On the theory elements and methods of parametrical regulation of the national economy evolution based on discrete stochastic dynamic models Abdykappar A. Ashimov1, Bahyt T. Sultanov2, Yuriy V. Borovskiy3, Dmitry A. Novikov4, Simon Ya. Serovajsky5, Askar A. Ashimov6 1) The head of state scientific-technical program, Kazakh National Technical University named after K. Satpayev. Address: KazNTU, 22 Satpayev str., 050013, Almaty city, Kazakhstan. E-mail: Ashimov37@mail.ru 2) The consultant of state scientific-technical program, Kazakh National Technical University named after K. Satpayev. Address: KazNTU, 22 Satpayev str., 050013, Almaty city, Kazakhstan. E-mail: sultanov_bt@pochta.ru 3) The vice-head of state scientific-technical program, Kazakh National Technical University named after K. Satpayev. Address: KazNTU, 22 Satpayev str., 050013, Almaty city, Kazakhstan. E-mail: yuborovskiy@gmail.com 4) The Deputy Director on scientific research, Institute of Control Sciences RAS. Address: 65 Profsoyuznaya st., 117997, Moscow, Russian Federation. Email: novikov@ipu.ru 5) The head of a chair, Al-Farabi Kazakh National University. Address: 39/47 Masanchi st., 050012, Almaty city, Kazakhstan. E-mail: serovajskys@mail.ru 6) The researcher of a state scientific-technical program, Kazakh National Technical University named after K. Satpayev. Address: KazNTU, 22 Satpaev str., 050013, Almaty city, Kazakhstan. E-mail: A.Ashimov@kcp.kz Preferred address for correspondence: Yuriy V. Borovskiy, 313 Furmanov str, apt. 22, 050059, Almaty city, Kazakhstan. E-mail: yuborovskiy@gmail.com The work presents results on development of the parametrical regulation theory for the class of discrete stochastic dynamic systems with additive noise. Within the framework of development of the parametrical regulation theory for discrete dynamic systems we formulated and proved the theorem about sufficient conditions for existence of solutions to the problems of variational calculus on synthesis of optimal laws of parametrical regulation and the theorem about sufficient conditions for continuous dependence of criterions' optimal values of the considered problem on values of unregulated parameters. The discrete stochastic model, taken as example, was obtained from the one deterministic computable model of general equilibrium of economic sectors by adding additive noise to the right parts of dynamic equations of the model. The problem of parametrical identification of the deterministic variant of the model was preliminary solved based on the statistical data for the Republic of Kazakhstan. For the stochastic and deterministic variants of the investigated model problems of optimal economic growth of national economy of the Republic of Kazakhstan were formulated and solved (applying the methods of parametrical regulation theory). Key words: dynamic system with additive noise, parametrical identification, parametrical regulation. 1 Introduction As it is known [1] presented in the literature models of the national economy reflects in mathematical form the most important properties of the economic system and do not account for a number of emerging shocks, such as violations of the supply side (production and supply of labor), violations on the demand side (preferences, specific investment, government spending), the effect of increasing costs or margins (margin, supplements to wages and salaries, the risk premium), violation of money circulation (interest rates, etc.). In this paper we adopted the assumption that these and other possible irregularities in the mathematical model of the national economy can be approximated by adding additive noise to the right sides of the dynamical equations of the corresponding mathematical model of economic system. Below the theory of parametrical regulation of the national economy (the effectiveness of which is shown in the