nπ/) einπ/)t x n t) k nπ/) sin n So the condition which guarantees a periodic particular solution is nπ ) k, for any n in which case x p t) n nπt bn sin ), bn b n k nπ/) 4 ) n+ nπ k nπ/) ) Since the characteristic polynomial ps) s + k s + ik)s ik) has purely imaginary roots, the general solution will also be periodic in this case. However, if for example, k π /4, so that it violates our condition when n, the solution will not be periodic, since we obtain x p t) t ) πt ) nπt π cos + bn sin where the resonant 3 term x t) was calculated using ERF : ) ) t t x t) Im p iπ/) eiπ/)t Im iπ/) eiπ/)t it ) ) Im cosπt/) + i sinπt/) t ) πt π π cos n 3 Resonance is the term we use for the response of a system of the form x + ω x ft) to an oscillator input ft) which has the same frequency ω as the homogeneous solutions. The particular solution called the resonant) solution, grows without bound becuase of the t in the amplitude. 6

17 3.3 Laplace Transform Methods The Laplace transform is a useful tool for solving initial value problems for a few reasons.. The transform itself automatically takes initial conditions into account, resulting in a particular solution satisfying the initial data without the need to go through a general solution first.. It is well suited to handle input functions which are piecewise, that is, f t) t < t < t ft) f t) t < t < t.. 3. In the case of purely zero initial conditions, solving by weight functions and convolution as in section 3.3. allows us to write down a direct general formula for the solution to an IVP with arbitrary ft). For general results and techniques regarding the Laplace transform, see appendix C. Here we will focus on its use in solving initial value problems. Example. Use the Laplace transform to solve the initial value problem x + 7x + x 6te t, x), x ) Solution. Taking the Laplace transform of both sides of the equation, we get s Xs) sx) x ) + 7 sxs) x)) + Xs) s + 7s + )Xs) 6 s + ) Note that the coefficient in front of Xs) is ps) s + 7s + s + )s + 5), as will always happen. We solve for Xs) to get Xs) ) 6 s + )s + 5) s + ) + We must use partial fraction decomposition to write 4 6 s + ) + s + )s + 5) s + ) s + ) 6 + s + ) s + ) 3 s + 5) A s + + B s + ) + C s + ) 3 + D s + 5 Heaviside coverup see appendix C) applies to C and D to give ) 6 + s + ) s + ) 3 s + 5) C ) + 5) ), D ) To obtain A and B we must put everything over a common denominator and compare sides it s best not to expand everything out yet): As + ) s + 5) + Bs + )s + 5) + Cs + 5) + Ds + ) s + ) Comparing terms of order s 3, we get A + D A 8/9 A This is a more complicated partial fraction decomposition than you would have to do on an exam. We do it for this problem so that we can compare with the solution obtained by earlier methods. 7

18 Comparing terms of order s count very carefully!), we get 5A + A) + B + 3D) B + 9A + 6D B + 38/9) B So Xs) 8 9 s + 3 s + ) + s + ) s + 5 and the inverse transform gives our solution xt) 8 9 e t 3 te t + t e t 8 9 e 5t Note that the particular solution x p t) t /3t)e t for this operator was computed by the method of undetermined coefficients in section 3... We can then easily check our Laplace transform answer with what we would get via the general solution. Since ps) s + )s + 5), the general solution will be x g t) c e t + c e 5t + t /3t)e t Matching the initial conditions x), x ) gives c 8/9, c 8/9, which agrees with the above. Example. Find the solution to x + x { t < t < π/ t > π/ x) x ) Solution. The input function is piecewise; we can write it in terms of step functions using the methods of appendix C. ft) tut) ut π/)) We take the Laplace transform to get s + )Xs) L tut) ut π/) t π/) + π/ )) s e π/s s + π/ ) s Xs) s s + ) e π/s s s + ) + π/ ) ss + ) We must do two partial fraction decompositions. First, Similarly, s s + ) s 3 : s : s : : π/ ss + ) s : s : : π/ A s + B s + Cs + D s + Ass + ) + Bs + ) + Cs + D)s A + C B + D A B A s + Bs + C s + π/ As + ) + Bs + C)s A + B C A A C, B, D C, A π/, B π/ 8

19 Thus, we get Xs) s s + e π/s s s + + π/ ) s π/) s s + ) xt) t sint) ut π/) t π/) sint π/) + π/ π/ cost π/) ) t sint) ut π/) t + cost) π/ sint) xt) { t sin t < t < π/ π/ ) sint) cost) t > π/ 3.3. Weight Function, Transfer Function, & Convolution To any given differential operator pd), we can associate the corresponding weight function, which is the solution to the initial value problem pd)wt) δt), w) w ) where δt) is the Dirac delta function. It is important here that the initial conditions are identically, sometimes referred to as rest initial conditions. Taking the Laplace transform of both sides yeilds ps)w s) Lwt)) W s) ps) We call W s) the transfer function. Note that knowing the weight function is equivalent to knowing the differential operator, since we can compute pd) from ps) /W s). Remark. An equivalent characterization of the weight function is as the solution to the initial value problem pd)wt), w) w ) w n ) ), w n ) ) /a where n is the order of the equation, and a is the top order coefficient of pd) a D n +. This follows by the Laplace transform, since the left hand side will have a term a w n ) ) coming from the transform of the highest derivative. The main feature of the weight function is that, once we know wt) for a given operator pd), we can solve the corresponding initial value problem with rest initial conditions) pd)xt) ft), x) x ) for any input ft), by taking the convolution see appendix C.) xt) w f)t) t wτ)ft τ) dτ Thus, we have a direct formula for the solution with arbitrary input function without having to re-solve the equation every time. Example. Write an integral formula in terms of ft) for the solution to Solution. We first determine wt), which satisfies x + 6x + 9x ft), x) x ) w + 6w + 9w δt), w) w ) or w + 6w + 9w, w), w ) 9

