Stochastic Dynamical Systems
Stock prices, the temperature of a protein in a cell, the motion of pollen in water, all governed by equations with noise. The Black-Scholes formula (Nobel Prize 1997) is the solution to an SDE. Without stochastic calculus there would be no mathematics for the $600 trillion financial derivatives market, nor modern statistical physics.
- **Finance:** Black-Scholes model dS = μS dt + σS dW, option pricing. VaR and Expected Shortfall are quantiles of SDE solution distributions
- **Molecular dynamics:** Langevin dynamics for protein folding, simulation at T > 0 requires stochastic thermostats
- **Neuroscience:** ion currents in neurons are stochastic (random channel opening/closing). Hodgkin-Huxley with noise more accurately describes spike timing
Предварительные знания
Stochastic Differential Equations
Tesla Autopilot operates as a stochastic dynamical system: 1,400 neurons process noisy data from 8 cameras at 36 fps , the Langevin equation governs each inference step. Real systems are subject to noise: temperature fluctuations, quantum randomness, measurement errors. A **Stochastic Differential Equation (SDE)** adds a random term to an ordinary ODE: the noise is modeled by white noise ξ(t), the derivative of Brownian motion W(t).
For multiplicative noise g(X), two interpretations exist: Ito (dW evaluated at start of interval) and Stratonovich (at midpoint). Ito: mathematically convenient, no standard chain rule. Stratonovich: physically natural for systems with colored noise. Difference in the 'drift' term: f_Ito = f_Strat - (g/2)·dg/dX.
In the SDE dX = f(X)dt + g(X)dW, the stochastic term g(X)dW with g = const is called:
Ito's Lemma and the Fokker-Planck Equation
Stochastic calculus requires **Ito's lemma**: the chain rule analog for functions of Brownian motion. The key difference from deterministic analysis: an extra second-order term appears because (dW)² = dt.
The Fokker-Planck equation is a deterministic PDE for the **probability density** of a random process. Instead of a single trajectory we track the entire ensemble: P(x, t), the probability density of being at x at time t.
In the Fokker-Planck equation for the Langevin oscillator dX = -X·dt + σ dW, the stationary distribution is:
Stochastic Bifurcations
Noise does not merely 'smear' deterministic bifurcations, it can **create new phenomena**: noise-induced transitions, stochastic resonance, and P-bifurcations (phenomenological, change in the shape of P_st).
D-bifurcation (dynamical): change in sign of Lyapunov exponents of a random dynamical system. P-bifurcation (phenomenological): change in topology of the stationary density P_st (number of modes). These two definitions do not always coincide!
Stochastic resonance is a phenomenon where:
Numerical SDE Solvers and Monte Carlo
Analytical solutions to SDEs are rare. In practice, **numerical schemes** are used: the Euler-Maruyama method (simplest) and Milstein (second-order). For statistics, simulate an ensemble of trajectories (Monte Carlo).
Milstein adds a correction: X_{n+1} = X_n + f·dt + g·dW + (g·∂g/∂X)/2·((dW)² - dt). This improves accuracy from O(dt^{0.5}) to O(dt) for multiplicative noise. For additive noise (g=const) both methods coincide.
The Euler-Maruyama method for SDEs converges at order O(dt^{0.5}), not O(dt) like the standard Euler for ODEs. Why?
Key Ideas
- **Ito SDE:** dX = f dt + g dW, adds Brownian noise to an ODE. Two types: additive (g=const) and multiplicative (g=g(X))
- **Ito's Lemma:** for Y=h(X) a correction term (g²/2)·h'' dt appears, difference from the standard chain rule
- **Fokker-Planck:** deterministic PDE for P(x,t). Stationary solution P_st(x) ∝ exp(2/σ² ∫f dx')
- **Stochastic resonance:** optimal noise level improves signal detection, exploited in biological sensors
Related Topics
Stochastic systems bridge dynamical systems with probability theory and statistical physics:
- Bifurcations — Noise smears deterministic bifurcations and creates new ones, P-bifurcations. Stochastic bifurcation threshold ≠ deterministic threshold
- Chaos and Strange Attractors — Stochastic chaos: noise in chaotic systems can stabilize or destabilize attractors
- Renormalization Group — Stochastic systems near critical points are described by the RG, fluctuations matter at all scales
Вопросы для размышления
- If Black-Scholes is based on Gaussian noise, why do financial crises happen 'too often'? What does the model miss?
- Stochastic resonance is used in biology (hair cells of the ear). Can this principle be deliberately exploited in engineering? What are the limitations?
- What is the fundamental difference between uncertainty in quantum mechanics (Heisenberg's principle) and stochasticity in SDEs?