Lecture - 2014 Stability PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Lecture Notes: Theoretical Methods in Industrial Physics Stability and Bifurcation

Thomas Christen FS 2014

Contents
1 Introduction 2 General Dynamic Systems 3 Bifurcation of Steady States 4 Linear Stability Analysis 5 Symmetries 6 Collective Coordinates 2 3 5 5 8 11

Introduction

When a technical system is investigated in the framework of a mathematical model, two important questions concern the existence and number of possible state(s), and the properties of these states. Of particular interest are the stability properties of a state. These are relevant in two respects. First, in many cases the states of interest must be stable in order that the system can perform its intended task. Instabilities must then be avoided. Secondly, certain technologies are intrinsically reliant on instabilities in order to be able to perform the intended task. Examples are devices which switch between two dierent stable states; the switching process runs then usually through an unstable state that separates the two stable states (see Fig. 1 a)). There exist technical systems which even work completely in unstable (controlled) states. (A prominent example is the unstable basic airframe of combat aircrafts, which require complex control systems; by this the aircraft can be lighter, smaller, more maneuverable, and more ecient. Another example is the high-voltage direct current circuit breaker based on passive resonant circuit instability. The instability is here used for creating a zero-crossing of the current, which is needed for current interruption and naturally occurs for alternating voltage but not for direct voltage). The minimum procedure of a stability investigation involves at least two steps: (i) the determination of the basic states that have to be investigated, and (ii) their stability analysis. In most cases the latter refers to a linear stability analysis, which involves innitesimal perturbations and to which we will restrict ourselves in the following. More detailed treatises than given in these short lecture notes can be found in Refs. [1, 2, 3]. Prior to a more general discussion, an illustrative example is in order.

a)
I

b)

Input In

Output On

In
n+1

v >1

vIn

feedback loop rOn

c)
II O III 1/r
State 3 State 1 State 2

v
O

Figure 1: a) A switching processes (I to III) from a stable state 1 to a stable state 2 (potential
minima) passes through an unstable state 3 (a saddle). b) The feedback model of Example 1 for a system that amplies an input In with amplication factor v and couples back the amount rOn of the output into the input at time n + 1. c) Plot of the steady state solutions O (solid curve: stable, dashed curve: unstable (and unphysical)) as a function of the control parameter product rv with divergence ar rv = 1. Dashed-dotted curve: growth rate (rv ) of a small perturbation of the steady-state solution.

Example 1 Feedback instability Consider the feedback loop sketched in Fig. 1 b). An input In ( 0) at (discrete) time n is linearly transformed by a device into an output On = vIn , where v (> 0). For v > 1 the input is amplied. A fraction r (0 r 1) gets back to the input at time n + 1, and is

, which gives In+1 = I + rOn . An example for such a system added to the external input I is audio-feedback, e.g., a microphone connected to a loudspeaker via an amplier. It is wellknown that if the microphone comes too close to the loudspeaker, the acoustic feedback (r) can be so strong that a shrill sound appears (feedback instability). Other examples that can be described by this feedback model exist in many disciplines, also non-technical ones like climate science, ecology, or even psychology. The basic model equation for this system is obviously + rvIn . In+1 = I (1) For step (i) we determine the steady-state (= time independent) solution In I of Eq. (1) (1 rv ), hence and nd I = I/ vI . (2) O= 1 rv The divergence for rv 1 corresponds to the mentioned feedback instability. For rv > 1 the output becomes negative which is unphysical; however from a mathematical point of view a solution is existing.1 One still has to investigate if the solution (2) can be realized at all in practice, or if it is unstable. For this second step (ii), one adds an (innitesimally) small perturbation I0 at n = 0 to I and determines the behavior of In for n . If In decays to zero, the state is stable; if In increases, it is unstable, and if it remains constant, it is called marginally + In in Eq. (1) leads to stable. Substitution of In = I In+1 = rv In , which can be solved with the ansatz In = I0 exp(n) , (4) (3)