20 by our favorite methods. We will do this two ways. Using the Laplace transform, and ps) s + 6s + 9 s + 3), we have ) wt) L s + 3) te 3t Alternatively, we can use the characteristic roots 3, repeated root), to obtain the general solution to w + 6w + 9w, w g t) c e 3t + c te 3t The initial conditions w) and w ) force c, and w ) c In any case, then the solution to the original equation in x can be written as xt) wt) ft) t τe 3τ ft τ) dτ t fτ)t τ)e 3t τ) dτ ft) wt) Example. Suppose a solution to is given by pd)wt) δt), w) w ) wt) e t where pd) is a constant coefficient linear differential operator. Find the solution to the initial value problem pd)xt) e t, x) x ) Solution. Even though we don t know the operator, we can write down the solution as xt) wt) e t e t e t t t e s e t s) ds e t e s ds e t e s t e t + e t since we have rest initial conditions. Alternatively, using the Laplace transform, we can recover pd) by so and we can solve by any other method. 4 Systems of Equations W s) ps) Le t ) s + pd) D + See section for definitions and terminology. We deal mostly with linear systems x t) Axt) + rt) The steps for solving such an equation are analogous to the ones for higher order linear equations discussed in section 3., namely

21 . Find the homogeneous solution x h t) c x t) + + c n x n t), as described in section 4., where x i t) solves x i Ax i.. If applicable), find a particular solution x p t) to x p Ax p + r, as described in section 4.4. Then the general solution is xt) x h t) + x p t) c x t) + + c n x n t) + x p t) 3. If applicable), Given an initial condition x) x, solve for the constants c i to get the solution to the initial value problem. For nonlinear systems, the best we can do is get a qualitative picture of the solutions using the graphical methods of section Relationship to Higher Order Equations You can always convert back and forth between nth order constant coefficient linear equations and first order n-dimensional systems of ODEs. Starting with the nth order equation y n) + a y n ) + + a n y + a n y ft) define new variables which will be the vector components) x y x y. x n y n ) Taking derivatives, we have by definition for the first n variables, x y x. x x 3.. x n x n The derivative of x n is y n), which we can solve for in the original equation, replacing derivatives of y by their corresponding x i variables x n a x n a n x + ft) So x t) Axt)+ft), xt) y y. y n ), A is called the companion matrix to the original equation a n a n a n a Example. Convert y + 3y + y e t to a two dimensional system, ft). ft)

22 Solution. We have So x y x y x y x x y 3y y + e t 3x x + e t Which we can write as [ x x [ 3 [ [ x + x e t We illustrate the conversion from a dimensional system to a second order equation by example. Example. Convert the system into a single, second order equation [ x x [ 3 [ x x + [ t e t Solution. Choose one of the two variables which will be the second order variable, say x. Start by taking the derivative of the corresponding equation x 3x + x + t 3x + x + x + e t ) + t where we substituted the other equation in for x. We still need to get rid of the x term, which we can do by using the first equation we ve only used the derivative of the first equation, not the equation itself, so this will not lead to any redundancy). Solving the first equation for x and substituting, we get Or, in more standard form x 3x + x + x 3x t ) + e t ) + t 4x 4x t + t + e t x 4x + 4x t + t + e t 4. Homogeneous Solutions Given a n n system x Ax we seek n linearly independent solutions x,..., x n. Supposing these solutions to have the form of an exponential in t times a constant vector in analogy with the higher order equations of section 3.), we guess and plug in. xt) e αt v e αt v) αe αt v Ae αt v αv Av We see that this will be a solution provided α is an eigenvalue with corresponding eigenvector v. For a discussion of eigenvalues and eigenvectors, see appendix D.. To sum, the general solution will have the form x h t) c x t) + + c n x n t) where x i t) is a solution corresponding to eigenvalue λ i. As for the individual solutions x i t), there are four cases to consider, and examples are provided below:

23 . λ,..., λ n are real and distinct. In this case, there will be n linearly independent eigenvectors v,..., v n, each corresponding to an eigenvalue, and each individual solution has the form x i t) v i e λit. Some of λ,..., λ n are complex. In this case, since the coefficients of A are real, the complex eigenvalues will come in conjugate pairs λ j a + ib, and λ k a ib. To get two linearly independent solutions for λ j and λ k, it is sufficient to choose one of them, say λ j a + ib, and compute its eigenvector v j which will have complex entries. Then we can take x j Re v j e a+ib)t), x k Im v j e a+ib)t) 3. λ i is repeated with multiplicity k, but it has k independent eigenvectors. This is called the complete case for repeated eigenvalues. In this case, the solution is identical to case, simply treating λ i as k distinct eigenvalues. Thus if v,..., v k are its eigenvectors, we have k independent solutions x i v e λit,..., x ik v k e λit 4. *Will not be on exam!)* λ i is repeated with multiplicity k, but it has fewer than k independent eigenvectors. This is called the defective case for repeated eigenvalues. In this case we can find chains of generalized eigenvectors see below). Say we have two independent eigenvectors v and v. Then we can find chains {v,..., v l } and {v,..., v m } where l+m k, so there are a total of k generalized eigenvectors. The equations defining the chains of eigenvectors are A λi)v l v l,..., A λi)v v, A λi)v ; so we can start with the beginning of the chain, which is a true eigenvector, and inductively solve for the successive elements of the chain. Once we have done this, we have independent solutions x v e λit x v t + v ) e λit t x v ) + v t + v 3 e λit. x v e λit. x v t + v ) e λit t x v ) + v t + v 3 e λit.. Example. Find the general solution to x Ax, where [ 4 A Solution. We must compute the eigenvalues and eigenvectors of A as described in appendix D.. To find the eigenvalues, we set det A λi) det λ 4 λ λ) λ) λ λ 3

24 Alternatively, we can use the direct formula for matrices given by see appendix D.) det A λi) λ tr Aλ + det A which agrees with our answer. The eigenvalues are and, and so are distinct. To find the eigenvector corresponding to λ, we solve into the equation A λi)v to get [ [ [ 3 4 v A )I)v 3v v + 4v so we can take [ 4 v 3 Similarly, for the eigenvalue λ we have [ 4 3 [ [ u u u So we can take The general solution is xt) c [ v 4 3 [ e t + c [ e t Example. Find the general solution to x Ax, where [ 5 A 3 Solution. The characteristic polynomial is λ 4λ + 8 λ ± i We have complex eigenvalues. We proceed to find two independent real valued solutions as described above: we choose one of these eigenvalues, find its eigenvector, and take real and imaginary parts at the end to get independent solutions. So take λ + i. Substituting into A λi)v we have the equation [ [ [ i 5 v i v This gives us two equations which are the same checking this is a good exercise in complex arithmetic), one is v + i)v, and we can take the solution [ [ [ i v i Then we have [ x Re e t cos t + i sin t) [ x Im e t cos t + i sin t) x c x + c x [ i [ i ) ) e t [ cos t + sin t cos t e t [ sin t cos t sin t 4

