Lecture 27: HW2 session: Thursday, April 2, 5:30 until we are done (with a break for O3). Recall defn of P (T , f ) via (n, δ)-separated sets, in particular topological entropy h(T ) : P (T , 0). We also outlined the original defn. of topological entropy using open covers, but we did not show the equivalence of the definitions. Recall defn. of expansiveness Prop: Expansiveness is a top. conjugacy invariant (but with possibly different expansive constants) Proof: let φ be a topological conjugacy from T on M to S on N and T be expansive with expansive constant δ. Since φ−1 is continuous, there exists > 0 s.t. if x, y ∈ N, d(x, y) < , then letting u = φ−1(x), v = φ−1(y), d(u, v) < δ. We claim that S is expansive constant . If not, the for some x, y ∈ N, x 6= y, and all n ∈ G, d(S nx, S ny) ≤ . Then for all n ∈ G, d(T nu, T nv) = d(T nφ−1x, T nφ−1y) = d(φ−1S nx, φ−1S ny) ≤ δ and u 6= v, a contradiction Note: expansiveness is not necessarily preserved under topological factor or topological extension. Recall: Example: a two-sided shift space is expansive as a Zd-action. Therefore horseshoe map on Ω is expansive. Theorem: If T is expansive with expansive constant δ > 0, then P (T , f ) = Pδ (T , f ). Proof of Theorem: We need a defn. and a lemma. 1 We first assume that the action is expansive as a Zd+-action. Defn: Let T be a a cts. Zd+-action. For a positive integer k, δ > 0, x ∈ M , let U (k, δ, x) := {y ∈ M : d(T i(y), T i(x)) ≤ δ, ∀i ∈ Dk } = ∩i∈Dk T −i(B δ (T i(x)) Note: E is (k, δ)-separated iff for all x, y ∈ E, x 6= y, we have y 6∈ U (k, δ, x). Lemma (a“finite” version of expansiveness): Let T be an expansive Zd+-action with expansive constant δ > 0. For all 0 < δ 0 < δ, there exists k = k(δ 0) s.t. for all x ∈ M , we have U (k, δ, x) ⊆ Bδ0 (x). Proof: Suppose not. The for some 0 < δ 0 < δ and all positive integers k, there exist xk , yk ∈ M s.t. d(T i(yk ), T i(xk )) ≤ δ, for all i ∈ Dk and d(xk , yk ) ≥ δ 0. By compactness, for some x, y ∈ M and some sequence kj , xkj → x and ykj → y. But then d(x, y) ≥ δ 0 > 0 and for all i ∈ Zd+, d(T i(y), T i(x)) ≤ δ (since Zd+ = ∪k Dk ), contradicting expansiveness with expansive constant δ. Proof of Theorem: We must show that for all δ 0 < δ, Pδ0 (T , f ) = Pδ (T , f ). It follows from the Lemma that for all positive integers n, 0 < δ 0 < δ and x ∈ M , U (n + k(δ 0), δ, x) ⊆ U (n, δ 0, x). So, any (n, δ 0)-separated set E is (n + k(δ 0), δ)-separated: for x, y ∈ E, x 6= y, we have y 6∈ U (n, δ 0, x) and so y 6∈ U (n + k(δ 0), δ, x). Thus, 2 Pn+k(δ0),δ (T, f ) ≥ Pn,δ0 (T, f ) Thus, Pδ (T , f ) ≥ Pδ0 (T , f ) and the reverse inclusion is obvious. For an expansive Zd-action, first change the proof by redefining: U (k, δ, x) := {y ∈ M : d(T i(y), T i(x)) ≤ δ, ∀i ∈ Bk } (where Bk = [−k, k]d). Then Lemma holds. And replace “any (n, δ 0)-separated set E is (n+k(δ 0), δ)-separated” by 0 “for any (2n, δ 0)-separated set E, we have that T −k(δ )E is (2n + 2k(δ 0), δ)-separated.” And observe that while Pδ is defined as a lim sup over Pn,δ , not as a limit, we get the same value if we take the lim sup over P2n,δ . This result