with = ln(rv ). Stability implies < 0, i.e. rv < 1. Consequently, the solution I is stable at rv < 1. The situation is sketched in Fig. 1 c). From this simple example one can learn several rather general facts (which will be partly discussed in subsequent sections): There are four dierent types of quantities: the time (n; later also continuous time and space variables), the state variables (In or On ; governed by model equations), control ; they can be externally adjusted), and small uctuations (I0 ; due parameters (v , r, I to noise or other disturbances which are not fully under control). The states of the system are characterized by relations between the system variables r, v ) or O = O(I, r, v )). A graph of O as a and the control parameters (I = I (I, function of a control parameter (see Fig. 1 c)) is an example for a bifurcation diagram (the origin of the name bifurcation (Verzweigung) will become clear below). Parameter values exist where solutions disappear or appear (here rv = 1). The stability behavior of I is characterized by a growth rate , which becomes also a function of the control parameters for parameterized I . Regions in control parameter space where is (positive) negative belong to (un)stable states. These regions are separated by boundaries where is zero. (This is a special (but often occurring) case. In general one has to look at the real part of the growth rate. Furthemore, may change its sign also via crossing a pole). The parameter values where solutions disappear or appear are related to these stability boundaries (critical parameter values).

General Dynamic Systems

Although the examples below will be simple and specic, the general framework is rst introduced. It holds for general system described by a (maybe multi-component) state
1 The mathematical divergence at rv = 1 is also unphysical; there will always be a saturation or limitation of O by eects not considered in this model. For example, the amplier has only limited power, or the loudspeaker will break, etc ...

variable2 (phase space or state space ) and a dynamic equation (the state may depend on time t) [] t = F (5) with initial conditions (t0 ) = 0 (and boundary conditions if Eq. (5) is a partial differential equation in space-time). In Eq. (5), (parameter space ) is a (maybe multi-component) control parameter (parameter values that can be controlled or tuned externally). Practically relevant questions are Existence: are there solutions (states) (t) ? Uniqueness/ambiguity: how many solutions exist ? Type of states: are they steady states (i.e., time independent solutions), time periodic states, non-periodic time dependent states (quasi-periodic, chaotic, ...) ? Stability: Which states can be realized ? Are they globally stable, metastable, ... ? Are there symmetries and conserved quantities ? Are there thermodynamic equilibrium states ? Are there steady (ow) states near thermodynamic equilibrium ? Existence of a stationarity principle (Ljapunov functional, potential, action, entropy production rate, ...) ?

(i)

(ii)

(iii)

V
>0 <0 <0 >0

Figure 2: Bifurcation diagrams for the cases discussed in Example 2 (stable: solid, unstable:
dashed, cf. Sect. 4). (i) Stability exchange at intersection points. (ii) Stability exchange at a turning point (saddle node bifurcation). (iii) Stability exchange at a pitch-fork bifurcation; along the solution = 0 the stability changes at the crossing point; along the other solution branch, the stability changes twice due to the turning point and the crossing point, hence the branch = 2 remains stable.
2 In

analogy to phase transition theory, is sometimes called order parameter.