25 Example. Find the general solution to x Ax, where 4 6 A 6 4 Solution. First we need to solve det A λi), which we calculate directly from the definition, using cofactor expansion see appendix D) to compute the determinant 4 λ 6 det λ 4 λ)det λ 6 4 λ 6 4 λ λ) 4 λ) For λ, we compute v v v 3 v v v 3 to get an eigenvector v. For λ 4, we have the eigenvector equation 6 6 v v 6 v 3 which amounts to the single equation v. Since we have two degrees of freedom, we can find two independent eigenvectors, and the matrix is complete. We take v, v 3 Then the general solution is xt) c x t) + c x t) + c 3 x 3 t) x e t v e t x e 4t v e 4t x 3 e 4t v 3 e 4t 4.3 Fundamental Matrices & Initial Value Problems Given an n n matrix A, a fundamental matrix for A is an n n matrix F t) satisfying F t) AF t), det F t) for all t 5

26 Writing the columns of F t) as vectors x t),..., x n t), F t) x x n the differential equation for F t) amounts to the n equations x it) Ax i, i,..., n and the condition det F t) is equivalent to x,..., x n linearly independent. Thus the following two things are equivalent F t) AF t), det F t) for all t F t) has columns consisting of linearly independent solutions x,..., x n to the equation x Ax. Remark. F t) is not unique, as indeed we could consider rearranging its columns or multiplying them by various constants to get a new fundamental matrix. In light of the above, we can express the general solution to x Ax as follows c.. xt) c x + + c n x n F t)c, c Suppose we want to find a solution satisfying the initial condition x) x. Since F ) is invertible because det F ) ), we can write Then x x) F )c c F ) x xt) F t)c F t)f ) x is the solution to the initial value problem. Note that while fundamental matrices are not unique, F t)f ) is always the same matrix, even when computed by different fundamental matrices. In fact c n F t)f ) e ta, for any fundamental matrix F t) where e ta is a matrix exponential, discussed in appendix D.. So we conclude The solution to the initial value problem x Ax, x) x is given by xt) e ta x We can compute e ta for any matrix A from any fundamental matrix F t) for A by setting e ta F t)f ) Example. Solve the initial value problem [ x Ax, x) [ 3 4, A 4 3 6

27 Solution. Eigenvalues and eigenvectors for A are calculated in an example in appendix D.. We have [ [ λ 5, v λ 5, v Thus we have linearly independent homogeneous solutions [ [ e x 5t e 5t e 5t, x e 5t and can form the fundamental matrix [ F t) e 5t e 5t and compute [ F ) [ 5 e ta F t)f ) [ [ e 5t e 5t 5 e 5t e 5t 5 The solution to the initial value problem is then xt) e ta x [ e 5t + e 5t) e 5t + e 5t) [ 5 e 5t + e 5t) 4e 5t + e 5t) 5 [ e 5t + e 5t) e 5t + e 5t) e 5t + e 5t) 4e 5t + e 5t) [ e 5t + e 5t) e 5t + e 5t) 4.4 Inhomogeneous Equations & Variation of Parameters We can also use a fundamental matrix to find particular solutions to an inhomogeneous system x t) Axt) + rt) by a method called variation of parameters. This method works for any input function 5 rt). The trick is as follows. First compute a fundamental matrix F t) for A. Then guess a particular solution of the form x p t) F t)vt) where vt) is a yet-to-be-determined vector function of t. Plug this into the equation and use the product rule 6 to get F t)vt)) F t)vt) + F t)v t) A F t)vt)) + rt) Since F t) is a fundamental matrix, we have F t) AF t), so the first terms on each side of the equation cancel, and we get F t)v t) rt) v t) F t) rt) using the fact that F t) is invertible since the other condition on a fundamental matrix is det F t) for all t). We integrate the lower limit of integration is unimportant; different choices will lead to different particular solutions) vt) t F τ) rτ) dτ x p t) F t)vt) F t) t F τ) rτ) dτ 5 For this reason, it is useful in the case of a single higher order inhomogeneous equation whose input function isn t amenable to the techniques of section 3.. We can transform such an equation into an inhomogeneous first order system, find a particular solution, and transform back. 6 It is straightforward and a good exercise!) to check that a product rule applies to matrix multiplication 7

28 We emphasize that this formula is valid for any fundamental matrix F t). In particular, if we use the matrix exponential F t) e ta, we have t x p t) e ta e τa rτ) dτ since e τa) e τa as discussed in appendix D.. Another fundamental matrix may give an easier integral however. Example. Find a particular solution to x t) Axt) rt), A [ [ e t, rt) Solution. First we find a fundamental matrix for A. Its eigenvalues and eigenvectors can be computed to get [ [ λ, v, λ, v So we have two independent homogeneous solutions x t) e λt v and x t) e λt v and can form the fundamental matrix [ e t e F t) t e t Using det F t) e 3t we compute see appendix D for the formula for the inverse of a matrix) [ [ F t) e 3t e t e t e t e e t t e t Guessing x p t) F t)vt) leads as above to the equation vt) t F τ) rτ) dτ t e t [ e t e t [ e t dτ t [ [ t dτ Then [ e t e x p t) F t)vt) t e t [ t [ te t Finally, using fundamental matrices and variation of parameters, we can express the complete solution to an inhomogeneous initial value problem as an integral. That is, suppose we want to solve x Ax + r, x) x The general solution can be written see section 4. for explanation of the x h t) term) x g t) c x t) + + c n x n t) +x p t) F t)c +F t) }{{}}{{} x h x h t) t F τ) rτ) dτ Here it is convenient to set the lower limit of integration to be ; we ll see why in a second. Solving for c which satisfies the initial condition as in 4., we compute x x) F )c + F ) F τ) rτ) dτ F )c c F ) x 8