class of models in the form of continuous or discrete dynamical systems [2], [3]), is developing to the class of discrete dynamic systems with additive noise, an important subclass of which are computable general equilibrium models (the so-called CGE-models [4]) with additive noise. In particular, in the framework of the parametrical regulation theory development in the present work for discrete dynamic systems with additive noise we formulated the following theorems: - about sufficient conditions of existence of solutions to problems of variational calculus on the choice of parameters’ optimal values in a given set of their values (on synthesis of optimal laws of regulation). In particular if these conditions are met, this ensures finiteness of mathematical expectation of phase trajectories on a finite time interval. - about sufficient conditions of continuous dependence (of variational calculus problems’ of synthesis laws of parametrical regulation) optimal values of criterion on values of uncontrolled parameters. In this paper the discrete stochastic model, obtained from deterministic computable model of the general equilibrium of economic sectors of Makarov [4] by adding additive noise to the right sides of the model’s dynamical equations, was considered as an example of theoretical results applications. The problem of parametrical identification of the researched model was solved based on statistical data on economy evolution of the Republic of Kazakhstan. The problems of optimal (in sense of some criterion) economic growth of national economy of the Republic of Kazakhstan were formulated and solved applying methods of the parametrical regulation theory. 1. Elements of the parametrical regulation theory based on the discrete dynamic system with additive noise We consider discrete stochastic regulated system ๐ฅ(๐ก + 1) = ๐(๐ฅ(๐ก), ๐ข(๐ก), ๐ผ) + ๐(๐ก), ๐ก = 0, … , ๐ − 1, x(0) ๏ฝ x0 , (1) (2) where t – time taking nonnegative integer values; x ๏ฝ x(t ) ๏ฝ x1 (t ),..., x m (t ) – function of the system (1) state, random vector-function of ๏จ ๏ฉ discrete argument (a vector random process); u ๏ฝ u (t ) ๏ฝ u1 (t ),..., u q (t ) – regulation, vector-function of discrete argument; ๏จ ๏ฉ 2 a ๏ฝ ๏จ a1 ,..., a s ๏ฉ – deterministic vector of uncontrollable parameters, ๐ ∈ ๐ด, ๐ด – given set, ๐ด ⊂ ๐ ๐ ; ๐ = ๐(๐ก) = (๐1 (๐ก), … , ๐ ๐ (๐ก)) – known vector random process, which expresses the noise (as such may be, for example, an additive Gaussian white noise); f – a known vector-function of its arguments; x0 ๏ฝ x01 ,..., x0m – initial state of the system, a deterministic vector. ๏จ ๏ฉ We define the optimality criterion to be the maximization for ๐ fixed: ๐พ๐ผ = ๐{∑๐๐ก=1 ๐น๐ก [๐ฅ(๐ก)]}. (3) Here Ft - are known functions, ะ – mathematical expectation, ๐ฅ(๐ก) - a solution to the system (1), (2) for a given ๐ผ. We introduce the phase constraints for the system (1), (2): E[ x(t )] ๏ X (t ), t ๏ฝ 1,..., n, (4) where X (t ) – given set. Hereinafter the mathematical expectation of a vector chance quantity means the vector of mathematical expectations of the coordinates of this value. In the below considered tasks, explicit constraints on regulation are assumed: u (t ) ๏ U (t ), t ๏ฝ 0,..., n ๏ญ 1, (5) where ๐(๐ก) – given set, ๐(๐ก) ⊂ ๐ ๐ . Sets X (t ) , U (t ) for all determined above t values are the closures of bounded open sets. The method of parametrical regulation is used in the formulation and solution of the following variational problem, so called the problem of variational calculus of the synthesis of the optimum law of parametrical regulation. Problem 1. Given the vector of unregulated parameters ๐ ∈ ๐ด find such regulation u, that satisfies the condition (5), so that corresponding to it the solution of dynamical system (1), (2) satisfies the condition (4) and provides the maximum of the functional (3). For the fixed ๐ ∈ ๐ด let us define the set of allowable regulations for the system (1) – (2) in the following manner: ๐๐ผ = {๐ข|๐ข(๐ก) ∈ ๐(๐ก), ๐ก = 0, … , ๐ − 1; ๐[๐ฅ(๐ก)] ∈ ๐(๐ก), ๐ก = 1, … , ๐} ⊂ ๐ ๐๐ , where x(t ) is the solution to the system (1), (2) corresponding to the regulation u and set ๐ parameter value. Then the problem 1 is reduced to maximizing the functional ๐พ = ๐พ๐ผ (๐ข) which is defined by the formula (3), on the set of allowable regulations of the system ๐๐ผ . We call the problem 1 non-trivial, if the set ๐๐ผ is non-empty and contains some open set. The theorem on sufficient conditions for existence of solution to the problem 1 (set in accordance with the method of parametrical regulation) and the theorem of continuous dependence of the optimal value of ๐พ๐ผ criterion of the problem 1 on the parameter a are presented below. Theorem 1. Suppose given fixed ๐ ∈ ๐ด in non-trivial problem 1 for any ๐ก = 1 ÷ ๐ chance quantity ๐(๐ก) are absolutely continuous and have zero mathematical expectations; functions f and Ft satisfy the Lipschitz condition. Functions f (for ๐ข ∈ ๐๐ผ ) and Ft by module 3 are constrained by some linear functions of |x|. That is, for some positive number b the inequalities |๐(๐ฅ, ๐ข, ๐ผ)| ≤ ๐(1 + |๐ฅ|), |๐น๐ก (๐ฅ)| ≤ ๐(1 + |๐ฅ|) are carried out. Then the problem 1 has a solution. Theorem 2. Suppose when ๐ ∈ ๐ด and ๐ก = 1 ÷ ๐ for non-trivial problem 1 chance quantity ๐(๐ก) are absolutely continuous and have zero mathematical expectations, functions f, Ft satisfy the Lipschitz condition. Functions f and Ft do not exceed a linear function with respect to |x| in some neighborhood of the point a. Then the mapping ๐ → max ๐พ๐ (๐ข) is ๐ข∈๐๐ continuous in A. Proofs of Theorem 1 and 2 are presented in Appendix. 2. Representation of computable general equilibrium models in the form of discrete stochastic dynamic systems and the problem of parametrical regulation of the national economy evolution on the basis of computable models Deterministic computable general equilibrium model in general form is represented by the following system of relations [4, Ch. 3]. 1) The subsystem of difference equations, binding the values of endogenous variables for two consecutive years: x(t ๏ซ 1) ๏ฝ f ( x(t ), y (t ), z (t ), u (t ), ๏ก ) , x(0) ๏ฝ x0 , (6) x (t ) ๏ฝ ( x(t ), y(t ), z (t )) ๏ R m – vector Here t – year number, discrete time, t ๏ฝ 0, 1, 2, ..., n ๏ญ 1 ; ~ of the system’s endogenous variables; x(t ) ๏ฝ ( x 1 (t ), x 2 (t ),..., x m1 (t )) ๏ X 1 (t ) , y(t ) ๏ฝ ( y 1 (t ), y 2 (t ),..., y m2 (t )) ๏ X 2 (t ) , z(t ) ๏ฝ ( z 1 (t ), z 2 (t ),..., z m3 (t )) ๏ X 3 (t ) ; x(t ) variables include values of fixed assets of producing sectors, balances in agents’ bank accounts, etc.; y (t ) include agents’ demand and supply values in different markets, etc.; z (t ) – different values of market prices and a budget share in the market with state prices for different economic agents; m1 ๏ซ m2 ๏ซ m3 ๏ฝ m ; u and ๏ก are vectors of exogenous parameters; u ๏ฝ (u 1 (t ), u 2 (t ),..., u q (t )) ๏ U (t ) ๏ R q – vector of controlled (regulated) parameters; X1(t), X2(t), X3(t), U(t) – compact sets with nonempty interior; ๏ก ๏ฝ ๏จ๏ก 1 , ๏ก 2 ,..., ๏ก s ๏ฉ ๏ A ๏ R m - vector of uncontrolled parameters, A – open connected set; f : X 1 (t ) ๏ด X 2 (t ) ๏ด X 3 (t ) ๏ด U (t ) ๏ด A ๏ฎ R m1 – continuous mapping for t ๏ฝ 0,1,2,..., n ๏ญ 1 . 2) The subsystem of algebraic equations, describing behavior and interaction of agents in different markets during the selected year, these equations allow the expression of variables y (t ) in terms of exogenous parameters and the remaining endogenous variables: y (t ) ๏ฝ g ( x(t ), z (t ), u (t ), ๏ก ) . Here g : X 1 (t ) ๏ด X 3 (t ) ๏ด U (t ) ๏ด A ๏ฎ R m2 continuous mapping, t ๏ฝ 0,1,2,..., n . 