Bifurcation of Steady States

All answers to the just listed questions are functions of . There are manifolds in dened by -parameterized solutions . The parameter space is divided in regions associated (n) with dierent properties of the system , i.e., number N of solutions (n = 1, ..., N ), dierent stability properties, etc. It is thus convenient to discuss the solutions in a so-called bifurcation diagram in the space (i.e., (, )). Bifurcation theory describes how solutions or states of a system appear and/or disappear when control parameters are varied. Typical application examples are related to stability issues of devices, or to switching processes, as will be discussed in the exercises 3 and 4, respectively. In the following, we will consider some simple cases that can be graphically illustrated and illuminate the mathematical background. Bifurcation theory is not restricted to steady states, but is much more general, and can be applied to general mathematical objects (including of course time dependent states). Example 2a Single-component state with one control parameter Consider = = R. The bifurcation diagram consists then of curves in the plane (, ) (Fig. 2). Equation (5) is an ordinary dierential equation for a real function (t). The = V function F can be expressed as derivative of a potential V () with a -dependent shape. Its optima are the steady states, which are functions of and are denoted by (n) () (n labels the dierent states). = 2 2 = (1) () = , (2) () = . There are (i) V = 2 3 /3 = F two solution branches which intersect at = 0. Such a situation is called transcritical bifurcation. = 2 = (1) () = , (2) () = . Here 0, (ii) V = + 3 /3 = F for negative there is no steady state solution. This bifurcation is called saddle-node bifurcation, because at = 0 a saddle and a minimum of the potential merge. = 3 = If < 0, only one solution exists: (iii) V = 2 /2 + 4 /4 = F (2) = 0. For 0 there are three solutions, (1) () = , (2) () = 0, (3) () = . This bifurcation is called pitchfork bifurcation. Example 3a Two-dimensional state space Consider = (1 , 2 ) and F = (F1 (1 , 2 ; ), F2 (1 , 2 ; )). The steady states are then given by the intersection points of the two implicit curves F1 = 0 and F2 = 0 (see Fig. 4 a)). Saddle node bifurcations correspond to the appearance or disappearance of intersection points when the curves move in the x1 -x2 -plane when is varied.

Figure 3: The spectrum of the linear stability analysis depends on the control parameter(s) . An instability occurs at a critical parameter c when the largest real part of the eigenvalues becomes positive.

Linear Stability Analysis

Once the basic states are found, their stability must be determined. In the framework of linear stability analysis, one has to investigate wether an innitesimal disturbance of the

a)
2

F2(1, 2)=0

c) (1)

Tr

(2)

F1(1, 2)=0
Saddle node bifurcation

(3)
Hopf bifurcation

b)
(4)

Im()
(3)

det (4)

Re()
(5) (5) (1) (4)

(1)
(1) (2) (2)

(5)
(3)

Figure 4: a) Determination of steady states for a dynamic system with two state variables (x1 , x2 ) via the intersection points of the curves F1 = F2 = 0. By changing control parameters, intersection points can disappear or appear (saddle-node bifurcations). b) For the two stability eigenvalues 1,2 ve dierent cases (1)-(5) exist, depending on if they are real or complex and on their location in the complex plane. c) Stability discussion in the plane of det(DF ) = 1 2 and Tr(DF ) = 1 + 2 ; at a saddle-node bifurcation det changes sign, while a sign-change of Tr at positive det corresponds to a Hopf bifurcation, where the complex conjugate eingenvalue pair cross the imaginary axis. state under investigation decays or increases. We again focus on steady states, i.e. time [] = 0. Linearization of Eq. (5) leads to t = DF [ ]. independent solutions of F Here, DF represents the rst derivative (the Jacobian), and terms of higher order in and k the associated eigenvectors are neglected. If k = k () are the eigenvalues of DF (eigen-modes), disturbances behave as k exp(k t). The index k labels the eigenvalue spectrum, which can be discrete or continuous. As usual, the spectrum is ordered according to the real parts of the eigenvalues, such that Re(k ) > Re(j ) for k < j . Hence, for a real discrete spectrum 0 > 1 > ... > k > k+1 > .... Only if for all real parts Re(k ) < 0 holds, all disturbances will decay and the state is stable. For the discrete real spectrum, the stability requirement is, for instance, 0 < 0. On the other hand, if at least one real part is positive, the state is unstable. The case of Re() = 0 will be discussed later. The eigenvalues depend on the control parameters . If a control parameter is varied, an instability is said to occur if the largest real part changes its sign from negative to positive. , Linear stability analyis can thus be summarized as follows. Determine the spectrum of DF i.e. solve the linear eigenvalue problem [ ] . = DF (6)

