Canard dynamics: applications to the biosciences and theoretical aspects

Transcription

1 Canard dynamics: applications to the biosciences and theoretical aspects M. Krupa INRIA Rocquencourt (France) Collaborators: M. Desroches (INRIA), T. Kaper (BU)

2 Outline 1 Canards in two dimensions 2 Mixed-mode oscillations 3 Spike adding in square wave bursters 4 Mixed-mode bursting oscillations 5 Analysis of the canard phenomenon 6 Other directions

3 Take VdP with α large and constant forcing a Second order ODE:... x + α(x 2-1)x + x = a Rewritten as first order system:. εx = y-x 3 /3+x. y = a-x where: 0 < ε=1/α 1 Long-term dynamics of when a is varied: from a=1.2 via a= to a= x 1.25 x 0 x time time time

4 Benoît, Callot, Diener & Diener (1981) Van der Pol oscillator. εx. = y y-x 3 /3+x a-x 2 O(exp( c ε )) O(ε)-away from the Hopf point the branch becomes almost vertical. 1 HB a ε =0.001

5 Benoît, Callot, Diener & Diener (1981) Van der Pol oscillator. εx. = y y-x 3 /3+x a-x O(exp( c ε )) 2. Hopf bifurcation at a=1 O(ε)-away from the Hopf point the branch becomes almost vertical. 1 HB a ε =0.001

6 Time-scale analysis: from ε > 0 to ε = 0. x O(1/ε) x is fast. y O(1) y is slow

7 Time-scale analysis: from ε > 0 to ε = 0. x O(1/ε) x is fast. y O(1) y is slow Limiting system for the slow dynamics: ε > 0 ε = 0: Reduced sys.. εx = y-x 3 /3+x 0 = y-x 3 /3 +x.. y = a-x y = a-x

8 Time-scale analysis: from ε > 0 to ε = 0. x O(1/ε) x is fast. y O(1) y is slow Limiting system for the slow dynamics: ε > 0 ε = 0: Reduced sys.. εx = y-x 3 /3+x 0 = y-x 3 /3 +x.. y = a-x y = a-x Limiting system for the fast dynamics: ε > 0 ε = 0: Layer sys. x = y-x 3 /3+x x = y-x 3 /3 +x y = ε(a-x) y = 0

9 Time-scale analysis: from ε > 0 to ε = 0 ẋ O(1/ε) x is fast. y O(1) y is slow Limiting system for the slow dynamics: ε = 0: Reduced sys. 0 = y-x 3 /3+x. y = a-x slow subsystem ODE defined on the cubic S:={y=x 3 -x} /3 Limiting system for the fast dynamics: ε = 0: Layer sys. x = y-x 3 /3+x y = 0 fast subsystem family of ODEs param. by y S is a set of equilibria

10 Time-scale analysis: dynamics from ε = 0 to ε > S a S r S a away from the slow curve S, the overall dynamics is fast in an ε-neighbourhood of S, the overall dynamics is slow Transition: bifurcation points of the fast dynamics Note S has 2 fold points different stability on each side: a r S is attracting and S is repelling

11 Fenichel theory: dynamics from ε = 0 to ε > For ε>0 there are Fenichel slow manifolds or rivers

12 Back to Benoît et al. The VdP system has limit cycles which follow the attracting part S of the cubic nullcline S... all the way down to the fold point and then... r continue along the repelling part S of S! a

13 Back to Benoît et al. The VdP system has limit cycles which follow the attracting part S of the cubic nullcline S... all the way down to the fold point and then... r continue along the repelling part S of S! a They have been called canards by the French mathematicians who discovered them

14 Back to Benoît et al. The VdP system has limit cycles which follow the attracting part S of the cubic nullcline S... all the way down to the fold point and then... r continue along the repelling part S of S! a They have been called canards by the French mathematicians who discovered them

15 Answer: Why do canard cycles exist in an exponentially small parameter range? S r is repelling so to follow it for a time t = O(1/ε) the solution (cycle) must be exponentially close to it.

