STOCHASTIC DIFFERENTIAL GAMES:THE LINEAR QUADRATIC ZERO SUM CASE

Similar documents
Bogoliubov Transformation in Classical Mechanics

Laplace Transformation

An Inequality for Nonnegative Matrices and the Inverse Eigenvalue Problem

Physics 741 Graduate Quantum Mechanics 1 Solutions to Final Exam, Fall 2014

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281

1 Bertrand duopoly with incomplete information

The Hassenpflug Matrix Tensor Notation

STOCHASTIC EVOLUTION EQUATIONS WITH RANDOM GENERATORS. By Jorge A. León 1 and David Nualart 2 CINVESTAV-IPN and Universitat de Barcelona

MATEMATIK Datum: Tid: eftermiddag. A.Heintz Telefonvakt: Anders Martinsson Tel.:

Convex Hulls of Curves Sam Burton

FILTERING OF NONLINEAR STOCHASTIC FEEDBACK SYSTEMS

HELICAL TUBES TOUCHING ONE ANOTHER OR THEMSELVES

The continuous time random walk (CTRW) was introduced by Montroll and Weiss 1.

Linear Quadratic Stochastic Differential Games under Asymmetric Value of Information

TRIPLE SOLUTIONS FOR THE ONE-DIMENSIONAL

A Constraint Propagation Algorithm for Determining the Stability Margin. The paper addresses the stability margin assessment for linear systems

6. KALMAN-BUCY FILTER

Lecture 17: Analytic Functions and Integrals (See Chapter 14 in Boas)

ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION. Xiaoqun Wang

Feedback Control Systems (FCS)

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004

696 Fu Jing-Li et al Vol. 12 form in generalized coordinate Q ffiq dt = 0 ( = 1; ;n): (3) For nonholonomic ytem, ffiq are not independent of

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS

Manprit Kaur and Arun Kumar

Michał Kisielewicz SOME OPTIMAL CONTROL PROBLEMS FOR PARTIAL DIFFERENTIAL INCLUSIONS

To appear in International Journal of Numerical Methods in Fluids in Stability analysis of numerical interface conditions in uid-structure therm

Digital Control System

EXTENDED STABILITY MARGINS ON CONTROLLER DESIGN FOR NONLINEAR INPUT DELAY SYSTEMS. Otto J. Roesch, Hubert Roth, Asif Iqbal

1 Routh Array: 15 points

A SIMPLE NASH-MOSER IMPLICIT FUNCTION THEOREM IN WEIGHTED BANACH SPACES. Sanghyun Cho

IEOR 3106: Fall 2013, Professor Whitt Topics for Discussion: Tuesday, November 19 Alternating Renewal Processes and The Renewal Equation

Optimal financing and dividend control of the insurance company with proportional reinsurance policy

UNIT 15 RELIABILITY EVALUATION OF k-out-of-n AND STANDBY SYSTEMS

Automatic Control Systems. Part III: Root Locus Technique

Online Appendix for Managerial Attention and Worker Performance by Marina Halac and Andrea Prat

Correction for Simple System Example and Notes on Laplace Transforms / Deviation Variables ECHE 550 Fall 2002

in a circular cylindrical cavity K. Kakazu Department of Physics, University of the Ryukyus, Okinawa , Japan Y. S. Kim

March 18, 2014 Academic Year 2013/14

Linear System Fundamentals

V = 4 3 πr3. d dt V = d ( 4 dv dt. = 4 3 π d dt r3 dv π 3r2 dv. dt = 4πr 2 dr

Multi-dimensional Fuzzy Euler Approximation

Chapter 4. The Laplace Transform Method

The Laplace Transform , Haynes Miller and Jeremy Orloff

Minimal state space realization of MIMO systems in the max algebra

A BATCH-ARRIVAL QUEUE WITH MULTIPLE SERVERS AND FUZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH

Domain Optimization Analysis in Linear Elastic Problems * (Approach Using Traction Method)

Reading assignment: In this chapter we will cover Sections Definition and the Laplace transform of simple functions

Nonlinear Single-Particle Dynamics in High Energy Accelerators

EE Control Systems LECTURE 14

Problem 1. Construct a filtered probability space on which a Brownian motion W and an adapted process X are defined and such that

On mild solutions of a semilinear mixed Volterra-Fredholm functional integrodifferential evolution nonlocal problem in Banach spaces

