5 LINEAR SYSTEMS 5.0 Introduction ‘As we’ ve seen, in one-dimensional phase spaces the flow is extremely confined— all trajectories are forced to move monotonically or remain constant. In higher- dimensional phase spaces, trajectories have much more room to maneuver, and so a wider range of dynamical behavior becomes possible. Rather than attack all this complexity at once, we begin with the simplest class of higher-dimensional sys- tems, namely linear systems in two dimensions. These systems are interesting in their own right, and, as we’fl see later, they also play an important role in the clas- sification of fixed points of nonlinear systems. We begin with some definitions and examples. 5.1 Definitions and Examples A two-dimensional linear system is a system of the form k= axtby jeextdy where a, b,c, d are parameters. If we use boldface to denote vectors, this system can be written more compactly in matrix form as 5.1 DEFINITIONS AND EXAMPLES 123Such a system is linear in the sense that if x, and x, are solutions, then so is any linear combination ¢,x, +¢,x,. Notice that x=0 when x=0, so x*=0 is al- ways a fixed point for any choice of A. The solutions of = Ax can be visualized as trajectories moving on the (x,)) plane, in this context called the phase plane. Our first example presents the phase plane analysis of a familiar system. EXAMPLE 5.1.1: As discussed in elementary physics courses, the vibrations of a mass hanging from a linear spring are governed by the linear differential equation mitke=0 wm where m is the mass, & is the spring constant, and x is the displacement of the mass from equilibrium (Figure 5.1.1). Give a phase plane analysis of this simple harmonic oscillator. . Solution: As you probably recall, it’s easy to solve (1) ana- lytically in terms of sines and cosines. But that’s precisely what makes linear equations so special! For the nonlinear equations Kk of ultimate interest to us, it’s usually impossible to find an ana- lytical solution. We want to develop methods for deducing the behavior of equations like (1) without actually solving them. The motion ir. the phase plane is determined by a vector x field that comes from the differential equation (1). To find this vector field, we note that the state of the system is char- acterized by its current position x and velocity v; if we know the values of both x and ¥, then (1) uniquely determines the future states of the system. Therefore we rewrite (1) in terms of x and v, as fol- lows: Figure 5.1.1 kev Qa) bate, (26) Equation (2a) is just the definition of velocity, and (2b) is the differential equation (1) rewritten in terms of v. To simplify the notation, let @* = k/m . Then (2) be- comes (a) wx. (3b) pe ?x) at each point (x,v), and there- . The system (3) assigns a vector (4,9) = (v,— fore represents a vector field on the phase plane. 124 LINEAR SYSTEMSFor example, let’s see what the vector field looks like when we're on the x-axis, ‘Then v=0 and so (,¥) = (0,-@*x). Hence the vectors point vertically downward for positive x and vertically upward for negative x (Figure 5.1.2). As x gets larger in magnitude, the vectors (0,—@*x) get longer. Similarly, on the v-axis, the vector field is (%,0)=(v,0), which points to the right when y>0 and to the left when v<0. As we move around in phase space, the vector shown in Figure 5.1.2. hange direction as Just as in Chapter 2, it is helpful to visualize the vector field in terms of the motion of an imaginary fluid. In the present case, we imagine that a fluid is flowing steadily on the phase plane with a local velocity given by (%,¥) = (v,-@°x). Then, to find the tra- a Jjectory starting at (x), ¥)), we place an imaginary particle or phase point at (Xs vp) and watch how it is carried Figure 5.1.2 around by the flow. The flow in Figure 5.1.2 swirls about the origin. The origin is special, like the eye of a hurricane: a phase point placed there would remain motionless, because (%,) = (0,0) when (x,v) = (0,0); hence the origin is a fixed point. But a phase point starting anywhere else would circulate around the origin and