29 and so the complete solution is xt) F t)f ) x + F t) or, using the matrix exponential e ta for our F t), t F τ) rτ) dτ t xt) e ta x + e ta e τa rτ) dτ Note the similarity with the integral formula you get for a single first order linear equation using integrating factors as in section Phase Diagrams for Linear Systems linear systems x Ax can be classified into 5 types depending on the behavior of solution curves in the x-y plane. These are Type Stability Eigenvalues Nodal source Unstable Both real, positive Nodal sink Stable Both real, negative Saddle Unstable Both real, opposite signs Spiral source Unstable Complex, positive real part Spiral sink Stable Complex, negative real part A system is stable if all of its solutions approach the origin as t. A picture of each type can be seen in figure 3, and examples are discussed below. Example. Determine the type and sketch the phase diagram for [ x 4 Ax, A Solution. We examined this equation in section 4., and found eigenvalues/eigenvectors [ [ 4 λ, v, λ 3, v This system is a saddle, and let us see why. Vector v lies along the line y 3/4x in the plane. Any solution to the equation which starts on this line will remain on it, since it will be given by the equation the initial condition will force c ) x t) c e t v, for some c So solutions along this line have exponentially decreasing magnitude from the e t term, and approach the origin. On the other hand, v lies along the x-axis, and any solution starting on this axis will remain on it, since it will be given by initial conditions forcing c this time) x t) c e t v, for some c These solutions have exponentially increasing magnitude, and move away from the origin as t. For this reason the saddle is unstable. This is the saddle depicted in figure 3.c. 9

30 Figure 3: Phase diagrams for a. Nodal source, b. Nodal sink, c. Saddle, d. Spiral source, e. Spiral sink Example. Determine the type and sketch the phase diagram for [ x 5 Ax, A 3 Solution. We examined this equation in section 4., and found that it had complex eigenvalues λ ± i and a general solution [ x Re e t cos t + i sin t) [ x Im e t cos t + i sin t) x c x + c x [ i [ i ) ) e t [ cos t + sin t cos t e t [ sin t cos t sin t Note that the coefficient in the exponent comes from the real part of the eigenvalues Re ± i). The magnitude of these solutions grows exponentially because of the e t term, so this is a spiral source, with solutions going away from the origin. For this reason it is unstable. The oscillatory functions contribute the spiraling behavior; to determine whether the spiral is clockwise or counterclockwise, we can simply check the following. Suppose a solution curve passes through the point, ) at time t, so [ x) then its velocity at time t is given by the differential equation: [ [ [ x 5 ) Ax) 3 this velocity vector points up and to the right, so we conclude that our solutions are spiraling counterclockwise. Some solution curves are sketched in figure 3.d. 3

31 Example. Determine the type and sketch the phase diagram for [ x 7 3 Ax, A 3 7 Solution. First we determine the eigenvalues and eigenvectors. det A λi) det 7 λ λ λ 4λ + 8 λ 8)λ 6) so we have two real, positive eigenvalues. For λ 8, we get the eigenvector [ [ [ [ 9 3 a A 8I)v v 3 b 3 and for λ 6, we get A 6I)v [ [ [ a b [ 3 v Solutions starting along the line y 3x through the vector v will remain on it, with magitude growing at a rate e 8t and hence travel away from the origin. Similarly, solutions starting on the line y /3x through the vector v will remain on the line, with magnitude growing at a rate e 6t and hence travel away from the origin. This system is therefore a nodal source, and is unstable. To examine the behavior of solutions not on either of the eigenvector lines, note that such solutions will be given by a linear combination xt) c e 8t v + c e 6t v as t, the second term clearly dominates, and so solutions will asymptotically approach the direction of v for large t. As t, the first term dominates, and so solutions asymptotically approach the direction of v near the origin since t ). The phase diagram is depicted in figure 3.a. 4.6 Nonlinear Systems A nonlinear autonomous) system of two equations has the general form x fx, y) y gx, y) We d like to get a qualitative picture of how various solutions behave, and we can do so in analogy with the phase diagrams of single equations in section.3.. Near a critical point x, y ) such that fx, y ) gx, y ), we can approximate the system by the linear system x x ) f x x, y )x x ) + f y x, y )y y ) y y ) g x x, y )x x ) + g y x, y )y y ) which follows by expanding f and g in Taylor series around the point x, y ) and throwing out any terms which are quadratic or higher in the variables x x ), y y ). We can write this in matrix notation as where Jx, y) is the Jacobian matrix Jx, y) x Jx, y )x [ fx x, y) f y x, y) g x x, y) g y x, y) We can classify this linear system as one of the 5 types discussed in section 4.5, and sketch the corresponding phase portrait near the critical point. Once we have done this for all of the critical points, we have a good idea of what global solution curves look like. To summarize, we develop a picture of solution curves in the x-y plane as follows 3

32 Figure 4: Phase diagram for the nonlinear example. Find the critical points, which are points x, y ) such that fx, y ) gx, y ).. Compute the Jacobian Jx, y ) at each critical point. 3. Classify each critical point according to the behavior of the linearized system x Jx, y )x 4. Analyze long term behavior showing the solutions are constrained in a particular region, etc.) Example. Examine the behavior of the nonlinear system x x x xy y y 4y + xy which models a population xt) of a prey species, and yt) of a preadator species. What are the possible eventual outcomes of the populations as t? What will be the outcome for any set of initial conditions < x) < and < y) <? Solution. First we find the critical points xx y) yy 4 + x) We have 4 possibilities: x, y ) {, ),, 4),, ), 3, )} The last point 3, ) comes from solving the linear system } x y) y 4 + x) x y x + y 4 The Jacobian matrix for this system is [ [ fx f Jx, y) y x y x g x g y y y 4 + x 3

33 At the critical point, ), we have J, ) [ 4 which has eigenvalues and eigenvectors [ λ, v [, λ 4, v So is a nodal sink. Note that for a source or sink, we don t care too much about the eigendirections, just that all solutions near the critical point are coming in). At, 4), we have J, 4) [ λ 6, λ 4 This has two real eigenvalues with opposite signs, so it is a saddle. In this case, we do care about the eigendirections, so we compute [ [ 5 v, v So solutions are approaching this critical point along the line y /5)x + 4 because of the negative eigenvalue), and departing from it along the y-axis corresponding to the positive eigenvalue). At, ), we have J, ) [ which is also a saddle. We compute the eigenvectors [ v, v λ, λ So solutions are approaching along y x ) and departing along the x-axis. At 3, ), we have using the formula from appendix D. that eigenvalues satisfy λ tr Aλ+det A ) [ 3 3 J3, ), λ tr J ± ) tr J) 4det J ) 4 ± 6 4 ± i So this is a spiral source because of the positive real part). To determine its direction we compute see the complex eigenvalue example in section 4.5 for an explanation of this) [ [ [ so the spiral is counterclockwise. The phase diagram is sketched in figure 4 The possible outcomes appear to be. xt), yt). xt), yt) 3. xt), yt), and possibly 4. xt), yt) [ 33