is, in some sense, an analogue of the Sinai Generator Theorem for ergodic theory and gives us some hope of computing pressure and topological entropy. Example: Let M = XF be a (two-sided) shift space (with shift action σ = σ|M ), which is expansive as a cts Zd-action with 1 as expansive constant. Let GAn = GAn(M ) be the set of configurations on Dn = [0, n−1]d which extend to an element of M (GA stands for “globally admissible”). Recall that E ⊂ M is (n, 1)-separated iff for all x, y ∈ M, x 6= y, we have x|Dn 6= y|Dn , i.e., x and y disagree on at least one site of Dn. Thus, a maximal (n, 1)-separated set has size |GAn|. Thus, h(σ) = P1(σ, 0) = lim (1/nd) log |GAn| n→∞ 3 Lecture 28: Recall: For any (two-sided) shift space M = XF , the shift action T = σ|M ) is expansive as a cts Zd-action with expansive constant = 1. And we showed: Prop: h(T ) = h(σ|M ) = P1(σ|M , 0) = lim (1/nd) log |GAn| n→∞ where GAn = GAn(M ) be the set of configurations on Dn = [0, n − 1]d which extend to an element of M . Idea: For each n, let En be a maximal (n, 1)-separated set En that achieves Pn,1(T , 0). The map En → x|Dn is a bijection from En onto GAn. Note that the limit exists because of subadditivity. 2 Example: For full (two-sided) shift on two symbols, |GAn| = 2n and so h(σ) = log 2. Example: For Ledrappier 3-dot, we have d = 2 and |GAn)| ≤ 22n and so h(σ|M ) = 0. Expansiveness is not always needed to compute topological entropy or pressure: Example: a rotation of the circle is not expansive: for all δ if d(x, y) < δ, then for all n, d(T nx, T ny) < δ. Claim: h(T ) = 0. Proof: Any (n, δ) separated set is (1, δ)-separated because T is an isometry and so if d(T i(x), T i(y)) > δ, then d(x, y) > δ. So, Pn,δ (T, 0) is constant in n. So for all δ > 0, Pδ (T, 0) = 0. So, h(T ) = 0. Back to shift spaces: 4 The problem of deciding if a given w ∈ F Dn belongs to GAn is undecidable. Defn: For a positive integer n, let LAn = LAn(M ) denote the set of configurations on Dn which do not contain an element of a forbidden list F that defines M = XF (LA stands for “locally admissible,”) Think of n.n. SFT defined by a list F of forbidden adjacencies: w ∈ LAn iff for all nearest neighbor configs 6∈ F. Clearly, GAn ≤ LAn. There are often LA configs that are not GA, even for n.n. SFT’s. But there is a decision procedure to decide membership in LAn(M ). Friedland’s Theorem: For a (two-sided) shift space M : h(σ|M ) = lim (1/nd) log |LAn(M )| n→∞ Limit exists again by subadditivity. Proof: Clearly, LHS ≤ RHS. For simplicity, consider d = 2. Let GAn, LAn be the same as GAn, LAn except that we replace Dn by Bn = [−n, n]d. Limits on both side of eqn. above are the same if you replace GAn, LAn by GAn, LAn and replace nd by (2n + 1)d. Claim: ∀k ∃a(k) such that for all u ∈ LAk , u ∈ GAk iff u extends to LAa(k). Proof: By compactness, for u ∈ LAk \ GAk , there exists a(u) s.t. u does not extend to an element of LAa(k). Set a(k) := maxu∈LAk \GAk a(u). 