eventually return to its starting point. Such trajectories form closed orbits, as shown in Figure 5.1.3. Figure 5.1.3 is called the phase portrait of the system—it shows the overall picture of trajectories in phase X space. What do fixed points and closed or- bits have to do with the original prob- fem of a mass on a spring? The answers are beautifully simple. The fixed point (x,v)=(0,0) corresponds to static equilibrium of the system: the mass is at rest at its equilibrium position and will remain there forever, since the spring is relaxed. The closed orbits have a more in- teresting interpretation: they correspond to periodic motions, i.c., oscillations of the mass. To see this, just look at some points on a closed orbit (Figure 5.1.4). When the displacement x is most negative, the velocity v is zero; this corre- sponds to one extreme of the oscillation, where the spring is most compressed (Figure 5.1.4). < v Figure 5.1.3 5.1 DEFINITIONS AND EXAMPLES 125Figure 5.1.4 In the next instant as the phase point flows along the orbit, it is carried to points where x has increased and v is now positive; the mass is being pushed back to- ward its equilibrium position. But by the time the mass has reached x = 0, it has a Jarge positive velocity (Figure 5.1.4b) and so it overshoots x= 0. The mass even- tually comes to rest at the other end of its swing, where x is most positive and v is, zero again (Figure 5.1.4c). Then the mass gets pulled up again and eventually com- pletes the cycle (Figure 5.1.4d). ‘The shape of the closed orbits also has an interesting physical interpretation. The orbits in Figures 5.1.3 and 5.1.4 are actually ellipses given by the equation ax? +v? =C, where C2 0 is a constant. In Exercise 5.1.1, you are asked to derive this geometric result, and to show that it is equivalent to conservation of energy. EXAMPLE 5.1.2: Graph the phase portrait Solve the linear system X= Ax, where A -( 126 LINEAR SYSTEMSas a varies from > to +0 , showing the qualitatively different cases. Solution: The system is x (: OYx a} lo -hyl Matrix multiplication yields kear ae y which shows that the two equations are uncoupled; there's no x in the y-equation and vice versa. In this simple case, each equation may be solved separately. The solution is X(f) = Xp (Ja) WOD= ye". (1b) The phase portraits for different values of a are shown in Figure 5.1.5. In each case, y(t) decays exponentially. When a <0, x(1) also decays exponentially and so all trajectories approach the origin as ¢ + 0. However, the direction of approach depends on the size of a compared to —1. @a<-l (b) a =-1 (©) -l0 Figure 5.1.5 5.1 DEFINITIONS AND EXAMPLES 127In Figure 5.1.5a, we have a<—1, which implies that x(t) decays more rapidly than y(¢). The trajectories approach the origin tangent to the slower direction (here, the y-direction). The intuitive explanation is that when a is very negative, the trajectory slams horizontally onto the y-axis, because the decay of x(t) is al- most instantaneous. Then the trajectory dawdles along the y-axis toward the ori- gin, and so the approach is tangent to the y-axis, On the other hand, if we look backwards along a trajectory (t > 9), then the trajectories all become parallel to the faster decaying direction (here, the x-direction). These conclusions are easily proved by looking at the slope dy/dx = j/% along the trajectories; see Exercise In Figure 5.1.5a, the fixed point x* is called a stable node. Figure 5.1.5b shows the case a = —1. Equation (1) shows that y(t)/ x(t) = yp /x)= constant, and so all trajectories are straight lines through the origin. This is a very special case—it occurs because the decay rates in the two directions are precisely equal. In this case, x * is called a symmetrical node or star. When -10 (Figure 5.1.5e), x* becomes unstable, due to the exponen- tial growth in the x-direction. Most trajectories veer away from x * and head out to infinity. An exception occurs if the trajectory starts on the y-axis; then it walks a tightrope to the origin. In forward time, the trajectories are asymptotic to the x- axis; in backward