The Perron-Frobenius operators, invariant measures and representations of the Cuntz-Krieger algebras

Similar documents
Chapter 3. Vector Spaces

2 Fundamentals of Functional Analysis

Math 270A: Numerical Linear Algebra

f p dm = exp Use the Dominated Convergence Theorem to complete the exercise. ( d φ(tx))f(x) dx. Ψ (t) =

A PROOF OF THE FUNDAMENTAL THEOREM OF CALCULUS USING HAUSDORFF MEASURES

REPRESENTATION THEORY OF PSL 2 (q)

Theoretical foundations of Gaussian quadrature

Math 520 Final Exam Topic Outline Sections 1 3 (Xiao/Dumas/Liaw) Spring 2008

Module 6: LINEAR TRANSFORMATIONS

4.5 JACOBI ITERATION FOR FINDING EIGENVALUES OF A REAL SYMMETRIC MATRIX. be a real symmetric matrix. ; (where we choose θ π for.

Results on Planar Near Rings

The Bochner Integral and the Weak Property (N)

arxiv: v1 [math.ra] 1 Nov 2014

STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

1.9 C 2 inner variations

New Expansion and Infinite Series

Fundamental Theorem of Calculus for Lebesgue Integration

SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL (1 + µ(f n )) f(x) =. But we don t need the exact bound.) Set

Necessary and Sufficient Conditions for Differentiating Under the Integral Sign

ON THE NILPOTENCY INDEX OF THE RADICAL OF A GROUP ALGEBRA. XI

MATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1

440-2 Geometry/Topology: Differentiable Manifolds Northwestern University Solutions of Practice Problems for Final Exam

Convex Sets and Functions

Notes on length and conformal metrics

Sturm-Liouville Eigenvalue problem: Let p(x) > 0, q(x) 0, r(x) 0 in I = (a, b). Here we assume b > a. Let X C 2 1

Linearly Similar Polynomials

Physics 116C Solution of inhomogeneous ordinary differential equations using Green s functions

Multivariate problems and matrix algebra

NOTES AND PROBLEMS: INTEGRATION THEORY

Lecture 10 :Kac-Moody algebras

arxiv: v1 [math.ca] 7 Mar 2012

STRUCTURED TRIANGULAR LIMIT ALGEBRAS

Are Deligne-Lusztig representations Deligne-Lusztig? Except when they are complex?

Math 6455 Oct 10, Differential Geometry I Fall 2006, Georgia Tech

STUDY GUIDE FOR BASIC EXAM

Chapter 14. Matrix Representations of Linear Transformations

Linearity, linear operators, and self adjoint eigenvalue problems

2 L. BILLINGS AND E.M. BOLLT 2 s well roken liner trnsformtions [5], wek unimodl mps [6]. Other closely relted res in the study the chotic ehvior of t

Advanced Calculus: MATH 410 Uniform Convergence of Functions Professor David Levermore 11 December 2015

A HELLY THEOREM FOR FUNCTIONS WITH VALUES IN METRIC SPACES. 1. Introduction

Elements of Matrix Algebra

Continuous Random Variables

Math 324 Course Notes: Brief description

g i fφdx dx = x i i=1 is a Hilbert space. We shall, henceforth, abuse notation and write g i f(x) = f

Chapter 3 Polynomials

On the Continuous Dependence of Solutions of Boundary Value Problems for Delay Differential Equations

On the free product of ordered groups

N 0 completions on partial matrices

W. We shall do so one by one, starting with I 1, and we shall do it greedily, trying

Mapping the delta function and other Radon measures

Review of Riemann Integral

Best Approximation in the 2-norm

Math Fall 2006 Sample problems for the final exam: Solutions

DUNKL WAVELETS AND APPLICATIONS TO INVERSION OF THE DUNKL INTERTWINING OPERATOR AND ITS DUAL

Chapter 3 MATRIX. In this chapter: 3.1 MATRIX NOTATION AND TERMINOLOGY

Chapter 5 : Continuous Random Variables

MATH 174A: PROBLEM SET 5. Suggested Solution

Variational Techniques for Sturm-Liouville Eigenvalue Problems

Contents. Outline. Structured Rank Matrices Lecture 2: The theorem Proofs Examples related to structured ranks References. Structure Transport