16 Applications aerodynamics M. Brøns, Math. Eng. Industry 2(1): 51-63, 1988 chemical reactions M. Brøns & K. Bar-Eli, J. Phys. Chem. 95: , 1991 B. Peng et al., Phil. Trans. R. Soc. Lond. A 337: , 1991 M. Brøns & K. Bar-Eli, Proc. R. Soc. London A 445: , 1994

17 Applications (...) J. Moehlis, J. Math. Biol. 52: , 2006 H.G. Rotstein, N. Kopell, A.M. Zhabotinsky, and I.R. Epstein, Canard phenomenon and localization of oscillations in the Belousov-Zhabotinsky reaction with global feedback. J. Chemical Physics, 2003;119:

18 4D Hodgkin-Huxley Cdv/dt = I I L I Na I K I (applied current) is const I L = g L (v V L ) I Na = g Na m 3 h(v V Na ) I K = g K n 4 (v V K ) x = m, h, n gating variables. dx/dt = 1 τ x (v) (x x (v)) 4D 2D reduction Known time scale separation: v and m are fast, h and m are slow. 4D 3D reduction: m = m (v) 3D 2D reduction (Rinzel): h(t) =0.8 n(t)

19 2D Hodgkin-Huxley (! (a) (! (b) Γ wh V V $%'! $%'! Γ m $%&!!*&( n!*(( H I appl H! "!! #!! Γ m Γ wh (c) Γ h S 0 S ε r V Γ h H $%&! $%'!! '! "! (! $%"( I appl $%'!!!!*( ' (! $%"( $%'!!!!*( ' (! (d) Γ wh Γ m!*+( $%)! $%"! #! V $%"( Γ h $%'!!!!*( ' t/t

20 Mixed mode oscillations Canard explosion with a drift

21 Mixed mode oscillations ' # v "'( " "'( # )*+ v # "'( " "'( # ),+ #'(!"" #""" #$"" #%"" #&"" #!"" t #'(!"" #""" #$"" #%"" #&"" #!"" t v )-+ Γ 6 Γ 7 L r S r ε ξ 5 ξ 6 ξ 7 S a ε h ' s

22 Mixed mode oscillations ' # v "'( " "'( # )*+ v # "'( " "'( # ),+ #'(!"" #""" #$"" #%"" #&"" #!"" t #'(!"" #""" #$"" #%"" #&"" #!"" t v )-+ Γ 6 Γ 7 L r S r ε ξ 5 ξ 6 ξ 7 S a ε h ' s M. Brons, M. Krupa and M. Wechselberger. Mixed mode oscillations due to the generalized canard phenomenon. Fields Inst. Comm. 49, (2006) M. Krupa, N. Popovic and N. Kopell. Mixed-mode oscillations in three timescale systems--a prototypical example. SIAM J. Appl. Dyn. Sys. 7, (2008)

23 Applications, mixed mode oscillations H. Rotstein, T. Oppermann, J. White, and N. Kopell The dynamic structure underlying subthreshold oscillatory activity and the onset of spikes in a model of medial entorhinal cortex stellate cells. J. Comput. Neurosci., 2006 J. Rubin and M. Wechselberger Giant squid-hidden canard: the 3D geometry of the Hodgkin Huxley model Biol Cybern (2007) 97:5 32 DOI /s M. Krupa, N. Popovic, N. Kopell and H. G. Rotstein. Mixed-mode oscillations in a three timescale model of a dopamine neuron. Chaos, 18, p (2008). Review: M. Deroches, J. Guckenheimer, B. Krauskopf, C. Kuehn, H. Osinga, M. Wechselberger. Mixed-mode oscillations with multiple time scales. Mixed-mode oscillations wit SIAM Rev. 54 (2012)

24 Spike adding canard explosion and mixed-mode bursting oscillations MMBOs References: D. Terman, Chaotic spikes arising from a model of bursting in excitable membranes, SIAM Journal on Applied Mathematics 51 (5) (1991) J. Guckenheimer, C. Kuehn, Computing slow manifolds of saddle type, SIADS 8, (2009) M. Desroches, T. J. Kaper and M. Krupa, Mixed-mode bursting oscillations: Dynamics created by a slow passage through spike-adding canard explosion in a squarewave burster. Chaos 23(4), pp (2013).

25 Square wave burster Context: Morris-Lecar type system (extra slow variable): v =I 0.5(v +0.5) 2w(v +0.7) 0.5(1 + tanh( v w =1.15(0.5(1 + tanh( v ) w) cosh(v ) I =ε(k v) ))(v 1) Two fast and one slow variable

26 Bursting (square wave burster)

27 The Hindmarsh-Rose burster x = y ax 3 + bx 2 + I z y = c dx 2 y z = ε(s(x x 1 ) z) x (b1) (b2) (b3) (a) 0 Ho HB LP Bifurcation diagram fast system LP Ho z