time, to the y-axis. Here x* = 0 is called a saddle point. The y-axis is called the stable manifold of the saddle point x*, defined as the set of initial conditions x, such that x(t) > x* as 1c. Likewise, the unstable mani- fold of x* is the set of initial conditions such that x(1) > x* as >» —ee. Here the unstable manifold is the x-axis. Note that a typical trajectory asymptotically ap- proaches the unstable manifold as fe, and approaches the stable manifold as 1 -e0. This sounds backwards, but it’s right! = Stability Language It’s useful to introduce some language that allows us to discuss the stability of different types of fixed points. This language will be especially useful when we an- alyze fixed points of nonlinear systems. For now we'll be informal; precise defini- tions of the different types of stability will be given in Exercise 5.1.10. We say that x* = 0 is an attracting fixed point in Figures 5.1.5a-c: all trajectories that start near x* approach it as t — co. That is, x(f) > x* as fe. In fact x* attracts ail trajectories in the phase plane, so it could be called globally attracting. ‘There’s a completely different notion of stability which relates to the behavior 128 LINEAR SYSTEMSof trajectories for all time, not just as t > ce . We say that a fixed point x * is Lia- punov stable if all trajectories that start sufficiently close to x * remain close to it for all time. In Figures 5.1.5a~d, the origin is Liapunov stable. Figure 5.1.5d shows that a fixed point can be Liapunoy stable but not attracting. This situation comes up often enough that there is a special name for it. When a fixed point is Liapunoy stable but not attracting, it is called neutrally stable. Nearby trajectories are neither attracted to nor repelled from a neutrally stable point. As a second example, the equilibrium point of the simple harmonic oscilla- tor (Figure 5.1.3) is neutrally stable. Neutral stability is commonly encountered in mechanical systems in the absence of friction. Conversely, it’s possible for a fixed point to be attracting but not Liapunov stable; thus, neither notion of stability im- plies the other, An example is given by the following vector field on the circl 6=1-cos@ (Figure 5.1.6). Here 6* =0 attracts all trajectories as ¢ > o , but it is not Liapunov stable; there are trajectories that start infini- tesimally close to @* but go ona very large excursion be- fore returning to @*. However, in practice the two types of stability often oc- cur together. If a fixed point is both Liapunov stable and at- tracting, we'll call it stable, or sometinies asymptotically stable. a Finally, x* is unstable in Figure 5.1.5e, because it is heither attracting nor Liapunov stable. A graphical convention: we'll use open dots to denote unstable fixed points, and solid black dots to denote Liapunovy stable fixed points. This convention is consis- tent with that used in previous chapters. Figure 5. 5.2 Classification of Linear Systems ‘The examples in the last section had the special feature that two of the entries in the matrix A were zero, Now we want to study the general case of an arbitrary 2x2 matrix, with the aim of classifying all the possible phase portraits that can occur. Example 5.1.2 provides a clue about how to proceed. Recall that the x and y axes played a crucial geometric role. They determined the direction of the trajecto- ries as ¢ > to. They also contained special straight-line trajectories: a trajectory starting on one of the coordinate axes stayed on that axis forever, and exhibited simple exponential growth or decay along it. For the general case, we would like to find the analog of these straight-line tra- jectories. That is, we seek trajectories of the form xQ=e%v, 2) 5.2 CLASSIFICATION OF LINEAR SYSTEMS 129where v #0 is some fixed vector to be determined, and A is a growth rate, also to be determined. If such solutions exist, they correspond to exponential motion along the line spanned by the vector v. To find the conditions on v and A, we substitute x(t) =e" into x= Ax, and obtain 2e“v = e* Av . Canceling the nonzero scalar