34 It appears that all solutions inside the box < x <, < y < approach, ) as t. We can verify this as follows. Provided the vector field fx, y)i + gx, y)j is always pointing into or along the box along the boundary of the box, we can conclude that no solution will ever exit the box. Along the boundary y, < x <, we have [ fx, ) gx, ) [ x x is a vector whose y component is and whose x component is negative, since x < x x xx ) <. Along the boundary x, < y <, we have [ f, y) g, y) [ y 4y points in the negative y direction since y < y 4y yy 4) <. Along the boundary x, < y <, we have [ f, y) g, y) [ y y y the x component of this vector field is negative, so it points into the box. Along the boundary y, < x <, we have [ fx, ) gx, ) [ x 4x 4 + x the y component of this vector field is negative, since < x < 4 + x <. We conclude that all solutions starting in the box must stay in the box. Moreover, the only two critical points which could be approached by solutions in the box are, ) and, ). However, the axis along which solutions approach, ) is y x ) which does not intersect the box, so in fact all solutions in this box approach the final value, ). A Complex Variables A complex number is a number z which can be written in the Cartesian form z x + iy, where x, y are real numbers. and i. The we define the real part Rez) x and imaginary part Imz) y to be the parts not multiplying i and multiplying i, respectively. We can also write z in polar form z re iτ where r and τ π. We can convert back and forth using Euler s formula, To go from Cartesian to polar form, we compute e iτ cos τ + i sin τ r x + y τ tan y x 34

35 Figure 5: Vectors in the complex plane and to go from polar to Cartesian we use See Figure 5. Also important are the reverse Euler formulas x r cos τ y r sin τ cos τ eiτ + e iτ sin τ eiτ e iτ i ie iτ e iτ ) which are obtained by adding or subtracting e iτ cos τ + i sin τ and e iτ cos τ i sin τ. A. Multiplication and Division For two complex numbers z and z, we can compute their product directly, using i by z z x + iy )x + iy ) x x + ix y + x y ) + i y y x x y y ) + ix y + x y ) Alternatively, in polar form, Division is easy in polar form: z z r e iτ )r e iτ ) r r e iτ+τ) It is slightly trickier in Cartesian form. We write z z r e iτ r e iτ r r e iτ e iτ r r e iτ τ) z z x + iy x + iy 35

36 but then we need to get rid of the i in the denominator. We can always do this by multiplying by the complex conjugate of z, denoted by z which is defined for any complex number by z x iy re iτ where we change the sign in front of i. Back in our division problem, we get A. Powers and roots z z x + iy x + iy x + iy x + iy x iy x iy x x + y y ) + ix y x y ) x + y Taking powers of a complex number z n, is done in the usual manner in Cartesian form, x + iy) n, but gets harder and harder as n gets large. It is most convienient to do these operations in polar form. Then to raise z re iτ to a power, simply compute z n re iτ ) n r n e inτ That is r r n and τ nτ. Roots are a bit trickier. We do the same thing, replacing n by /n, but we must remember that there are n different complex roots of any complex number. To get them all, we note that e iτ e iτ+πi e iτ+4πi.... Thus, z /n re iτ ) /n re iτ+kπ)) /n r /n e iτ+kπ)/n, k,,,... n Example. Find the cube roots of i, that is i) /3. Solution. First write i in polar form: since Rei) x and Imi) y, we get, using the formulas above, The cube roots of i are then given by r x + y τ tan y/x tan π/ i) /3 { /3 e iπ/)/3, /3 e iπ/+π)/3, /3 e iπ/+4π)/3} {e iπ/6, e i5π/6, e i3π/} If we want to express these in Cartesian form, we can compute e iπ/6 3/ + i/, e i5π/6 3/ + i/ and e i3π/ i, so that { i) /3 ± } 3/ + i/, i A.3 Trigonometric Identities Most trigonometric identities can be derived quite simply from Euler s formula and the reverse Euler formula, using multiplication of complex exponentials. cosα+β) Re e iα+β)) Re e iα e iβ) Re cos α + i sin α)cos β + i sin β)) cosα) cosβ) sinα) sinβ) 36

37 and sinα+β) Im e iα+β)) Im e iα e iβ) Im cos α + i sin α)cos β + i sin β)) sinα) cosβ)+sinβ) cosα) The double angle, half angle identities, and so on are proved similarly. Using the reverse Euler formula gives a convienient way to express powers of the trig functions sin n τ and cos n τ in terms of cos kτ and sin kτ for k n, and vice versa. Example. Write sin 3 τ in terms of sin kτ, cos kτ for k, 3. Solution. By the reverse Euler formula, sin 3 τ 3 ie iτ e iτ ) ) 3 i 3 8 e 3iτ 3e iτ + 3e iτ e 3iτ ) i 8 [ e 3iτ e 3iτ ) 3 e iτ e iτ ) 4 sin 3τ sin τ Example. Write sin 4τ in terms of powers of cos τ, sin τ Solution. We have sin 4τ Im e 4iτ ) [ e iτ Im ) 4 Im [cos τ + i sin τ) 4 Expanding out using the binomial theorem, and throwing away everything which doesn t multiply i, we get sin 4τ Im cos 4 τ 6 cos τ sin τ + sin 4 τ + 4icos 3 τ sin τ cos τ sin 3 τ) ) 4 cos 3 τ sin τ 4 cos τ sin 3 τ A.4 Sinusoidal Identity There is a nice trick for writing a linear combintation a cos ωτ + b sin ωτ in the form The trick is to write A cos ωτ φ) a cos ωτ + b sin ωτ Re a ib)cos ωτ + i sin ωτ)) Converting to polar form, we get a ib) Ae iφ where A a + b ) / and φ tan b/a tan b/a so φ tan b/a. That is, a cos ωτ + b sin ωτ Re a ib)cos ωτ + i sin ωτ)) Re Ae iφ e iωτ ) Re Ae iωτ φ)) A cos ωτ φ) where A a + b ) / and φ tan b/a 37