The state is stable if the spectrum k () lies in the left half of the complex plane. At instability the control parameter value is called critical, c . If the dimension is larger than one, not critical points but rather critical hyper-surfaces in exist, where the largest real part of the eigenvalue spectrum vanishes. The states on this surface are the marginally stable states. The eigenvector (unstable mode) belonging to the critical eigenvalue characterizes the physical disturbance which grows exponentially when the state becomes unstable. Below we will see that eigenvalues can identically vanish due to symmetries, which must be distinguished from a zero eigenvalue at an instability. Example 2b Single-component state with one control parameter = F = V . Hence We return to Example 2a and Fig. 2. Linearization yields DF 6

(a)

(b)

Re()

(ii)

> c

kc

(i)
> c

< c = c

k
< c

(c) c

(i)

(ii)

Figure 5: a) Subcritical bifurcation, where the unstable state is curved into the parameter region below c . At instability, the critical perturbation is not immediately saturated be grows to a new state far from the initial state. (Thin arrows: vector eld, thick arrows, hysteresis loop). b) In spatially extended systems with translation symmetry, the stability problem of the uniform state is characterized by (k ), where k is the wave number of the perturbation. The zero-crossing of the maximum of Re((k)) is sketched in (i) for dierent control parameter values. The kc of the unstable mode determines the spatial period of the growing structure. For kc = 0 the unstable mode is uniform, for nite kc and Im((kc )) it is a wave. It can also happen (see (ii)) that the maximum of Re((k)) is a k = 0 and the dispersion is diusive, Re() = D k2 , and the eective diusion constant becomes negative for > c (diusion instability). c) In spatially extended systems with ow, one distinguishes between (i) absolute instabilities, where the amplitude of a small perturbation growths at given locations, and convective instabilities, where the perturbation growths in a co-moving frame (but decays at every x for suciently large times). minima of V are stable and maxima (and saddle points) are unstable. One nds for the above examples (i) F = 2 = (1) () = 2, (2) () = 2. One concludes that (1) is unstable and (2) is stable for < 0, and they exchange their stability properties at the crossing point ( = 0). In Fig. 2, stable and unstable states are indicated by solid and dashed curves, respectively. (ii) F = 2 = (1) () = 2 , (2) () = 2 (recall 0). The stability behavior of (1,2) is exchanged at the turning point, = 0. (iii) F = 32 = For < 0, the only solution is stable, (2) () = . For positive , the three solutions have the following eigenvalues: (1) () = 2, (2) () = , (3) () = 2. Obviously, (1,3) are stable, while (2) is now unstable at positive . Note that = 0 is both an intersection point and a turning point. While the solution (2) that is only intersected, changes stability, the solution that is intersected at its turning point keeps its stability behavior. If the dimension of the state space is larger than one, a potential V () with F = V does not exist in general. It is then not possible to illustrate the bifurcation behavior in terms of maxima, minima, or saddle points of a global potential. However, due to topological reasons there are still general bifurcation rules associated with stability change, and which follow from the theorem of implicit functions (see also Example 3). We list here the most important ones. The stabilities of solutions change at an intersection point in the bifurcation diagram 7