factor e” yields Av=Ay, (3) which says that the desired straight line solutions exist if v is an eigenvector of A with corresponding eigenvalue 2.. In this case we call the solution (2) an eigen- solution. Let’s recall how to find eigenvalues and eigenvectors. (If your memory needs more refreshing, see any text on linear algebra.) In general, the eigenvalues of a matrix Aare given by the characteristic equation det(A— AI) =0, where I is the identity matrix. For a 2x2 matrix a-(Z a the characteristic equation becomes ofe4 8). a Expanding the determinant yields B-ta+h=0 @ where T=trace(A)=a+d, A= det(A) =ad-be. Then 1. = Ve cit =4A : are the solutions of the quadratic equation (4). In other words, the eigenvalues de- pend only on the trace and determinant of the matrix A. The typical situation is for the eigenvalues to be distinct: A, # A, . In this case, a theorem of linear algebra states that the corresponding eigenvectors v, and v, are linearly independent, and hence span the entire plane (Figure 5.2.1). In particular, any initial condition x, can be written as a linear combination of eigenvectors, say Xp = 6, $O,¥5. 130 LINEAR SYSTEMSXq =p +OV9 This observation allows us to write down the general solution for x(t)—it is simply x(t)=ce"'v, +e,e"'v,. (6) Why is this the general solution? First of all, it is a linear combination of solu- tions to X= Ax, and hence is itself a solution, Second, it satisfies the initial condi- tion x(0)=x,, and so by the existence and uniqueness theorem, it is the only solution. (See Section 6.2 for a general statement of the existence and uniqueness theorem.) EXAMPLE 5.2.1: Solve the initial value problem ¢=x+y, j=4x—2y, subject to the initial condition (x.y) = (2-3). Solution: The corresponding matrix equation is x) (1 life 3) 4 ahyl First we find the eigenvalues of the matrix A . The matrix has T=—1 and A=-6, so the characteristic equation is 2° +A —6 =0. Hence 4, Next we find the eigenvectors. Given an eigenvalue A, the corresponding eigenvector v= (¥,,¥,) satisfies (0 salt) lec cak 0 For 4, =2, this yields (: “e)-(o} which has a nontrivial solution 5.2 CLASSIFICATION OF LINEAR SYSTEMS 131(¥;,¥2) = (LD), or any scalar multiple thereof. (Of course, any multiple of an cigen- vector is always an eigenvector; we try to pick the simplest multiple, but any one will 4 1¥y,) (0 do.) Similarly, for A, = -3, the eigenvector equation becomes ( iI ‘)-(3} = which has a nontrivial solution (9, v.) = (1,—4). In summary, vel eG) Next we write the general solution as a linear combination of eigensolutions. From (6), the general solution is a] Finally, we compute c, and c, to satisfy the initial condition (9, yp) = (2.—3). At 1=0, (7) becomes [a)- >, and the stable manifold as 1» —co, Figure 5.2.2 shows the phase portrait. = De Figure 5.2.2 EXAMPLE 5.2.3: Sketch a typical phase portrait for the case A, < A, <0. Solution: First suppose 2, <2, <0. Then both eigensolutions decay exponen- tially. The fixed point is a stable node, as in Figures 5.1.5a and 5.1.5c, slow eigendirection except now the eigenvectors are not mutually perpendicular, in general. Trajectories typically approach the origin tangent to the slow eigendirec- tion, defined as the direction spanned by the eigenvector with the smaller fast eigendiretion Al. In backwards time (1 —» -s»), the trajectories become parallel to the fast eigendirection. Figure 5.2.3 shows the phase portrait. (If we reverse all the arrows in Figure 5.2.3, we obtain a typical phase portrait for an unstable node.) w Figure 5.2.3 EXAMPLE 5.2. What happens if the eigenvalues are complex numbers? 