38 B Fourier Series It is a rather deep) mathematical fact that we can express any periodic function as a series in sines and cosines 7. A periodic function ft) is one such that ft) ft + P ), for all t and P is the period of f 8. Setting L P/ to be the half-period, we can write f by its Fourier series where ft) a + a n cos nπt L n a L a n L b n L L L L L L L ft) dt + n ft) cos nπt L dt ft) sin nπt L dt b n sin nπt L If ft) is even ft) f t)), or odd ft) f t)), we have the following If ft) is even, then b n for all n and If ft) is odd, then a n for all n and B. Periodic Extensions a n L b n L L L ft) cos nπt L dt ft) sin nπt L dt Given a function ft) which is defined on an interval [, L, we can construct various periodic extensions ft), which are periodic functions agreeing with ft) on [, L. These are constructed as Fourier series with coefficients computed using ft). We have the even extension of period P L, f ev t) f ev t), where f ev t) a + a n cos nπt L, n a n L L ft) cos nπt L dt the odd extension of period P L, f od t) f od t), where f od t) n b n sin nπt L, b n L L ft) sin nπt L dt 7 This can be stated by saying that sin nπt nπt and cos form an orthonormal basis in the space of periodic functions with L L period L. The inner product is given by integration over a single period. 8 There are many periods; for instance, P and 3P are also periods. We call the smallest such P the period. 38

39 Example. Find the Fourier sine series for the odd, period 4 function ft), such that ft) t on [,. Solution. The odd extension contains only sine terms, so we compute b n. In this case L is the half-period. b n t sin nπt dt t nπ cos nπt + nπ cos nπt dt 4 [ nπt cos nπ + sin nπ nπ nπ But sin nπt vanishes at both endpoints t and t, and cos nπ ) n. Thus we have ft) n b n sin nπt where b n 4 )n+ nπ B. Operations on Series Given a function ft) and its Fourier series, we can perform various operations on it to obtain the Fourier series of related functions. For instance, we can get the series for the following transformations of f: ft) cft) by multiplying each term in the series by c. ft) fct) by replacing t by ct in each term ft) ft ± L) by changing the sign of each term with n odd. This is because ) { nπt ± L) nπt cos nπt cos cos L L ± nπ L n odd n even and similarly for sines. ft) ft L/) by using the formulas cos and similarly for sines. nπt L/)) L ) nπt cos L nπ cos nπt L cos nπt L n, 4, 8,... cos nπt L n, 6,,... sin nπt L n, 5, 9,... sin nπt L n 3, 7,,... ft) t ft ) dt by integrating each term. The constant term if any) will have to be computed explicitly see example below). ft) f t) by taking the derivative of each term. Note that this is only valid if the original series has no points of discontinuity, e.g. if ft) is everywhere continuous. Example. Find the Fourier cosine series for the even extension of t on [, using the example above. Solution. Since t t dt t /, we will integrate the sine series 4 ) n+ n nπ sin nπt 39

40 term-wise to get our result. We compute t b n sin nπt dt b n nπ cos nπt t t b n nπ b n nπ Integrating a sine series therefore gives a cosine series with a n given by Collecting the constant terms, we have a n b n nπ 8 )n n π a n 8 ) n+ n π which isn t very explicit. We can compute a directly, using the definition 9 a t dt t dt In any case, the desired cosine series for the even extension of t on [, is a + n a n cos nπt, a 4 3, a n 8 )n n π cos nπt C Laplace Transform For a function ft), the Laplace transform Lf) F s) is defined by Note that L is linear, that is Lf) F s) e st ft) dt L aft) + bgt)) al ft)) + bl gt)) You should think of L as a map which takes a function ft) of t as its input, and gives a function F s) of s as its output. The Laplace transform takes most of the basic functions of t, such as t n, cos t, sin t, e αt and transforms them into rational functions of s, that is F s) typically qs) ps), qs), ps) polynomials in s Furthermore, it transforms the operation of differentiation in t into multiplication in s, thus transforming differential equations in t into algebraic ones in s. A table of Laplace transforms follows ft) aft) + bgt) F s) e st ft) dt af s) + bgs) e at ft) F s a) tft) ut a)ft a) f t) F s) e as F s) sf s) f) f t) s F s) sf) f ) ft) gt) F s)gs) ft) s F s) t n n! s n+ e at s a sin ωt ω s +ω cos ωt s +ω δt) 9 As an interesting corollary to our computation therefore, we find the value of the series P 8 ) n+ n n π 4 3 4

41 Example. Verify the identities Solution. Integrating by parts, we have Lt k ) e st t k dt tk e st s Lt k k! ) s k+ L e αt gt) ) Gs α) L sin βt) s + β + k s Repeating, we have Lt k ) k! L), so now we need only compute L): s k L) e st dt e st s s So t k e st dt + k s Ltk ) Lt k ) k! s k+ Now assume given a function gt) with Laplace transform Gs). We have L e αt gt) ) e st e αt gt) dt In particular since L) /s t ), we have Finally, we can use the reverse Euler identity L e αt) s α e s α)t) gt) dt Gs α) to get sin βt eiβt e iβt i Lsin βt) Le iβt ) Le iβt ) ) i i s iβ ) s + iβ ) s + iβ) s iβ) i β s + β s + β To compute the inverse Laplace transform of a function F s), we typically use partial fraction decomposition to split F s) up into a sum of terms whose inverse Laplace transform is obvious by examining the table. In the examples that follow, we will use the Heaviside cover up method for partial fraction decomposition, which allows us to compute the coefficients of some of the partial fractions quickly. In particular, it applies to linear factors only, and in the case of repeated linear factors, only to the one with the highest degree. See the examples for details. 4

42 Example. Find L s +4)s ) ). Solution. The denominator is already factored into the irreducible factors s + 4) and s ). Therefore we want A, B and C such that s + 4)s ) As + B s C s The Heaviside cover up method is applicable to find C, since s ) is a linear factor, and no higher powers of s ) appear in the denominator of any terms. We cover up the s ) in the denominator, and evaluate at s. That is, C s + 4) 5 s We then proceed as usual to find A and B. As + B s /5 s As + B)s ) + /5s + 4) s + 4)s ) s + 4)s ) Looking at coefficients of s in the numerator, we find So A /5, and examining coefficients of s, So B A /5. Thus, A + /5)s s B A)s s s + 4)s ) s + 5 s ) s We know Le t ) s, but the other term needs a little more nudging. We write s + s + 4 s s s + 4 Then it is obvious by looking at the table above that ) L s cos t ) + 4)s ) 5 sin t + et Example. Find L s+ s 4 s 3 +s ). Solution. Examining the denominator, we first notice that we can pull out a factor of s : s 4 s 3 + s s s s + ) and since s s + factors as s ), we have the rational function s + s s ) A s + B s + C s + D s ) The Heaviside cover up method applies to B and D, since the denominators of those terms contain the highest powers of their respective linear factors. We have B s + s ) s 4

