Questions & explanations
1. Give an example of a stochastic system where the Lyapunov exponent is negative but the system is not stable in probability. Explain why.
Consider dx = -x dt + x dW. The Lyapunov exponent is -1 + 0.5σ^2 (by Ito's formula). For σ^2 < 2, the exponent is negative, but the system is stable in probability. However, if the noise is large, the exponent can be negative but the system may still have finite-time blow-up? Actually, for this linear system, negative exponent implies almost sure stability. A counterexample: a system with negative Lyapunov exponent but non-zero probability of large excursions? Typically, negative exponent implies stability. So this is tricky. Instead, consider a system where the Lyapunov exponent is negative but the system is not stable in probability due to rare large noise events? Actually, Khasminskii's theorem ensures stability. So perhaps a better example: a system with negative exponent but the equilibrium is not stable in probability because the noise can push it away? That contradicts. Let's correct: For linear systems, negative Lyapunov exponent implies almost sure stability. For nonlinear systems, a negative exponent does not guarantee stability in probability because the system may have mu
2. Compare the behavior of a chaotic Hamiltonian system with a dissipative chaotic system in terms of phase space volume.
In a dissipative chaotic system (like the Lorenz system), phase space volume contracts over time because the divergence of the flow is negative. Trajectories are attracted to a strange attractor of zero volume (a fractal set). In contrast, a Hamiltonian system has no attractors: volume is preserved, so the chaotic motion fills a region of phase space (like a chaotic sea) that has the same volume as the initial conditions. There is no attracting set; instead, the system is ergodic (it explores all accessible states over time). Also, in dissipative chaos, initial conditions in a basin of attraction all go to the same attractor, while in Hamiltonian chaos, different initial conditions can lead to different regions (like islands of regular motion surrounded by chaos). The conservation of energy also restricts motion to a constant-energy surface, which is a lower-dimensional manifold.
3. What is a practical problem when applying Takens' Embedding Theorem to real-world data, and how can you partly solve it?
A common problem is choosing the right time delay τ and embedding dimension m. If τ is too small, the coordinates are almost the same (redundant); if too large, they become unrelated (irrelevant). You can estimate τ using the first minimum of the mutual information (a measure of how much one value tells about another) between x(t) and x(t+τ). For m, you can use the false nearest neighbors method: start with m=1, increase it, and see when adding dimensions stops reducing the number of points that appear close only because of projection. Another issue is noise: real data has measurement errors. You can reduce noise by filtering or using robust methods like singular spectrum analysis (a technique to separate signal from noise). Still, the theorem only works for deterministic systems, so if the noise is too large, the reconstruction may be unreliable.
4. What does Takens' Embedding Theorem allow us to do with a single time series from a dynamical system?
Takens' Embedding Theorem says we can rebuild the full state space (the set of all possible states) of a system using just one measured variable over time. For example, if we measure the number of fish in a lake each month, we can create a 'shadow' of the whole ecosystem's dynamics. We do this by plotting delayed versions of the same measurement: for a delay τ, we plot x(t) against x(t-τ). The theorem guarantees that if the delay is chosen well and the dimension of the embedding (number of delays) is large enough, the reconstructed space has the same shape (topology) as the original state space. This lets us study chaos and other behaviors without measuring all variables. The key condition is that the measurement function must be generic (not special in a bad way) and the system's attractor (the set it settles onto) must have a finite dimension.
5. Give an example of a factor map that is not a conjugacy.
Consider the full shift on two symbols {0,1}^ℕ with the shift map. Define a factor map π that sends a sequence to its first coordinate. Then the factor system is just a single point with the identity map? Actually, the factor system would be the space {0,1} with the identity map, but that's trivial. A better example: the shift on {0,1}^ℕ factors onto the shift on {0}^ℕ? No. Instead, take the shift on {0,1}^ℤ and factor by mapping each sequence to the sequence of differences mod 2? That might be complicated. A simple example: the tent map on [0,1] factors onto the logistic map at parameter 4 via the conjugacy? Actually, they are conjugate. A clear example: the map f(x)=2x mod 1 on the circle factors onto the map g(y)=2y mod 1 on the circle via the map π(x)=2x mod 1? That is a factor map but not one-to-one, so not a conjugacy.
6. Give an example of a simple Hamiltonian system that can exhibit chaos, and explain why it is chaotic.
A classic example is the double pendulum: two pendulums connected end to end. Its Hamiltonian has two degrees of freedom (angles and angular momenta). For small energies, the motion is regular (quasiperiodic). But for larger energies, the system becomes chaotic: small differences in initial angles lead to wildly different trajectories. The chaos arises because the system is nonlinear (the equations involve sines of angles) and non-integrable (there are not enough conserved quantities to solve exactly). The phase space volume is preserved (Liouville), but the trajectories become sensitive to initial conditions. This is seen in the way the double pendulum flips unpredictably. Another example is the Henon-Heiles system, which models the motion of a star in a galaxy, and shows chaotic orbits at high energies.