OPTIMAL STOPPING FOR SHEPP S URN WITH RISK AVERSION

THE PARAMETERIZATION OF ALL TWO-DEGREES-OF-FREEDOM SEMISTRONGLY STABILIZING CONTROLLERS. Tatsuya Hoshikawa, Kou Yamada and Yuko Tatsumi

Factor Analysis with Poisson Output

Unbounded solutions of second order discrete BVPs on infinite intervals

REPRESENTATION OF ALGEBRAIC STRUCTURES BY BOOLEAN FUNCTIONS. Logic and Applications 2015 (LAP 2015) September 21-25, 2015, Dubrovnik, Croatia

Stochastic Perishable Inventory Control in a Service Facility System Maintaining Inventory for Service: Semi Markov Decision Problem

arxiv: v1 [math.ac] 30 Nov 2012

2 where. x 1 θ = 1 θ = 0.6 x 2 θ = 0 θ = 0.2

Given the following circuit with unknown initial capacitor voltage v(0): X(s) Immediately, we know that the transfer function H(s) is

Control Systems Analysis and Design by the Root-Locus Method

ON A CERTAIN FAMILY OF QUARTIC THUE EQUATIONS WITH THREE PARAMETERS. Volker Ziegler Technische Universität Graz, Austria

Social Studies 201 Notes for November 14, 2003

Jump condition at the boundary between a porous catalyst and a homogeneous fluid

ON A CERTAIN FAMILY OF QUARTIC THUE EQUATIONS WITH THREE PARAMETERS

ON THE SMOOTHNESS OF SOLUTIONS TO A SPECIAL NEUMANN PROBLEM ON NONSMOOTH DOMAINS

Robust Decentralized Design of H -based Frequency Stabilizer of SMES

New bounds for Morse clusters

arxiv: v3 [quant-ph] 23 Nov 2011

Problem Set 8 Solutions

AMS 212B Perturbation Methods Lecture 20 Part 1 Copyright by Hongyun Wang, UCSC. is the kinematic viscosity and ˆp = p ρ 0

Long-term returns in stochastic interest rate models

Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions

The Study of a Class of the Fractional Brownian Motion Base on Wavelet. 1 Introduction. 2 Basic definition

A Likelihood Ratio Formula for Two- Dimensional Random Fields

The Secret Life of the ax + b Group

February 5, :53 WSPC/INSTRUCTION FILE Mild solution for quasilinear pde

Learning Multiplicative Interactions

Lecture 8: Period Finding: Simon s Problem over Z N

Flag-transitive non-symmetric 2-designs with (r, λ) = 1 and alternating socle

Clustering Methods without Given Number of Clusters

NCAAPMT Calculus Challenge Challenge #3 Due: October 26, 2011

DIFFERENTIAL EQUATIONS

Example: Amplifier Distortion

Reformulation of Block Implicit Linear Multistep Method into Runge Kutta Type Method for Initial Value Problem

Reading assignment: In this chapter we will cover Sections Definition and the Laplace transform of simple functions

Logic, Automata and Games

Coupling of Three-Phase Sequence Circuits Due to Line and Load Asymmetries

Volume 29, Issue 3. Equilibrium in Matching Models with Employment Dependent Productivity

A Short Note on Hysteresis and Odd Harmonics

List coloring hypergraphs

Euler-Bernoulli Beams

/University of Washington Department of Chemistry Chemistry 453 Winter Quarter 2009

AN EXAMPLE FOR THE GENERALIZATION OF THE INTEGRATION OF SPECIAL FUNCTIONS BY USING THE LAPLACE TRANSFORM

PHYS 110B - HW #6 Spring 2004, Solutions by David Pace Any referenced equations are from Griffiths Problem statements are paraphrased

PARAMETRIC ESTIMATION OF HAZARD FUNCTIONS WITH STOCHASTIC COVARIATE PROCESSES

Equivalent POG block schemes

Advanced methods for ODEs and DAEs

NULL HELIX AND k-type NULL SLANT HELICES IN E 4 1

CONTROL SYSTEMS. Chapter 5 : Root Locus Diagram. GATE Objective & Numerical Type Solutions. The transfer function of a closed loop system is

Transcription:

Sankhyā : he Indian Journal of Statitic 1995, Volume 57, Serie A, Pt. 1, pp.161 165 SOCHASIC DIFFERENIAL GAMES:HE LINEAR QUADRAIC ZERO SUM CASE By R. ARDANUY Univeridad de Salamanca SUMMARY. hi paper conider the problem of finding optimal trategie, with two player, baed on linear tochatic differential ytem dξ =(Aξ +B 1 u 1 +B 2 )dt+σdw. Uing a minimax principle we found that the feedback optimal trategie are linear in the tate of the ytem and may be calculated by olving a Ricatti matrix differential equation. We alo formulate a eparation principle that relate the olution in the determinitic cae with the olution in the tochatic cae, a in a linear tochatic control problem. 1. Introduction Let u conider the linear tochatic differential ytem: dξ =(Aξ + B 1 u 1 + B 2 )dt + σdw... (1) where ξ i n-dimentional, W i a m-dimenional Wiener Proce, u 1 i the trategy of player i, anr i -dimenional column (i =1, 2); one aume that there are no retriction on the value of u i ; A, B 1,B 2 and σ are matrice of appropriate dimenion and, in general, function of time. Player 1 attempt to maximize J and player 2 to minimize it, where: J(x,, u 1, )=E x, (ξ u 1 ) W ξ 00 W 01 W 02... W01 W 11 W u 1 W02 W dt + ξ(τ) Fξ(τ) W 22... τ...(2) where τ i the final moment of the game that one aume to be fixed and 0 τ; moreover: Paper received. May 1993. AMS (1985) claification. 93E05, 90D05, 90D25. Key word and phrae. Stochatic differential game, tochatic optimization, eparation principle.

162 r. ardanuy W 00 = W 00 (t) i a ymmetric quare matrix of order n. W 11 = W 11 (t) i a negative definite ymmetric quare matrix of order r 1. W 22 = W 22 (t) i a poitive definite ymmetric quare matrix of order r 2. F i a ymmetric quare matrix. indicate the tranpoition. With a view to implifying the notation, let u et: f(x, t, u 1, )=Ax + B 1 u 1 + B 2...(3) a = σσ...(4) L(x,, u 1, )=(x u 1 ) W x 00 W 01 W 02... W01 W 11 W u 1 W02 W W 22... = x W 00 x + u 1 W 11 u 1 + W 22 +2x W 01 u 1 +2x W 02 +2u 1 W...(5) H(x, t, u 1,,p)=pf(x, t, u 1, )+L(x, t, u 1, )...(6) herefore, taking into account that there are no retriction on u 1 (x, t), (x, t), except for the analytical condition neceary for (1) to have a unique olution and having fixed an initial condition, and by virtue of the Minimax Principle (ee Ardanuy and Alcalá (1991, 1992)), one mut earch for a function V (x, ) of cla C 2,1 (R n (0,τ)), uch that { 0= V + 1 2 tr [V xxa]+h ( x,, u 1,,V x ) ;0 τ V (x, τ) =x Fx where the pair u 1, i a addle point for the Hamiltonian H: max u 1 min...(7) H ( x,, u 1, ),V x = min max H ( x,, u 1, ) (,V x = H x,, u 1, ),V x u 1...(8) 2. Determination of the addle point o determine the addle point, it i neceary to olve the following ytem of equation: { ( ) Hu 1 ( x,, u 1,,V x ) =0 x,, u 1,...(9),V x =0 H After tranpoing term, ytem (9) may be expreed in matrix term a: [ ][ ] [ W11 W u 1 1 ] W W 22 = 2 B 1 Vx + W01x 1 2 B 2 Vx + W02x...(10)

tochatic differential game 163 Since W 11 i negative definite and W 22 poitive definite, there exit the invere of the matrix: [ ] W11 W W = W...(11) W 22 where: [ ] W 1 Γ11 Γ =Γ= Γ...() Γ 22 with: Γ 11 = ( W 11 W W22 1 W ) 1 Γ 22 = ( W 22 W W 11 1 W ) 1...(13) Γ = W11 1 W Γ 22 = Γ 11 W W22 1 and hence: [ ] { [ ] [ ] } u 1 1 B = Γ 1 2 B2 Vx W + 01 W02 x...(14) that i: { [ u 1 = Γ 1 11 2 B 1 Vx + W01x ] [ Γ 1 2 B 2 Vx + W02x ] [ = Γ 1 22 2 B 2 Vx + W02x ] [ Γ 1 2 B 1 Vx + W01x ]...(15) In view of the tructure of H and ince W 11 i negative definite and W 22 poitive definite, it i eay to check that u 1, i a addle point for H. 3. Determination of the value of the game V (x, ) mut atify the partial differential equation (7) that may be rewritten in ymmetrical form: 0 = V + 1 2 tr [V xxa]+ 1 2 V x Ax +[B 1 : B 2 ] u1 + 1 2 x A +[u 1 ] B 1 B2 V x + x W 00 x...(16) +[u 1 ]W u1 + x [W 01 : W 02 ] u1 +[u 1 : ] W 01 W02 with the boundary condition V (x, τ) =x Fx, and where u 1, are a given in (14) or (15). By analogy with the problem of linear-quadratic control, we potulate the following expreion for V (x, ): V (x, ) =x P ()x + g()...(17)