5 By defn of a(k), if u ∈ LAk extends to LAa(k), then u ∈ GAk . Clearly, if u ∈ GAk , then extends to LAa(k). Fix k. Choose n large and partition Bn into translates of Bk , reserving a border of width a(k). |LAn| ≤ |GAk |( 2n+1−2a(k) 2 ) 2k+1 |F |(8n+4)a(k) log |LAn| 2n + 1 − 2a(k) 2 log |GAk | ≤ ( )( ) (2n + 1)2 2k + 1 (2n + 1)2 (8n + 4)a(k) log |F | ) + (2n + 1)2 log |LAn| log |GAk | ≤ . n→∞ (2n + 1)2 (2k + 1)2 lim (1) Let k → ∞. For nn Zd SFT M . Let Φ be a nn interaction on M , i.e., a real-valued function defined on all allowed (i.e., non-forbidden) configurations on edges and vertices. For w ∈ LAn(M ), define the energy function X Φ U (w) := Φ(w(η)), edges, vertices η contained in Dn Define the partition function: X Φ Φ Zn := e−U (w). w∈LAn Example: Ising model: M is full Zd shift action, with F = {1, −1} and Φ(wiwj ) = −βwiwj , for all adjacent vertices i, j in Zd. Define d X f Φ(x) := −Φ(x0) − Φ(x0xei ) i=1 6 Observe that Snf Φ(x) depends only on x|Dn+1 . Prop: (a special case of Ruelle, Theorem 3.4) For nn Zd SFT and f = f Φ for nn potential Φ, P (T , f ) = lim (1/nd) log ZnΦ n→∞ 7 Lecture 29: Recall statement of: Prop: (a special case of Ruelle, Theorem 3.4) For n.n. Zd SFT M and a n.n. potential Φ, P (σ|M , f Φ) = lim (1/nd) log ZnΦ n→∞ Note: The RHS is often taken to be the defn of the pressure, P (Φ), of a n.n. interaction on a n.n. SFT. Note: Special case Φ = 0 reduces to computation of topological entropy as in Friedland’s theorem. Proof: For simplicity, consider d = 2. Write T = σ|M , f = f Φ, U = U Φ, Zn = ZnΦ. Let En be a maximal (n, 1)-separated set En that achieves Pn,1(T , f ). The map En → x|Dn is a bijection from En onto GAn. Let M1 = max Φ, M0 = min Φ. Then X −U (w) −2nM1 e e w∈GAn ≤ X Sn f (x) e X ≤ x∈En e−U (w)e−2nM0 w∈GAn So 2 P (T , f ) = P1(T , f ) = lim sup(1/n ) log( n→∞ = lim (1/n2) log( n→∞ X eSnf (x)) x∈En X e−U (w)) w∈GAn and by a modification of proof of Friedland’s Theorem, X 2 = lim (1/n ) log( e−U (w)) = lim (1/n2) log Zn n→∞ n→∞ w∈LAn 8 Modification of Friedland: X e−U Φ (w) w∈LAn X ≤ (( e−U (w))( 2n+1−2a(k) 2 ) 2k+1 ((e−M0 )8k+4)( 2n+1−2a(k) 2 ) 2k+1 (|F |e−3M0 )a(k)(8n+4) w∈GAk Take log, divide by (2n + 1)2 and let n → ∞: X X Φ 2 −U Φ (w) 2 lim (1/(2n+1) ) log e ≤ (1/(2k+1) ) log e−U (w) n→∞ w∈LAn w∈GAk −(4M0)/(2k + 1) Let k → ∞. Next, we compute this explicitly for d = 1. Will use: Perron-Frobenius Theorem: Let A be an irreducible matrix with spectral radius λA := max eigenvalues of A |λ|. Then 1. λA > 0 and is an eigenvalue of A. 2. λA has a (strictly) positive eigenvector and is the only eigenvalue with a nonnegative eigenvector. Terminology: λA is called the Perron eigenvalue of A and its corresponding eigenvectors are called Perron eigenvectors. Brief discussion of proof of P-F: Recall the geometric argument that for a primitive stochastic matrix P , with strictly positive eigenvector π, we have Pijk → πj : Let W be the unit simplex in Rn. Define g : W → W, 9 x 7→ xA xA1 In the primitive case, g contracts W into a single point, giving a unique eigenvalue λ with positive eigenvector. And λ has maximum modulus; for otherwise we would have leakage out of W . Modify to take care of the general irreducible case: if A is irreducible, then for > 0, A + I is primitive. 10