4 (7) 3) The subsystem of recurrent relations for iterative calculations of equilibrium values of market prices in different markets and of budget shares in markets with state prices for the different economic agents: z (t )[Q ๏ซ 1] ๏ฝ h( z (t )[Q], y (t )[Q], L, u (t ), ๏ก ) . (8) Here Q ๏ฝ 0, 1, 2, ... – iteration number; L – set of positive numbers (adjustable iteration constants (When their values decrease the economic system reaches the equilibrium state faster, but the risk that the price go to the negative domain increases); h : X 2 (t ) ๏ด X 3 (t ) ๏ด (0,๏ซ๏ฅ) m3 ๏ด U (t ) ๏ด A ๏ฎ R m3 – continuous mapping (which is contracting when x(t ) ๏ X 1 (t ), u (t ) ๏ U (t ) , ๏ก ๏ A are fixed and some fixed L. In this case h mapping has a single fixed point, to which the iterative process (7), (8) converges), t ๏ฝ 0,1,2,..., n . Computable model (6), (7), (8) given fixed values of exogenous variables for each x (t ) that correspond to the demand moment of t time defines values of endogenous variables ~ and supply equilibrium (in the markets of agents’ goods and services) in the framework of the algorithm below. 1) On the first step it is assumed that t=0 and the initial values of x(0) variables are set. 2) On the second step the initial values of z (t )[ 0] variables are set for the current t in different markets and for different agents; with the help of (7) the values of y (t )[0] ๏ฝ G ( x(t ), z (t )[0], u (t ), a) are calculated (initial values of demand and supply of agents in markets of goods and services). 3) On the third step the iteration process (8) is run for current t. Meanwhile current values of demand and supply for every Q are found from (7): y (t )[Q] ๏ฝ G ( x(t ), z (t )[Q], u, a) through the refinement of market prices and budget shares of economic agents. The condition for completion of iteration process is the equality of demand and supply values in different markets. As a result equilibrium values of market prices are determined in every market and budget shares in markets with government prices for different economic agents. Q index is omitted for such equilibrium values of endogenous variables. 4) On the next step values of xt ๏ซ1 variables for the next moment of time are found in accordance with the obtained equilibrium solution for t moment with the help of difference equations (6). The value of t increases by unity. Transition to the step 2. The number of reiteration of steps 2, 3, 4 is defined according to the problems of calibration, forecasting and regulation for time intervals selected in advance. Stochastic computable general equilibrium model (stochastic computable model) obtained from the deterministic model (6), (7), (8), is a model to the dynamical equation’s (6) right side of which we added an additive noise ๐(๐ก): ๐ฅ(๐ก + 1) = ๐(๐ฅ(๐ก), ๐ฆ(๐ก), ๐ง(๐ก), ๐ข(๐ก), ๐ผ) + ๐(๐ก), ๐ก = 0, … , ๐ − 1; ๐ฅ(0) = ๐ฅ0 , (9) i.e., the model of type (9), (7), (8). We formulate the problem of variational calculus on the synthesis of the parametrical regulation optimal law for a stochastic computable model. Problem 2. Given the vector of unregulated parameters ๐ ∈ ๐ด find such a control u(t), which satisfies the condition (5), so that the corresponding to it the solution of the dynamical system (9), (7), (8) satisfies the following condition ๐[๐ฅฬ(๐ก)] ∈ ๐1 (๐ก) × ๐2 (๐ก) × ๐3 (๐ก), ๐ก = 1, … , ๐ 5 and provides maximum to the functional ๐พ๐ผ = ๐{∑๐๐ก=1 ๐น๐ก [๐ฅฬ(๐ก)]}. It is not difficult to check, that formulated above theorems 1, 2 hold true for the computable model with the continuous mappings f, g, h and with convergent iterative process (7), (8). 