Numerical analysis of nonlinear dynamics Ricardo Alzate Ph.D. Student University of Naples FEDERICO II (SINCRO GROUP)
Download ReportTranscript Numerical analysis of nonlinear dynamics Ricardo Alzate Ph.D. Student University of Naples FEDERICO II (SINCRO GROUP)
Numerical analysis of nonlinear dynamics Ricardo Alzate Ph.D. Student University of Naples FEDERICO II (SINCRO GROUP) Numerical analysis of nonlinear dynamics 2/45 Outline • Introduction • Branching behaviour in dynamical systems • Application and results R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 3/45 Introduction R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Study of dynamics Elements for extracting dynamical features: • Mathematical representation • Parameters and ranges • Convenient presentation of results (first insight) • Careful quantification and classification of phenomena • Validation with real world R. Alzate - UN Manizales, 2007 4/45 Numerical analysis of nonlinear dynamics Dynamics overview How to predict more accurately dynamical features on system? R. Alzate - UN Manizales, 2007 5/45 Numerical analysis of nonlinear dynamics 6/45 References Chronology: [1]. Seydel R. “Practical bifurcation ans stability analysis: from equilibrium to chaos”. 1994. [2]. Beyn W. Champneys A. Doedel E. Govaerts W. Kutnetsov Y. and Sandstede B. “Numerical continuation and computation of normal forms”. 1999. [3]. Doedel E. “Lecture notes on numerical analysis of bifurcation problems”. 1997. R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 7/45 References (2) [4]. Keller H.B. “Numerical solution of bifurcation and nonlinear eigenvalue problems”. 1977. [5]. MATCONT manual. 2006. and Kutnetsov Book Ch10. [6]. LOCA (library of continuation algorithms) manual. 2002. R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Why numerics? Nonlinear systems Dynamics - complex behaviour - closed form solutions not often available - discontinuities !!! Computational resources - availability – technology - robust/improved numerical methods R. Alzate - UN Manizales, 2007 8/45 Numerical analysis of nonlinear dynamics How numerics? Brute force simulation - heavy computational cost - tracing of few branches and just stable cases - jumps into different attractors (suddenly) - affected by hysteresis, etc.. Continuation based algorithms - a priori knowledge for some solution - a priori knowledge for system interesting regimes R. Alzate - UN Manizales, 2007 9/45 Numerical analysis of nonlinear dynamics 10/45 Numeric bugs Hysteresis Branch jump R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Branching behaviour in dynamical systems R. Alzate - UN Manizales, 2007 11/45 Numerical analysis of nonlinear dynamics 12/45 General statement sin( f ) 0 xA d J f K x A tan( f ) y A d 0 d1 d 2 cos 2 ( ) cos( ) cos ( ) ( t ) f f f x1 (t ) equilibrium pts x2 (t ) f ( t ) x2 (t ) f ( x, ) equilibrium orbits x3 (t ) t x3 (t ) x1 (t ) f ( t ) In general, it is possible to study the dependence of dynamics (solutions) in terms of parameter variation (implicit function theorem). R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Implicit function theorem Establishes conditions for existence over a given interval, for an implicit (vector) function that solves the explicit problem Given the equation f(y,x) = 0, if: f(y*,x*) = 0, f is continuously differentiable on its domain, and fy(y*,x*) is non singular Then there is an interval x1 < x* < x2 about x*, in which a vector function y = F(x) is defined by 0 = f(y,x) with the following properties holding for all x with x1<x<x2 : - f(F(x),x) = 0, F(x) is unique with y* = F(x*), F(x) is continuously differentiable, and fy(y,x)dy/dx + fx(y,x) = 0 . R. Alzate - UN Manizales, 2007 13/45 Numerical analysis of nonlinear dynamics 14/45 Implicit function theorem (2) x2 y 2 1 0 g ( y, x) y 2 x2 1 1 2 2 f1 ( x) 1 x y f ( x) 1 2 f ( x) 1 x 2 2 dg J dx dg 2 x 2 f ( x) dy 2 0 g ( f ( x), x) 1 x 2 x 2 1 1 2 Then, singularity condition on gy(f(x),x) excludes x = ±1 as part of function domain in order to apply the theorem. R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 15/45 Branch tracing The goal is to detect changes in dynamical features depending on parameter variation: f ( y, ) df ( y, ) f ( y, ) dy f ( y, ) d y d Then, by conditions of IFT: f ( y, ) df ( y, ) dy 0 f ( y, ) dy d y Behaviour evolution as function of λ, not defined for singularities on fy(y,λ) (system having zero eigenvalues) R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 16/45 Branch tracing (2) In general, there are two main ending point type for a codimension-one branch namely turning points and single bifurcation points. R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Parameterization In order to avoid numerical divergence closing to turning points: - Convenient change of parameter, - Defining a new measure along the branch, e.g. the arclength R. Alzate - UN Manizales, 2007 17/45 Numerical analysis of nonlinear dynamics 18/45 Arclength Augmented system with additional constraints: y y(s) f ( y, ) 0 ( s) s 2 y 2 dy d 1 ds ds 2 2 df dy d 0 fy f ds ds ds R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 19/45 Tangent predictor Tangential projection of solution: y j 1 y j h j v j R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 20/45 Tangent predictor (2) Tangent unity vector: J ( y j )v j 0 J(y j) F y y y j J j 0 j 1 T v (v ) 1 R. Alzate - UN Manizales, 2007 v j 1 , v j 1 Numerical analysis of nonlinear dynamics 21/45 Root finding Newton-Raphson method for location of equilibria: f ( y0 dy) f ( y0 ) f y0dy h.o.t. f ( y0 ) f y0dy f ( y0 ) 0 f ( y ) f dy dy f y0 0 0 y f ( y1 ) y y dy f ( y ) 0? y y ... 