43 and D s + s s Then A s + s + C s + s ) Ass ) + s ) + Cs s ) + s s s ) and the numerator of this must be equal to s +. Expanding the numerator, we have Ass ) + s ) + Cs s ) + s A + C)s A C)s + A )s + s + Comparing terms, we see that A + C) and A ), so that A 3 and C 3. Thus, s + s s ) 3 s + s 3 s + s ) To get the inverse Laplace transform of the last term, note the identity L e αt gt)) Gs α) from the table. We then write ) s + L s s ) 3 + t 3e t + te t since L te t ) s ). Example. Find L s+) s +4s+5 Solution. Complete the square in the denominator to get So we have the function ). s + 4s + 5 s + ) s + ) + s + ) s + ) + Remember that L e αt gt)) Gs α). The above is just the Laplace transform for cosine, but shifted by. Indeed we have L e t cos t ) s + ) s + ) + so ) s + ) L s e t cos t + 4s + 5 C. Convolution and the Laplace Transform Given two functions ft) and gt), we can form their convolution product, which is a new function of t given by f g)t) t fτ)gt τ) dτ This product is both associative, f g h) f g) h; and commutative, f g g f. The Laplace transform takes convolution in t into ordinary multiplication in s: L ft) gt)) F s)gs) This can sometimes be useful in computing otherwise difficult convolutions. 43

44 Example. Compute t m t n Solution. The integral t m t n t τ m t τ) n dτ is somewhat difficult to compute directly, involving m integration by parts. Using the Laplace transform on the other hand, we compute ) ) m! n! L t m t n ) s m+ s n+ m!n! ) s m+n+ m!n! m + n + )! m + n + )! s m+n+ so t m t n L m!n! m + n + )! )) m!n! m + n + )! sm+n+ m + n + )! tm+n+ C. Piecewise Functions & Delta Functions The Laplace transform is often used in conjunction with piecewise functions which are functions of the form f t) t < t < t ft) f t) t < t < t.. To express piecewise functions on one line, we typically use the unit step function defined as { t < ut) t > We can easily create a box function which is equal to on an interval [a, b and equal to elsewhere by writing t < a ft) ut a) ut b) a < t < b t > b Think of the first term as a switch which turns on a at time t a, and the second term as a switch which turns on a at time t b and so cancels out the first term. Using such box functions, we can write f t) t < t < t ft) f t) t < t < t.. as ft) ut t ) ut t ) ) f t) + ut t ) ut t ) ) f t) + This is in a more useful form for taking the Laplace transform, for instance see example in section 3.3) When we take derivatives of piecewise functions, we often obtain spikes. The primordial jump is the Dirac delta function δt) which is not really a function but has the following properties δt), fort, δt) dt ɛ ɛ δt) dt 44

45 so it has a finite amount of area under the curve at the point t. More generally, δt) satisfies δt a)ft) dt fa) that is, if we integrate a function ft) against a delta function living at t a, we get the value of the function at a. The unit step function has a jump of height at t, and its derivative u t) has a spike of total area at t, namely u t) δt) This is true more generally. Any time a piecewise function has a jump of height h negative if the jump is down ), its generalized derivative will have a spike of area h given by a term hδt). Example. Let ft) be given by t < t < t < ft) < t < 3 t > 3 Write ft) in terms of step functions and find its generalized derivative f t). Solution. To write ft) in terms of step functions, we think of creating boxes defined on [, and [, 3. Then ft) ut) ut ) ) t + ut ) ut 3) ) Note that ft) has two jumps of height since we jump down as we go from left to right) at t and at t 3. It s generalized derivative is most easily thought of as follows. On the regions where f t) is well-defined, we have t < f < t < t) < t < 3 t > 3 But we also have to include the spikes at t and t 3 of area -. We obtain f t) ut) ut ) ) δt ) δt 3) D Matrices Here we recall some facts about matrices. Recall that matrices can be multiplied together AB makes sense if the number of columns of A equals the number of rows of B), to get a muliplication which is associative: ABC) AB)C, but not neccessarily commutative: AB BA in general. The unit element is the identity matrix I...., meaning that AI IA A for all matrices A for which either product makes sense. We call this the generalized derivative since the function doesn t technically have a derivative in the usual sense. Any time you differentiate something to get something with jumps or spikes, you re using generalized derivatives. 45

46 An n n matrix A is invertible if and only if its determinant is nonzero: det A. Recall that the determinant of a matrix is given by det A det a b c d ad bc For a 3 3 matrix, you can evaluate its determinant by expanding along a row or column. For each entry you have a term equal to that entry times the determinant of the matrix obtained by deleting the corresponding row and column. Then you add the three terms up, with an alternating sign + starting in the upper left, and changing each time you step right or down). For example, det A det a b c d e f g h i adet e h f i bdet d g So provided det A, there is a matrix A such that AA A A I f i + cdet d g For matrices, there is a handy formula for the inverse of a matrix: [ a b c d [ d b det A c a e h aei hf) bdi gf) + cdh ge) where the diagonal entries switch places, the off diagonal enteries get opposite signs, and we multiply the whole thing by the scalar quantity /det A. D. Eigenvalues & Eigenvectors Given a matrix A, it frequently happens that we are interested in finding scalars λ and vectors v such that the action of A on v is just the same as multiplying v by λ. Av λv Solutions to this equation are called eigenvalues for the scalars λ, and eigenvectors for the vectors v. The above equation can be rewritten as A λi)v, and of course the only way we can get a nonzero vector v satisfying this equation is if A λi) is not invertible otherwise we d have v A λi) ). So we determine the possible eigenvalues from the eigenvalue equation det A λi) Then, for a given eigenvalue λ, we can try to find an associated eigenvector by solving the eigenvector equation A λi)v In the case of a matrix A, the eigenvalue equation has a special form which may be worth memorizing: [ a b A det A λi) λ tr A + det A, tr A a + d), det A ad bc) c d where tr A a + d is called the trace of A, defined for any matrix as the sum of the diagonal entries. Be careful to note the minus sign in front of the trace in the quadratic polynomial for λ. Remark. The case of repeated eigenvalues is discussed in section 4. and will not be covered here. 46

