Measuring Topological Chaos Jean-Luc Thiffeault http://www.ma.imperial.ac.uk/ jeanluc Department of Mathematics Imperial College London Measuring Topological Chaos p.1/22
Mixing: An Overview A fundamental problem in fluid dynamics. Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. A similar regime occurs in microfluidics. Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. A similar regime occurs in microfluidics. Two related goals: Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. A similar regime occurs in microfluidics. Two related goals: Optimisation faster, cheaper (chemical engineering) Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. A similar regime occurs in microfluidics. Two related goals: Optimisation faster, cheaper (chemical engineering) Diagnostic what is doing the mixing? (geophysics) Measuring Topological Chaos p.2/22
Mixing: An Overview A fundamental problem in fluid dynamics. Different regimes: Turbulent flows: mixing is usually easy, but efficiency and energetics an issue. Very viscous, non-newtonian, granular flows: mixing is tough! Flows are typically slow. A similar regime occurs in microfluidics. Two related goals: Optimisation faster, cheaper (chemical engineering) Diagnostic what is doing the mixing? (geophysics) For fluid dynamics, mixing is one of the best reasons to study chaos, since sensitivity to initial conditions leads directly to good mixing. Measuring Topological Chaos p.2/22
Several Approaches Stochastic approach: Characterise behaviour of system through probabilistic methods. Lyapunov exponents, correlation lengths and times. Very generic, but domain of applicability unclear. Measuring Topological Chaos p.3/22
Several Approaches Stochastic approach: Characterise behaviour of system through probabilistic methods. Lyapunov exponents, correlation lengths and times. Very generic, but domain of applicability unclear. Chaotic advection approach: Focus on specific features of a flow. Spatial distribution of Lyapunov exponents, stable and unstable manifolds. Difficult to implement, especially for aperiodic flows; often mostly used as diagnostic. Diffusivity doesn t really enter. Somewhat generic. Measuring Topological Chaos p.3/22
Several Approaches Stochastic approach: Characterise behaviour of system through probabilistic methods. Lyapunov exponents, correlation lengths and times. Very generic, but domain of applicability unclear. Chaotic advection approach: Focus on specific features of a flow. Spatial distribution of Lyapunov exponents, stable and unstable manifolds. Difficult to implement, especially for aperiodic flows; often mostly used as diagnostic. Diffusivity doesn t really enter. Somewhat generic. Eigenfunction approach: solve infinite-dimensional eigenvalue problem for the advection diffusion operator. Dominant eigenmodes. Highly non-generic, very difficult, often little insight gained. Measuring Topological Chaos p.3/22
Several Approaches Stochastic approach: Characterise behaviour of system through probabilistic methods. Lyapunov exponents, correlation lengths and times. Very generic, but domain of applicability unclear. Chaotic advection approach: Focus on specific features of a flow. Spatial distribution of Lyapunov exponents, stable and unstable manifolds. Difficult to implement, especially for aperiodic flows; often mostly used as diagnostic. Diffusivity doesn t really enter. Somewhat generic. Eigenfunction approach: solve infinite-dimensional eigenvalue problem for the advection diffusion operator. Dominant eigenmodes. Highly non-generic, very difficult, often little insight gained. Today: focus on chaos and topology. Measuring Topological Chaos p.3/22
Experiment of Boyland et al. [P. L. Boyland, H. Aref, and M. A. Stremler, J. Fluid Mech. 403, 277 (2000)] (movie by Matthew Finn) Measuring Topological Chaos p.4/22
Four Basic Operations ements PSfrag replacements σ 1 1 σ 2 σ 1 2 σ 1 σ 1 σ 1 1 σ 1 2 σ 2 ements σ 1 σ 2 σ 1 2 PSfrag replacements σ 1 1 σ 1 σ 1 1 σ 2 σ 1 2 σ 1 and σ 2 are referred to as the generators of the 3-braid group. Measuring Topological Chaos p.5/22
Two Stirring Protocols σ 1 σ 2 protocol σ 1 1 σ 2 protocol [P. L. Boyland, H. Aref, and M. A. Stremler, J. Fluid Mech. 403, 277 (2000)] Measuring Topological Chaos p.6/22
Braiding σ 1 σ 2 protocol σ 1 1 σ 2 protocol [P. L. Boyland, H. Aref, and M. A. Stremler, J. Fluid Mech. 403, 277 (2000)] Measuring Topological Chaos p.7/22
Matrix Representation of σ 2 I I II II Let I and II denote the lengths of the two segments. After a σ 2 operation, we have ( ) ( ) ( ) ( ) ( ) I I + II 1 1 I I II = = = σ 2. II 0 1 II II Hence, the matrix representation for σ 2 is ( ) 1 1 σ 2 =. 0 1 Measuring Topological Chaos p.8/22
Matrix Representation of σ 1 1 II I II I Similarly, after a σ1 1 operation we have ( ) ( ) ( ) ( ) I I 1 0 I II = = I + II 1 1 II = σ 1 1 ( ) I II. Hence, the matrix representation for σ1 1 is ( ) σ1 1 1 0 =. 1 1 Measuring Topological Chaos p.9/22