Best Approximation. Chapter The General Case

S. S. Dragomir. 2, we have the inequality. b a

FUNDAMENTALS OF REAL ANALYSIS by. III.1. Measurable functions. f 1 (

Chapter 4. Lebesgue Integration

Semigroup of generalized inverses of matrices

MTH 5102 Linear Algebra Practice Exam 1 - Solutions Feb. 9, 2016

Energy Bands Energy Bands and Band Gap. Phys463.nb Phenomenon

Frobenius numbers of generalized Fibonacci semigroups

Entrance Exam, Real Analysis September 1, 2009 Solve exactly 6 out of the 8 problems. Compute the following and justify your computation: lim

WHEN IS A FUNCTION NOT FLAT? 1. Introduction. {e 1 0, x = 0. f(x) =

Pavel Rytí. November 22, 2011 Discrete Math Seminar - Simon Fraser University

Math Advanced Calculus II

11 An introduction to Riemann Integration

Approximation of functions belonging to the class L p (ω) β by linear operators

Chapter 4 Contravariance, Covariance, and Spacetime Diagrams

Properties of the Riemann Integral

Research Article Moment Inequalities and Complete Moment Convergence

Hilbert Spaces. Chapter Inner product spaces

BIFURCATIONS IN ONE-DIMENSIONAL DISCRETE SYSTEMS

1 1D heat and wave equations on a finite interval

RIEMANN-LIOUVILLE AND CAPUTO FRACTIONAL APPROXIMATION OF CSISZAR S f DIVERGENCE

A tutorial on sequential functions

Quadratic Forms. Quadratic Forms

The final exam will take place on Friday May 11th from 8am 11am in Evans room 60.

1 2-D Second Order Equations: Separation of Variables

Introduction to Some Convergence theorems

Review of Gaussian Quadrature method

A GENERAL INTEGRAL RICARDO ESTRADA AND JASSON VINDAS

Calculus of Variations

ODE: Existence and Uniqueness of a Solution

HW3, Math 307. CSUF. Spring 2007.

Bypassing no-go theorems for consistent interactions in gauge theories

Here we study square linear systems and properties of their coefficient matrices as they relate to the solution set of the linear system.

Jian-yi Shi East China Normal University, Shanghai and Technische Universität Kaiserslautern

The Regulated and Riemann Integrals

Homework Problem Set 1 Solutions

A Convergence Theorem for the Improper Riemann Integral of Banach Space-valued Functions

Geometric Sequences. Geometric Sequence a sequence whose consecutive terms have a common ratio.

STURM-LIOUVILLE THEORY, VARIATIONAL APPROACH

THE JOHN ELLIPSOID THEOREM. The following is a lecture given in the functional analysis seminar at the University of South Carolina.

ON THE C-INTEGRAL BENEDETTO BONGIORNO

Transcription:

The Perron-Frobenius opertors, invrint mesures nd representtions of the Cuntz-Krieger lgebrs Ktsunori Kwmur Reserch Institute for Mthemticl Sciences Kyoto University, Kyoto 606-8502, Jpn For trnsformtion F on mesure spce (X, µ), we show tht the Perron-Frobenius opertor of F cn be written by representtion (L 2 (X, µ), π) of the Cuntz-Krieger lgebr O A ssocited with F when F stisfies some ssumption. Especilly, when O A is the Cuntz lgebr O N nd (L 2 (X, µ), π) in the bove is some irreducible representtion of O N, then there is n F -invrint mesure on X which is bsolutely continuous with respect to µ.. Introduction Invrint mesures (especilly, Hr mesures) ply n importnt role in the representtion theory of Lie groups nd hrmonic nlysis. On the other hnd, invrint mesures of non invertible trnsformtions re studied in [4, 5, 6] by the Perron-Frobenius opertors of dynmicl systems. By using the Perron-Frobenius opertors, the chrcteriztion of given dynmicl system nd the construction of invrint mesure re obtined. We show their roles in representtion theory of opertor lgebrs in this pper. Let L p (X, µ) be the set of ll complex-vlued mesurble functions φ on mesure spce (X, µ) stisfying φ Lp < nd let L p (X, µ; R) be the subset of ll rel-vlued functions in L p (X, µ) for p =, 2,. For nonsingulr trnsformtion F on X(tht is, µ(f (A)) = 0 if µ(a) = 0 for A X), P F is the Perron-Frobenius opertor (or the Frobenius-Perron opertor, the trnsfer opertor) of F if P F is the opertor on L (X, µ) which stisfies (.) (P F ψ)(x) dµ(x) = ψ(x) dµ(x) ( ψ L (X, µ)) A F (A) for ech mesurble subset A of X ([5]). By (.), P F ψ is uniquely determined s n element in L (X, µ) for ech ψ L (X, µ). For ψ L (X, µ) nd θ L (X, µ), we obtin X θ(f (x))ψ(x) dµ(x) = X θ(x)(p F ψ)(x) dµ(x). From this, P F is bounded liner opertor on L (X, µ) nd P F ψ L e-mil:kwmur@kurims.kyoto-u.c.jp.

ψ L for ech ψ L (X, µ; R). Further, positive function ρ L (X, µ) stisfies P F ρ = ρ if nd only if ρ is the density of n F -invrint mesure, tht is, the following holds for ny ψ L (X, µ): (.2) ψ(f (x))ρ(x) dµ(x) = ψ(x)ρ(x) dµ(x). X In order to describe both the Perron-Frobenius opertors nd representtions of the Cuntz-Krieger lgebrs simultneously, we introduce brnching function systems on mesure spce (X, µ). A fmily f = {f i } N i= of mps on X is semibrnching function system if there is finite fmily {D i } N i= of mesurble subsets of X such tht f i is mesurble mp from D i to R i f i (D i ), µ(x \ R R N ) = 0, µ(r i R j ) = 0 when i j nd there is the Rdon-Nikodým derivtive Φ fi of µ f i with respect to µ nd Φ fi > 0 lmost everywhere in D i for i =,..., N. A mp F on X is clled the coding mp of semibrnching function system f = {f i } N i= if F f i = id Di for i =,..., N. For semibrnching function system f = {f i } N i= with the coding mp F, define fmily {S(f i )} N i= of opertors on L 2(X, µ) by (.3) (S(f i )φ)(x) χ Ri (x) {Φ F (x)} /2 φ(f (x)) (φ L 2 (X, µ)) where χ Ri is the chrcteristic function of R i. Then S(f i ) is prtil isometry with the initil spce L 2 (D i, µ) nd the finl spce L 2 (R i, µ), nd S(f i )S(f j ) = S(f i f j ) when D j R i. For N 2, let A be n N N mtrix which consists of elements 0 or nd ny column nd row re not 0. A semibrnching function system f = {f i } N i= is n A-brnching function system if µ(d i \ j; ij = R j) = 0 for ech i =,..., N. For n A-brnching function system f = {f i } N i=, (.4) π f (s i ) S(f i ) (i =,..., N), defines representtion (L 2 (X, µ), π f ) of the Cuntz-Krieger lgebr O A. X Theorem.. For n A-brnching function system f = {f i } N i= coding mp F, the following holds: with the (P F ψ)(x) = {(π f (s ) ψ)(x)} 2 + + {(π f (s N) ψ)(x)} 2 for ny positive function ψ L (X, µ) where ψ(x) ψ(x). Theorem.2. Assume tht F is the coding mp of n A-brnching function system f = {f i } N i= on mesure spce (X, µ) nd b i Φ fi is constnt for i =,..., N nd µ(x) <. Then the following holds: (i) Define subspce V Lin < {χ R,..., χ RN } > of L 2 (X, µ). Then P F V V where R i is the imge of f i. 2

