Umnouy Ponsukcharoen Project IV: Mapping Regions for a Double Well System Mapping Regions for a Double Well System: A In this project, you will investigate the evolution of phase space regions for the damped driven motion of a particle in the double well. The idea is to take a whole bunch of points (at least a thousand) bunched into a fairly small area, propagate them all forward in time using the equations of motion, and see what the resulting region looks like. Can you think of a way of estimating the area of the resulting region? In any case, try to determine if the area seems to decrease exponentially as derived in Chapter 7. The rate of shrinkage of the area of the region after a long time should be independent of the value of F0. However, the shape of the region does depend on F0. Investigate the evolution of the shapes of the regions in the different regimes. This project is quite open-ended, and we expect you to do interesting and imaginative work. A. Introduction In Chapter 6, we have studied the equation of motion of an object in a double well, which is d2x dV dx m 2 F cost dt dx dt 1 1 , where V k1 x 2 k 2 x 4 . 2 2 From this equation, one can treat (x(t), p(t)) as a 2-dimensional object and study its In Chapter 7, we use the fact that there is a natural periodicity behavior in phase space. given by the periodic driving force in order to construct iterated function: xi1 : x(t ), pi1 : p(t ) with xi : x(t), pi : p(t). In order to see the chaotic behavior (i.e. period doubling), one may plot x i after number of iterations as a function of driving force amplitude F. Figure 7.1 in Chapter 7 illustrates the bifurcate (chaotic) behavior of this double well system reduced to 1 dimension [1]. in this project, we are going to study (x(t),p(t)) as a 2-dimensional object However, to examine the directly via phase-space plot. First, as a preliminary, we are going bifurcate behavior of the double well system via the phase space plot for different F. Then, we will study the behavior of colony of points in phase-space propagating in time by measuring area and length of the colony. Also, we will study the chaotic behavior by considering two close points in phase-phase in x-coordinate. Finally, we will discuss how chaos accounts for the result. B. Preliminary: Bifurcate behavior of trajectory in phase-space In Chapter 6 and 7, we have used ‘Oscillator.java’ to represent the motion of the object by graphs x(t) and p(t). Now, we are going to represent the motion of the object by phasespace plot, i.e. p(x). For simple harmonic motion, for example, the phase-space plot is just an ellipse. Here, we will do the phase-space plot for our system in the same way, but we will start plotting after some time in order to eliminate the initial values dependence (this is just same as what we did with bifurcation plot). This method yields “attractor” for each case. The revised ‘Oscillator.java’ is used to produce figures below. More details and comments are given in the program. Initial condition: x(0) = -1.2, p(0) = 0.0, 1.0 . Other variables are set as same as ones in default, i.e., dt = 0.01, A =1.0, , mass =1.0. These will be fixed throughout the project unless I note a change in particular section. 1 of 14 Umnouy Ponsukcharoen F =0.5 F= 1.5 F=1.75 F=2.0 2 of 14 Umnouy Ponsukcharoen F=2.2 F=2.25 F=2.3 F=2.35 3 of 14 Umnouy Ponsukcharoen F=2.375 F=2.4 F=2.5 F=4.0 Figure 1: Phase space plots for different F’s. 4 of 14 Umnouy Ponsukcharoen These phase space plots clearly illustrate bifurcate behavior of the system. If one draws a straight-line p = 0 on the graph, the number of intersections between p=0 line and object’s trajectory in phase-space are just doubling from 2, 4, 8, … as F increases. Under some specific values of parameters, we can draw a conclusion 0.5 < F < 1.5 2 distinct local x maximum/minimum (i.e. at p = 0) 1.75 < F < 2.2 4 distinct local x maximum/minimum 2.25 < F < 2.35 8 distinct local x maximum/minimum F = 2.375 16 distinct local x maximum/minimum 2.4 < F < 2.5 16 or more distinct local x maximum/minimum F = 4.0 very large number of distinct local x maximum/minimum (equivalent to graph beyond r in one dimensional bifurcation plot – chaos appears here). From this information, we can estimate roughly what F is. The value of F’s corresponding to “bifurcations” are F2 1.75 , F3 2.25 , F4 2.375. Hence, similar to F F3 2.375 2.25 0.25 . The value of F is about analysis to Project II, 4 F3 F2 2.25 1.75 F 1.75 (2.25 1.75) (2.25 1.75)(0.25)1 .... 