Random dynamical system

In the mathematical field of dynamical systems, a random dynamical system is a dynamical system in which the equations of motion have an element of randomness to them. Random dynamical systems are characterized by a state space S, a set of maps [math]\displaystyle{ \Gamma }[/math] from S into itself that can be thought of as the set of all possible equations of motion, and a probability distribution Q on the set [math]\displaystyle{ \Gamma }[/math] that represents the random choice of map. Motion in a random dynamical system can be informally thought of as a state [math]\displaystyle{ X \in S }[/math] evolving according to a succession of maps randomly chosen according to the distribution Q.^[1]

An example of a random dynamical system is a stochastic differential equation; in this case the distribution Q is typically determined by noise terms. It consists of a base flow, the "noise", and a cocycle dynamical system on the "physical" phase space. Another example is discrete state random dynamical system; some elementary contradistinctions between Markov chain and random dynamical system descriptions of a stochastic dynamics are discussed.^[2]

Motivation 1: Solutions to a stochastic differential equation

Let [math]\displaystyle{ f : \mathbb{R}^{d} \to \mathbb{R}^{d} }[/math] be a [math]\displaystyle{ d }[/math]-dimensional vector field, and let [math]\displaystyle{ \varepsilon \gt 0 }[/math]. Suppose that the solution [math]\displaystyle{ X(t, \omega; x_{0}) }[/math] to the stochastic differential equation

[math]\displaystyle{ \left\{ \begin{matrix} \mathrm{d} X = f(X) \, \mathrm{d} t + \varepsilon \, \mathrm{d} W (t); \\ X (0) = x_{0}; \end{matrix} \right. }[/math]

exists for all positive time and some (small) interval of negative time dependent upon [math]\displaystyle{ \omega \in \Omega }[/math], where [math]\displaystyle{ W : \mathbb{R} \times \Omega \to \mathbb{R}^{d} }[/math] denotes a [math]\displaystyle{ d }[/math]-dimensional Wiener process (Brownian motion). Implicitly, this statement uses the classical Wiener probability space

[math]\displaystyle{ (\Omega, \mathcal{F}, \mathbb{P}) := \left( C_{0} (\mathbb{R}; \mathbb{R}^{d}), \mathcal{B} (C_{0} (\mathbb{R}; \mathbb{R}^{d})), \gamma \right). }[/math]

In this context, the Wiener process is the coordinate process.

Now define a flow map or (solution operator) [math]\displaystyle{ \varphi : \mathbb{R} \times \Omega \times \mathbb{R}^{d} \to \mathbb{R}^{d} }[/math] by

[math]\displaystyle{ \varphi (t, \omega, x_{0}) := X(t, \omega; x_{0}) }[/math]

(whenever the right hand side is well-defined). Then [math]\displaystyle{ \varphi }[/math] (or, more precisely, the pair [math]\displaystyle{ (\mathbb{R}^{d}, \varphi) }[/math]) is a (local, left-sided) random dynamical system. The process of generating a "flow" from the solution to a stochastic differential equation leads us to study suitably defined "flows" on their own. These "flows" are random dynamical systems.

Motivation 2: Connection to Markov Chain

An i.i.d random dynamical system in the discrete space is described by a triplet [math]\displaystyle{ (S, \Gamma, Q) }[/math].

[math]\displaystyle{ S }[/math] is the state space, [math]\displaystyle{ \{s_1, s_2,\cdots, s_n\} }[/math].
[math]\displaystyle{ \Gamma }[/math] is a family of maps of [math]\displaystyle{ S\rightarrow S }[/math]. Each such map has a [math]\displaystyle{ n\times n }[/math] matrix representation, called deterministic transition matrix. It is a binary matrix but it has exactly one entry 1 in each row and 0s otherwise.
[math]\displaystyle{ Q }[/math] is the probability measure of the [math]\displaystyle{ \sigma }[/math]-field of [math]\displaystyle{ \Gamma }[/math].

The discrete random dynamical system comes as follows,

The system is in some state [math]\displaystyle{ x_0 }[/math] in [math]\displaystyle{ S }[/math], a map [math]\displaystyle{ \alpha_1 }[/math] in [math]\displaystyle{ \Gamma }[/math] is chosen according to the probability measure [math]\displaystyle{ Q }[/math] and the system moves to the state [math]\displaystyle{ x_1=\alpha_1(x_0) }[/math] in step 1.
Independently of previous maps, another map [math]\displaystyle{ \alpha_2 }[/math] is chosen according to the probability measure [math]\displaystyle{ Q }[/math] and the system moves to the state [math]\displaystyle{ x_2=\alpha_2(x_1) }[/math].
The procedure repeats.

