Emblem, main page link Modeling in Nonlinear Dynamics Group Переключение на русский Swich to English

  Reference materials

Dynamical modeling

Before construction of a model, it is important to formulate the requirements to its predictive abilities. This choice determines the kind of necessary mathematical tools. If precise and unique forecast of the future states based on the current one is not desired then the probabilistic approach is traditionally used. Hence, at least some of the quantities describing an object of modeling are claimed to be random, i.e. unpredictable in principle, stochastic. There can be different justifications for such a "verdict". If it is "signed", then the tools of the theory of probabilities and mathematical statistics should be used.

An alternative approach is called "dynamical" and relies on the concept of determinism. Determinism is the doctrine claiming regularity and causation in all natural and social phenomena. To realize this approach, one needs mathematical constructions providing a unique forecast. Its classical (but not unique) exponent are differential equations.

Thus, the dynamical approach is based on the idea that the entire future behavior of an object is uniquely determined by its state (x) at the initial time instant. The state (or state vector) is a set of D values: x=(x1, x2, ..., xD); D is its "dimension". The rule which uniquely determines the object evolution from the initial state x(t0) is called "evolution operator" Фt. The state at any future time instant t0+t is found as x(t0+t)=Фt(x(t0)). Evolution operator can be specified with the aid of equations, maps, matrices, graphs, and in other ways, which meet the "unique forecast" condition. The entire model construction described is called "dynamical system".

Let us consider an ordinary cat as an example. If the goal of modeling is description of its health dynamics, one can use the mass M=x3, height H=x2, and hair density (number of hairs per square unit) N=x1 as dynamical variables. The set x=(x1, x2, x3) is the state vector; the dimension is D = 3. We cannot derive an expression for the evolution operator, but it is obvious that cat development depends, e.g., on its food (a1 - mass of the food per a day, kg/day) and life conditions (a2 - duration of walks per a day, hour/day). In contract to dynamical variables, the quantities a1, a2 and similar to them (which can be held constant or controlled) are called "parameters". Coefficients in evolution equations are parameters of dynamical systems.

Separation of characterizing quantities into variables and parameters is dictated by the formulation of the modeling problem. If the goal of modeling of the cat were description of its mechanical movement in space (rather than health), it would be logical to take coordinate of its mass center (x1=x, x2=y, x3=z) and Euler angles of its "longitudinal axis" (x4=α, x5=β, x6=γ) as dynamical variables, while its mass M, downiness N and size H which are constant during a jump but influence the "flight" should be considered as parameters (a1=M, a2=>N, a3=H). Note that the quantities serving as dynamical variables in one problem setting turned into parameters for the other.

Dynamical modeling suggests very wide possibilities of visualization of information, e.g., portraits in state space (see Figures) - phase portraits, representation of the parameters space as a chart showing the regions where different kinds of motions exist, etc.

Rambler's Top100