Reduced System Computing and MMOs. Warren Weckesser. Department of Mathematics Colgate University

Transcription

1 Reduced System Computing and MMOs Warren Weckesser Department of Mathematics Colgate University

2 Acknowledgments NSF Grant DMS-54468: RUI: Reduced System Computing for Singularly Perturbed Differential Equations

3 Acknowledgments NSF Grant DMS-54468: RUI: Reduced System Computing for Singularly Perturbed Differential Equations Undergraduate students: Brian Kinney Tomas Gruszka Dimitar Simeonov

4 Slow and Fast Subsystems, x R m, y R n εẋ = f(x,y) ẏ = g(x,y)

5 Slow and Fast Subsystems, x R m, y R n εẋ = f(x,y) ẏ = g(x,y) ε = Slow Subsystem (DAE) = f(x,y) ẏ = g(x,y)

6 Slow and Fast Subsystems, x R m, y R n εẋ = f(x,y) ẏ = g(x,y) t εt ẋ = f(x,y) ẏ = εg(x,y) ε = Slow Subsystem (DAE) = f(x,y) ẏ = g(x,y)

7 Slow and Fast Subsystems, x R m, y R n εẋ = f(x,y) ẏ = g(x,y) t εt ẋ = f(x,y) ẏ = εg(x,y) ε = Slow Subsystem (DAE) = f(x,y) ẏ = g(x,y) ε = Fast Subsystem ẋ = f(x,y) ẏ =

8 Slow and Fast Subsystems, x R m, y R n εẋ = f(x,y) ẏ = g(x,y) t εt ẋ = f(x,y) ẏ = εg(x,y) ε = Slow Subsystem (DAE) = f(x,y) ẏ = g(x,y) ε = Fast Subsystem ẋ = f(x,y) ẏ = Critical Manifold = f(x,y)

9 Reduced System Computing Goal: Create computational tools for the study of the reduced system of a singularly perturbed differential equation.

10 Reduced System Computing Goal: Create computational tools for the study of the reduced system of a singularly perturbed differential equation. Simple Idea: Concatenate solutions to fast and slow subsystems. Create zero th order approximations.

11 Reduced System Computing Goal: Create computational tools for the study of the reduced system of a singularly perturbed differential equation. Simple Idea: Concatenate solutions to fast and slow subsystems. Create zero th order approximations. Complications:

12 Reduced System Computing Goal: Create computational tools for the study of the reduced system of a singularly perturbed differential equation. Simple Idea: Concatenate solutions to fast and slow subsystems. Create zero th order approximations. Complications: Canards

13 Reduced System Computing Goal: Create computational tools for the study of the reduced system of a singularly perturbed differential equation. Simple Idea: Concatenate solutions to fast and slow subsystems. Create zero th order approximations. Complications: Canards Fast periodic orbits (and more general ω limit sets of the fast subsystem)

14 Slow Subsystem - A Few More Details = f(x,y) ẏ = g(x,y)

15 Slow Subsystem - A Few More Details = f(x,y) ẏ = g(x,y) Differentiate = f(x,y), solve for ẋ to obtain ẋ = (D x f) (D y f)g(x,y). Multiply by det(d x f). Desingularized slow equations: ẋ = (adjd x f)(d y f)g(x,y) ẏ = det(d x f)g(x,y)

16 Slow Subsystem - A Few More Details = f(x,y) ẏ = g(x,y) Differentiate = f(x,y), solve for ẋ to obtain ẋ = (D x f) (D y f)g(x,y). Multiply by det(d x f). Desingularized slow equations: ẋ = (adjd x f)(d y f)g(x,y) ẏ = det(d x f)g(x,y) Fold points: f(x,y) =, det(d x f) = Fold points are saddle-node equilibria of the fast subsystem.

17 Example: Two Coupled Neurons v = v + tanh(σ v ) q ω f(v 2 )(v + 4) v 2 = v 2 + tanh(σ 2 v 2 ) q 2 ω f(v )(v 2 + 4) q = ε( q + v ) q 2 = ε( q 2 + v 2 ) f(v) = +e 4(v /75) Each (v i,q i ) is a relaxation oscillator. When one is firing, the other s v nullcline is depressed ( reciprocal inhibition ). This system has two fast variables and a two dimensional critical manifold.

18 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

19 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

20 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

21 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

22 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

23 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

24 Fast/Slow System Definition File: cpldosc.fs # Definitions for the coupled oscillator fast/slow system. cpldosc # Fast variables: v, v2 2 v v2 # Slow variables: q, q2 2 q q2 # Parameters: 3 omega sigma sigma2 # Vector field for the fast variables -v+tanh(sigma*v) - q - omega*(v+4)/(+exp(-4*(v2-/75))) -v2+tanh(sigma2*v2) - q2 - omega*(v2+4)/(+exp(-4*(v-/75))) # Vector field for the slow variables -q + v -q2 + v2

25 Computer Code Generation C code is generated, to be used with the SUNDIALS suite [ SUNDIALS includes CVODE for ODEs and IDA for DAEs.

26 Computer Code Generation C code is generated, to be used with the SUNDIALS suite [ SUNDIALS includes CVODE for ODEs and IDA for DAEs. Jacobians are generated automatically (via GiNaC)

27 Computer Code Generation C code is generated, to be used with the SUNDIALS suite [ SUNDIALS includes CVODE for ODEs and IDA for DAEs. Jacobians are generated automatically (via GiNaC) Some MATLAB code is also generated. Other target languages or libraries could be implemented.