The random variable [math]\displaystyle{ X_n }[/math] is constructed by means of composition of independent random maps, [math]\displaystyle{ X_n=\alpha_n\circ \alpha_{n-1}\circ \dots \circ \alpha_1(X_0) }[/math]. Clearly, [math]\displaystyle{ X_n }[/math] is a Markov Chain.

Reversely, can, and how, a given MC be represented by the compositions of i.i.d. random transformations? Yes, it can, but not unique. The proof for existence is similar with Birkhoff–von Neumann theorem for doubly stochastic matrix.

Here is an example that illustrates the existence and non-uniqueness.

Example: If the state space [math]\displaystyle{ S=\{1, 2\} }[/math] and the set of the transformations [math]\displaystyle{ \Gamma }[/math] expressed in terms of deterministic transition matrices. Then a Markov transition matrix[math]\displaystyle{ M =\left(\begin{array}{cc} 0.4 & 0.6 \\ 0.7 & 0.3 \end{array}\right) }[/math] can be represented by the following decomposition by the min-max algorithm, [math]\displaystyle{ M =0.6\left(\begin{array}{cc} 0 & 1 \\ 1 & 0 \end{array}\right)+0.3 \left(\begin{array}{cc} 1 & 0 \\ 0 & 1 \end{array}\right)+ 0.1\left(\begin{array}{cc} 1 & 0 \\ 1 & 0 \end{array}\right). }[/math]

In the meantime, another decomposition could be [math]\displaystyle{ M = 0.18 \left(\begin{array}{cc} 0 & 1 \\ 0 & 1 \end{array}\right)+ 0.28\left(\begin{array}{cc} 1 & 0 \\ 1 & 0 \end{array}\right) +0.42\left(\begin{array}{cc} 0 & 1 \\ 1 & 0 \end{array}\right)+0.12\left(\begin{array}{cc} 1 & 0 \\ 0 & 1 \end{array}\right). }[/math]

Formal definition

Formally,^[3] a random dynamical system consists of a base flow, the "noise", and a cocycle dynamical system on the "physical" phase space. In detail.

Let [math]\displaystyle{ (\Omega, \mathcal{F}, \mathbb{P}) }[/math] be a probability space, the noise space. Define the base flow [math]\displaystyle{ \vartheta : \mathbb{R} \times \Omega \to \Omega }[/math] as follows: for each "time" [math]\displaystyle{ s \in \mathbb{R} }[/math], let [math]\displaystyle{ \vartheta_{s} : \Omega \to \Omega }[/math] be a measure-preserving measurable function:

[math]\displaystyle{ \mathbb{P} (E) = \mathbb{P} (\vartheta_{s}^{-1} (E)) }[/math] for all [math]\displaystyle{ E \in \mathcal{F} }[/math] and [math]\displaystyle{ s \in \mathbb{R} }[/math];

Suppose also that

[math]\displaystyle{ \vartheta_{0} = \mathrm{id}_{\Omega} : \Omega \to \Omega }[/math], the identity function on [math]\displaystyle{ \Omega }[/math];
for all [math]\displaystyle{ s, t \in \mathbb{R} }[/math], [math]\displaystyle{ \vartheta_{s} \circ \vartheta_{t} = \vartheta_{s + t} }[/math].

That is, [math]\displaystyle{ \vartheta_{s} }[/math], [math]\displaystyle{ s \in \mathbb{R} }[/math], forms a group of measure-preserving transformation of the noise [math]\displaystyle{ (\Omega, \mathcal{F}, \mathbb{P}) }[/math]. For one-sided random dynamical systems, one would consider only positive indices [math]\displaystyle{ s }[/math]; for discrete-time random dynamical systems, one would consider only integer-valued [math]\displaystyle{ s }[/math]; in these cases, the maps [math]\displaystyle{ \vartheta_{s} }[/math] would only form a commutative monoid instead of a group.