1 fy 1 0 1 R. Alzate - UN Manizales, 2007 2 1 Numerical analysis of nonlinear dynamics Root finding (2) In general: fYj 1dY j 1 f (Y j 1 ) f ( y, ) f (Y ) j j 1 j 1 Y Y dY rank ( fYj 1 ) n 1 i.e. nonsingularity of Jacobian at solution F j 1 f (Y ) j 1 g (Y ) Allowing implementation of method. R. Alzate - UN Manizales, 2007 22/45 Numerical analysis of nonlinear dynamics 23/45 Correction Additional relation gj(y) defines an intersection of the curve f(y) with some surface near predicted solution (ideally containing it): - Natural continuation: g j ( y) yio yioj 1 f (Y ) F 0 j 1 Y Y io io R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 24/45 Correction (2) - Pseudo-arclength continuation: g j ( y ) y y j 1 , v j R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 25/45 Correction (3) - Moore-Penrose continuation (MATCONT): g kj ( y ) y Y k 1 ,V k J (Y k 1 )V k 0 R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 26/45 Moore-Penrose Pseudo-inverse matrix: A nx n1 A( n1)x n AT AAT 1 Y Y Y Yf f Y 1 0 Y Y f 1 0 0 Y f Y f 0 Y 0 R. Alzate - UN Manizales, 2007 0 Y f f Y 0 Y nx n 1 0 Y 0 Numerical analysis of nonlinear dynamics Step size control Basic and effective approach (there are many !!!): - Step size decreasing and correction repeat if non converging - Slightly increase for step size if quick conversion - Keep step size if iterations are moderated R. Alzate - UN Manizales, 2007 27/45 Numerical analysis of nonlinear dynamics Test functions Detection of stability changes between continued solutions: - In general are developed as smooth functions zero valued at bifurcations, i.e. ( y0 , 0 ) 0 ( y j , j ) ( y j 1, j 1 ) 0 R. Alzate - UN Manizales, 2007 28/45 Numerical analysis of nonlinear dynamics 29/45 Test functions (2) Usual chooses: : max 1,2 ,...,n : det f y y0 , 0 f ( y, ) F (Y ) : ( y , ) R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 30/45 Branch switching When there is a single bifurcation point, there are more than one trajectories for the which (y0,λ0) is an equilibrium: R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 31/45 Branch switching (2) How to track such new trajectory? f y s , s 0 df dy d f y0 f 0 0 ds ds ds 0 v range f y dy 0 v 1h 0 ds h null f y 2 - Algebraic branching equation (Keller 1977 !!!) b ac 0 a : T f yy0 hh 2 2 a 1 2b 1 0 c 0 0 b : T f yy0 v f y0 h T 0 0 0 c : f yy vv 2 f y v f 0 null range f y R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 32/45 An algorithm R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Application and results R. Alzate - UN Manizales, 2007 33/45 Numerical analysis of nonlinear dynamics 34/45 Continuation of periodic orbits P. Piiroinen – National University of Ireland (Galway): - Single branch continuation - Extrapolation prediction based - Parameterization by orbit period - Step size increasing if fast converging - Step size reducing if non converging - Newton-Raphson correction based R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 35/45 Continuation of periodic orbits (2) R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics Tracing a perioud-doubling R. Alzate - UN Manizales, 2007 36/45 Numerical analysis of nonlinear dynamics Eigenvalue evolution R. Alzate - UN Manizales, 2007 37/45 Numerical analysis of nonlinear dynamics On unit circle R. Alzate - UN Manizales, 2007 38/45 Numerical analysis of nonlinear dynamics Sudden chaotic window R. Alzate - UN Manizales, 2007 39/45 Numerical analysis of nonlinear dynamics On set – brute force R. Alzate - UN Manizales, 2007 40/45 Numerical analysis of nonlinear dynamics On set – continued R. Alzate - UN Manizales, 2007 41/45 Numerical analysis of nonlinear dynamics 42/45 Conjectures How to explain such particularly regular cascade? - development of local maps R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 43/45 Open tasks - Improvement of numerical approximation for map - Theoretical prediction (or validation): A. Nordmark (2003) p( x) a bx cx2 dx3 ex4 f x R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 44/45 Conclusion A general description about numerical techniques for branching analysis of systems has been developed, with promising results for a particular application on the cam-follower impacting model. By the way, is not possible to think about a standard or universal procedure given inherent singularities of systems, then researcher skills constitute a valuable feature for success purposes. R. Alzate - UN Manizales, 2007 Numerical analysis of nonlinear dynamics 45/45 ...? http://wpage.unina.it/r.alzate Grazie e arrivederci !!! R. Alzate - UN Manizales, 2007