5.2 CLASSIFICATION OF LINEAR SYSTEMS 133Solution: If the eigenvalues are complex, the fixed point is either a center (Fig- ure 5.2.4a) or a spiral (Figure 5.2.4b). We've already seen an example of a center in the simple harmonic oscilla- tor of Section 5.1; the origin is surrounded by a family of > closed orbits. Note that centers Cc are neutrally stable, since nearby trajectories are neither attracted to nor repelled from Oe Oo the fixed point. A spiral would occur if the harmonic oscillator Figure 5.2.4 were lightly damped. Then the wajectory would just fail to close, because the oscillator loses a bit of energy on each cycle. To justify these statements, recall that the eigenvalues are A, a(e +4 ). ‘Thus complex eigenvalues occur when vT-4A<0. To simplify the notation, let’s write the eigenvalues as A. =atio where a@=t/2, w=tV4a—7' By assumption, w #0. Then the eigenvalues are distinct and so the general solu- tion is still given by x(t) = cje"'v, +c,e"v, . But now the c’s and v’s are complex, since the A’s are. This means that x(¢) in- volves linear combinations of e'“*®”. By Euler’s formula, e“” = cos aot + isin ar. Hence x(t) is a combination of terms involving e” cosox and e” sine, Such terms represent exponentially decaying oscillations if a@ = Re(A) <0 and growing oscillations if o> 0. The corresponding fixed points are stable and unstable spi- rals, respectively. Figure 5.2.4b shows the stable case. If the eigenvalues are pure imaginary (ot = 0), then all the solutions are periodic with period T= 2z/@. The oscillations have fixed amplitude and the fixed point is acenter. For both centers and spirals, it’s easy to determine whether the rotation is clock- wise or counterclockwise; just compute a few vectors in the vector field and the sense of rotation should be obvious. = 134 LINEAR SYSTEMSEXAMPLE 5.2.5: In our analysis of the general case, we have been assuming that the eigenvalues are distinct. What happens if the eigenvalues are equal? Solution: Suppose A, = 2, = A. There are two possibilities: either there are two independent eigenvectors corresponding to 2, or there’s only one. If there are two independent eigenvectors, then they span the plane and so every vector is an eigenvector with this same eigenvalue A. To see this, write an arbitrary vector xX, as a linear combination of the two eigenvectors: Xy =G\¥, +¢,¥). Then AX, = ACC, V2) = GAY, +e,Av, = AK, so X, is also an eigenvector with eigenvalue A. Since multiplication by A sim- ply stretches every vector by a factor A, the matrix must be a multiple of the identity: HB) Then if 4#0, all trajectories are straight lines through the origin (x(t) =e”x,) and the fixed point is a star node (Figure 5.2.5) Figure 5.2.5 On the other hand, if 2 =0, the whole plane is filled with fixed points! (No sur- prise—the system is x = 0.) The other possibility is that there’s only one eigenvector (more accurately, the eigenspace corresponding to 2 is one-dimensional.) For example, any matrix of the form A=| 5 ; , with b#0_ has only a one-dimensional eigenspace (Exer- cise 5.2.11). When there’s only one eigendirection, the fixed point is a degenerate node. A 5.2 CLASSIFICATION OF LINEAR SYSTEMS 135typical phase portrait is shown in Figure 5.2.6. As t > +0 and also as t > —o, eigendirection all trajectories become paral- lel to the one available eigendirection. A good way to think about the degenerate node is to imagine that it has been cre- Figure 5.2.6 ated by deforming an ordi- nary node. The ordinary node has two independent eigendirections; all trajectories are parallel to the slow eigendirection as t —> © , and to the fast eigendirection as ¢ > —oo (Figure 5.2.7a). fast slow (a) node (b) degenerate node Figure 5.2.7 Now suppose we start changing the parameters of the system in such a way that the two eigendirections are scissored together. Then some of the trajectories will get squashed in the collapsing region between the two eigendirections, while the sur- viving trajectories get pulled around to form the degenerate node (Figure 5.2.7b). Another way to get intuition about this case is to realize that the degenerate node is on the borderline between a spiral and a node. The trajectories are trying to wind around in a spiral, but they don’t quite make it. = Classification of Fixed Points By now you’re probably tired of all the examples and ready for a simple classi- fication scheme. Happily, there is one. We can show the type and stability of all the different fixed points on a single diagram (Figure 5.2.8). 