(ii) For digonl mtrix B dig(b,..., b N ) M N (R), the following identity of mtrices holds: P F V = BA where P F V is the mtrix representtion over the bsis χ R,..., χ RN of V nd the rhs is the product of mtrices. In Theorem.2 (ii), the eigenvlues of the Perron-Frobenius opertor ssocited with F depend on not only A but lso B. In this sense, eigenvlues of the Perron-Frobenius opertor hve the informtion of representtion of the Cuntz-Krieger lgebr. It is importnt problem to construct the invrint mesure for given dynmicl system. For exmple, Lsot-York theorem shows construction of invrint mesure by using the Perron-Frobenius opertor of dynmicl system ([4]). We show the condition of existence of invrint mesure from the viewpoint of representtion theory of the Cuntz lgebr. f = {f i } N i= is brnching function system if f = {f i} N i= is n A- brnching function system for mtrix A = ( ij ), ij = for ech i, j =,..., N. In this cse, (L 2 (X, µ), π f ) is representtion of the Cuntz lgebr O N. For z = (z i ) N i= SN {y R N : y = }, (H, π) is GP (z) of O N if there is unit cyclic vector Ω H such tht (.5) π(z s + + z N s N )Ω = Ω. We cll Ω by the GP vector of (H, π). In this cse, there is g O(N) U(N) such tht (π α g )(s )Ω = Ω where α is the cnonicl ction of U(N) on O N. This implies tht (H, π α g ) is n irreducible permuttive representtion of O N ([]). Hence GP (z) of O N exists uniquely up to unitry equivlence nd it is irreducible. Further we see tht GP (z) GP (y) if nd only if z = y where mens the unitry equivlence. Theorem.3. Let F be the coding mp of brnching function system f on mesure spce (X, µ). If there is z S N such tht (L 2 (X, µ), π f ) is GP (z) with the GP vector Ω L 2 (X, µ; R), then there is probbilistic F -invrint mesure ν on X which is bsolutely continuous with respect to µ nd it is given s follows: dν(x) {Ω(x)} 2 dµ(x) (x X). In 2, we show the min theorems. It is explined tht (.5) implies the eigeneqution of the Perron-Frobenius opertor. In 3, we show concrete exmples. 2. Proofs of the min theorems For N 2, let M N ({0, }) be the set of ll N N mtrices such tht ech element is 0 or nd ny row nd column is not 0. For A = ( ij ) M N ({0, }), 3

O A is the Cuntz-Krieger lgebr by A if O A is C -lgebr which is universlly generted by genertors s,..., s N nd they stisfy s i s i = N j= ijs j s j for i =,..., N nd N i= s is i = I ([3]). Especilly, when ij = for ech i, j =,..., N, O A is the Cuntz lgebr O N ([2]). In this pper, ny representtion is unitl nd -preserving. Proof of Theorem.. L 2 (X, µ) is s follows: By (.3), the djoint opertor S(f i ) of S(f i ) on (2.) (S(f i ) φ)(x) = χ Di (x) {Φ fi (x)} /2 φ(f i (x)) (φ L 2 (X, µ)). For the coding mp F of semibrnching function system f = {f i } N i=, we hve N (2.2) (P F ψ)(x) = χ Di (x) Φ fi (x) ψ(f i (x)) (ψ L (X, µ)). i= By (2.) nd (2.2), we hve (2.3) (P F ψ)(x) = {(S(f ) ψ)(x)} 2 + + {(S(f N ) ψ)(x)} 2 for ny positive function ψ L (X, µ). By (.4) nd (2.3), the sttement holds. Proof of Theorem.2. Define v i χ Ri for i =,..., N. (i) We see tht χ Di = N k= ikv k. By (2.), (S(f i ) φ)(x) = b i χ Di (x)φ(f i (x)). From this, S(f i ) v j = N k= b /2 i c (j) ik v k for i =,..., N where c (j) ik = δ ij b i ik. Hence S(f i ) V V. By (2.2), the following holds: (2.4) P F = b /2 S(f ) + + b /2 N S(f N). Therefore the sttement is proved. (ii) By the proof of (i), we see tht S(f i ) V = (b /2 i c (i) jk ) s mtrix with respect to v,..., v N. From (2.4), the sttement holds. Corollry 2.. Let X be bounded closed intervl of R nd A M N ({0, }). Assume tht f = {f i } N i= is n A-brnching function system on X nd b i Φ fi is constnt for ech i =,..., N. Then the eigenvlue of BA becomes tht of the Perron-Frobenius opertor of the coding mp F of f where B dig(b,..., b N ). Proof of Theorem.3. Assume tht Ω L 2 (X, µ; R) stisfies π f (z s + + z N s N )Ω = Ω. Define ρ(x) (Ω(x)) 2 for x X. Then ρ L (X, µ). By Theorem. nd π f (s i ) Ω = π f (s i ) π f (z s + + z N s N )Ω = z i Ω, we hve (P F ρ)(x) = N i= {(π(s i) Ω)(x)} 2 = N i= {z iω(x)} 2 = ρ(x). Hence 4