164 r. ardanuy where P i ymmetric and atifie the boundary condition P (τ) =F, while g() i a real function with g(τ) = 0. From (17), one ha: V = x Px+ġ; V x =2x P and V xx =2P...(18) where with the point we indicate derivative with repect to time; by ubtituting in (14), one obtain the optimal trategie; u1 = Γ B 1 P + W 01 B2 W02 x...(19) that are linear in the tate of the ytem. o determine P and g we ubtitute (18) and (19) in (16), and taking into account expreion (), one obtain: where: 0=ġ +tr[pa]+x { P + PG 1 P + PG 2 + G 2 P + G 3 } x...(20) G 1 = [B 1 B 2 ]Γ[B 1 B 2 ] G 2 = A [B 1 B 2 ]Γ[W 01 W 02 ]...(21) G 3 = W 00 [W 01 W 02 Γ[W 01 W 02 ] However, for P and g to fulfill (20) with the repective boundary condition, it i only neceary to olve the ordinary matrix differential equation: P + PG 1 P + PG 2 + G 2 P + G 3 =0...(22) with the boundary condition P (τ) =F, and take: τ g() = tr [P (t)a(t)]dt...(23) In ummary, the calculation method conit in olving, by integrating backward in time, the Ricatti matrix equation (22); by determining P (S), 0 τ, one arrive at (23) and calculate g(), 0 τ. With thee two function, one thu determine the value of the game (17) at each moment and for each tate of the ytem, together with the optimal trategie given in (19). Upon tudying thee equaiton one ee that in the cae where a(t) = 0, with which the game i determinitic, the optimal trategie with feedback continue to be the ame, and only the value of the game varie; i.e.: V (x, ) determinitic = x P ()x...(24) ince by (23), in the determinitic cae g() i equal to zero, and one can thu offer the following eparation principle.

tochatic differential game 165 heorem 1: o olve the linear-quadratic tochatic game, it uffice to olve the correponding linear-quadratic determinitic game (σ = 0); when thi i done, the optimal feedback trategie are linear and in both cae coincide, wherea the value of the game will be given by: τ V (x, ) tochatic = V (x, ) determinitic + tr [P (t)a(t)]dt Regarding the numerical technique to olve the Ricatti equation, among other the following are of particular interet: i) Method of direct integration: the implet of thee i the Euler method. However, owing to numerical error, the value of P (t) i detroyed, uch that it i appropriate to ymmetrize at each point; i.e. ubtitute the P calculated by (P + P )/2. Alternatively, it i poible to take advantage of the ymmetry of P to reduce the matrix equation (22) to a ytem of n(n +1)/2 differential equation of firt order, which coniderably reduce computation time. A lengthy dicuion of thi method of direct integration can be found in Bucy and Joeph (1968). ii) he Kalman-Englar Method, thi i uually applied to time-invariant Ricatti equation (ee, for example, Kalman and Englar (1966), Vaughan (1966), etc.). Reference Ardanuy, R. and Alcala, A. (1991). Minimax Principle for Stochatic Differential Game, Extracta Mathematicae, 6, No. 2-3, pp. 184-186. (1992). Weak infiniteimal Operator and Stochatic Differential Game, Stochatica, 13, No. 1, pp. 5-. Bucy, R. S. and Joeph, P.D. (1968). Filtering for Stochatic Procee with Application to Guidance, J. Wiley, New York. Kalman, R. E. and Englar,.S. (1966). A Uer Manual for the Automatic Synthei Program. NASA, Report CR-475. Vaughan, D. R. (1969). A Negative Exponential Solution for the Matrix Ricatti Equation. IEEE ran. Autom. Control. 14, No. 1, pp. 72-75. Depto. de Matemática Pura y Aplicada Plaza de la Merced, 1 E37008 - Salamanca, Spain.