28 Computer Code Generation C code is generated, to be used with the SUNDIALS suite [ SUNDIALS includes CVODE for ODEs and IDA for DAEs. Jacobians are generated automatically (via GiNaC) Some MATLAB code is also generated. Other target languages or libraries could be implemented. Input file: cpldosc.fs Output files: fs_cpldosc.c fs_cpldosc_cvode.c fs_cpldosc_ida.c fs_cpldosc.m General C functions C functions for CVODE C functions for IDA MATLAB functions

29 Computer Code Generated The generated code includes:

30 Computer Code Generated The generated code includes: Fast vector field and IVP solver Fast vector field, to be used with CVODE int cpldoscf_cv(realtype t, N_Vector Xvec, N_Vector Xdotvec, void *params) Fast vector field Jacobian int cpldoscfx_cv(long int N, DenseMat Fx, realtype t, N_Vector Xvec, N_Vector Fvec, void *params, N_Vector tmp, N_Vector tmp2, N_Vector tmp3) Solve the fast subsystem IVP void cpldosc_fast(file *fastfile, double X[], double Y[], double params[], SolverParams *solver_params)

31 Computer Code Generated The generated code includes: Slow subsystem DAE, configured for IDA Compute residuals for the IDA DAE solver int cpldosc_idares(realtype t, N_Vector Zvec, N_Vector Zdotvec, N_Vector rvec, void *params) Jacobian for IDA DAE solver int cpldosc_idajac(long int Neq, realtype t, N_Vector Zvec, N_Vector Zdotvec, N_Vector rvec, realtype c_j, void *params, DenseMat jacmat, N_Vector tmp, N_Vector tmp2, N_Vector tmp3) Solve the slow subsystem DAE IVP void cpldosc_slow_dae(file *slowfile, double X[], double Y[], double params[], SolverParams *solver_params)

32 Computer Code Generated The generated code includes: Desingularized slow subsystem (for CVODE) Desingularized slow subsystem vector field int cpldosc_slowde(realtype t, N_Vector Zvec, N_Vector Zdotvec, void *params) Desingularized slow subsystem IVP solver void cpldosc_slow_des(file *slowfile, double X[], double Y[], double params[], SolverParams *solver_params)

33 Computer Code Generated The generated code includes: Fold function and Jacobian Fold function realtype cpldosc_foldfunc_nv(n_vector Zvec, void *params) Fold function gradient (with respect to all variables) void cpldosc_foldfunc_grad_nv(double *grad, N_Vector Zvec, void *params)

34 Example: Initial Value Problem for the Coupled Oscillator System v 2 q v.8 v v.5.5 (ω =.5, σ =.5, σ 2 = 2.6)

35 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2.5 q v v 2 v σ =.2

36 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2.5 q v v 2 v σ =.3

37 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2 q v v 2 v σ =.4

38 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2 q v v 2 v σ =.5925

39 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2 q v v 2 v σ =.625

40 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2 q v v 2 v σ =.625

41 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2 q v v 2 v σ =.6325

42 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2.5 q v v v.5.5 σ =.8

43 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2.5 q v v v.5.5 σ = 2.5

44 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) v 2.5 q v v v.5.5 σ = 3

45 Periodic Orbits of the Coupled Oscillator System (ω =.5, σ 2 =.2) σ =.2.5 σ = v v t t σ =2.5 σ = v v t t

46 Bifurcation Diagram: Stable Periodic Orbits ( Poor man s continuation; not a complete diagram)..5 L2 norm (slow subsystem) σ

47 Bifurcation Diagram: Stable Periodic Orbits ( Poor man s continuation; not a complete diagram)..5 L2 norm (slow subsystem) σ

48 Observations Bifurcations observed at:

49 Observations Bifurcations observed at: Fold-to-fold fast transition (SN-SN connection in fast subsystem)

50 Observations Bifurcations observed at: Fold-to-fold fast transition (SN-SN connection in fast subsystem) (Fold-to-fold examples where the slow flow is towards the fold or away from the fold have been observed.)

51 Observations Bifurcations observed at: Fold-to-fold fast transition (SN-SN connection in fast subsystem) (Fold-to-fold examples where the slow flow is towards the fold or away from the fold have been observed.) Periodic orbits homoclinic to a folded saddle.

52 Observations Bifurcations observed at: Fold-to-fold fast transition (SN-SN connection in fast subsystem) (Fold-to-fold examples where the slow flow is towards the fold or away from the fold have been observed.) Periodic orbits homoclinic to a folded saddle. The investigation is not complete!

53 Current and Future Development

54 Current and Future Development RSC boundary value problem solver that handles canards.

55 Current and Future Development RSC boundary value problem solver that handles canards. Computing and continuing periodic orbits.

56 Current and Future Development RSC boundary value problem solver that handles canards. Computing and continuing periodic orbits. Find fast/slow orbit homoclinic to folded singular points. These play an import role in the bifurcations of periodic orbits: J. Guckenheimer, K. Hoffman, W. Weckesser, Bifurcations of relaxation oscillations near folded saddles, Int. J. Bif. Chaos 5(), (25) M. Brøns, M. Krupa, M. Wechselberger, Mixed Mode Oscillations Due to the Generalized Canard Phenomenon, Fields Institute Communications Vol. 49, (26)