P F ρ = ρ. This implies the sttement. Corollry 2.2. Let X be mesurble subset of R. Assume tht piecewise C -clss mp F on X is the coding mp of brnching function system {f i } N i= on the mesure spce (X, dx) where dx is the Lebesgue mesure. If φ 0 L 2 (X, dx; R) stisfies (2.5) F (x) φ 0 (F (x)) = Nφ 0 (x) (.e. x X), then dµ(x) {φ 0 (x)} 2 dx is n invrint mesure on X with respect to F. Proof. By (.3) nd (.4), we see tht (π f (s + + s N )φ)(x) = F (x) φ(f (x)) for ech φ L 2 (X, dx). From this nd (2.5), π f (N /2 s + + N /2 s N )φ 0 = φ 0. By Theorem.3 for z = (N /2,..., N /2 ) S N, the sttement holds. In 6.5 of [5], it is explined tht intertwiners mong dynmicl systems bring new invrint mesures from known ones. We show its unitry version s follows: Proposition 2.3. Let F be the coding mp of brnching function system f = {f i } N i= on mesure spce (X, µ). Assume tht (L 2(X, µ), π f ) is GP (z) for z S N with the GP vector Ω L 2 (X, µ; R). If ζ is mesure spce isomorphism from (X, µ) to other (Y, ν) nd G ζ F ζ, then ρ (S(ζ)Ω) 2 is the density of probbilistic G-invrint mesure on Y which is bsolutely continuous with respect to ν where S(ζ) is unitry opertor from L 2 (X, µ) to L 2 (Y, ν) defined by (S(ζ)φ)(y) {Φ ζ (y)} /2 φ(ζ (y)). Proof. Define brnching function system g = {g i } N i= by g i ζ f i ζ. Then G is the coding mp of g. We see tht S(ζ)π f ( )S(ζ) = π g ( ), π g (z s + + z N s N )Ω = Ω for Ω S(ζ)Ω L 2 (Y, ν; R). Hence (L 2 (Y, ν), π g ) is GP (z) with the GP vector Ω. By Theorem.3, we hve the sttement. 3. Exmples Exmple 3.. Let 0 < < nd X [0, ]. (i) Define mp F on X by F (x) x/ on R [0, ] nd F (x) (x )/( ) on R 2 [, ]. 5