While true in most applications, it is not usually part of the formal definition of a random dynamical system to require that the measure-preserving dynamical system [math]\displaystyle{ (\Omega, \mathcal{F}, \mathbb{P}, \vartheta) }[/math] is ergodic.

Now let [math]\displaystyle{ (X, d) }[/math] be a complete separable metric space, the phase space. Let [math]\displaystyle{ \varphi : \mathbb{R} \times \Omega \times X \to X }[/math] be a [math]\displaystyle{ (\mathcal{B} (\mathbb{R}) \otimes \mathcal{F} \otimes \mathcal{B} (X), \mathcal{B} (X)) }[/math]-measurable function such that

for all [math]\displaystyle{ \omega \in \Omega }[/math], [math]\displaystyle{ \varphi (0, \omega) = \mathrm{id}_{X} : X \to X }[/math], the identity function on [math]\displaystyle{ X }[/math];
for (almost) all [math]\displaystyle{ \omega \in \Omega }[/math], [math]\displaystyle{ (t,x) \mapsto \varphi (t, \omega,x) }[/math] is continuous;
[math]\displaystyle{ \varphi }[/math] satisfies the (crude) cocycle property: for almost all [math]\displaystyle{ \omega \in \Omega }[/math],

[math]\displaystyle{ \varphi (t, \vartheta_{s} (\omega)) \circ \varphi (s, \omega) = \varphi (t + s, \omega). }[/math]

In the case of random dynamical systems driven by a Wiener process [math]\displaystyle{ W : \mathbb{R} \times \Omega \to X }[/math], the base flow [math]\displaystyle{ \vartheta_{s} : \Omega \to \Omega }[/math] would be given by

[math]\displaystyle{ W (t, \vartheta_{s} (\omega)) = W (t + s, \omega) - W(s, \omega) }[/math].

This can be read as saying that [math]\displaystyle{ \vartheta_{s} }[/math] "starts the noise at time [math]\displaystyle{ s }[/math] instead of time 0". Thus, the cocycle property can be read as saying that evolving the initial condition [math]\displaystyle{ x_{0} }[/math] with some noise [math]\displaystyle{ \omega }[/math] for [math]\displaystyle{ s }[/math] seconds and then through [math]\displaystyle{ t }[/math] seconds with the same noise (as started from the [math]\displaystyle{ s }[/math] seconds mark) gives the same result as evolving [math]\displaystyle{ x_{0} }[/math] through [math]\displaystyle{ (t + s) }[/math] seconds with that same noise.

Attractors for random dynamical systems

The notion of an attractor for a random dynamical system is not as straightforward to define as in the deterministic case. For technical reasons, it is necessary to "rewind time", as in the definition of a pullback attractor.^[4] Moreover, the attractor is dependent upon the realisation [math]\displaystyle{ \omega }[/math] of the noise.

References

↑ Bhattacharya, Rabi; Majumdar, Mukul (2003). "Random dynamical systems: a review". Economic Theory 23 (1): 13–38. doi:10.1007/s00199-003-0357-4.
↑ Ye, Felix X.-F.; Wang, Yue; Qian, Hong (August 2016). "Stochastic dynamics: Markov chains and random transformations". Discrete and Continuous Dynamical Systems - Series B 21 (7): 2337–2361. doi:10.3934/dcdsb.2016050.
↑ Arnold, Ludwig (1998). Random Dynamical Systems. ISBN 9783540637585. https://books.google.com/books?id=W5AY5A3S2kQC.
↑ Crauel, Hans; Debussche, Arnaud; Flandoli, Franco (1997). "Random attractors". Journal of Dynamics and Differential Equations 9 (2): 307–341. doi:10.1007/BF02219225. Bibcode: 1997JDDE....9..307C.

0.00

(0 votes)

Original source: https://en.wikipedia.org/wiki/Random dynamical system. Read more

[Bhattacharya2003-1] Bhattacharya, Rabi; Majumdar, Mukul (2003). "Random dynamical systems: a review". Economic Theory 23 (1): 13–38. doi:10.1007/s00199-003-0357-4.

[2] Ye, Felix X.-F.; Wang, Yue; Qian, Hong (August 2016). "Stochastic dynamics: Markov chains and random transformations". Discrete and Continuous Dynamical Systems - Series B 21 (7): 2337–2361. doi:10.3934/dcdsb.2016050.