(as in Fig. 2 (i)). The stability of a solution changes at a turning point in the bifurcation diagram (as in Fig. 2 (ii)). If a stability exchange occurs twice at the same point (e.g. crossing at a turning point), the stability of the corresponding solution branch remains (as in Fig. 2 (iii)). Example 3b Two-dimensional state space The Jacobian needed for stability analysis of a steady state given by F 1 = F2 = 0 is (Df )kl = Fk /l and is characterized by the two invariants trace T r and determinant det. The stability eigenvalues are then 1,2 = T r/2 T r2 /4 det. Two cases can appear, 1,2 are both real, or they are complex conjugates. The cases are shown in Fig. 4 b) and c). Stability requires that the real parts of 1,2 are negative, i.e. T r < 0 and det > 0. If det = 1 2 crosses zero as a function of a control parameter, a single eigenvalue changes sign, hence a saddle-node bifurcation occurs. If T r crosses zero but det does not, two complex conjugate eigenvalues cross the imaginary axis, where the sign of Re() changes. This case is called a Hopf bifurcation and can result in time-periodic states (limit cycle). There is a large zoo of dierent bifurcation scenarios involving generic (structurally stable) and non-generic (structurally unstable) cases. An extensive specialist dictionary exists [3]. In the following some practically relevant terms are mentioned. One distinguishes between super- und subcritical bifurcations. Supercritical bifurcations are characterized by the continuous growth of the bifurcating steady state solution from zero when c is crossed, as in Fig. 2 (iii). In a sub-critical bifurcation, the new state extends to the sub-critical parameter region and is thus unstable near bifurcation. At instability, a perturbation increases quickly to another state which is far from the initial state. A typical example is the instability which occurs in systems with hysteresis loops, as is shown in Fig. 5 a). Often systems spatially extended are to be considered, and the states can be space dependent. In linear stability analysis, perturbations of uniform (spatially constant) steady states exp(t + ikx), where for simplicity a single can be decomposed into Fourier components, space dimension is assumed (in general, other sets of eigenfunctions may be used). The growth rate is a function of wave number k , hence = (k ) (dispersion relation). Instabilities can then be classied according to the values of k and i (or frequency = Im()) at instability. The unstable mode behaves as follows: kc = 0; c = 0: uniform, non-oscillating kc = 0; c = 0: uniform, oscillating kc = 0; c = 0: spatially periodic, non-oscillating kc = 0; c = 0: spatially periodic, oscillating (growing wave) Localized spatio-temporal instabilities can be classied according to there convection behavior. An instability is called absolute if a perturbation growths at given location x, and is convective if perturbations decay at every x for t but grow in a co-moving frame. Convective instabilities are common in hydrodynamic systems.

Symmetries

The presence of symmetries can (i) lead to simplications when solving for the ground state, and (ii) can have important consequences for the stability and other properties of a state which breaks the symmetry. For instance, case (iii) of example 2 had mirror symmetry . This symmetry group has two elements, namely the inversion and the identity . It is obvious, that any solution which breaks this symmetry (i.e., which is not invariant if mirrored; to be concrete: = 0), can be used to create another solution by its mirror image. The number of elements of the group properties of the symmetry group are related to the number of solutions, which can be generated by the group operations from a symmetry breaking solution. Most important is that if a control parameter dependent state () breaks a continuous symmetry of the system , then an identically vanishing eigenvalue exists, i.e., () = 0 . The associated eigenvector is called Goldstone mode. 8

(with R2 ) with rotation symmetry. g is a rotation with angle Figure 6: Plane vector eld F between 0 and 2 .

Goldstone modes are not only relevant in the context of stability analysis. They are useful for describing and modelling the behavior of complex systems with the help of collective coordinates, as will be discussed in the next section. Due to its relevance, we will now briey explain in more detail the eect of symmetries and the origins of Goldstone modes. as a velocity eld, which denes Denition: According to Eq. (5), one may consider F trajectories that are the solutions of Eq. (5) (we drop here the index ). A dieomorphism , if F is invariant under g (i.e., bijectiv, g and g 1 dierentiable) is called a symmetry of F g , i.e., F [g] = g F []. The symmetry maps solutions (trajectories) of (5) to solutions (trajectories) of this equation. The symmetry operations of an equation form a group, the symmetry group. be symmetric under the (Lie) group {gs | s S dierentiable manifold; g0 = 1}. Let F Examples are the group of continuous translations in d-dimensional space, or rotations in R2 or R3 . If 0 , with 0 = s gs 0 for s = 0 (this inequality denes symmetry breaking) is [] = 0, then also s is a stationary solution, F [s ] = 0 (s S ). a stationary solution of F In other words, if a symmetry breaking solution is known, one can create more solutions [s ]/ds = 0 and (the symmetry manifold) by the action of the symmetry group. From dF the chain rule one obtains [s ]s s = 0 , DF (7) i.e., GM s s is a Goldstone mode. The meaning of this zero eigenvalue is that a displacement of an otherwise stable state in direction of the Goldstone mode (i.e., tangential to the symmetry manifold) does not lead to a restoring force. As will be shown in Sect. 6, a weak symmetry breaking force can lead to a slow motion of the state on the symmetry manifold. Example 3 Translational symmetry 2 [] = f () + x Consider F , with x, t, and (x, t) real (and one-dimensional). If 0 (x) is a spatially inhomogeneous steady state, then also gs 0 (x) 0 (x s) is a steady state for all s R. The spectrum of the linear stability problem has an eigenvalue 0 with eigenvector 0 . It is clear that = 0 s is the innitesimal displacement of 0 . The equation t (0 s) = 0 implies that a spatial displacement of a stable solution 0 (x) is not restored. In the presence of weak noise, for instance, this can lead to a Brownian motion of s, i.e., a diusion of the solution in x-space. The eigenvalue spectrum has the form 0 = 0 > 1 > ... > k > k+1 > .... Example 4 Generalized third law of Keppler Appropriate elaboration of symmetry properties can provide relevant information on the behavior of a system. As an example, consider Newtons law x (= d2 x/dt2 ) = F with d homogeneous force F (kx) = k F (x). Consider the group gs acting on (x, t) and dened 9