3. Example. Application of the parametrical regulation theory by the example of a stochastic computable model of economic sectors 3.1. Results of the parametrical identification of the deterministic computable model of economic sectors The considered model according to the statistical data of the Republic of Kazakhstan is presented by nineteen economic agents (sectors), among which there are 16 agents-producers, and there is a sector – an aggregate consumer, a government sector and a banking sector. The model under consideration is presented in the framework of the general expressions of relations (6), (7), (8) respectively m1 ๏ฝ 67 , m2 ๏ฝ 597 , m3 ๏ฝ 34 relations, with the help of which the values of its 698 endogenous variables are calculated. The model also contains 2045 exogenous evaluated parameters. In this case the problem of identification (calibration) of exogenous parameters ω of the model comes to finding the global minimum of some objective function, which is set by the model itself. The constraints on the set of optimization ( ๏ ) are also set by using the same model. To solve the problem of parametric identification of the considered computable model (based on the apparent assumption that the mismatch (in general case) of points of minimum two different functions) the two criteria of the following types were proposed: K IA (๏ท ) ๏ฝ 1 n๏ก T ๏ฆ y i (t ) ๏ญ y i* (t ) ๏ถ ๏ง ๏ท ๏ก ๏ฅ๏ฅ i i* ๏ง ๏ท y ( t ) t ๏ฝ1 i ๏ฝ1 ๏จ ๏ธ T 2 nA K IB (๏ท ) ๏ฝ 1 n๏ข T ๏ฆ y i (t ) ๏ญ y i* (t ) ๏ถ ๏ง ๏ท ๏ข ๏ฅ๏ฅ i i* ๏ง ๏ท y ( t ) t ๏ฝ1 i ๏ฝ1 ๏จ ๏ธ 2 nB T Here y i (t ) , y i* (t ) – calculated and observed values of the model’s output variables respectively, K IA (๏ท ) – auxiliary criterion, K IB (๏ท ) – basic criterion; n B ๏พ n A ; ๏ก i ๏พ 0 ะธ ๏ข i ๏พ 0 – some weight coefficients, which values are determined by solving the problem of nA parametric identification of the dynamic systems; ๏ฅ๏ก i ๏ฝ1 i ๏ฝ n๏ก , nB ๏ฅ๏ข i ๏ฝ1 i ๏ฝ n๏ข . The algorithm of solution to the problem of parametrical identification of the model was selected in the form of the following steps. 1. The problems A and B (problems of finding minimums of the functions K IA (๏ท) and K IB (๏ท ) respectively) are solved simultaneously for some vector of initial values of parameters ๏ท1 ๏ ๏ , as a result points ๏ท K0 IA ะธ ๏ท K0 IB are found. 2. If K IB (๏ท K0 IB ) ๏ผ ๏ฅ , then the problem of parametrical identification of the model (6, 7, 8) is considered to be solved. 3. Otherwise, the problem A is solved taking the point ๏ท K0 IB as a starting point ๏ท 1 and the problem B is solved taking the point ๏ท K0 IA as a starting point ๏ท 1 . Proceed to Step 2. 6 Sufficiently large number of iterations of steps 1, 2, 3 gives an opportunity for searched values of parameters to come out from neighborhood of points of nonglobal minimums of one criterion with the help of another criterion and thus to solve the problem of parametrical identification. As a result of simultaneous solving of the problems A and B applying the stated algorithm with the help of the Nelder-Mead algorithm [5] the values K IA ๏ฝ 0.015 and K IB ๏ฝ 0.0129 were obtained. Meanwhile the relative magnitude of the deviations of calculated values of the variables used in the basic criteria from the corresponding observed values was less than 1.29%. 3.2. Finding optimal values of regulated parameters based on the stochastic computable model of economic sectors The stochastic computable model of economic sectors was obtained from the corresponding deterministic model (with found by solving the problem of parametrical identification estimate values of exogenous parameters) by adding discrete white noise to the right side of the dynamic equations of the model. Values to the order of magnitude smaller compared to the values of deterministic parts of the corresponding dynamic equations were taken as the standard deviations of Gaussian random variables defining the white noise. In computational experiments with the stochastic computable model as the optimization criterion we used the criterion ๏ฉ 1 2015 ๏น K 1 ๏ฝ E ๏ช ๏ฅ Y (t )๏บ ๏ฎ max ๏ซ 6 t ๏ฝ 2010 ๏ป (10) – average value of gross output of the country in prices of 2000 for 2010 - 2015 and according