7. What is the Center Manifold Theorem used for in dynamical systems?
The Center Manifold Theorem helps us simplify a system near an equilibrium point (a steady state) where some directions are stable (return to equilibrium), some unstable (move away), and some are neutral (neither). It says there exists a special surface called the center manifold that is tangent to the neutral directions. The dynamics on this manifold determine the long-term behavior of the whole system near the equilibrium. By projecting the system onto the center manifold, we reduce the number of variables we need to study. For example, in a system with one neutral and two stable directions, we can ignore the stable ones and focus on a one-dimensional equation on the center manifold. This makes analysis much easier, especially for bifurcations (qualitative changes) where neutral directions appear.
8. What does Liouville's Theorem say about the volume of phase space in Hamiltonian systems?
Liouville's Theorem states that for a Hamiltonian system (a system described by a Hamiltonian function that gives the total energy), the volume of any region in phase space (the space of positions and momenta) remains constant as the system evolves in time. This means that if you take a set of initial conditions, the 'cloud' of points moves and deforms but its total volume does not change. This is a consequence of the fact that Hamiltonian flow is incompressible (divergence-free). For example, in a pendulum, the area of a small patch of initial angles and angular velocities stays the same over time. This property is key to understanding statistical mechanics and also leads to the possibility of chaos: even though volume is preserved, the shape can become very stretched and folded, leading to mixing.
9. What does the Poincaré-Bendixson Theorem tell us about the long-term behavior of a continuous dynamical system in two dimensions?
The Poincaré-Bendixson Theorem says that in a two-dimensional continuous system (like a flow on a plane), if a trajectory stays within a bounded region (does not go to infinity) and does not approach a fixed point, then it must approach a limit cycle (a closed periodic orbit) or a set of fixed points connected by trajectories. In other words, chaos (strange attractors) cannot occur in two-dimensional continuous systems. The theorem applies only to systems on the plane or on a sphere (compact two-dimensional manifolds). It is a powerful result because it limits the possible long-term behaviors to just fixed points, periodic orbits, or cycles of homoclinic/heteroclinic orbits (orbits connecting fixed points). This is why chaos requires at least three dimensions in continuous systems.
10. What does the Hartman-Grobman Theorem say about the behavior of a system near a hyperbolic fixed point?
The Hartman-Grobman Theorem states that near a hyperbolic fixed point (where all eigenvalues have non-zero real parts), the nonlinear system behaves like its linearization (the linear approximation). More precisely, there is a continuous change of coordinates that makes the nonlinear system exactly equal to the linear one in a small neighborhood. This means the local dynamics are topologically the same: the same number of stable and unstable directions, and the same flow patterns (like a saddle or a node). For example, if the linearization has a saddle (one stable, one unstable direction), the nonlinear system also has a saddle nearby. The theorem does not apply when eigenvalues have zero real parts (non-hyperbolic), because then the linear part does not capture all the behavior.
11. If you have a time series of daily stock prices, how would you use Takens' Embedding Theorem to check for chaos?
First, choose a time delay τ (say 1 day) and an embedding dimension m (say 3). Then create vectors [price(t), price(t-τ), price(t-2τ)] from the data. These vectors form a reconstructed state space. To check for chaos, you can compute the correlation dimension (a measure of how the points fill space) or the largest Lyapunov exponent (how fast nearby points separate). If the correlation dimension is low and non-integer, and the Lyapunov exponent is positive, the data may come from a chaotic system. However, stock prices are noisy, so you must be careful: the theorem assumes the underlying system is deterministic (no randomness). In practice, you would also test for nonlinearity and stationarity (statistical properties not changing over time) before concluding chaos.
12. How can a Hamiltonian system be chaotic if Liouville's Theorem says phase space volume is preserved?
Chaos in Hamiltonian systems comes from stretching and folding of phase space regions, not from volume change. Liouville's Theorem ensures volume is constant, but the shape can become extremely complicated. For example, a small ball of initial conditions can get stretched into a long thin filament that folds back on itself, like kneading dough. Over time, nearby points separate exponentially (positive Lyapunov exponent), which is the hallmark of chaos. This is possible because the stretching in some directions is balanced by compression in others, keeping the total volume fixed. So chaos coexists with volume preservation. Examples include the motion of asteroids in the solar system (the three-body problem) and the dynamics of charged particles in magnetic fields.