1 n 1.75 (2.25 1.75) 0.25 1.75 0.5 2.42 1 0.25 This is not the only way to see the chaotic behavior of the double-well system. Next, we will show how chaos appears in a colony of points in phase-space, not just a trajectory of a single point in phase-space. C. Behavior of colony of points in phase-space propagating in time: Area Suppose there is a colony or a blob of points in phase-space. Since we know from previous section that there is chaos in the system, we might suspect that colony of points after propagating in time behaves chaotically. Still, what kind of behavior is counted as “chaotic”? One of properties that might relate to the chaos is area of the colony. As time propagates, area of the colony may increase in order to create unpredictability, which is a main feature of chaos. That is, if the area increases as time forwards, it will be more and more difficult to predict the position of object in the next time step. However, as we will show by both analytically and numerically, this is not the case. Area of the colony shrinks as time increases without dependence of F. In order to prove analytically that the area of colony in phase-space decreases as time increases, one may refer to the proof of Liouville’s Theorem or the proof in Chapter 7. I will just quote the result. It is found that, dA A or A(t ) A0 exp( t ) dt ,where is the damping factor in the dissipative force term [1]. Notice that the derivation is given in general, so there is no chaos behavior involved here. Now, in order to show that the area decreases as time forwards independence of chaos numerically, we may write a program to plot the colony and to compute the colony’s area at different time. In ‘Blobplot.java’, we start with simplest colony of initial points: 101x101 points distributes uniformly on the 0.1x0.1 square unit in phase-space with 5 of 14 Umnouy Ponsukcharoen lower-left coordinate (-1.2,0.0). Then, we use same algorithm as one in ‘Oscillator.java’ to find positions of those points in phase-space. Then, we can sum the area of parallelogram in which the corner of each parallelogram are consecutive points in the blob. More details and comments are given in the program ‘Blobplot.java’. First, let’s test the relation between area and time Figure 2: The plot of natural log-value of area versus time in seconds for different ’s. All other parameters are fixed. F = 4.0. The plot above agrees with the relation between area of colony and time propagated that has been proposed. Note that at F = 4.0 and =1.0 is in chaos as we have shown in Section B. Now, let’s make another plot for F = 2.0. Figure 3: The plot of natural log-value of area versus time in seconds for different ’s. All other parameters are fixed. F = 2.0. 6 of 14 Umnouy Ponsukcharoen The case also yields the same law of area shrinkage. Observe that on Figure 2, setup F = 4.0 and =1.0 represents chaotic case as illustrated in Section B, while on Figure 3, setup F = 2.0 and =1.0 represents non-chaotic case. The exponential of area shrinkage applies to all cases here. Remark: more deviation from the expected value of small ’s occurs because the method to find the area is accurate up to a certain size of blob. When the points in colony disperse without shrinking the area ( =0.0), the deviation becomes significant. So far, we have shown that, no matter it is chaotic case or not, as time forwards, the area of colony of points in phase-phase shrinks as long as the dissipative factor exists. The how does unpredictability appear if the area shrinks down to zero? next question is then, The answer is quite fascinating: zero area does not mean zero information. We will see how the unpredictability arises more closely in the next section. D. Behavior of colony of points in phase-space propagating in time: Length Let’s revisit the phase space plot from the previous section setup 1) F = 4.0 and =1.0 2) F = 1.5 and =1.0 Here ‘Blobplot.java’ is used in order to make the plot of blob evolved in time. Notice that the plot is made every 0.20 s time step. 7 of 14 Umnouy Ponsukcharoen Figure 4: The evolution of colony of size 0.1x0.1 unit in phase space. The first graph corresponds to F = 4.0 and =1.0 and the second graph corresponds to F = 1.5 and =1.0 First, two graphs confirm the area shrinkage in both chaotic and non-chaotic case. However, what distinguishes two graphs is the perimeter of the colony. As time passes, into a line. The line tends to grow longer for colony of points in both case are squeezed chaotic case, while the line tends to shrink shorter for non-chaotic case. Let’s investigate this effect more thoroughly by constructing a new kind of colony – line-like instead of square blob. For simplicity, we start with initial points: 10001 points distributes uniformly on a circle with center (-1.2,0.0) and diameter 0.1 unit in phase-space. Then, we use same algorithm as one in ‘Oscillator.java’ to find positions of those points in phase-space. Then, we can sum the length of line connecting consecutive points in the colony. More details and comments are given in the program ‘Lineplot.java’. First, let’s find the relation between length and time under different F (and same ). In order to avoid effect of initial setup, we do same analysis to new circle with new center (-0.6,0.8) but same diameter. 