Dynamics on Bratteli Diagrams Kostya Medynets Wright State University April 22, 2016 1 / 22 Autonomous Differential Equations • The Lennard-Jones Oscillator • Consider an autonomous differential equation x′ = f (x), x′ = p, x ∈ X ⊂ Rn p′ = 12(x−13 − x−7 ) (1) • Fix a point x0 ∈ X . Denote by φ(t, x0 ) the value x(t), where x(t) is the unique solution of (1) subject to x(x) = x0 . • Group Property: φ(t2 , φ(t1 , x0 )) = φ(t1 + t2 , x0 ). • Flow on the phase space Φt (x) = φ(x, t), Φt : X → X 2 / 22 Discrete Dynamical Systems • Let Φt : X → X be a flow on the phase space. For example, the phase space could consist of all possible positions and velocities of a particle. • Set T = Φ1 : X → X . • Then T n (x0 ) represents the position/velocity of the particle whose initial location/velocity was x0 n time units into the future. • The transformation T : X → X is the called the evolution law. • Abstract Dynamical Systems: Let X be a compact metric space and T : X → X be a homeomorphism. The pair (X, T ) is called a • • (discrete) dynamical system. (X, T ) is called minimal if every T -orbit, {T n (x0 ) : n ∈ Z} is dense in X . (Zorn’s Lemma:) Every system has a minimal subsystem. 3 / 22 Example: Irrational Rotation • Let X = S1 and T : X → X be the rotation of X by an angle θ. • If θ/(2π) ∈ Q, the T -orbits are finite. • If If θ/(2π) ∈ R \ Q, each T -orbit is infinite and dense. The system is minimal. Theorem 1. If a dynamical system (X, T ) is equicontinuous, dist(x, y) < δ implies dist(n (x), T n (x)) < ε, then (X, T ) is conjugate to a translation on a compact Abelian group [Auslander, Ellis]. 4 / 22 Symbolic Systems • Phase Space. Let A = {0, 1, . . . , n − 1}. Consider Xn the space of two-sided sequences with values in A. • A is given the discrete topology and Xn the product topology. Two sequences are close if the share a long common word centered at 0. Xn is the Cantor set. • Dynamics. Let T : Xn → Xn be the left Bernoulli shift. For example, for n = 2, T (· · · 101|001 · · · ) = · · · 1010|01 · · · . • Subshifts. Let Y ⊂ Xn be closed and shift-invariant, that is, T (Y ) = Y . Then (Y, T ) is called a subshift. Theorem 2. A dynamical system (X, T ) is topologically conjugate to a subshift if it is expansive, that is, there exists c > 0 such that for any x 6= y ∈ X , dist(T n (x), T n (y)) > c for some n ∈ Z [Perry]. 5 / 22 Lind-Marcus: Applications to Coding • • • • Computers store data as 0’s and 1’s. Polarity reversals along the track generate voltage pulses. A pulse is a 1. The absence of a pulse is a 0. Problems: Intersymbol Interference and Clock Drift. • Fix: no more than consecutive n 1’s and m 0’s. Gives rise to a subshift. • A subshift is good for coding/decoding if its entropy ≥ log(2). 6 / 22 Bratteli diagrams: Definition • Bratteli diagram = infinite graded • • • • • graph Single root Edges connect consecutive levels No isolated vertices Phase Space = path space of infinite paths Phase Space is a Cantor set 7 / 22 Bratteli Diagrams: Ordering • • • • 3 0 Consider an arbitrary vertex. Consider all incoming edges. Enumerate these edges. Repeat for every vertex. 21 8 / 22 Bratteli Diagrams: Max Paths • x = (xn ) is maximal if every • 0 0 1 1 1 1 0 3 0 3 0 12 0 12 • edge xn is maximal amongst the edges with the same range. Denote by Xmax and Xmin the set of all maximal and minimal paths. Xmax 6= ∅ and Xmin 6= ∅. 9 / 22 Bratteli Diagrams: Action • Define T : X → X. • Fix x ∈ X . • Find the first non-maximal edge xk . 0 1 1 0 3 0 12 • Take the successor of xk . • Connect the successor to the top vertex by the unique path labeled with 0’s. • T : X \ Xmax → X \ Xmin is 0 1 1 0 3 0 12 continuous. If T can be extended to T : X → X , (X, T ) is called a Bratteli-Vershik system. 10 / 22 Bratteli-Vershik Models Theorem 3 (Herman-Putnam-Skau 1995). Every minimal system (X, T ) on the Cantor set X is conjugate to a minimal Bratteli-Vershik system. Minimal systems correspond to simple Bratteli diagrams. Theorem 4 (M. 2005). Every aperiodic, no finite orbits, (X, T ) is conjugate to a Bratteli-Vershik system. 