PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis 2nd Order Nonlinear DE – Intro to General Phase Plane Analysis We will generalize what was done in lectures for the single nonlinear pendulum (for a complete description of fixed point topology see for example course textbooks Ott, Drazin) We have 2 kinds of information: global and local. Global information 1) Express 2nd order DE as two coupled 1st order DE (which we can’t in general integrate) dx = F ( x, y ) dt dy = G ( x, y ) dt Some notation: autonomous nonautonomous c.f. pendulum x ≡θ F, G not explicitly t dependent F, G are explicitly t dependent dθ ≡y dt F ≡ y, G ≡ −ω 2 sin x pendulum – dynamics given by trajectory in phase space i.e. y,θ space 2) Can then write dy G = dx F hence there is a uniquely defined tangent to the y ( x) “trajectory” everywhere except at F = 0 or G = 0 - the fixed points x, y . So generally – trajectories can’t cross except at the fixed points. For the pendulum, other global information came from the symmetries of y (θ ) . So generally look at symmetries in F , G to understand topology. N.B.: This framework encompasses Hamiltonian systems where one has p(q) trajectories in p, q phase space and for each component: dpi ∂H =− dt ∂qi p, q dqi ∂H =− dt ∂pi H - Hamiltonian here are canonical coordinates and H = ( p, q ) = ∑ j p j2 2m (H constant) - the pendulum is of this class. 1 + V (q j ) PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis Local information: We can – i) find fixed points F = 0 , G = 0 : - x, y ii) linearize about fixed points to classify them. - To consider the general case – let’s consider that a fixed point x , y exists – classify all possible behaviour. Classification of fixed point behaviour Let F = 0 , G = 0 have solution x , y consider vicinity of this fixed point x = x +δ x y = y +δ y we can Taylor expand F, G ∂F ( x , y ) ∂F ( x , y ) δx+ δ y + O(δ x 2 , δ y 2 , δ xδ y ) ∂x ∂y ∂G ( x , y ) ∂G ( x , y ) δx+ δ y + O(δ x 2 , δ y 2 , δ xδ y ) G ( x, y ) = G ( x , y ) + ∂x ∂y F ( x, y ) = F ( x , y ) + but F ( x , y ) , G ( x , y ) = 0 - assumed that F , G are twice differentiable. So now need to solve equations of the form: F = aδ x + bδ y G = cδ x + d δ y - nonlinear problem now reduced to linear algebra. a, b, c, d are the differentials from Taylor expansion a = So in the vicinity of x , y dδ x = aδ x + bδ y dt dδ y = cδ x + dδ y dt where ⎛δ x ⎞ ⎟ ⎝δ y ⎠ δx=⎜ or dδ x = J iδ x dx ⎛a b ⎞ J =⎜ ⎟ ⎝c d ⎠ 2 ∂F ( x , y ) ∂x etc. PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis Use the method of normal modes – i.e. assume solutions of the form δ x = e st u , where ⎛u ⎞ is a constant vector (eigenvector) and s is eigenvalue (one eigenvalue per eigenvector). u =⎜ ⎟ ⎝v ⎠ then J ⋅ u = s u . To rearrange, write the RHS as a dot product, i.e.: ⎛ s 0⎞⎛u ⎞ su = ⎜ ⎟ ⎜ ⎟ = s iu so that J iu = s iu ⎝0 s ⎠⎝ v ⎠ b ⎞ ⎛a − s or ( J − s ) . u = 0 so J − s = ⎜ ⎟ d − s⎠ ⎝ c For nontrivial solutions ( u ≠ 0 ) we have J − s = 0 working out the determinant: 0 = ( a − s )( d − s ) − bc = s 2 − s ( a − d ) + ad − bc which we will write as s = giving s = 1 2 {p ± 1 2 i.e. 0 = {( a + d ) ± a−s b c d −s (a + d ) 2 − 4 ( ad − bc ) } } p 2 − 4q for short. There are 2 eigenvalues s+ , s− so there are also 2 eigenvectors u + , u − ⎛δ x ⎞ then the general solution for δ x = ⎜ ⎟ is δ x = C1e s+t u + + C2 e s−t u − and C1 , C2 are arbitrary constants. ⎝δ y ⎠ In components this is: δ x = C1e s+t u+ + C2e s−t u− = Ae s+t + Be s−t δ y = C1e s+t v+ + C2 e s−t v− = Ce s+t + De s−t ⎛u ⎞ and u ± both satisfy J ⋅ u = su so since u + = ⎜ + ⎟ etc, we have that ⎝ v+ ⎠ au± + bv± = s± u± if you need to find the u + , u − cu ± + dv± = s± v± Now use this solution to classify all fixed points. 3 PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis Classification All depends on nature of the roots: s± = 1. 1 2 {p ± p 2 − 4q unstable δy } p 2 > 4q, q > 0 δx s± both real, distinct, same sign distinct so take s+ > s− (say) then for t → ∞, δ x C1e s+t u + dominates, implying s± > 0 -explosive growth, δ x → ∞ s± < 0 - decay δ x → 0 stable for p > 0 for p < 0 unstable, δy [special case p 2 = 4q] → star point δx 2. p 2 > 4q q < 0 s± real, distinct –of opposite sign consider δ x = C1e s+t u + + C2 e s−t u − and take s+ > 0 > s− then as t → ∞ so t → ∞, δ x ≈ C1e s+t u + , as δ x = c1e s+t u+ δ y = c1e s+t v+ δy= t → −∞ v+ δx u+ δ x ≈ C2 e s−t u − ditto t → −∞, δ y = v− δx u− δy moves from one characteristic to the other as t → −∞ to t → +∞ δ y = v− μ δ x − Saddle point δx (only time you need to evaluate u,v) δ y = v+ μ δ x + 4 PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis 3. p 2 < 4q p≠0 complex conjugate pair s± = α ± i β p 2 − 4q α = p2 β = 2 so that δ x = C1eα t e + iβ t u + + C2 eα t e − iβ t u − = eα t [C1e + iβ t u + + C2 e − iβ t u − ] if p, α + ve then we have explosive growth if - ve -then decay in addition - there is an oscillating part with ‘frequency’ β 1 2 spiral fixed point δy δx 4. p 2 < 4q p = 0 s± = ± i q now (special case of (3)) so q > 0 Substitute into case 3 above α = 0 β = q then δ x = C1e + iqt u + + C2 e −iqt u − oscillates with ‘frequency’ q. δy δx No growth or decay – centre fixed point c.f. simple pendulum 5 PX391 Nonlinearity, Chaos, Complexity- General Phase Plane Analysis Summary p=a+d ∂F ∂G p= + ∂x ∂y Recall i.e. q = ad − bc ∂F ∂G ∂F ∂G q= ⋅ − ⋅ ∂x ∂y ∂y ∂x all evaluated at the fixed point x , y Nomenclature Attractors, Attractive fixed points: stable nodes Repellors, Repulsive fixed points: unstable nodes centre, saddle– not an attractor or repellor 6