by gs (x, t) = (x exp(s), t exp(s)). This stretching transforms the acceleration as x exp(( 2)s) x, and the force F exp(sd)F . The Newton equation remains invariant if (1 d) = 2, i.e. if this relation between and holds, gs is a continuous symmetry group s. If a particle needs a time T1 for a trajectory of length L1 , then a particle moving on the stretched trajectory with length L2 = L1 exp( ) needs the time T2 = T1 exp() with = (1 d)/2. Consequently, d ( ) 1 L2 2 T2 = . T1 L1 For d = 2 this gives Kepplers third law T 2 L3 , which is proven to be a consequence of the specic symmetry of the force. Exercise 3 Stability analysis for thermal runaway Consider the one-dimensional heat conduction problem (
2 ct T = x T +

U L

)2

with boundary conditions x T = T at x = L/2. Here , c, , and U are mass density, specic heat, heat conductivity, and applied voltage. The conductivity is given by (T ) = 0 (1 + T /T0 ). Assume rst homogeneous Dirichlet boundary conditions, i.e., an innite heat transfer coecient . a) Transform all quantities and the heat equation to dimensionless units and determine the control parameter which appears as the factor in front of the dimensionless heat source term. b) Calculate the solution T (x), sketch the bifurcation diagram (, Tmax ), and determine the critical parameter value c . c) Perform a stability analysis by solving the associated eigenvalue problem, and determine again c . How changes the critical value c as a function of ? (d) Determine the order of magnitude of the critical voltage and the characteristic time scale for a typical polymeric electric insulation material. Exercise 4 Interruption of an electric current If one tries to interrupt an electrical current by separating two contacts, a conductive plasma channel (electric arc) is created, the ionization of which is sustained by the ohmic heating U I . Here, U and I are voltage and current of the arc, respectively. A simple model for the arc conductance G = I/U , which reproduces qualitatively the arc behavior, is given by (the Mayr model) ( ) dG G UI = 1 dt K where is a typical charge-carrier generation-recombination time, and K is the cooling power (heat radiation, convection, etc.); both parameters are assumed to be constant. a) Give a physical interpretation of this equation. Determine the steady states for (i) imposed current I (current control) and (ii) imposed voltage U (voltage control) and discuss their stability properties. b) Consider a series connection of the arc, a voltage source U0 , and a series resistance R. (i) Show with a sketch in the I -U diagram, that a saddle-node bifurcation occurs. (ii) Which limit cases for U0 and R correspond to the cases (i) and (ii) of a)? (iii) Write the dynamic equation for the unitless variable g = RG, determine the relevant control parameter , nd all steady state solutions, and discuss the bifurcation diagram in the g plane. What happens at current interruption when the driving voltage U0 is lowered? c) Discuss interruption of an alternating current near the zero crossing for large R and 0 , with U0 = U 0 sin(t), by linearizing the voltage in time and comparing the conlarge U

10

ductance decay time and the traversal time of the monostable U0 -region (near t = 0).3

(0) with direction of Goldstone mode s s and perpendicular Figure 7: Symmetry manifold (0) s

direction t s .