28 xx x Spike-adding via canards zt tt z zzz 0.75 t x xx x x x t z t t z (g) xxx (g) x z z 0.75z 0.75 t t (h) t z t t z (f) (g) (f) (e) (e) zzz t t x t (e) x x (e) (f) x x x xx x 0.75 t 0.75 z tt z (d) (d) (h) (h) z 0.75 zz t 0.75 t (i)

29 Spike-adding via canards (cont) t x 0 x z z

30 Adding a spike within canard explosion maximal canard 1 spike intermediate canard 2 spikes maximal canard 2 spikes

31 Combination of MMO and bursting (II) + =...? MMOs slow oscillations (+ 1 spike) Bursting fast oscillations (+ quiescent phase)

32 Combination of MMO and bursting (II) + =...? MMOs slow oscillations (+ 1 spike) Bursting fast oscillations (+ quiescent phase) different time scales, fast and slow complex oscillations Minimal system: 2 fast & 2 slow variables

33 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables

34 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: ε 1 x 1 = f 1 ε 2 x 2 = f 2

35 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2

36 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2 MMOs

37 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2 MMOs

38 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2 Bursting

39 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2 Bursting

40 MMOs + Bursting = Mixed-Mode Bursting Oscillations (MMBOs) 2 fast variables 2 slow variables 2 fast: 2 slow: ε 1 x 1 = f 1 ε 2 x 2 = f 2 y 1 = g 1 y 2 = g 2 Combine theories... and questions: what patterns of oscillations? what organising centres?

41 Our strategy to construct and analyse a 4D slow-fast system with MMBOs start from a burster (Hindmarsh-Rose in our case) add a slow variable similarly to the MMO case, we want MMBOs to be the result to a slow passage through a canard explosion we will construct a slow passage through a spike-adding canard explosion 1.5 x 1 (a) 1.5 x 1 (b) t z x = y ax 3 + bx 2 + I z y = c dx 2 y z = ε(s(x x 1 ) z) Hindmarsh-Rose I = ε(k h x (x x fold ) 2 h y (y y fold ) 2 h I (I I fold ))

42 Understanding MMBOs as a slow passage x = y ax 3 + bx 2 + I z y = c dx 2 y z = ε(s(x x 1 ) z) I = ε(k h x (x x fold ) 2 h y (y y fold ) 2 h I (I I fold ))

43 x = y ax 3 + bx 2 + I z y = c dx 2 y z = ε(s(x x 1 ) z) Controlling the number of SAOs using folded node theory I = ε(k h x (x x fold ) 2 h y (y y fold ) 2 h I (I I fold )) (a) 2 ε = 10 4 (b) 2 ε = 10 3 x (c) ! x 10 4 (d) t 2 x I I t 4.5! x t 2 x 10 4

44 x = y ax 3 + bx 2 + I z y = c dx 2 y z = ε(s(x x 1 ) z) I = ε(k h x (x x fold ) 2 h y (y y fold ) 2 h I (I I fold )) Reducing the value of epsilon to match theoretical formulas ε = 10 5 (a) 2 x 1 (b) 2 x t 2 x t 2 x 10 5

45 Analysis: canard phenomenon and MMOs Naive approach: rescaling We first translate the singularity to the origin: ẋ = y x x3 ẏ = λ x λ = a 1 Now rescale: x = ε x, y = εȳ λ = ε λ

46 Rescaled equations (we drop the bars): ẋ = ε(y x εx 3 ) ẏ = ε(λ x) After time rescaling: ẋ = y x ẏ = λ x εx 3

47 What is the problem? x = ε x, y = εȳ λ = ε λ If ( x, ȳ) were assumed uniformly bounded with respect to ε then the corresponding neighborhood in (x, y) is O(ε) in size. Too small for Fenichel theory!

48 Possible approaches: Nonstandard analysis E. Benoît, J.-L. Callot, F. Diener and M. Diener, Chasse au canard, Collectanea Mathematica 32 (1-2): , Classical analysis, using stretch variables W. Eckhaus, Standard chase on French Ducks, Springer LNM Vol. 985: , 1983 Blow-up F. Dumortier and R. Roussarie, Canard cycles and center manifolds, Memoirs of the American Mathematical Society 121(577), M. Krupa and P. Szmolyan, Relaxation oscillations and canard explosion, Journal of Differential Equations 174(2) , 2001.