considered realizations of a random process. The value of K1 criterion for the basic computational variant (applying the values of exogenous parameters, obtained as a result of the model’s parametrical identification) equals to K1 = 0.9891โ1013. During the experiments with the optimization criterion (10) the constraints on growth of consumer prices of the following type was used: E( P(t )) ๏ฃ 1.09E( P (t )), t ๏ฝ 2010 ๏ธ 2015 . Here P (t ) – calculated consumer price level of the model without parametrical regulation, P (t ) – consumer price level with parametrical regulation. In the computational experiments the regulation of 1536 exogenous parameters – j-th agent-producer’s budget shares, assigned to purchase of goods and services that are produced by i-th agent-producer for the years of 2010-2015 was carried out: Oi j (t ); t ๏ฝ 2010 ๏ธ 2015; i, j ๏ฝ 1 ๏ธ 16 . Here 16 ๏ฅO i i ๏ฝ1 j (t ) ๏ฃ 1 for the selected values of t. Basic values of the selected shares, obtained as a result of solution of the parametrical identification problem according to the data for 2000-2008 are expressed in terms of Oi j ; i, j ๏ฝ 1 ๏ธ 16 . The following problem of finding the values of regulated vectors parameters was considered. Based on the stochastic computable model of economic sectors find values of 7 agent-producers budget shares ( Oi j (t ); t ๏ฝ 2010 ๏ธ 2015; i, j ๏ฝ 1 ๏ธ 16 ), that would provide the upper boundary of K1 criterion under the following additional constraints on these shares: 0.5 ๏ฃ Oi j (t ) / Oi j ๏ฃ 2; i, j ๏ฝ 1 ๏ธ 16; t ๏ฝ 2010 ๏ธ 2015 . The solutions of such optimization problems were obtained with the help of NelderMead algorithm [5]. After application of the parametrical regulation of the stochastic model’s budget shares, the values of the criterion turned out to be K1 = 1.2453โ1013 its value increased by 25.89 % as compared to the basic variant. The analogous problem of the parametrical regulation with corresponding constraints 1 2015 and with the criterion (10) analogue K 2 ๏ฝ ๏ฅ Y (t ) was solved on the basis of the 6 t ๏ฝ 2010 . deterministic CGE model of economic sectors. After application of the parametrical identification of agent producer’s budget shares the criterion value of the deterministic model turned out to be equal to K2 = 1.6283โ1013, the value of the criterion K2 increased by 33.14 % as compared to the basic variant. If to compare the results of the problem of the variational calculus on the basis of stochastic and deterministic computable models of general equilibrium, one can say that there is a reduction in the estimated value of the functional of the variational problem, taking into account the disturbing violations in the deterministic computable general equilibrium model in the form of additive noise. Conclusion 1. Some results on the development of the theory of parametrical regulation for a class of discrete stochastic dynamic models are presented. We prove theorems on sufficient conditions for the existence of solution of the stated optimization problem and on the continuous dependence of optimal values of the criterion of this problem on uncontrollable parameters. 2. We show the application efficiency of the parametrical regulation theory by the example of a stochastic computable model of economic sectors. A method for estimating the optimal values of regulated parameters of economic policy based on the considered mathematical model is offered and the optimal values of regulated parameters are found. 