[3] Arnold, Ludwig (1998). Random Dynamical Systems. ISBN 9783540637585. https://books.google.com/books?id=W5AY5A3S2kQC.

[4] Crauel, Hans; Debussche, Arnaud; Flandoli, Franco (1997). "Random attractors". Journal of Dynamics and Differential Equations 9 (2): 307–341. doi:10.1007/BF02219225. Bibcode: 1997JDDE....9..307C.

[1]

[2]

[3]

[4]

v t e Stochastic processes
Discrete time	Bernoulli process Branching process Chinese restaurant process Galton–Watson process Independent and identically distributed random variables Markov chain Moran process Random walk Loop-erased Self-avoiding Biased Maximal entropy
Continuous time	Bessel process Birth–death process Brownian motion Bridge Excursion Fractional Geometric Meander Cauchy process Contact process Continuous-time random walk Cox process Diffusion process Empirical process Feller process Fleming–Viot process Gamma process Geometric process Hunt process Interacting particle systems Itô diffusion Itô process Jump diffusion Jump process Lévy process Local time Markov additive process McKean–Vlasov process Ornstein–Uhlenbeck process Poisson process Compound Non-homogeneous Point process Schramm–Loewner evolution Semimartingale Sigma-martingale Stable process Superprocess Telegraph process Variance gamma process Wiener process Wiener sausage
Both	Branching process Galves–Löcherbach model Gaussian process Hidden Markov model (HMM) Markov process Martingale Differences Local Sub- Super- Random dynamical system Regenerative process Renewal process Stochastic chains with memory of variable length White noise
Fields and other	Dirichlet process Gaussian random field Gibbs measure Hopfield model Ising model Potts model Boolean network Markov random field Percolation Pitman–Yor process Point process Cox Poisson Random field Random graph
Time series models	Autoregressive conditional heteroskedasticity (ARCH) model Autoregressive integrated moving average (ARIMA) model Autoregressive (AR) model Autoregressive–moving-average (ARMA) model Generalized autoregressive conditional heteroskedasticity (GARCH) model Moving-average (MA) model
Financial models	Black–Derman–Toy Black–Karasinski Black–Scholes Chen Constant elasticity of variance (CEV) Cox–Ingersoll–Ross (CIR) Garman–Kohlhagen Heath–Jarrow–Morton (HJM) Heston Ho–Lee Hull–White LIBOR market Rendleman–Bartter SABR volatility Vašíček Wilkie
Actuarial models	Bühlmann Cramér–Lundberg Risk process Sparre–Anderson
Queueing models	Bulk Fluid Generalized queueing network M/G/1 M/M/1 M/M/c
Properties	Càdlàg paths Continuous Continuous paths Ergodic Exchangeable Feller-continuous Gauss–Markov Markov Mixing Piecewise deterministic Predictable Progressively measurable Self-similar Stationary Time-reversible
Limit theorems	Central limit theorem Donsker's theorem Doob's martingale convergence theorems Ergodic theorem Fisher–Tippett–Gnedenko theorem Large deviation principle Law of large numbers (weak/strong) Law of the iterated logarithm Maximal ergodic theorem Sanov's theorem
Inequalities	Burkholder–Davis–Gundy Doob's martingale Kunita–Watanabe
Tools	Cameron–Martin formula Convergence of random variables Doléans-Dade exponential Doob decomposition theorem Doob–Meyer decomposition theorem Doob's optional stopping theorem Dynkin's formula Feynman–Kac formula Filtration Girsanov theorem Infinitesimal generator Itô integral Itô's lemma Karhunen–Loève_theorem Kolmogorov continuity theorem Kolmogorov extension theorem Lévy–Prokhorov metric Malliavin calculus Martingale representation theorem Optional stopping theorem Prokhorov's theorem Quadratic variation Reflection principle Skorokhod integral Skorokhod's representation theorem Skorokhod space Snell envelope Stochastic differential equation Tanaka Stopping time Stratonovich integral Uniform integrability Usual hypotheses Wiener space Classical Abstract
Disciplines	Actuarial mathematics Econometrics Ergodic theory Extreme value theory (EVT) Large deviations theory Mathematical finance Mathematical statistics Probability theory Queueing theory Renewal theory Ruin theory Statistics System on Chip design Stochastic analysis Time series analysis Machine learning
List of topics Category