49 Blow-up Singular coordinate transformation: Φ:R + S 4 R 4 x = r x, y = r 2 ȳ, ε = r 2 ε, λ = r λ -it contains the rescaling -it covers a neighborhood of fixed size (wrt to ε)

50 Charts of the blow-up, i.e. charts of the sphere, correspond to parts of the phase space y K1 K1 K2 x Chart K1 connects to the slow flow Chart K2 is the rescaling Chart K3 connects to the fast flow

51 2D problems lead to 3D problems Case 1: Simple fold εẋ = y + x 2 ẏ = g(x, y), g(0, 0) < 0. S a S r

52 Case II: canard point Canard point is a degenerate fold defined by the condition g(0, 0) = 0 The following equations give an example: εẋ = y + x 2 ẏ = x λ λ 0 S a S r

53 ε =0 Unfoldings of a canard point, ε>0 S a S r S a S r S a S r

54 Case III: folded node (two slow one fast dimensions) εẋ = y + x 2 ẏ = x z ż = µ

55 The slow subsystem 0= y + x 2 ẏ = x z ż = µ reduced equations

56 Explanation: λ unfoldings of a canard point y λ>0! > 0 x y λ =0! > 0 x y λ<0! < 0 x

57 Explanation: λ unfoldings of a canard point!

58 Folded node! εẋ = y + x 2 ẏ = x z ż = µ Folded node is a canard point with a drift

59 Folded node! ε>0 The green trajectory and the magenta trajectory are called The green trajectory and the magenta trajectory are called primary canards primary canards The black trajectory is a secondary canard The black trajectory is a secondary canards

60 Computed slow manifolds and canards Computed slow manifolds and canards z C + 2 z 1 C + 0. C η 1 η 3 η 2 γ s!1 γ s γ w C0!2!3!1 1 3 x 5 C 0 Σ 0 C. y x computed by Mathieu Desroches

61 Secondary canard obtained by continuation. ξ r 6 L r S r Σ fn F ξ a 6 L a. S a computed by Mathieu Desroches

62 Selected references on folded node E. Benoît, Canards et enlacements, Publications Mathématiques de l IHES 72(1): 63 91, P. Szmolyan and M. Wechselberger, Canards in R 3, Journal of Differential Equations 177(2): , M. Wechselberger, Existence and bifurcations of canards in R 3 in the case of the folded node, SIAM Journal on Applied Dynamical Systems 4(1): , M. Brøns, M. Krupa and M. Wechselberger, Mixed-mode oscillations due to the generalized canard phenomenon, in Bifurcation Theory and spatio-temporal pattern formation, Fields Institute Communications vol. 49, pp , 2006.

63 Multiple secondary canards. Σ fn L r S r ε v S a ε ξ 12 ξ 11 ξ 10 ξ 9 ξ 8 ξ 7 ξ 6 ξ 5 ξ 4 n. h Computed by M. Desroches

64 References on secondary canards M. Wechselberger, Existence and bifurcations of canards in Rin 3 the case of the folded node, SIAM Journal on Applied Dynamical Systems 4(1): , M. Krupa, N. Popovic and N. Kopell, Mixed-mode oscillations in the three time-scale systems: A prototypical example, SIAM Journal on Applied Dynamical Systems 7(2): , M. Krupa and M. Wechselberger, Local analysis near a folded saddle-node singularity, Journal of Differential Equations 248(12): , M. Krupa, A. Vidal, M. Desroches and F. Clément, Multiscale analysis of mixed-mode oscillations in a phantom bursting model, SIAM Journal on Applied Dynamical Systems (2012)

65 Torus canards: transition from spiking to bursting (related to discrete canards?) Early work by Izhikievich, SIAM Review, 43, , Canonical system: ż =(u + iω)z +2z z 2 z z u = ε(a z 2 )+... Canard explosion for amplitude equations

66 Transition from bursting to spiking bursting: relaxation oscillation modulated spiking: canard with head modulated spiking: small canard fast spiking

67 Activity of a Purkinje cell The big picture Burke, Barry, Kramer, Kaper, Desroches Increasing excitation (J) Bursting Bursting [fold fold-cycle] [canard] AM Spiking [canard] Bif Diag (Fast system) V (Full system) Fixed point Slow nullcline & attracting FP of fast system intersect J = (AM spiking) Torus canard explosion Torus bifurcation (TR) in full system

68 Noise driven canards Other directions Extensions to infinite dimensions, delay eqs, PDES Krupa, Touboul Fine aspects of the dynamics, e.g. chaotic MMOs Krupa, Popovic, Kopell, SIADS 2008 Systems with more than two timescales Krupa, Vidal, Desroches, Clémént, SIADS 2012 Networks of canard oscillators, synchronization Canards in boundary value problems Torus canards? Touboul, Krupa, Desroches, submitted 2013