8 of 14 Umnouy Ponsukcharoen Figure 5: The plot of natural log-value of length versus time in seconds for different ’s. All other parameters are fixed. =1.0. The first graph is for a circle with center (-1.2,0.0) and diameter of 1 unit, and the second graph with center (-0.6,0.8) and diameter of 1 unit. Note that data in the range t=0 to t=10 s is neglected since it does not behave linearly to Lyapunov spectrum, which I will explain it in the discussion). Here one can (this is due see that, for extreme cases of chaos (F =3.0, 4.0), the slope is definitely positive. On the other hand, for extreme cases of non-chaos (F= 0.0,0.5,1.0) the slope is definitely negative. Hence, the rate of the length between points in phase-space, or, in other words, the rate of the separation of close trajectories is an indicator of chaos. One way to understand the picture is that, the points on circumference of circle in phase-space can be spread out or squeeze in as time increases as they moves in the phase-space. If we connect the line of each point in time, that line is just the trajectory of point. Therefore, 9 of 14 Umnouy Ponsukcharoen even though the length defined in phase-space sounds unphysical unlike area in phasespace, it can still be used as indicator of chaos. Now, can one find F from this analysis? Let’s make a plot of slope in each line versus the corresponding driving force. Figure 6: The relation between the driving force strength F and the slope of plot between log L vs. t. The data is taken from graphs in Figure 5. Here one can see two regimes separated by x-axis. The regime far above x-axis represents chaotic system, while the regime far below x-axis represents non-chaotic system. From this graph, one may find the x-intercept and claim that it is F . However, two difficulties arise. First, we do not know exactly what the relation actually look like. It may be linear, polynomial, or even not smooth at x-intercept. Secondly, the determination of slope near the critical point becomes poor. As we have seen in logistic map before, it is If the number of difficult to distinguish between F and FN where N is very large. iterations (i.e. time to run the program) is not large enough, one might get the slope of graph log L vs. greater than 0 and misunderstand that the critical has already passed. For example, in our plot, F = 2.3 and F = 2.35 gives the slope value greater than 0. However, they are still less than F >2.4as we know from Section B. Therefore, in practice, the rate of perimeter change in time in phase-space fits in some extension but it is a good way to extract value of F . E. The study of two close points in phase-space: x-coordinate perspective In this section, instead of length in phase-space, we are going to investigate in only one length measurement in x-coordinate only corresponds to direct coordinate. This measurement in real life. First, let set up two points, which are very close in position, and 10 of 14 Umnouy Ponsukcharoen both of them are at rest. Then, let the system evolves. Similar to previous section, we might expect the chaotic behavior when F is large enough. The distance in x-coordinate between two close points should grow larger and larger. The program “testchaos.java” is written in order to demonstrate this phenomenon. More details and comments are given in the program. F=2.35 F=2.375 F=2.4 F=2.5 11 of 14 Umnouy Ponsukcharoen F=3.0 F=4.0 Figure 7: The plot of distance in x-coordinate between 2 points and time under different F’s. The initial positions of two points are x= -1.2 and x= -1.2001. =1.0. For F= 2.35 and F=2.375, the non-chaotic behavior becomes clear. The distance between two points does not grow; it even shrinks. However, for F=2.5 or up, the distance between two points grows from 0.0001 to the order of 1. Before it reaches the “saturation point” where the distance is bound by the overall size of trajectory [2], the distance between two points grows exponentially over time. One might check this by plotting Log(delta x) vs. time during this period. 