11 / 22 Examples: Odometer • Let Z2 be the group of 2-adic • • integers, X2 ∼ = {0, 1}Z . T : Z2 → X2 , T (x) = x + 1. Consider x0 = (0, 0, 0, . . .) T (x0 ) = (1, 0, 0, . . .) T 2 (x0 ) = (0, 1, 0, . . .) • (Z2 , T ) is called an odometer. 12 / 22 Examples: Stationary Diagrams Theorem. Dynamics on stationary Bratteli diagrams correspond to either odometers or to substitution dynamical systems [Durand, Host, Skau, 1999, minimal systems] and [Bezuglyi, Kwiatkowski, M., 2009, general aperiodic systems]. • Let A be a finite alphabet and σ : A → A+ . Let Xσ be the set of sequences x ∈ AZ such that every x[−n, n] is a subword of σ k (a) for some a ∈ A, k > 0. Then (Xσ , T ) is a substitution system. • Example (Fibonacci Substitution) σ(0) = 01, σ(1) = 0. 0 7→ 01 7→ 010 7→ 01001 · · · 7→ Fibonacci Word. • The length of σ n (0) = Fn+2 , the (n + 2)th Fibonacci number. 13 / 22 Examples: Stationary Diagrams (Fibonacci) 14 / 22 Invariant Measures • Consider an autonomous Hamiltonian system x′ = f (x), x ∈ X ⊂ Rn and the associated one-parameter flow Φt : X → X. • Lioville’s Theorem asserts if D ⊂ X , open, then V ol(D) = V ol(Φt (D)) for every t. • Let (X, T ) be a dynamical system. • A probability measure µ on X is called T -invariant if µ(A) = µ(T (A)) for every Borel set A ⊂ X . • The invariant measure µ is ergodic if µ(A) = 0, 1 for any Borel set A such that T (A) = A. 15 / 22 Ergodic Theorem • Invariant measures form a simplex. • Ergodic measures are precisely extreme elements of that simplex. • Any measure is a convex combination (possibly, infinite uncountable) of ergodic measures. Theorem 5 (Birkhoff). Let (X, T ) be a dynamical system and µ be an ergodic measure. Then for every f ∈ L1 (X, µ) and almost every x ∈ X , Z n−1 X 1 lim f (T k x) = f (x)dµ(x). n→∞ n X k=0 In particular, if f = 1A , then card{0 ≤ k ≤ n − 1 : T k (x) ∈ A} = µ(A). f req(x, A) = lim n→∞ n 16 / 22 Invariant Measures on Bratteli Diagrams • µ is an invariant measure. • v ∈ Vn and e ∈ E(v0 , v). (n) • Denote by Xv (e) the set {x : x[1, n] = e}. (n) (n) • pv = µ(Xv (e)). (n) pn = (pv )v∈Vn . v 17 / 22 Invariant Measures on Bratteli Diagrams (n) • pv (n) = µ(Xv (e)). (n) pn = (pv )v∈Vn . • Then (n) (n+1) (n+1) (n+1) +p2 +2p3 . pv = p1 • pn = An pn+1 , here An is the v incidence matrix between levels n and n + 1. 18 / 22 Invariant Measures: Perron-Frobenius Theory Assume that {An } are N × N , a uniformly bounded diagram. • pn = An pn+1 • pn ≥ 0 . Set core(An ) = T N A A · · · A R n n+1 n+k +. k≥1 Theorem 6 (Geometrical P.-F. Theory). (1) The cones core(An ) are polyhedral (finitely generated). (2) The number of extreme rays does not exceed N . (3) Ergodic measures correspond to sequences of extreme vectors {p̄n }. Corollary 7. For stationary diagrams, when A = An , ergodic measures correspond to non-negative eigenvectors of A. 19 / 22 Invariant Measures: Example 1 2 1 2 1 1 0 A= 1 1 2 0 0 2 0 x = (1/2, 1/2, 0) 1 2λ 1 2λ 0 and λ=2 core(A) = {βx : β ≥ 0}. 20 / 22 Invariant Measures: Example 2 Consider two Bratteli diagrams determined by incidence matrices An = 2 n 1 1 n2 and Bn = n 1 1 n . {An } has two ergodic measures, whereas {Bn }n≥1 is uniquely ergodic. Matrices are contractions on the positive projective space (Birkhoff’s contraction coefficient). • Handelman (1999), Markov chains (uniqueness of stationary distributions). • Bezuglyi, Kwiatkowski, Medynets, Solomyak (2014), Trans. AMS. 21 / 22 Bratteli diagrams: Mixing Properties • If a Bratteli diagram has K vertices at every level, then the associated system (for any ordering) has fewer than K + 1 ergodic measures. • A measure-preserving dynamical system (X, T, µ) is called strongly mixing if for any two sets A, B ⊂ X , µ(A ∩ T n (B)) = µ(A)µ(B) as n → ∞. In other words, any two sets become asymptotically independent. Strong mixing implies ergodicity. • A full Bernoulli shift is strongly mixing. One can derive the Strong Law of Large Numbers from the (pointwise) Ergodic Theorem for Bernoulli shifts. Theorem 8 (Bezuglyi-Kwiatkowski-M.-Solomyak, 2014). Dynamical systems on uniformly bounded (finite rank) Bratteli diagrams cannot be strongly mixing. 22 / 22