3. The obtained results can be used in the development and implementation of effective state economic policy. The authors acknowledge N.Yu Borovskiy and D.B. Nurseitov for helping to conduct computational experiments. References [1] Samarskiy A.A., Mikhailov A.P. Mathematical modeling: Ideas. Methods. Examples, Physmatlit, Moscow, 2002 (in Russian). [2] Ashimov A.A., Sultanov B.T., Adilov Zh.M., Borovskiy Yu.V., Novikov D.A., Nizhegorodcev R.V., & Ashimov As.A., Macroeconomic analysis and economic policy based on parametrical regulation, Physmatlit, Moscow, 2010 (in Russian). [3] Ashimov A.A., Sagadiyev K.A., Borovskiy Yu.V., Iskakov N.A. & Ashimov As.A., On the market economy development parametrical regulation theory. Kybernetes, The 8 international journal of cybernetics, systems and management sciences. Vol. 37, โ5, 2008, 623-636. [4] Makarov V.L., Bakhtizin A.R., & Sulakshin S.S. The use of computable models in public administration, Scientific Expert, Moscow: 2007. [5] Nelder J.A. & Mead R. A simplex method for function minimization. The Computer Journal. โ. 7, 1965, 308-313. Appendix Proof of theorem 1. According to the Weierstrass theorem, a continuous function on a nonempty closed bounded set reaches its maximum. Thus, we need to show that a defined with the help of (3) the function of many variables ๐พ = ๐พ๐ผ (๐ข) is continuous and the set ๐๐ผ is closed and bounded. Its non-emptiness is a part of the theorem condition. We show that there are mathematical expectations of values in the phase constraint (4). Indeed, according to equation (1), we have ๐[๐ฅ(๐ก + 1)] = ๐[๐(๐ฅ(๐ก), ๐ข(๐ก), ๐ผ)] + ๐[๐(๐ก)] The second term on the right side of this equation makes sense under the conditions of the theorem, and the first is calculated by the formula ๐[๐(๐ฅ(๐ก), ๐ข(๐ก), ๐ผ)] = ∫ ๐(๐, ๐ข(๐ก), ๐ผ)๐๐ฅ(๐ก) (๐)๐๐, ๐ ๐ if the last integral is absolutely convergent (here the probability density function of a random variable ๐ฅ(๐ก) is expressed in terms of ๐๐ฅ(๐ก) ). The latter fact is indeed the case due to constraints on the growth of f function and due to the existence of the mathematical expectation of x(t) value for any t=1,…,n (this fact is verified by induction). The existence of mathematical expectation on the right side of this equation (3) follows from the constraints on growth of Ft function and the existence of the mathematical expectation of x (t ). value. We prove K continuous dependence on u in topology of ๐ ๐๐ . Suppose we have the convergence of vectors uk ๏ฎ u , ๐ข๐ ∈ ๐๐ผ . From equation (1) it follows that |๐ฅ๐ (๐ก + 1) − ๐ฅ(๐ก + 1)| = |๐(๐ฅ๐ (๐ก), ๐ข๐ (๐ก), ๐) − ๐(๐ฅ(๐ก), ๐ข(๐ก), ๐)| where xk and ั are problem (1), (2) solutions under regulations uk and ะธ respectively. Then the following relation xk (t ๏ซ 1) ๏ญ x(t ๏ซ 1) ๏ฃ L f ๏ฉ๏ซ xk (t ) ๏ญ x(t ) ๏ซ uk (t ) ๏ญ u (t ) ๏น๏ป , is true where L f is a Lipschitz constant of the function f. Repeating similar arguments and taking into account that, in virtue of condition (2) xk (0) ๏ฝ x(0) , we have |๐ฅ๐ (๐ก + 1) − ๐ฅ(๐ก + 1)| ≤ (๐ฟ๐ )2 |๐ฅ๐ (๐ก − 1) − ๐ฅ(๐ก − 1)| + +(๐ฟ๐ )2 |๐ข๐ (๐ก − 1) − ๐ข(๐ก − 1)|+๐ฟ๐ |๐ข๐ (๐ก) − ๐ข(๐ก)| ≤ 9 ๐ก ≤ ∑(๐ฟ๐ )๐ +1 |๐ข๐ (๐ก − ๐ ) − ๐ข(๐ก − ๐ )| ≤ ๐๐ ๐ =0 where ๏ฅ k ๏ฎ 0 when k ๏ฎ ๏ฅ. Denoting the maximum of the Lipschitz constant of Ft functions, for t = 1,…,n through LF we get the estimate Ft ๏ xk (t )๏ ๏ญ Ft ๏ x(t )๏ ๏ฃ LF ๏ฅ k . Having calculated the mathematical expectations of both sides of this inequality, we obtain the inequality ะ{|๐น๐ก [๐ฅ๐ (๐ก)] − ๐น๐ก [๐ฅ(๐ก)]|}≤๐ฟ๐น ๐๐ . It follows that ะ{|๐น๐ก [๐ฅ๐ (๐ก)] − ๐น๐ก [๐ฅ(๐ก)]|}→0 and convergence of the considered sequence: E ๏ปFt ๏ xk (t )๏๏ฝ ๏ฎ E ๏ปFt ๏ x(t )๏๏ฝ . Hence, this convergence and (3) imply the continuity of the functions ๐พ๐ผ from u. Boundedness of the set ๐๐ผ follows from the boundedness of sets U (t ) . Closure of the set ๐๐ผ follows from the continuous mapping ๐๐ผ → ๐, set with the help of definition of the set ๐๐ผ and compactness of the set ะฅ (the theorem on the closure of a complete preimage of a compact under the continuous mapping). Now the existence of problem 1 solution follows from the Weierstrass theorem. The theorem is proved. Tentatively we formulate the following definition and auxiliary statement used in the