12 of 14 Umnouy Ponsukcharoen Figure 8: The plot of Log(delta x) vs. t for F= 2.5, F= 3.0 and F=4.0 respectively. The estimate slope of each plot is 0.125, 0.25 and 0.375 respectively. Overall, the plot of Log(delta x) vs. t shows linear trend even though there are some periodic shift. If we consider only a specific time in period of driving force (i.e. consider in Poincare section), we will have clearer linear trend. From this graph, one can estimate the slope of this relation. The values obtained here is same order as in Figure 6 in Section E. This is not surprising since the distance of two points in x-coordinate is just the project of a certain length in phase-space. Chaos appears in both phase-space and x-coordinates, which we can observe directly in real life. F. Discussion and Conclusion At this point, we have seen chaotic behavior in the double well system in many ways: shape of attractor, length in phase-space and distance in x-coordinates. Is there any theory to explain these phenomena systematically and quantitatively? Here Lyapunov exponent is introduced in order to quantify the dynamic system systematically. Remark: this analysis turns out to be similar to discussion in Chapter 7a [1]. The Lyapunov exponent of a dynamical system is quantity that characterizes the rate of separation of infinitesimally close trajectories: v v Z (t) et Z0 For n-dimensional phase space, there is a set of n Lyapunov exponents called “spectrum of Lyapunov exponents” [3]. Now, what can we say about Lyapunov spectrum in our system? In the double well problem here, there are 2 Lyapunov exponents since the phasespace is 2 dimensional. Let’s define it to be 1 and 2 . Each exponent is responsible to the separation in each direction in phase-space. Now, the separation in phase-space is be written as v v v Z (t ) e1Z(t)1 e2Z (t)2 ,where is small time-step. After many iterations, the contribution in the direction with smaller Lyapunov exponent will be suppressed by the contribution in the direction with bigger Lyapunov exponents due to the exponential factor. Therefore, in Section D Figure t=10 s is removed from the analysis since the contribution from small 5, data from t=0 to Lyapunov exponent is still significant. Also, if those two Lyapunov exponents are not 13 of 14 Umnouy Ponsukcharoen much different, it took longer time to diminish the effect from the smaller Lyapunov exponent. That might also be the reason why the data at the boundary between chaotic and non-chaotic case (Figure 6) is not quite consistent with our prediction. At this point, one can see that the maximal Lyapunov exponent is just the quantitative measure of chaos. For chaotic case, Lyapunov is positive, and for non-chaotic, it is negative. Section D and E are just how to measure the maximal Lyapunov exponent in different projection. After all, the length of perimeter of close loop in phase-space or the distance in xcoordinate reflects the same rate of separation. However, near the regime where the maximal Lyapunov exponent changes from negative to positive, the methods to find it becomes less accurate as we have already seen. Other than length, one can understand the v v Lyapunov spectrum via the area. The area of parallelogram formed by Z (t) 1 and Z (t) 2 develops in time proportional to e 1 t e 2 t e( 1 2 )t . According to the derivation in Chapter 7a [1], we obtain the relation of Lyapunov exponents as 1 2 exponent, , is positive, the minimal Lyapunov Therefore, when the maximal Lyapunov m exponent is negative with even higher magnitude, m . In other words, the separation shrinks in one direction faster than the separation grows in another direction. However, when the maximal Lyapunovexponent is negative, the magnitudes of two exponents are bound by . Figure 4 agrees with this argument. Under the same and same point of time, the blob in chaotic case tends to be thinner than the blob in non-chaotic case. Here the notion of Lyapunov exponent is quite powerful in order to explain chaotic behavior in phase-space of double well system. G. Bibliography [1] Benjamin Blander, Marko Kleine Berkenbusch, Susan Coppersmith, Leo P. Kadanoff, Amy J. Kolan, Marcelo Magnasco, Michael J. Vinson, Thomas A. Witten, and Scott Wunsch. (2010) Lecture notes 2010 Phys 251/CS 279/Math 292 Chaos, Complexity and Computers. [2] Hilborn, Robert C. (2000). Chaos and Nonlinear Dynamics, An Introduction for Scientists and Engineers. Oxford University Press. [3] Lyapunov Exponent in Wikipedia. Retrieved March 11, 2010, from http://en.wikipedia.org/wiki/Lyapunov_exponent 14 of 14