136 LINEAR SYSTEMS7-44 =0 unstable nodes “0 Is saddle points stable spirals A Stable nodes non-isolated fixed points stars, degeneraté nodes Figure 5.2.8 ‘The axes are the trace 7 and the determinant A of the matrix A. All of the infor- mation in the diagram is implied by the following formulas: dya=4 (reve -4a), ie err The first equation is just (5). The second and third can be obtained by writing the characteristic equation in the form (2-4, )(A-A,)=# - 12+ A=0. To arrive at Figure 5.2.8, we make the following observations: If A <0, the eigenvalues are real and have opposite signs; hence the fixed point isa saddle point. If A>0, the eigenvalues are either real with the same sign (nodes), or complex conjugate (spirals and centers). Nodes satisfy t?-4A>0 and spirals satisfy 1? 4A <0. The parabola 1? 4A =0 is the borderline between nodes and spi- rals; star nodes and degenerate nodes live on this parabola. The stability of the nodes and spirals is determined by t. When 7 < 0, both eigenvalues have negative zeal parts, so the fixed point is stable. Unstable spirals and nodes have t > 0. Neu- trally stable centers live on the borderline t = 0 , where the eigenvalues are purely imaginary. If A=0, at least one of the eigenvalues is zero. Then the origin is not an iso- lated fixed point. There is either a whole line of fixed points, as in Figure 5.1.5d, or a plane of fixed points, if A=0. Figure 5.2.8 shows that saddle points, nodes, and spirals are the major types of fixed points; they occur in large open regions of the (A,t) plane. Centers, stars, degenerate nodes, and non-isolated fixed points are borderline cases that occur along curves in the (A,t) plane. Of these borderline cases, centers are by far the most important. They occur very commonly in frictionless mechanical systems where energy is conserved. 5.2 CLASSIFICATION OF LINEAR SYSTEMS 137EXAMPLE 5.2.6: io Classify the fixed point x* = 0 for the system X= Ax, where A -( i): Solution: ‘The matrix has A=~2, hence the fixed point is a saddle point. = EXAMPLE 5.2.7: eee Redo Example 5.2.6 for A= ( a Solution: Now A=5 and t=6. Since A>0 and t? -4A=16>0, the fixed point is a node. It is unstable, since T > 0. u 5.3 Love Affairs To arouse your interest in the classification of linear systems, we now discuss a simple model for the dynamics of love affairs (Strogatz 1988). The following story illustrates the idea. Romeo is in love with Juliet, but in our version of this story, Juliet is a fickle lover. The more Romeo loves her, the more Juliet wants to run away and hide. But when Romeo gets discouraged and backs off, Juliet begins to find him strangely at- tractive. Romeo, on the other hand, tends to echo her: he warms up when she loves him, and grows cold when she hates him, Let R(t) = Romeo’s love/hate for Juliet at time 1 J(t) = Juliet’s love/hate for Romeo at time r. Positive values of R, J signify love, negative values signify hate. Then a model for their star-crossed romance is Raal J=-bR where the parameters @ and b are positive, to be consistent with the story. The sad outcome of their affair is, of course, a neverending cycle of love and hate; the governing system has a center at (R, J) = (0,0). At least they manage to achieve simultaneous love one-quarter of the time (Figure 5.3.1). 138 LINEAR SYSTEMSxl Figure 5.3.1 Now consider the forecast for lovers governed by the general linear system R=aR+bs J=cR+d) where the parameters a, b, c, d may have either sign. A choice of signs specifies the romantic styles. As named by one of my students, the choice a>0, b>0 means that Romeo is an “eager beaver”—he gets excited by Juliet’s love for him, and is further spurred on by his own affectionate feelings for her. It’s entertaining to name the other three romantic styles, and to predict the outcomes for the various pairings. For example, can a “cautious lover” (a <0, b > 0) find true love with an eager beaver? These and other pressing questions will be considered in the exer cises. EXAMPLE 5.3.1: What happens when two identically cautious lovers get together? Solution: The system is R=aR+bs J=bR+aJ with a<0, b>0. Here a is a measure of cautiousness (they each try to avoid throwing themselves at the other) and b is a measure of responsiveness (they both get excited by the other’s advances), We might suspect that the outcome depends on the relative size of a and b. Let’s see what happens. The corresponding matrix is a(t? “(ba which has t=2a<0, Aza’-b?, t?