(1)

Collective Coordinates

If the stable state of a system breakes a continuous symmetry, there exists a manifold of non-equivalent solutions described by the parameter s. Often, physical systems (Eq. (5)) =F (0) + F (1) , where F (0) is symmetric with respect to a can be decomposed such that F (1) (1) acts only as symmetry group gs , while F is not symmetric, but is small such that F a weak disturbance. In that case, one can make the ansatz for the solution
(1) = (0) s + s + ... (0) (0)

(8)

(0) [s ] = 0. where s is a symmetry breaking steady state of the undisturbed system, i.e., F The weak symmetry breaking force has two eects. First, it weakly deforms the shape of (1) (0) s . This deformation, which is described by s in leading order, is only weak if the force is small compared to the restoring forces, which are related to the nite eigenvalues (0) of the stability spectrum of the undisturbed system. Secondly, it drives the state s in direction of the Goldstone mode, where the restoring force vanishes. In other words, the symmetry parameter s becomes a slow dynamic variable and is thus time dependent, s = s(t). Slow means, that the velocity s is of order . Insertion of the expansion in Eq. (5) yields
2 (1) [(0) (0) (1) s s + t (1) s ] + O ( ) . s = D F [s ]s + F

In lowest order of , rearrangement gives


(1) (1) [(0) (1) [(0) DF F s ]s t s = s s s ] . (1)

(9)

The previous equation can be understood as a linear inhomogeneous equation for s of the (1) form A s = h. There exists a solvability condition (Fredholm alternative) which requires . In our context, this means that that h is perpendicular to the kernel of the adjoint A scalar product of the right hand side of Eq. (9) and the (adjoint) Goldstone mode vanishes. The geometrical meaning is that the dynamic equation (5) is projected onto the symmetry (0) . The result is an ordinary dierential equation for the collective coordinate manifold of F s given by (1) > < s s | F s = . (10) < s s | s s > The following exercise should clarify these issues.
3 This problem can also be analytically calculated under the assumption, that 1, and that for t < 0 is imposed (current control), and for t > 0 the voltage is imposed (voltage control). I = It

11

Exercise 5 One-dimensional domain wall subject to a weak force Consider a 1-d system (space coordinate x) with state variable (x, t), and boundary con2 [] = x ditions (x ) = 1 and (x ) = 1. Assume F dW/d + f with 2 2 W () = ( 2), and f (, x) is a weak disturbance which breaks the symmetry of the unperturbed system. a) What are the symmetries of the unperturbed system. Sketch a steady state (x, s) of the undisturbed system (f = 0) (Hint: use the analogy of the steady state equation with Newtons equation for coordinate and time x). Discuss the stability of the solution, and show that the zero eigenvalue corresponds to the translation symmetry (Hint: for the stability spectrum, use the analogy to the Schr odinger equation). b) Derive the equation for the domain wall motion in the form s = dH (s)/ds from Eq. (10). c) Sketch H (s) for f = vx a (x), where (x) is the Dirac delta function. Discuss the behavior of the domain wall for a > 0 and v > 0. What is the condition for a stable steady state? Exercise 6 One-dimensional domain wall between metastable and stable states In the previous exercise, the potential was symmetric such that the two stable states had the same energy. Determine the domain wall velocity between two bistable states 1 and 2 with energy dierence W = W (1 ) W (2 ) (but now f 0).

References
[1] H. Haken, Synergetics, an Introduction, Springer, New York, 1983. [2] H. Thomas, Nonlinear Dynamics in Solids, Springer, Berlin, 1992. [3] J. D. Crawford, Introduction to bifurcation theory, Rev. Mod. Phys. 63, 991, 1991.

12

You might also like