47 Example. Find the eigenvalues and corresponding eigenvectors for the matrix [ 3 4 A 4 3 Solution. The eigenvalue equation is det A λi) det 3 λ λ 3 λ) 3 λ)+6 λ 5 λ tr A+det A λ+5)λ 5) so we have eigenvalues λ 5 and λ 5. For λ 5, we want to solve A 5I)v [ [ a b [ which gives the equation a + 4b the second equation 4a 8b is equivalent to the first as it must be to get a nonzero solution). We can choose any convenient value for a and b which solves this equation, say a, b. So [ λ 5, v is an eigenvalue, eigenvector pair. Similarly, for λ 5 we solve A 5)I)v So we can take a, b, for instance and get for our second pair. D. Matrix Exponentials λ 5, v [ [ [ [ a b Another important quantity associated to a matrix A is the matrix exponential e ta I + ta + t! A + t3 3! A3 + n t n n! An obtained by applying the taylor series e x n xn n! to the matrix quantity ta. The result is a matrix which is a function of t, and is important in the theory of homogeneous systems of equations see section 4.3). It is a solution to the differential equation characterizing fundamental matrices for A see section 4.3), and is the unique one satisfying the additional initial condition e ta ) Ae ta, e A I We can verify this by examining the series: e ta ) d ) I + ta + t dt! + e A I + A + A + I A + t! A + 3t 3! A3 A I + It can be shown that the resulting series of matrices converges for all t ) t A + t!! A + 47

48 In some cases we can evaluate the series directly, for instance when a matrix is diagonal for which entries are zero except along the diagonal), or strictly upper/lower triangular for which the only nonzero entries are either above the main diagonal or below it). In other cases, we can compute the matrix exponential using fundamental matrices, which is discussed in section 4.3. Example. Compute e ta and e tb, where A [ [, B Solution. A is a diagonal matrix, as defined above. As we compute its successive powers, we see that we end up just taking the corresponding powers of the entries along the diagonal [ [ [ A, A 3 3 3, A n n n, etc. Then for the exponential series, we obtain e ta n t n n! An n t n n! [ n n [ n t) n n! n t) n n! [ e t e t B is a strictly upper triangular matrix. Such matrices have the property called nilpotent) that B n for all sufficiently large n. In this case, [ B, B 3 BB,..., B n, for n > So in the exponential series for B, we have e tb I + tb + t! B + I + tb [ t Finally, we might expect the usual rule for multiplication of exponentials to hold for matrices, that is, e A+B e A e B. This is not true in general, however. For this to hold, we need A and B to commute; that is, e A+B e A e B only if AB BA The reason for this can be seen by examining the series: whereas e A+B I + A + B) + A + B) + I + A + B) + e A e B A + B + AB + BA ) + ) ) I + A + I A + + B + B + I + A + B) + A + B + AB ) + so the two series are only equal up to second order at least, but a similar thing happens for all orders) if AB + BA AB AB BA. As a useful corollary to this, note that e ta ) e ta that is, the inverse matrix to e ta is just the matrix exponential itself, evaluated at t. This follows from using the multiplication rule, which applies here since A clearly commutes with itself AA A AA), so which is the defining property for the inverse matrix e ta e ta e ta A) e t I 48

49 Example. Compute e ta where A [ [ }{{} B [ + } {{ } C Solution. A is the sum of a diagonal matrix B and a strictly upper triangular matrix C. Using the techniques from the previous example, we compute [ [ e tb e t e t, e tc t In order to use the multiplication rule, we need to check commutativity: [ [ [ BC [ [ [ CB Since BC CB, we can compute e ta e tb+c) e tb e tc [ e t [ t e t [ e t te t e t 49

50 Index amplitude response, 5 autonomous equation, 8 cartesian form, 34 change of variables, 6 example, 6 characteristic polynomial, companion matrix, complete case, 3 complex conjugate, 36 complex number, 34 constant coefficient, convolution, 9 example, 9,, 43 convolution, 43 critical point, 3 critical point, 8 defective case, 3 delta function, 9 delta function, 44 determinant, 45 diagonal matrix, 47 differential operator, eigenvalue, eigenvalue, 46 eigenvalue equation, 46 eigenvector, eigenvector, 46 eigenvector equation, 46 Euler s formula, 34 Euler s method example, Euler s method, even function, 38 exponential response example, 3 exponential response formula, first order equation example, 4 6 Fourier series example, 6, 39 Fourier series, 38 fundamental matrix, 47 example, 6 fundamental matrix, 5 general solution, example,, 3 5 system, general solution, generalized derivative, 45 example, 45 Heaviside cover up, 4 Heaviside coverup, 7 homogeneous equation, example, homogeneous equation, homogeneous solution, system, 3, homogeneous solution, homogeneous solutions system, homogeneous system example, 3 5 homogeneous system, 3 identity matrix, 45 imaginarypart, 34 inhomogeneous system of equations, 7 initial condition, 3 initial value problem, example, 7, 6 system, initial value problem, 3 input, integrating factor example, 5 integrating factor, 5 inverse of a matrix, 46 invertible matrix, 45 isoclines example, 7 isoclines, 7 Jacobian matrix, 3 jump, 45 Laplace transform, 7 example, 7, 8, 4 43 linear equation, example, 5 linear equation, linear system system, linear system, 3 5

51 matrix, 3 matrix exponential, 6 example, 48, 49 matrix exponential, 47 nodal source example, 3 nonlinear system, 3 example, 3 odd function, 38 order, oscillatory input example, 5, 3 particular solution,, example, 3 6, 8 system, 3, particular solution, period, 38 periodic, 5 periodic extension example, 39 periodic function, 38 periodic solution, phase diagram example, 9 3 phase portrait example, 9 phase portrait, 8 phase shift, 5 piecewise function, 7, 8 example, 8, 45 piecewise functions, 44 polar form, 34 pseudofrequency, pseudofrequency, pseudoperiod, spiral source example, 3 stability example, 9 stable, 9 critical point, 9 step size, system of equations, 3 trace, 46 transfer function, 9 triangular matrix, 47 undetermined coefficients example, 4, 5 undetermined coefficients, 4 unit step function, 44 unstable critical point, 9 variation of parameters example, 8 variation of parameter, 7 vector, 3 weight function example, 9, weight function, 9 real part, 34 resonance, 6 response, rest initial conditions, 9 reverse Euler formula, 35 saddle, 9 example, 9 semistable critical point, 9 seperable equation example, 4 seperable equation, 4 spike, 44 5