-4A=4b> >0. 5.3 LOVE AFFAIRS 139Hence the fixed point (R, J) = (0,0) is a saddle point if a’ b° . The eigenvalues and corresponding eigenvectors are Azatb, v,=(), a,=a-b, v,=(-1). Since a+b>a—b, the eigenvector (1,1) spans the unstable manifold when the origin is a saddle point, and it spans the slow eigendirection when the origin is a stable node. Figure 5.3.2 shows the phase portrait for the two cases. aoe ech Figure 5.3.2 If a? > b*, the relationship always fizzles out to mutual indifference. The lesson seems to be that excessive caution can lead to apathy. If a® ee, and parallel to the x-direction as t > 20. (Hint: Examine the slope dy/dx = j/i.) Write the following systems in matrix form. Sat 5.16 <= Sketch the vector field for the following systems. Indicate the length and direction of the vectors with reasonable accuracy. Sketch some typical trajectories. 5.7 f=x,yaxty 5.18 4=-2y, pax Xe 5.1.9 Consider the system * =~-y, ) a) Sketch the vector field. b) Show that the trajectories of the system are hyperbolas of the form x? — y = C. (Hint: Show that the governing equations imply xé— yy = 0 and then integrate both sides.) ©) The origin is a saddle point; find equations for its stable and unstable mani- folds. d) The system can be decoupled and solved as follows. Introduce new variables w and v, where w= x+y, y=x—y. Then rewrite the system in terms of w and v. Solve for u(t) and v(t), starting from an arbitrary initial condition (ty, Vp). ) What are the equations for the stable and unstable manifolds in terms of w and ve £) Finally, using the answer to (d), write the general solution for x(¢) and y(t), starting from an initial condition (x),y). 5.1.10 (Attracting and Liapunov stable) Here are the official definitions of the various types of stability. Consider a fixed point x * of a system x = f(x). We say that x* is attracting if chere is a 6 > 0 such that timx() = x* when- ever || x(0)~x*||< 8. In other words, any trajectory that starts within a distance 5 of x* is guaranteed to converge to x* eventually. As shown schematically in Fig- ure 1, trajectories that start nearby are allowed to stray from x * in the short run, but they must approach x* in the long run, In contrast, Liapunov stability requires that nearby trajectories remain close for all time. We say that x * is Liapunov stable if for cach € > 0, there isa 6 > 0 such that ||x(r)-x*||<€ whenever 20 and | x(0)-x*||<6. Thus, trajectories that start within 6 of x* remain within € of x* for all positive time (Figure 1). EXERCISES 141radius = 6 radius =€ J) x Attracting Liapunoy stable Figure 1 Finally, x* is asymptotically stable if it is both attracting and Liapunov stable. For each of the following systems, decide whether the origin is attracting, Lia- punov stable, asymptotically stable, or none of the above a) & =-4x, 5.1.11 (Stability proofs) Prove that your answers to 5.1.10 are correct, using the definitions of the different types of stability. (You must produce a suitable 5 to prove that the origin is attracting, or a suitable 6(€) to prove Liapunov stability.) 5.1.12 (Closed orbits from symmetry arguments) Give a simple proof that orbits are closed for the simple harmonic oscillator = v, = —x, using only the symme- try properties of the vector field. (Hint: Consider a trajectory that starts on the v- axis at (0,—v,), and suppose that the trajectory intersects the x-axis at (x,0) . Then use symmetry arguments to find the subsequent intersections with the v-axis and a-axis.) 5.1.13| Why do you think a “saddle point” is called by that name? What's the connection to real saddles (the kind used on horses)? 