f 2 F f 0 0 Then F is the coding mp of brnching function system f {f, f 2 } defined by f i (F Ri ) for i =, 2. Then (π f (s )φ)(x) = /2 χ R (x)φ(x/), (π f (s 2 )φ)(x) = ( ) /2 χ R2 (x)φ( (x )/( )) for φ L 2 (X, dx). (P F ψ)(x) = ψ(x) + ( )ψ( ( )x + ) (ψ L (X, dx)). The Lebesgue mesure dx is the probbilistic invrint mesure of X with respect to F. (ii) Define f (x) x nd f 2 (x) ( )x 2 + on X. The coding mp F of f = {f, f 2 } is given by F (x) = x/ on [0, ], F (x) = (x 2 )/( ) on [, ]. Then function Ω(x) 2x on [0, ] stisfies π f ( s + s 2 )Ω = Ω. Hence the probbilistic F -invrint mesure on X is 2xdx. (L 2 (X, dx), π f ) in both (i) nd (ii) is GP (, ) of O 2. Both invrint mesures re independent in the prmeter. Exmple 3.2. For, b R, 0, define F (x) (x b) 2 / + b 2 on X [ 2 +b, 2 +b]. Define brnching function system f = {f, f 2 } on X by f i (F Ri ), i =, 2 for R [ 2 + b, b] nd R 2 [b, 2 + b]. Then π f (s + s 2 )Ω = Ω for Ω(x) π /2 {4 2 (x b) 2 } /4. Hence ρ(x) π 4 2 (x b) 2 is the density of probbilistic invrint mesure on X with respect to F. When = /4 nd b = /2, we hve F (x) = 4x( x), ρ(x) = π x( x). This ws first obtined by Ulmn nd von Neumnn ([8]). 6

F /2 f 2 f 0 /2 0 Exmple 3.3. For 0 < <, define mp F : [0, ] [0, ] s follows: 0 Define R [0, ], R 2 [, ], D( [0,) ], D 2 [0, ]. Then F is the coding mp of the following A = -brnching function system on 0 X = [0, ]: f i : D i R i, f (x) = x for x [0, ] nd f 2 (x) = ( )x/+ for x [0, ]. Define v χ [0,], v 2 χ [,], V Lin < {v, v 2 } >. Then the mtrix representtion of P F with respect to v, v 2 is given s follows: P F V = ( ( )/ 0 Hence its eigenvlue re nd. Their normlized eigenvectors re given s follows: w = χ [0,] χ ( ) [,], w 2 = χ[0,] + χ ( + 2 [,]. ) Especilly w 2 is the density of the invrint mesure on [0, ] with respect to F. Exmple 3.4. We show pplictions of Proposition 2.3. The following (X, F, µ) is trnsformtion F on X R nd probbilistic invrint mesure µ on X: (i) For b R \ [, 0], X [0, ], F (x) ). 2b 2 /(b x) 2b (x D ), 2b 2 { + 3b ( + b) 2 ( + 3b)x + b(b ) 7 } (x D 2 ),

where D [0, b/(2b + )) nd D 2 [b/(2b + ), ]. Then dµ(x) = (ii) For 0 k <, X [, ], F (x) b(b + ) dx (x [0, ]). (x + b) 2 2x 2 dx k 2 ( x 2. Then dµ(x) = ) 2 2K ( x 2 )( k 2 ( x 2 )) where K is the positive constnt defined by K (iii) For n integer N 2, X [0, ] nd 0 ( x 2 )( k 2 x 2 ) dx. F (x) ( N x [ N x ] ) 2, we hve dµ(x) = where [ ] is the gretest integer less thn equl x. (iv) For rel number >, X [, ] nd dx 2 x F (x) /x 2 (x [, )), F (x) x 2 / (x [, ]), we hve dµ(x) = dx/x. (v) For X R \ (, ) nd F (x) 2/(2 x ), dµ(x) = dx 2x 2. Exmple 3.5. Let F be mp defined by the following grph: 2/3 /3 0 2/3 /3 Define A 0, B dig (/3, /2, ), R [0, /3], R 2 0 0 [/3, 2/3], R 3 [2/3, ], D [0, ], D 2 [/3, ], D 3 [0, /3]. The A-brnching function system f = {f, f 2, f 3 }, f i : D i R i, i =, 2, 3, with the coding mp F is given by f (x) = x/3, f 2 (x) = (x )/2 + /3, f 3 (x) = x +. From this nd Theorem.2 (ii), P F W = BA where W Lin< {χ R, χ R2, χ R3 } >. We see tht 0, /6, re eigenvlues of P F W. Hence they re eigenvlues of P F. Exmple 3.6. For A M N ({0, }) nd n A-brnching function system f = {f i } N i= on mesure spce (X, µ), µ(x) <, ssume tht b i Φ fi is constnt for ech i =,..., N. Then b i = r i /( N j= ijr j ) 8