proof of Theorem 2. Definition. Let the family of functions ๐พ๐ (๐ข), ๐ ∈ ๐ด, ๐ข ∈ ๐ be defined for the family of subsets {๐๐ผ } of some set U of Euclidian space with the parameter ะฐ (from some subset ะ of Euclidian space). The family {๐๐ผ } is K-continuous on the set ะ, if for any ๐ > 0 there is a number ๐ฟ > 0, so that when performing the inequality |๐ − ๐| ≤ ๐ฟ, ๐, ๐ ∈ ๐ด for any ๐ข๐ ∈ ๐๐ there is such point ๐ข๐ ∈ ๐๐ , so that |๐พ๐ (๐ข๐ ) − ๐พ๐ (๐ข๐ )| < ๐ inequality holds. According to this definition the family of sets {๐๐ผ } is K-continuous, if in the case of enough proximity of the parameter b to ะฐ, for any element from the set ๐๐ there is arbitrarily close (for a function value) to it element of the set ๐๐ผ . The following lemma is auxiliary to prove the continuity of optimal values of the problem of variational calculus on the synthesis of optimal laws of parametrical regulation. Lemma 1. Let A and U – be some subsets of Euclidian spaces; A – open set; U - compact; the family of closed subsets {๐๐ผ } belongs to U. Let the mapping (๐, ๐ข) → ๐พ๐ (๐ข) be continuous in the product of ๐ด × ๐. Let the family of subsets {๐๐ผ } be Kcontinuous in some neighborhood of a point ๐0 ∈ ะ. Then the mapping ๐ → max ๐พ๐ (๐ข) is ๐ข∈๐๐ continuous in the point ๐0 . Proof of Lemma 1. Suppose the convergence of some sequence ๐๐ → ๐0 holds, where ๐๐ ∈ ๐ด Let ๐ข๐ denote the maximum point of the function ๐พ๐๐ on the set ๐๐๐ , and ๐ข0 denote the maximum point of the function ๐พ๐0 on the set ๐๐0 . Taking into account ๐พ-continuity of the family of sets {๐๐ } and continuity of the function ๐๐ (๐ฅ) on ๐ด × ๐, we conclude that for any value ε > 0 we can find such a number ๐0 , 10 that when ๐ > ๐0 there are points ๐ข′๐ ∈ ๐๐0 for which the inequalities |๐พ๐๐ (๐ข′๐ ) − ๐พ๐๐ (๐ข๐ )| ≤ ε hold and additionally max|๐พ๐๐ (๐ฆ) − ๐พ๐ะพ (๐ฆ)| ≤ ε holds. ๐ฆ∈๐ As a result, when ๐ > ๐0 we get the following inequalities ๐พ๐0 (๐ข0 ) ≥ ๐พ๐0 (๐ข′ ๐ ) ≥ ๐พ๐๐ (๐ข′ ๐ ) − ๐ ≥ ๐พ๐๐ (๐ข๐ ) − 2๐. (11) In the same way the following relation is checked ๐พ๐๐ (๐ข๐ ) ≥ ๐พ๐0 (๐ข0 ) − 2๐. (12) From (11) and (12) it follows that under sufficiently large ๐ the inequality |๐พ๐๐ (๐ข) − ๐พ๐0 (๐ข0 )| ≤ 2๐ holds, that provides the convergence of the sequence ๐พ๐๐ (๐ข๐ ) → ๐พ๐0 (๐ข0 ). Lemma is proved. Proof of Theorem 2. In the proof of theorem 1 we established that sets U a are closed and bounded and K a function is continuous. There we also established continuity of the mapping u ๏ฎ E ๏ฉ๏ซ xau (t ) ๏น๏ป from which the K-continuity of the family of sets follows. Here ๐ฅ๐ผ๐ข (๐ก) denotes the solution of the system (1), (2) for selected ๐ผ and ๐ข. We show the uniform on u continuity of the mapping a ๏ฎ Ka (u ). Suppose that there is convergence ak ๏ฎ a where ๐, ๐๐ ∈ ๐ด. From the condition (1) we get the following estimate ๏จ ๏ฉ xauk (t ๏ซ 1) ๏ญ xau (t ๏ซ 1) ๏ฝ f xauk (t ), u (t ), ak ๏ญ f ๏จ xau (t ), u (t ), a ๏ฉ ๏ฃ ๏ฃ L f ๏ฉ๏ซ xauk (t ) ๏ญ xau (t ) ๏ซ ak ๏ญ a ๏น๏ป , where L f is the Lipschitz constant of f function. Given the initial state of the system, we obtain the inequality xauk (t ) ๏ญ xau (t ) ๏ฃ ๏ฅ ๏จ L f ๏ฉ ak ๏ญ a . n s s ๏ฝ1 Let LF denote the maximum of Lipschitz constants of Ft functions. Then the following estimate is true Ft ๏ฉ๏ซ xauk (t ) ๏น๏ป ๏ญ Ft ๏ฉ๏ซ xau (t ) ๏น๏ป ๏ฃ LF xauk (t ) ๏ญ xau (t ) ๏ฃ LF ๏ฅ ๏จ L f n s ๏ฝ1 ๏ฉ s ak ๏ญ a . Having proceeded to the mathematical expectations of the right and left sides in the last inequality, we obtain that ะ{|๐น๐ก [๐ฅ๐๐ข๐ (๐ก)] − ๐น๐ก [๐ฅ๐๐ข (๐ก)]|}→0. From this it follows that, when ๐ข ๐ข k ๏ฎ ๏ฅ for any t = 1,…,n there is convergence ะ{๐น๐ก [๐ฅ๐๐ (๐ก)]}→ะ{๐น๐ก [๐ฅ๐ (๐ก)]}. uniformly on ะธ from the sum of all sets U a and hence, uniform on ะธ continuity of the mapping a ๏ฎ Ka (u ). and continuity on (๐, ๐ข) of the function ๐พ๐ (๐ข). Having applied Lemma 1 we obtain the desired result. The theorem is proved. 11