5.2 Classification of Linear Systems Consider the system ¢=4x-y, p=2xt+y. a) Write the system as X=Ax. Show that the characteristic polynomial is 4 —5A+6, and find the eigenvalues and eigenvectors of A. b) Find the general solution of the system. c) Classify the fixed point at the origin. d) Solve the system subject to the initial condition (x,.¥) = 3,4). 5.2.2 (Complex eigenvalues) This exercise leads you through the solution of a 142 LINEAR SYSTEMSlinear system where the eigenvalues are complex. The system is += .x—-y, pexty a) Find A and show that it has eigenvalues A, = 1+i, 2, =1—i, with eigenvectors v, =(i,1), ¥; =(-i,1). (Note that the eigenvalues are complex conjugates, and so are the eigenvectors—this is always the case for real A with complex eigen- values.) b The general solution is x(t)=c,e*"v, +c,e"’v,. So in one sense we're done! But this way of writing x(t) involves complex coefficients and looks unfamil- iar. Express x(¢) purely in terms of real-valued functions. (Hint: Use e'® =coswr+isinwr to rewrite x(#) in terms of sines and cosines, and then separate the terms that have a prefactor of j from those that don’t.) Plot the phase portrait and classify the fixed point of the following linear systems. If the eigenvectors are real, indicate them in your sketch. 5.2.3 5.24 i=5x+l0y, p=-x-y 5.2.5 5.2.6 ~3xt2y, 5.2.7 5.2.8 %=-3x4+4y, p=-2x43y 5.2.9 5.2.10 =ox-2y. Ab 5.2.11 Show that any matrix of the form A = (3 i} with b # 0, has only a one- dimensional eigenspace corresponding to the eigenvalue A. Then solve the system x = Ax and sketch the phase portrait. 5.2.12 (LRC circuit) Consider the circuit equation Li + Ri +1/C=0, where L,C>Oand R20. a) Rewrite the equation as a two-dimensional} linear system. b) Show that the origin is asymptotically stable if R>O and neutrally stable if R=0. c) Classify the fixed point at the origin, depending on whether R’°C—4L is posi- tive, negative, or zero, and sketch the phase portrait in all three cases. 5.2.13 (Damped harmonic oscillator) The motion of a damped harmonic oscilla- tor is described by m¥ + bi +kx =0, where b> 0 is the damping constant a) Rewrite the equation as a two-dimensional linear system. b) Classify the fixed point at the origin and sketch the phase portrait. Be sure to show all the different cases that can occur, depending on the relative sizes of the parameters. c) How do your results relate to the standard notions of overdamped, critically damped, and underdamped vibrations? 5.2.14 (A project about random systems) Suppose we pick a linear system at EXERCISES 143random; what’s the probability that the origin will be, say, an unstable spiral? To ab be more specific, consider the system x= Ax, where A -( ‘| Suppose we : pick the entries a,b,c\d independently and at random from a uniform distribution on the interval [~1,1]. Find the probabilities of all the different kinds of fixed points. To check your answers (or if you hit an analytical roadblock), try the Monte Carlo method. Generate millions of random matrices on the computer and have the machine count the relative frequency of saddles, unstable spirals, ete. Are the answers the same if you use a normal distribution instead of a uniform distribution? 5.3 Love Affairs 45.3.1 (Name-calling) Suggest names for the four romantic styles, determined by the signs of a and b in R=aR+bJ. 5.3.2 Consider the affair described by R= J, J =-R+J a) Characterize the romantic styles of Romeo and Juliet. b) Classify the fixed point at the origin. What does this imply for the affair? ¢) Sketch R(s) and J(r) as functions of f, assuming R(Q)=1, J(0) = 0. In each of the following problems, predict the course of the love affair, depending j on the signs and relative sizes of a and b. 5.3.3 (Out of touch with their own feelings) Suppose Romeo and Juliet react to \ each other, but not to themselves: R= aJ, J = bR. What happens? te Sy 5.3.4 (Fire and water) Do opposites attract? Analyze R= aR +bJ, J =-bR—al. 5.3.5 (Peas in a pod) If Romeo and Juliet are romantic clones (R= aR +bJ, J = bR+aJ), should they expect boredom or bliss? ~p #536 (Romeo the robot) Nothing could ever change the way Romeo feels about Juliet: R = 0, J = aR + bJ. Does Juliet end up loving bim or hating him? 144 LINEAR SYSTEMS b F