where r i µ(r i ). When A dig ( r r r 2 +r 3, 2 r +r 3, 0 0, we hve B = (b, b 2, b 3 ) = r 3 r +r 2 +r 3 ). For 0 < < b <, consider the cse F on X = [0, ] which grph is given s follows: F b 0 b b 0 b F is the coding mp of n A-brnching function system given s follows: f i : D i R i, i =, 2, 3, f (x) = f 2 (x) = (x ) (x D ), b b+ x + b, (x R ), b b+ (x ) + (x R 3), f 3 (x) = ( b)x + b (x [0, ]), where R [0, ], R 2 [, b], R 3 [b, ], D [, ], D 2 [0, ] (b, ], D 3 [0, ]. From these, we hve B = dig (/( ), (b )/( b + ), b). } } } f 3 f 2 f dx 4 x 2 Exmple 3.7. Let F (x) x 3 3x on X [ 2, 2]. Then π is probbilistic invrint mesure with respect to F. For brnching function system f = {f, f 2, f 3 } defined by f (F [ 2, ] ), f 2 (F [,] ), f 3 (F [,2] ), (L 2 [ 2, 2], π f ) is GP (3 /2, 3 /2, 3 /2 ) of O 3. Exmple 3.8. Let X be the closed bounded region in R 2 which is clipped by 4-curves L {( x 2, x) : x [, ]}, L 2 {(x, ) : x [, ]}, L 3 {(2 x 2, x) : x [, ]}, L 4 {(x, ) : x [, ]}. Define mp F on X by F (x, y) ( 2x 2 + 4xy 2 2y 2 4x +, 2y 2 ) ((x, y) X). Then F (X) = X. There re the following four subregions R,..., R 4 of X: 9

0 X 2 R R 2 R 3 R 4 where the new center curve in the right figure is {( x 2, x) : x [, ]}. Then X = R R 4 nd F Ri is bijection from R i to X for ech i =,..., 4. Then dµ(x, y) = dxdy π 2 ( y 2 )(x + y 2 )(2 x y 2 ) is probbilistic invrint mesure on X with respect to F. For brnching function system f = {f i } 4 i= defined by f i (F Ri ), (L 2 (X, dxdy), π f ) is GP (/4, /4, /4, /4) of O 4. Acknowledgement: I would like to thnk Mkoto Mori for his tlk in [7]. References [] O.Brtteli nd P.E.T.Jorgensen, Iterted function Systems nd Permuttion Representtions of the Cuntz lgebr, Memories Amer. Mth. Soc. No.663 (999). [2] J.Cuntz, Simple C -lgebrs generted by isometries, Comm. Mth. Phys. 57 (977), 73-85. [3] J.Cuntz nd W.Krieger, A clss of C -lgebr nd topologicl Mrkov chins, Invent.Mth., 56 (980), 25-268. [4] A.Lsot nd J.A.Yorke, Exct dynmicl systems nd the Frobenius-Perron opertor, Trns.Amer.Mth.Soc. 273, (982), 375-384. [5] A.Lsot nd M.C.Mckey, Chos, Frctls nd Noise, Stochstic Aspects of Dynmics, Second Edition, Springer-Verlg (99). [6] M.Mori, On the convergence of the spectrum of Perron-Frobenius opertors, Tokyo J. Mth. 7 (994), -9. [7] M.Mori, Frctl nd Perron-Frobenius opertor (Jpnese), Representtions of Cuntz lgebrs nd their pplictions in mthemticl physics, Symposium in RIMS Kōkyuroku No.333 (2003). [8] S.M.Ulm nd J.von Neumnn, On combintion of stochstic nd deterministic processes, Bull. Am. Mth. Soc., 53:20, (947). 0