Matrix Representation of the Braid Group We now invoke the faithfulness of the representation to complete the set, ( ) ( ) 1 0 1 1 σ 1 = ; σ 2 = ; 1 1 0 1 σ 1 1 = ( ) 1 0 1 1 ; σ 1 2 = ( ) 1 1. 0 1 Our two protocols have representation ( ) 1 1 σ 1 σ 2 = ; σ1 1 1 0 σ 2 = ( ) 1 1 1 2. Measuring Topological Chaos p.10/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle The second has eigenvalues (3 ± 5)/2 = 2.6180 for the larger eigenvalue. Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle The second has eigenvalues (3 ± 5)/2 = 2.6180 for the larger eigenvalue. So for the second protocol the length of the lines I and II grows exponentially! Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle The second has eigenvalues (3 ± 5)/2 = 2.6180 for the larger eigenvalue. So for the second protocol the length of the lines I and II grows exponentially! The larger eigenvalue is a lower bound on the growth factor of the length of material lines. Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle The second has eigenvalues (3 ± 5)/2 = 2.6180 for the larger eigenvalue. So for the second protocol the length of the lines I and II grows exponentially! The larger eigenvalue is a lower bound on the growth factor of the length of material lines. That is, material lines have to stretch by at least a factor of 2.6180 each time we execute the protocol σ 1 1 σ 2. Measuring Topological Chaos p.11/22
The Difference between the Protocols The matrix associated with each generator has unit eigenvalues. The first stirring protocol has eigenvalues on the unit circle The second has eigenvalues (3 ± 5)/2 = 2.6180 for the larger eigenvalue. So for the second protocol the length of the lines I and II grows exponentially! The larger eigenvalue is a lower bound on the growth factor of the length of material lines. That is, material lines have to stretch by at least a factor of 2.6180 each time we execute the protocol σ 1 1 σ 2. This is guaranteed to hold in some neighbourhood of the rods (Thurston Nielsen theorem). Measuring Topological Chaos p.11/22
Freely-moving Rods in a Cavity Flow [A. Vikhansky, Physics of Fluids 15, 1830 (2003)] Measuring Topological Chaos p.12/22
Particle Orbits are Topological Obstacles Choose any fluid particle orbit (green dot). Material lines must bend around the orbit: it acts just like a rod! The idea: pick any three fluid particles and follow them. How do they braid around each other? Measuring Topological Chaos p.13/22
Detecting Braiding Events σ 2 σ 1 2 σ 1 2 ements PSfrag replacements σ 2 In the second case there is no net braid: the two elements cancel each other. Measuring Topological Chaos p.14/22
Random Sequence of Braids We end up with a sequence of braids, with matrix representation Σ (N) = σ (N) σ (2) σ (1) where σ (µ) {σ 1, σ 2, σ 1 2 events detected after a time t. 1, σ 1 } and N is the number of braiding Measuring Topological Chaos p.15/22
Random Sequence of Braids We end up with a sequence of braids, with matrix representation where σ (µ) {σ 1, σ 2, σ 1 2 events detected after a time t. Σ (N) = σ (N) σ (2) σ (1) 1, σ 1 } and N is the number of braiding The largest eigenvalue of Σ (N) is a measure of the complexity of the braiding motion, called the braiding factor. Random matrix theory says that the braiding factor can grow exponentially! We call the rate of exponential growth the braiding Lyapunov exponent or just braiding exponent. Measuring Topological Chaos p.15/22
Non-braiding Motion ents First consider the motion of of three points in concentric circles with irrationally-related frequencies. y 1 0.5 0 0.5 1 1 0 1 x Eigenvalue of Σ (N) 20 15 10 5 0 0 100 200 300 400 500 The braiding factor grows linearly, which means that the braiding exponent is zero. Notice that the eigenvalue often returns to unity. t Measuring Topological Chaos p.16/22
Blinking-vortex Flow To demonstrate good braiding, we need a chaotic flow on a bounded domain (a spatially-periodic flow won t do). Aref s blinking-vortex flow is ideal. Active Vortex Inactive Vortex First half of period Second half of period The only parameter is the circulation Γ of the vortices. Measuring Topological Chaos p.17/22
Blinking Vortex: Non-braiding Motion ents For Γ = 0.5, the blinking vortex has only small chaotic regions. y 1 0.5 0 0.5 1 1 0 1 x Eigenvalue of Σ (N) 80 60 40 20 0 0 50 100 150 200 One of the orbits is chaotic, the other two are closed. t Measuring Topological Chaos p.18/22
Blinking Vortex: Braiding Motion ents For Γ = 13, the blinking vortex is globally chaotic. y 1 0.5 0 0.5 1 1 0 1 x Eigenvalue of Σ (N) 10 20 10 10 10 0 0 50 100 150 200 The braiding factor now grows exponentially. In the same time interval as for Γ = 0.5, the final value is now of order 10 20 rather than 80! t Measuring Topological Chaos p.19/22
Averaging over many Triplets 400 Log of eigenvalue of Σ (N) replacements 350 300 250 200 150 100 50 slope = 0.187 Γ = 13 0 0 500 1000 1500 2000 Averaged over 100 random triplets. t Measuring Topological Chaos p.20/22
Comparison with Lyapunov Exponents Lyapunov exponent g replacements 2.6 2.4 2.2 2 1.8 1.6 1.4 Line stretching exponent Lyapunov exponent 0.15 0.2 0.25 Braiding Lyapunov exponent Γ varies from 8 to 20. Measuring Topological Chaos p.21/22
Conclusions Topological chaos involves moving obstacles in a 2D flow, which create nontrivial braids. The complexity of a braid can be represented by the largest eigenvalue of a product of matrices the braiding. factor. Any triplet of particles can potentially braid. The complexity of the braid is a good measure of chaos. No need for infinitesimal separation of trajectories or derivatives of the velocity field. For instance, can use all the floats in a data set (J. La Casce). Test in 2D turbulent simulations (F. Paparella). Higher-order braids! Measuring Topological Chaos p.22/22